Datasets:

Hendrik-a

/

arxiv

like 0

the behavior of the proton magnetic form factor is studied within the modified hard scattering picture , which takes into account gluonic radiative corrections in terms of transverse separations . we parallel the analysis given previously by li and make apparent a number of serious objections . the appropriate cut - off needed to render the form - factor calculation finite is both detailed and analyzed by considering different cut - off prescriptions . the use of the maximum interquark separation as a common infrared cut - off in the sudakov suppression factor is proposed , since it avoids difficulties with the-singularities and yields a proton form factor insensitive to the inclusion of the soft region which therefore can be confidently attributed to perturbative qcd . results are presented for a variety of proton wave functions including also their intrinsic transverse momentum . it turns out that the perturbative contribution , although theoretically self - consistent for larger than about gev to gev , is too small compared to the data . = 0

hep-ph9405340

the proton magnetic form factor at large momentum transfer has been extensively analyzed within perturbative quantum chromodynamics ( pqcd ) over the last decade @xcite . the theoretical basis of these calculations is the hard scattering formula @xcite in which the proton form factor is generically expressed as a convolution of a hard - scattering amplitude @xmath5 and proton distribution amplitudes ( da ) @xmath6 which represent valence quark fock state wave functions integrated over quark transverse momenta ( defined with respect to the momentum of their parent proton ) : @xmath7 \int_{0}^{1}[dx^{\prime } ] |f_{n}(\mu _ { f})|^{2}\ , \phi ^{\star}(x^{\prime},\mu _ { f } ) t_{h}(x , x^{\prime},q,\mu ) \phi ( x,\mu _ { f } ) , \label{eq : gm}\ ] ] where @xmath1 is the invariant momentum transfer squared and @xmath8=dx_{1}dx_{2}dx_{3}\delta ( 1-\sum_{}^{}x_{i})$ ] , @xmath9 being the momentum fractions carried by the valence quarks . the renormalization scale is denoted by @xmath10 and the factorization scale by @xmath11 . the latter scale defines the interface between soft physics absorbed in the wave function and hard physics , treated explicitly within pqcd . the dimensionful constant @xmath12 represents the value of the proton wave function at the origin of the configuration space and has to be determined nonperturbatively @xcite . the residual ( mainly perturbative ) scale dependence of @xmath12 and that of the proton da is controlled by the evolution equation @xcite . to lowest order the hard scattering amplitude is calculated as the sum of all feynman diagrams for which the three quark lines are connected pairwise by two gluon propagators . this allows the quarks in the initial and final proton to be viewed as moving collinearly up to transverse momenta of order @xmath11 . it is then easy to show that @xmath13 , wherein @xmath0 is the running strong coupling constant in the one - loop approximation . the pauli form factor @xmath14 and hence the electric form factor @xmath15 can not be calculated within the hard scattering picture ( hsp ) , since they require helicity - flip transitions which are not possible for ( almost ) massless quarks . these form factors are dominated by sizeable higher twist contributions as we know from experiment @xcite . ( [ eq : gm ] ) is obtained by taking the + component of the electromagnetic vertex and represents the helicity - conserving part of the form factor . the choice of the renormalization scale in the calculation of the proton form factor is a crucial point . most authors @xcite use a constant @xmath0 outside the integrals over fractional momenta , with an argument rescaled by the characteristic virtualities for each particular model da . choosing @xmath10 that way and using das calculated by means of qcd sum rules distributions whose essential characteristic is a strong asymmetry in phase space results for @xmath16 have been obtained @xcite that compare fairly well with the data @xcite . on the other hand , the so - called `` asymptotic '' da @xcite @xmath17 into which any da should evolve with @xmath18yields a vanishing result for @xmath19 . however , for a renormalization scale independent of x , large contributions from higher orders are expected in the endpoint region , @xmath20 . indeed , for the pion form factor this has been shown explicitly , at least for the next - to - leading order @xcite . such large higher - order contributions would render the leading - order calculation useless . a more appropriate choice of the renormalization scale would be , e.g. , @xmath21 , since such a scale would eliminate the large logarithms arising from the higher - order contributions . unfortunately , this is achieved at the expense that @xmath0 becomes singular in the endpoint regions . it has been conjectured @xcite that gluonic radiative corrections ( sudakov factors ) will suppress that @xmath0-singularity and , therefore , in practical applications of the hsp one may handle this difficulty by cutting off @xmath0 at a certain value , typically chosen in the range 0.5 to 0.7 . another , semi - phenomenological recipe to avoid the singularity of @xmath0 is to introduce an effective gluon mass @xcite which cut - offs the interaction at low @xmath1 values . besides the extreme sensitivity of the form factors on the utilized da and besides the problem with higher - order contributions and/or the singularity of @xmath22 , there is still another perhaps more fundamental difficulty with such calculations . indeed , the applicability of ( [ eq : gm ] ) at experimentally accessible momentum transfer , typically a few gev , is not _ a priori _ justified . it was argued by isgur and llewellyn - smith @xcite and also by radyushkin @xcite that the hsp receives its main contributions from the soft endpoint regions , rendering the perturbative calculation inconsistent . recently , this criticism has been challenged by sterman and collaborators @xcite . based on previous works by collins , soper , and sterman @xcite , they have calculated sudakov corrections to the hard - scattering process taking into account the conventionally neglected transverse momentum , @xmath23 , of the quarks . the sudakov corrections damp those contributions from the endpoint regions in which transverse momenta of the quarks are not large enough to keep the exchanged gluons hard . moreover , as presumed , the sudakov corrections cancel the @xmath22-singularity without introducing additional _ ad hoc _ cut - off parameters as for instance a gluon mass . thus the modified hsp provides a well - defined expression for the form factor which takes into account the perturbative contributions in a self - consistent way , even for momentum transfers as low as a few gev however , an important element has not been considered in the analyses of refs . this concerns the inclusion of the intrinsic transverse momentum of the hadronic wave function . as it was recently shown by two of us @xcite for the case of the pion form factor , the inclusion of the transverse size of the pion extends considerably the self - consistency region of the perturbative contribution down to values of momentum transfer unreachable by the sudakov corrections alone . on the other hand , the incorporation of the @xmath23-dependence leads to a substantial decrease of the magnitude of the ( leading - order ) pion form factor . unfortunately , a clear - cut comparison with the available data is not possible because of their low quality and the uncertainty in the determination of the pion - nucleon coupling constant @xcite . nevertheless , it seems reasonable to expect that the pion form factor receives considerable soft contributions in the presently accessible gev region . the aim of the present paper is to perform an analysis for the proton form factor within the modified hsp . one of our objectives is to critically examine li s approach @xcite and to enlarge the theoretical framework by including the intrinsic @xmath23-dependence of the proton wave function . at the same time we want to clarify several technical points , which are absent in the pion case and are first encountered in the more complicated calculation of the proton form factor . the purpose of our analysis is to investigate how reliably the perturbative contribution to the proton form factor can be calculated and to answer the question whether there is a proton wave function modeled on the basis of qcd sum rules @xcite which is capable of providing , in a theoretically self - consistent way , a good agreement with the data within the modified hsp . it is clear that being able to identify the leading - order perturbative contribution reliably allows us to estimate the size of soft contributions to the proton form factor , contributions which are not accounted for in the modified hsp formalism . [ note that the @xmath23-dependent effects taken into account in the modified hsp represent also soft contributions of higher - twist type . ] sudakov suppression ( which can be viewed as the perturbative part of the transverse wave function ) and intrinsic @xmath23-dependence of the wave function may also have a lot of interesting consequences in other exclusive reactions . thus , for instance , sotiropoulos and sterman @xcite have applied these elements to near - forward proton - proton elastic scattering claiming that their interplay drives the transition of the fixed @xmath24 differential cross section from the @xmath25 behavior at moderate @xmath26 to the @xmath27 behavior at larger @xmath26 , as predicted by dimensional counting rules @xcite . the outline of the paper is as follows . in sec . ii we discuss the proton wave function . the modified hsp is treated in sec . the discussion of the infrared ( ir ) cut - off prescription in the sudakov factor and its effect on the @xmath0-singularities is given in sec . the numerical results are presented in sec . v and our conclusions are contained in sec . similarly to sotiropoulos and sterman @xcite , we write the valence quark component of the proton state with positive helicity in the form @xmath28 \int_{}^ { } [ d^{2}k_{\perp } ] \bigl\ { & \phantom { } & \!\!\!\!\!\ ! \psi _ { 123}\,{\cal m}_{+-+}^{a_{1}a_{2}a_{3 } } + \psi _ { 213}\,{\cal m}_{-++}^{a_{1}a_{2}a_{3 } } \nonumber \\ & - & \bigl(\psi _ { 132}\ , + \ , \psi _ { 231}\bigr){\cal m}_{++-}^{a_{1}a_{2}a_{3 } } \bigr\ } \epsilon _ { a_{1}a_{2}a_{3 } } , \label{eq:|p,+>}\end{aligned}\ ] ] where we have assumed the proton to be moving rapidly in the @xmath29-direction . hence , the ratio of transverse to longitudinal momenta of the quarks is small . the measure over the transverse momentum integration is defined by @xmath30 = \frac{1}{\left(16\pi ^{3}\right)^{2 } } \ , \delta ^{(2 ) } \left ( \sum_{i=1}^{3}\vec{k}_{\perp i } \right ) d^{2}k_{\perp 1 } d^{2}k_{\perp 2 } d^{2}k_{\perp 3}.\ ] ] in the zero binding energy limit , which is characteristic for the parton picture , one has @xmath31 the three quark state with helicities @xmath32 and colors @xmath33 is given by @xmath34 since the orbital angular momentum is assumed to be zero , the proton helicity is the sum of the quark helicities . the quark states are normalized as follows : @xmath35 from the permutation symmetry between the two u quarks and from the requirement that the three quarks have to be coupled to give an isospin @xmath36 state it follows that eq . ( [ eq:|p,+ > ] ) can be expressed in terms of only one independent scalar function @xcite . in the sequel , @xmath37 denotes the momentum space wave function . the subscripts on @xmath37 refer to the order of momentum arguments , for example @xmath38 . note that , in general , the wave function depends on the factorization scale @xmath11 . we make the following convenient ansatz for the wave function : @xmath39 the distribution amplitude @xmath40 ( in the notation of @xcite ) is defined in such a way that @xmath41\ , \phi _ { 123}(x,\mu _ { f } ) = 1 , \label{eq : danorm}\ ] ] where an obvious abbreviated notation has been introduced . the da can be expressed in terms of the eigenfunctions of the evolution equation @xcite , @xmath42 , which are linear combinations of appell polynomials . then the proton da can be cast into the form @xmath43 where the notations of @xcite are adopted . @xmath44 is the asymptotic da mentioned in the introduction . the exponents @xmath45 , driving the evolution behavior of the da , are related to the anomalous dimensions of trilinear quark operators with isospin @xmath36 ( see @xcite ) and resemble the @xmath46 in the brodsky - lepage notation @xcite . because they are positive fractional numbers increasing with n , higher - order terms in ( [ eq : phi ] ) are gradually suppressed . the constants @xmath45 are given in table 1 ; @xmath47 for three flavors . constraints on the da are obtained implicitly by restricting their few first moments within intervals determined from qcd sum rules @xcite , which are evaluated at some self - consistently determined normalization point @xmath11 of order 1 gev ( see , e.g. , @xcite ) : @xmath48x_{1}^{n_{1}}x_{2}^{n_{2}}x_{3}^{n_{3 } } \phi _ { 123}(x,\mu _ { 0})\ ] ] [ eq : moments ] in most model calculations , mentioned above , the moment constraints provided by qcd sum rules are used to determine the first five expansion coefficients @xmath49 , where @xmath50 due to normalization ( 2.7 ) . however , since the moments are burdened by errors , these expansion coefficients although mathematically uniquely determined by the moments of corresponding order @xcite in practice their numerical values can not be fixed precisely giving rise to different options for the proton da . in our calculation of form factors we employ amplitudes complying with the chernyak - ogloblin - zhitnitsky ( coz ) sum - rule moment constraints . it was shown in @xcite that such amplitudes constitute a finite orbit in the @xmath51 plane ranging from coz - like amplitudes @xcite with @xmath52 to the recently proposed @xcite heterotic one with @xmath53 . for the convenience of the reader , the qcd sum - rules constraints and the expansion coefficients @xmath49 of selected model amplitudes are compiled in table 1 . the @xmath23-dependence of the wave function is contained in the function @xmath54 which is normalized according to @xmath55 \omega _ { 123}(x,\vec{k}_{\perp } ) = 1 . \label{eq : omeganorm}\ ] ] due to ( [ eq : danorm ] ) and ( [ eq : omeganorm ] ) , @xmath12 is the value of the da at the origin of the configuration space . its evolution behavior is given by @xmath56 and its value has been determined to be @xmath57 @xcite . in eq . ( [ eq : ansatz ] ) @xmath58 represents the soft part of the proton wave function , which results by removing the perturbative part and absorbing it into the hard - scattering amplitude @xmath5 . the perturbative tail of the full wave function behaves as @xmath59 for large @xmath23 , whereas the soft part vanishes as @xmath60 or faster . the nonperturbative or intrinsic @xmath23-dependence of the soft wave function , being related to confinement , is parametrized as a simple gaussian according to @xmath61 . \label{eq : blhm - omega}\ ] ] this parametrization of the intrinsic @xmath23-dependence of the wave function , which is due to brodsky , huang , lepage , and mackenzie @xcite , seems to be more favorable than the standard form of factorizing @xmath62- and @xmath23-dependencies . at least for the case of the pion wave function , this has recently been effected by zhitnitsky @xcite on the basis of qcd sum rules . he finds that a factorizing wave function is in conflict with some general theoretical constraints which any reasonable wave function should comply . zhitnitsky s qcd sum - rule analysis of the pion wave function seems to indicate that the @xmath23-distribution may also show a double - hump structure , which means that small and large values of @xmath23 are favored relative to intermediate values . it is likely that the proton wave function may exhibit a similar behavior , though this kind of analysis has yet to be done . for the purposes of the present work we ignore this possibility in ( [ eq : blhm - omega ] ) the parameter @xmath63 controls the root mean square transverse momentum ( r.m.s . ) , @xmath64 , and the r.m.s . transverse radius of the proton valence fock state . from the known charge radius of the proton , we expect the r.m.s . transverse momentum to be larger than about @xmath65 mev . the actual value of @xmath64 may be much larger than @xmath65 mev , e.g. , 600 mev or so . indeed , sotiropoulos and sterman @xcite show that application of the modified hsp to proton - proton elastic scattering leads to an approximate @xmath25-behavior of the differential cross section at moderate @xmath66 . the behavior @xmath67 , predicted by dimensional counting , appears only at very large @xmath66 . at precisely which value of @xmath66 the transition from the @xmath25 to the @xmath27 behavior occurs , depends on the transverse size of the valence fock state of the proton . since the isr @xcite and the fnal @xcite data are rather compatible with a @xmath25-behavior of the differential cross section , sotiropoulos and sterman conclude that the transverse size of the proton is small , perhaps @xmath68 fm . correspondingly , the r.m.s . transverse momentum is larger than 600 mev . it is worth noting that such a value is supported by the findings of the emc group @xcite in a study of the transverse momentum distribution in semi - inclusive deep inelastic @xmath69 scattering . a phenomenologically successful approach to the hsp , in which baryons are viewed as bound states of a quark and an effective diquark , also uses a value of this size for @xmath64 @xcite . there is a second constraint on the wave function , _ viz . _ the probability for finding three valence quarks in the proton : @xmath70 \frac{2\left(\phi _ { 123}(x)\right)^{2 } + \phi _ { 132}(x ) \phi _ { 231}(x)}{x_{1}x_{2}x_{3 } } \leq 1.\ ] ] in our numerical analysis to be presented in sec . 5 , we make use of two different values of the r.m.s . transverse momentum , namely , one which is obtained by the requirement @xmath71 for a given wave function . [ this corresponds to the minimum value of the r.m.s . transverse momentum . ] the other option for the r.m.s . transverse momentum we consider is the rather large value of 600 mev . in the latter case , the probability for the valence quark fock state depends on the wave function . following li @xcite , we write the proton form factor in the form @xmath72[dx ' ] \int_{}^{}[d{}^{2}k_{\perp}][d{}^{2}k_{\perp}^{\prime } ] \sum_{j=1}^{2 } t_{h_{j}}(x , x',\vec{k}_{\perp},\vec{k}'_{\perp},q,\mu ) y_{j}(x , x',\vec{k}_{\perp},\vec{k}'_{\perp},\mu _ { f } ) . \label{eq : g_m}\ ] ] note , however , that our notation is slightly different compared to that of li . making use of the symmetry properties of the proton wave function under permutation , the contributions from the 42 diagrams involved in the calculation of the proton form factor in lowest order can be arranged into two reduced hard scattering amplitudes of the form @xmath73 \left [ x_{2}x_{2}'q^{2 } + ( \vec{k}_{\perp 2 } - \vec{k}_{\perp 2}')^{2 } \right ] } , \label{eq : t_h_1}\ ] ] @xmath74 \left [ x_{2}x_{2}'q^{2 } + ( \vec{k}_{\perp 2 } - \vec{k}_{\perp 2 } ' ) \right ] } , \label{eq : t_h_2}\ ] ] where @xmath75 is the casimir operator of the fundamental representation of @xmath76 . in the hard scattering amplitudes only the @xmath23-dependence of the gluon propagators is included , whereas that of the quark propagators has been neglected . it is expected that this technical simplification introduces only a minor error of about @xmath77 in the final result . for the case of the pion form factor this has been explicitly demonstrated by li @xcite . the functions @xmath78 in ( [ eq : g_m ] ) are short - hand notations for linear combinations of products of the initial and final state wave functions @xmath79 , weighted by @xmath9-dependent factors arising from the fermion propagators , namely : @xmath80 @xmath81 ignoring the transverse momenta in the hard scattering amplitudes ( [ eq : t_h_1 ] ) and ( [ eq : t_h_2 ] ) , and inserting ( [ eq : ansatz ] ) and ( [ eq : omeganorm ] ) , one arrives at the standard hsp result for the magnetic form factor . although this expression is correct in the asymptotic momentum domain , the transverse degrees of freedom are an essential ingredient of the formalism and neglecting them leads to inconsistencies in the endpoint regions , where one of the fractional momenta @xmath9 or @xmath82 tends to zero . after all , it is precisely this approximation that is responsible for the inconsistencies mentioned in the introduction . the power of combining the transverse momentum dependence of the hard scattering amplitude and radiative corrections in the form of sudakov form factors was realized by sterman and collaborators @xcite . ultimately , it leads to a suppression of contributions from the dangerous soft regions , where both the longitudinal and transverse momenta of the quarks are small . in order to include the sudakov corrections , it is advantageous to reexpress eq . ( [ eq : g_m ] ) in terms of the variables @xmath83 , which are canonically conjugate to @xmath84 and span the transverse configuration space . then @xmath72[dx ' ] \int_{}^{}\frac{d{}^{2}b_{1}}{(4\pi ) ^{2 } } \frac{d{}^{2}b_{2}}{(4\pi ) ^{2 } } \sum_{j}^{}\ , \hat{t}_{j}(x , x',\vec{b},q,\mu ) \hat{y}_{j}(x , x',\vec{b},\mu _ { f } ) \ , { \rm e}^{-s_{j } } , \label{eq : g_m(b)}\ ] ] where the fourier transform of a function @xmath85 is defined by @xmath86 since the hard scattering amplitudes depend only on the differences of initial and final state transverse momenta , there are only two independent fourier - conjugate vectors @xmath87 and @xmath88 . they are , respectively , the transverse separation vectors between quarks 1 and 3 and between quarks 2 and 3 . accordingly , the transverse separation of quark 1 and quark 2 is given by @xmath89 [ note that sotiropoulos and sterman @xcite define the transverse separations in a cyclic way which results in the interchange @xmath90 , as compared to our definition . ] the fact that there are only two independent transverse separation vectors is a consequence of the approximation made in the treatment of the hard scattering amplitudes ( [ eq : t_h_1 ] ) and ( [ eq : t_h_2 ] ) which disregards the @xmath23-dependence of the quark propagators . this approximation is justified by the enormous technical simplification it entails , given that the thereby introduced errors are very small . then by virtue of rotational invariance of the system with respect to the longitudinal axis , the form factor ( [ eq : g_m(b ) ] ) can be expressed in terms of a seven - dimensional integral instead of an eleven - dimensional one . physically , the relations @xmath91 , @xmath92 mean that the physical probe ( i.e. , the photon ) mediates only such transitions from the initial to the final proton state , which have the same transverse configurations of the quarks . the fourier - transformed hard scattering amplitudes appearing in eq . ( [ eq : g_m(b ) ] ) read @xmath93 @xmath94 where @xmath95 is the modified bessel function of order 0 and @xmath96 denotes the length of the corresponding vector . we have now chosen the renormalization scale in such a way that each hard gluon carries its own individual momentum scale @xmath97 as the argument of the corresponding @xmath0 . the @xmath97 are defined as the maximum scale of either the longitudinal momentum or the inverse transverse separation , associated with each of the gluons : @xmath98 , \nonumber \\ & t_{21 } & = { \rm max } \left [ \sqrt{x_{1}x_{1}^{\prime}q } , 1/b_{1 } \right ] , \nonumber \\ & t_{12 } & = t_{22 } = { \rm max } \left [ \sqrt{x_{2}x_{2}^{\prime}q } , 1/b_{2 } \right ] , \label{eq : t_ij}\end{aligned}\ ] ] one may think of other choices . however , they are not expected to lead to very different predictions for the form factor @xcite . the quantities @xmath99 contain the same combinations of initial and final state wave functions as those in ( [ eq : y_1 ] ) and ( [ eq : y_2 ] ) , the only difference being that now the products @xmath100 are replaced by corresponding products of fourier - transformed wave functions : @xmath101 . using ( [ eq : ansatz ] ) and ( [ eq : blhm - omega ] ) , the fourier transform of the wave function reads @xmath102 where the fourier - transform of the @xmath23-dependent part is given by @xmath103 \right\}. \label{eq : fourieromega}\ ] ] the exponentials @xmath104 in ( [ eq : g_m(b ) ] ) are the sudakov factors , which incorporate the effects of gluonic radiative corrections . because of this , ( [ eq : g_m(b ) ] ) is not simply the fourier transform of ( [ eq : g_m ] ) but an expression comprising an additional physical input . thus ( [ eq : g_m(b ) ] ) may be termed the `` modified hard - scattering formula '' . on the ground of previous works by collins and soper @xcite , botts and sterman @xcite have calculated a sudakov factor using resummation techniques and having recourse to the renormalization group . they find sudakov exponents of the form @xmath105 \nonumber \\ & + & \sum_{l=1}^{3 } \left [ s(x_{l}^{\prime},\tilde{b}_{l},q ) + \int_{1/\tilde{b}_{l}}^{t_{j2 } } \frac{d\bar{\mu}}{\bar{\mu } } \gamma _ { q}(g(\bar{\mu } ^{2 } ) ) \right ] , \label{eq : s}\end{aligned}\ ] ] wherein the sudakov functions @xmath106 are given by @xmath107 \nonumber \\ & - & \left [ \frac{a^{(2)}}{4\beta _ { 1}^{2 } } - \frac{a^{(1)}}{4\beta _ { 1 } } { \rm ln}\bigl({\rm e}^{2\gamma -1}/2\bigr ) \right ] { \rm ln}\left ( \frac{\hat{q}_{l}}{\hat{b}_{l } } \right ) \nonumber \\ & - & \frac{a^{(1)}\beta _ { 2}}{32\beta _ { 1}^{3 } } \left [ { \rm ln}^{2}(2\hat{q}_{l } ) - { \rm ln}^{2}(2\hat{b}_{l } ) \right ] . \label{eq : s}\end{aligned}\ ] ] here @xmath108 or @xmath109 ( @xmath110 ) and the variables @xmath111 and @xmath112 are defined as follows : @xmath113\ ] ] @xmath114.\ ] ] the coefficients @xmath115 and @xmath116 are @xmath117 @xmath118 where @xmath119 is the number of quark flavors and @xmath120 is the euler - mascheroni constant . in the sequel @xmath121 @xmath122 is the anomalous quark dimension in the axial gauge @xcite . the sudakov function , @xmath106 , in ( [ eq : s ] ) takes into account leading and next - to - leading gluonic radiative corrections of the form shown in fig . [ fig : gluonlines ] . the quantities @xmath123 ( @xmath110 ) are infrared cut - off parameters , naturally related to , but not uniquely determined by the mutual separations of the three quarks @xcite . a physical perspective on the choice of the ir cut - off is provided by the following analogy to ordinary qed . one expects that because of the color neutrality of a hadron , its quark distribution can not be resolved by gluons with a wave length much larger than a characteristic quark separation scale ; meaning that long wave length gluons probe the color singlet proton and hence radiation is damped . radiative corrections with wave lengths between the ir cut - off and an upper limit ( related to the physical momentum q ) yield to suppression ; it is understood that still softer gluonic corrections are already taken care of in the hadron wave function , whereas harder gluons are considered as part of @xmath5 . different choices of the ir cut - off have been used in the literature : thus , li @xcite chooses @xmath124 ( this choice hereafter is termed the `` l '' prescription ) , whereas hyer @xcite in his analysis of the proton - antiproton annihilation into two photons and of the time - like proton form factor as well as sotiropoulos and sterman @xcite take @xmath125 , @xmath126 , @xmath127 ( this choice is denoted the `` h - ss '' prescription ) . still another possibility , and the one proposed in the present work for reasons that will be explained below is to use as ir cut - off the maximum of the three interquark separations , i.e. , to set @xmath128 this choice , designated by `` max '' , is analogous to that in the meson case , wherein the quark - antiquark distance naturally provides a secure ir cut - off . the specific features of each particular cut - off choice will be discussed in detail in sec . [ sec : singularities ] . the integrals in ( [ eq : s ] ) arise from the application of the renormalization group equation ( rge ) . the evolution from one scale value to another is governed by the anomalous dimensions of the involved operators . the integrals combine the effects of the application of the rge on the wave functions and on the hard scattering amplitude . the range of validity of ( [ eq : s ] ) for the sudakov functions is limited to not too small @xmath129 values . whenever @xmath130 is large relative to the hard ( gluon ) scale @xmath131 , the gluonic corrections are to be considered as higher - order corrections to @xmath5 and hence are not contained in the sudakov factor but are absorbed in @xmath5 . for that reason , li @xcite sets any sudakov function @xmath106 equal to zero whenever @xmath132 . moreover , li holds the sudakov factor @xmath104 equal to unity whenever it exceeds this value , which is the case in the small @xmath123-region . actually , the full expression ( [ eq : s ] ) shows in this region a small enhancement resulting from the interplay of the next - to - leading logarithmic contributions to the sudakov exponents and the integrals over the anomalous dimensions . we follow the same lines of argument in our analysis . the ir cut - offs @xmath130 in the sudakov exponents mark the interface between the nonperturbatively soft momenta , which are implicitly accounted for in the proton wave function , and the contributions from soft gluons , incorporated in a perturbative way in the sudakov factors . obviously , the ir cut - off serves at the same time as the gliding factorization scale @xmath11 to be used in the evolution of the wave function . for that reason , li @xcite as well as sotiropoulos and sterman @xcite take @xmath133 . the `` max '' prescription ( [ eq : max ] ) , adopted in the present work , naturally complies with the choice of the evolution scale proposed in @xcite . it is well known that the inclusion of an @xmath62-dependent renormalization scale in the argument of @xmath0 within the standard hsp of brodsky - lepage @xcite presents the difficulty that the value of @xmath0 becomes singular in the endpoint regions . to render the form factors ( eq . ( [ eq : gm ] ) ) finite , additional external parameters , like an effective gluon mass @xcite or a cut - off prescription have to be introduced . technically , such parameters play the rle of ir regulators serving to regularize one of the gluon propagators , which may become soft along the boundaries of phase space ( see , e.g. , @xcite ) . one of the crucial advantages of the modified hsp , proposed by sterman and collaborators @xcite , is that there is no need for external regulators because the sudakov factor may suppress the singularities of the `` bare '' ( one - loop ) @xmath135 inherently . indeed in the pion case , it was shown @xcite that the transverse quark - antiquark separation is tantamount to an ir regulator which suffices to cancel all singularities from the soft region . concerning the proton form factor , the situation is much more complicated because more scales are involved and hence the choice of the appropriate ir cut - off parameters @xmath123 is not obvious , as discussed in sec . iii . as we shall effect in the following , the cancellation of the @xmath0-singularities by the sudakov factor depends sensitively on that particular choice . in fig . [ fig : sudakov ] we display the exponential of the sudakov function @xmath136 $ ] for @xmath137 by imposing li s requirement @xcite : @xmath138 whenever @xmath139 . ultimately , the cancellation of the @xmath0-singularities relies on the fact that whenever one of the @xmath0 tends to infinity ( owing to the limit @xmath140 ) , the sudakov factor @xmath104 rapidly decreases to zero . as it can be observed from fig . [ fig : sudakov ] this is not the case in the region determined by @xmath141 and simultaneously @xmath142 , where @xmath136 $ ] is fixed to unity . in the pion case this does not matter , since the other @xmath143\to 0 $ ] faster than any power of @xmath144 $ ] and , consequently , the sudakov factor drops to zero . in contrast , the treatment of the proton form factor is more subtle . in that case , @xmath104 does not necessarily vanish fast enough to guarantee the cancellation of the @xmath0-singularities . this can be illustrated by the following configuration : if , say , @xmath145 and @xmath146 then @xmath147 and @xmath148 can have any value between 0 and @xmath149 . since @xmath150 is unrestricted within the limits @xmath151 and @xmath152 , the corresponding exponentials of the sudakov functions @xmath153 $ ] and @xmath154 $ ] do not automatically fall off to zero in order to yield sufficient suppression of the @xmath0-singularities , unless all three @xmath123 are coerced to be equal . if the three @xmath123 are allowed to be different , then the sudakov factor provides suppression only through the contributions of the anomalous dimensions . according to the `` l '' and `` h - ss '' prescriptions , which , in general , allow for different @xmath123 in the sudakov functions , the integrand in ( [ eq : g_m(b ) ] ) has singularities behaving as @xmath155 for @xmath156 and @xmath157 hold fixed . the maximum degree of divergence is given by @xmath158 where the first term @xmath159 comes from the evolution of @xmath12 , ( [ eq : f_n ] ) and the constant @xmath160 is related to the anomalous dimension driving the evolution behavior of the proton da , see ( [ eq : phi ] ) and table 1 : @xmath160 is the maximum value of the @xmath161 within a given polynomial order of the expansion of the da . we reiterate that the @xmath45 are positive fractional numbers increasing with @xmath162 . thus the singular behavior of the integrand becomes worse as the expansion in terms of appell polynomials extends to higher and higher orders . the term @xmath163 in ( [ eq : kappa ] ) stems from the integrations over the anomalous dimensions in the sudakov factor @xmath104 ( see ( [ eq : s ] ) ) . finally , the term @xmath152 originates from that @xmath164 which becomes singular in ( [ eq : g_m(b ) ] ) , c.f . , ( [ eq : t_ij ] ) . which one of the @xmath0 couplings becomes actually singular , depends on the prescription imposed on the ir cut - off parameters @xmath123 . the integral ( [ eq : g_m(b ) ] ) does not exist if @xmath165 . as table 1 reveals , this happens already for proton das which include appell polynomials of order 1 , i.e. , for all das except for the asymptotic one : @xmath17 . thus application of the `` l '' and `` h - ss '' prescriptions on the choice of the ir cut - off parameters @xmath123 to the proton form factor entails the modified hsp to be invalid . in view of these results , li s analysis of the proton form factor @xcite seems to be seriously flawed . a simple recipe to bypass the singular behavior of the integrand is to ignore completely the evolution of the da or to `` freeze '' it at any ( arbitrary ) value larger than @xmath166 . hyer @xcite suggested to take for the factorization scale @xmath167 . in this case , the @xmath160 appears in ( [ eq : kappa ] ) only if all three @xmath123 tend to @xmath168 at once . but then at least one of the @xmath136 $ ] drops to @xmath151 faster than any power of @xmath169 . apparently , hyer s choice of the factorization scale avoids singularities of the form ( [ eq : sing ] ) , but seems to us physically implausible . since he only presents numerical results for the proton form factor in the time - like region , we can not compare with his results directly . another option , and actually the one proposed in this work , is to use a common ir cut - off not only for the evolution of the wave function but also in the sudakov exponent . for a common cut - off @xmath170 , the sudakov factors always cancel the @xmath0-singularities ; if , for a given @xmath171 , we are in the dangerous region , @xmath172 , @xmath173 , at least one of the other two sudakov functions lies in the region @xmath174 , @xmath173 ( @xmath175 ) and therefore provides sufficient suppression , as outlined above . in particular , we favor @xmath176 as the optimum choice ( `` max '' prescription ) , since it does not only lead to a regular integral but also to a non - singular integrand . the sudakov factor @xmath177 subject to the `` l '' and `` max '' prescriptions is plotted for a specific quark configuration in fig . [ fig : bcplots ] . this figure makes it apparent that the sudakov factor in connection with the `` max '' prescription is unencumbered by singularities in the dangerous soft regions . as a consequence of the regularizing power of the `` max '' prescription , the perturbative contribution to the proton form factor ( [ eq : g_m(b ) ] ) saturates in the sense that the results become insensitive to the inclusion of the soft regions . a saturation as strong as possible is a prerequisite for the self - consistency of the modified hsp , as will be discussed in sec . v. to demonstrate the amount of saturation , we calculate the proton form factor through ( [ eq : g_m(b ) ] ) , employing a cut - off procedure to the @xmath96-integrations at a maximum value @xmath178 . in fig . [ fig : g_m(b_c ) ] the dependence of @xmath16 on @xmath178 for the three choices , labeled : `` l '' , `` h - ss '' , and `` max '' is shown using , for reasons of comparison with previous works , the coz da and ignoring evolution . [ evolution has been dispensed with to avoid the concomitant singularity in @xmath179 as @xmath180 when imposing the `` l '' and `` h - ss '' prescriptions . ] as one sees from the figure , the `` max '' prescription leads indeed to saturation ; the soft region @xmath181 does not contribute to the form factor substantially . in fact , already @xmath182 of the result are obtained from the regions with @xmath183 . note that @xmath0 increases to a value of @xmath184 at @xmath185 . this indicates that a sizeable fraction of the contributions to the form factor is accumulated in the perturbative region . unfortunately , this saturation is achieved at the expense of a rather strong damping of the perturbative contribution to the proton form factor . using the two other prescriptions ( `` l '' and `` h - ss '' ) and ignoring evolution , we have found larger results for @xmath16 , but no indication for saturation : the additional contributions to the form factor gained this way are accumulated exclusively in the soft regions , i.e. , for values of @xmath186 near @xmath152 . these findings are in evident contradiction to li s results ( figure 5 in @xcite ) for which an acceptable saturation has been claimed . on the other hand , we can qualitatively confirm the saturation behavior of the proton form factor calculated by hyer @xcite in the time - like region . since we regard a saturation behavior as a stringent test for the self - consistent applicability of pqcd , calculations which accumulate large contributions from soft regions ( large @xmath178 ) can not be considered as theoretically legitimate . the rle of the evolution effect subject to the `` max '' prescription is also exhibited in fig . [ fig : g_m(b_c ) ] . it shows that the effect of evolution is large , although finite , owing to the strong suppression provided by the sudakov factor . note that according to our discussion in sec . iii , the factorization scale is @xmath187 . the significant feature of the evolution effect is that it tends to neutralize the influence of the ir cut - off . thus one obtains larger values of the proton form factor at the expense of a slightly worse saturation . in this section we give numerical results for the proton form factor . in these calculations we throughout employ the `` max '' prescription with evolution included , using @xmath188180 mev and @xmath189 gev . before proceeding with the presentation of our final results , let us investigate the effect of including the intrinsic transverse momentum in our calculations . the @xmath23-dependence of the proton wave function effectively introduces a confinement scale in the formalism , the importance of which may be appreciated by looking at fig . [ fig : g_m(q^2 ) ] . this figure shows results , obtained for the coz da without @xmath23-dependence and for two different values of @xmath190 . to describe the intrinsic @xmath23-dependence , one can use ( [ eq : blhm - omega ] ) or , after transforming to the transverse configuration space , ( [ eq : fourieromega ] ) . notice that in li s approach the gaussian in ( [ eq : fourieromega ] ) has been replaced by unity . the oscillator parameter @xmath63 is determined in such a way that either the normalization of the wave function @xmath191 is unity ( resulting into @xmath192 mev for the coz da ) , or by inputing the value of the r.m.s . transverse momentum . in the second case , we use a value of @xmath193 mev ( see the discussion in sec . ii ) , which implies @xmath194 . as can be seen from this figure , the predictions for the form factor are quite different for the three cases . the intrinsic @xmath23-dependence of the wave function leads to further suppression of the perturbative contribution , which becomes substantial if the r.m.s . transverse momentum is large . on the other hand , this suppression is accompanied by an increasing amount of saturation , since also the gaussian ( [ eq : fourieromega ] ) suppresses predominantly contributions from the soft regions , viz . , the large @xmath195-regions . in contrast to the sudakov factor , however , this suppression is @xmath196-independent . the interplay of the two effects , sudakov suppression and intrinsic transverse momentum , leads to a different @xmath196-behavior of the form factor depending on the value of the r.m.s transverse momentum , as can be seen from fig . [ fig : g_m(q^2 ) ] . the @xmath196-dependence beyond @xmath4 gev@xmath3 is rather weak , being approximately compatible with dimensional counting ( modulo logarithmic corrections ) . for very large values of @xmath196 beyond @xmath197 gev@xmath3 the three curves have approached each other within @xmath77 accuracy . this happens when the sudakov factor dominates the gaussian ( [ eq : fourieromega ] ) and selects those configurations with small interquark separations . in this region , which one may consider as the pure perturbative region , the results for the form factor are independent of the confinement scale introduced by the r.m.s . transverse momentum . the penalty of the additional suppression of the perturbative contribution caused by the gaussian ( [ eq : fourieromega ] ) is mitigated by the advantage that the perturbative contribution becomes more self - consistent than by the sudakov factor alone . this is indicated in the enhanced amount of saturation with increasing r.m.s . transverse momentum . adapting the criterion of self - consistency , originally suggested by li and sterman @xcite for the pion case , namely that @xmath198 of the results are accumulated at moderate values of the coupling constant , say , @xmath199 , we find self - consistency for @xmath200 gev@xmath3 ( for the coz da ) . finally , in fig . [ fig : strip ] , we demonstrate the effect of different proton das on the form factor . to this end , we investigate a set of @xmath201 das @xcite , which all respect the qcd sum - rule constraints @xcite . the results for the various das or more precisely wave functions , since we include their intrinsic transverse momentum dependence obtained under the `` max '' prescription with evolution included , form the shaded area shown in the figure . all wave functions are normalized to unity and the corresponding r.m.s . transverse momenta vary between @xmath202 mev and @xmath203 mev ( see table 1 ) . the theoretical form - factor predictions span a `` band '' congruent to the `` orbit '' of solutions found in @xcite . the upper bound of the `` band '' corresponds to the da coz@xmath204 , which yields the maximum value of the form - factor ratio @xmath205 in the standard hsp . the lower limit of the `` band '' is obtained using the da `` low '' ( sample 8 in @xcite ) with @xmath206 . explicitly shown are the results for the coz da @xcite , its optimized version ( with respect to the sum - rule constraints ) and the `` heterotic '' da , recently proposed by two of us in @xcite . we note that the differences among these curves practically disappear already at about @xmath207 gev@xmath3 , despite the fact that these amplitudes have distinct geometrical characteristics @xcite . since the true valence fock state probability is likely much smaller , or invariably the r.m.s . transverse momentum larger than of order of @xmath208 mev , the `` band '' describes rather _ maximal _ expectations for the ( leading - order ) perturbative contributions to the form factor ; at least for proton wave functions of the type we utilize . comparison with the experimental data reveals that the theoretical predictions amount , at best , to approximately @xmath198 of the measured values . this is the benchmark against which we have to discern novelties and aberrations . closing this discussion we note that , since we are calculating only the helicity - conserving part of the current matrix element it is not obvious whether we should compare the theoretical predictions with the data for the sachs form factor @xmath16 or the dirac form factor @xmath209 . therefore we have exhibited in fig . [ fig : strip ] both sets of data @xcite for comparison . since the two sets of data differ by only @xmath77 , our conclusions concerning the smallness of the theoretical results remain unaffected . the various model wave functions led to self - consistency of the perturbative contribution , i.e. , @xmath198 of the results are accumulated in regions where @xmath199 , in the range of @xmath1 between @xmath2 and @xmath4 gev@xmath3 . the objective of the present work has been to derive the proton magnetic form factor within the modified version ( sec . iii ) of the standard brodsky - lepage hsp @xcite , a scheme which takes into account gluonic radiative corrections @xcite in terms of transverse separations . this is done by incorporating in the formalism the sudakov factor , calculated by botts and sterman @xcite . there are already some interesting applications of the modified hsp @xcite . the significant element of this type of analyses is that the @xmath0-singularities , arising from hard - gluon exchange and evolution , can be cancelled without introducing free external parameters . we emphasize that in contrast to pure phenomenological recipies ( e.g. , the introduction of a gluon mass ) , the modified hsp provides an explicit scheme how the ir protection of the `` bare '' @xmath0 proceeds through gluonic radiation accumulated in the sudakov factor . thus , in the modified hsp , one may conceive of the ( finite ) ir - protected @xmath0 as being the effective coupling . by this procedure the potentially dangerous soft regions of momenta are suppressed entailing also a reduction of the perturbative contribution to the form factor . while in the pion case @xcite , it is fortunate that the cancellation of the @xmath0-singularities comes out naturally , li s approach to the proton form factor @xcite leads to a lack of complete cancellation of the @xmath0-singularities ( see sec . iv ) . without evolution of the proton wave function the emerging singularities in ( [ eq : g_m(b ) ] ) are still integrable , but logarithmic corrections due to evolution yield ultimately to uncompensated singularities . on the grounds of our discussion , we are reasonably confident that li s treatment can be cured within the modified hsp . we suggest to use a a common ir cut - off in the sudakov exponents ( [ eq : s ] ) and sudakov functions ( [ eq : s ] ) : viz . , the maximum transverse separation . this `` max '' prescription provides sufficient ir protection , since even with evolution , the integrand in ( [ eq : g_m(b ) ] ) remains finite . a significant feature of this treatment is that the proton form factor saturates , i.e. , it becomes insensitive to the contributions from large transverse separations . the other choices of the ir cut - off ( `` l '' , `` h - ss '' ) , we have discussed , do not lead to saturation . however , this reliable saturation and ir protection of the form factor is achieved at the expense of a strong reduction of the perturbative contribution to the form factor . the damping of the proton form factor becomes even stronger if one takes into account the intrinsic transverse momentum dependence of the proton wave function ( see fig . [ fig : g_m(b_c ) ] and fig . [ fig : g_m(q^2 ) ] ) . this has been done by assuming a non - factorizing @xmath62 and @xmath23-dependence of the wave function of the brodsky - lepage - huang - mackenzie @xcite type and fixing the value of @xmath190 either via the valence quark probability @xmath191 or by inputing the value @xmath193 mev by hand @xcite . a remarkable finding is that the form factor calculated within the modified hsp , appropriately extended to include the intrinsic transverse momentum of the proton wave function , shows only a mild dependence on the particular model da . the perturbative contribution to the form factor becomes self - consistent in all cases for momentum transfers larger than @xmath2 to @xmath4 gev@xmath3 . the actual value of the onset of self - consistency depends on the particular wave function and the r.m.s . transverse momentum chosen . self - consistency is defined such that @xmath198 of the result are accumulated in regions where @xmath210 is smaller than @xmath211 ( sec . v ) . comparing our theoretical results with the data , it turns out that they fall short by at least @xmath198 . this is true not only for the coz da , ( which we have exemplarily used to facilitate comparison with previous works ) but actually for the whole spectrum of amplitudes determined in @xcite and found to comply with the coz sum - rule requirements . depending on the actual value of the r.m.s . transverse momentum , the reduction of the perturbative contribution may be even stronger than @xmath198 . the fact that in all considered cases the self - consistently calculated perturbative contribution to the proton form factor fails to reproduce the existing data , is perhaps a signal that soft contributions ( higher twists ) not accounted for so far by the modified hsp should be included . such contributions comprise , e.g. , improved and/or more complicated wave functions , orbital angular momentum , higher fock components , quark - quark correlations ( diquarks ) , radiative corrections to the quark and gluon condensates , quark masses , etc . also remainders of genuine soft contributions , like vector - meson - dominance terms or the overlap of the soft parts of the wave functions ( feynman contributions ) , may still be large at accessible momentum transfers . the rather large value of the pauli form factor @xmath14 around @xmath4 gev@xmath3 , as found experimentally @xcite , indicates that sizeable higher - twist contributions still exist in that region of momentum transfer @xcite . one may suspect similar or even larger higher - twist contributions to the helicity non - flip current matrix element controlling @xmath209 and @xmath16 . large ( perturbative ) higher - order corrections to the hard - scattering amplitude can not be excluded as well , since their size has not yet been estimated . in analogy to the drell - yan process , these corrections may be condensed in a k - factor multiplying the leading - order perturbative result . however , with our choice of the renormalization scale , the k - factor is expected to be close to unity . at least for the case of the pion form factor , calculations of the k - factor to one - loop order exist @xcite , which indicate that choosing the renormalization scale analogously to ours , the value of the k - factor is indeed close to unity . in conclusion we note that it was not our primary aim to use the modified hsp to obtain best agreement with the data , although from our point of view this scheme represents a decisive step towards a deeper understanding of the electromagnetic form factors . in the present work the focus has been placed on theoretical problems , overlooked previously . lepage , s.j . brodsky : phys . d22 ( 1980 ) 2157 v.l . chernyak , i.r . zhitnitsky : nucl . b246 ( 1984 ) 52 m. gari , n.g . stefanis : phys . b175 ( 1986 ) 462 ; phys . d35 ( 1987 ) 1074 i.d . king , c.t . sachrajda : nucl . b279 ( 1987 ) 785 c.e . carlson , m. gari , n.g . stefanis : phys . 58 ( 1987 ) 1308 c .- r . sill , r.m . lombard - nelsen : phys . d36 ( 1987 ) 165 n.g . stefanis : phys . d40 ( 1989 ) 2305 ; d44 ( 1991 ) 1616(e ) v.l . chernyak , a.a . ogloblin , i.r . zhitnitsky : z. phys . c42 ( 1989 ) 569 a. schfer : phys . b217 ( 1989 ) 545 n.g . stefanis , m. bergmann : phys . d47 ( 1993 ) r3685 s.j . brodsky , g.r . farrar : phys . d1 ( 1975 ) 1309 p. bosted _ et al . 68 ( 1992 ) 3841 p. kroll : proceedings of workshop on hadron structure 93 , s. dubnika , a. z. dubniov ( eds . ) , bansk tiavnica , slovakia , sept . 5 - 10 , 1993 n.g . stefanis , m. bergmann : proceedings of workshop on exclusive reactions at high momentum transfer , c.e . carlson , p. stoler , m. taiuti ( eds . ) , elba , italy , 24 - 26 june , 1993 , world scientific , singapore , 1994 ; proceedings of workshop on quantum field theoretical aspects of high energy physics , b. geyer , e .- ilgenfritz ( eds . ) , kyffhuser , germany , september 20 - 24 , 1993 s. rock _ et al . _ : phys . 49 ( 1982 ) 1139 ; r.g . et al . _ : phys . 57 ( 1986 ) 174 ; s. platchkov _ et al . _ : nucl . a510 ( 1990 ) 740 ; s. rock _ et al . _ : phys . d46 ( 1992 ) 24 ; a. lung _ et al . _ : phys . 70 ( 1993 ) 718 ; a.f . d48 ( 1993 ) 29 r.d . b186 ( 1981 ) 429 f.m . dittes , a.v . radyushkin : yad . 34 ( 1981 ) 529 [ sov . j. nucl . 34 ( 1981 ) 293 ] j.m . cornwall : phys . d26 ( 1982 ) 1453 n. isgur , c.h . llewellyn - smith : phys . 52 ( 1984 ) 1080 ; nucl . phys . b317 ( 1989 ) 526 a.v . radyushkin : nucl . a527 ( 1991 ) 153c ; _ ibid . _ a532 ( 1991 ) 141c ; j. botts , g. sterman : nucl . phys . b325 ( 1989 ) 62 h .- li , g. sterman : nucl . b381 ( 1992 ) 129 h .- li : phys . d48 ( 1993 ) 4243 j.c . collins , d.e . soper : nucl . b193 ( 1981 ) 381 ; _ ibid . _ b194 ( 1982 ) 445 ; j.c . collins , d.e . soper , g. sterman : nucl . b261 ( 1985 ) 104 r. jakob , p. kroll : phys . b315 ( 1993 ) 463 c.e . carlson , j. milana : phys . 14 ( 1990 ) 1717 m.g . sotiropoulos , g. sterman : stony brook report itp - sb-93 - 83 ( january 1994 ) s.j . brodsky , g. farrar : phys . ( 1973 ) 1153 m. peskin : phys . lett . 88b ( 1979 ) 128 m. bergmann , n.g . stefanis : phys . d48 ( 1993 ) r2990 m. bergmann , n.g . stefanis : phys . b325 ( 1994 ) 183 g.p . lepage , s.j . brodsky , t. huang , p.b . mackenzie : banff summer institute , particles and fields 2 , a.z . capri , a.n . kamal ( eds . ) ( 1983 ) a.r . zhitnitsky : dallas preprint ( smu - hep-93 - 25 ) ( december 1993 ) e. nagy _ et al . _ : nucl . b150 ( 1979 ) 221 j.l . et al . _ : phys . 39 ( 1977 ) 979 ; w. feissler _ et al . d23 ( 1981 ) 33 j.j . : phys.lett . 95b ( 1980 ) 306 ; a. schlagbhmer : ph.d thesis , freiburg ( 1981 ) p. kroll : proceedings of the adriatico research conference on _ spin and polarization dynamics in nuclear and particle physics _ , trieste , italy , 1988 ; p. kroll , m. schrmann , w. schweiger : z. phys . a338 ( 1991 ) 339 ; _ ibid . _ a342 ( 1992 ) 429 ; p. kroll , m. schrmann , w. schweiger : int . phys . a6 ( 1991 ) 4107 p. kroll _ et al . _ : phys . b316 ( 1993 ) 546 r. jakob _ et al . _ : z. phys . a347 ( 1993 ) 109 d.e . soper : nucl . b163 ( 1980 ) 93 t. hyer : phys . d47 ( 1993 ) 3875 t. gousset , b. pire : saclay report dapnia / sphn-94 - 09 ( 03/1994 ) c. corian , h .- li : phys . b 309 ( 1993 ) 409 lrrrrrr @xmath212 & @xmath213 & @xmath214(coz@xmath215 ) & @xmath214(coz ) & @xmath214(coz@xmath216 ) & @xmath214(het ) & @xmath214(low ) @xmath217 @xmath218 & 20/9 & 3.2185 & 3.6750 & 3.5268 & 3.4437 & 4.1547 @xmath217 @xmath219 & 24/9 & 1.4562 & 1.4840 & 1.4000 & 1.5710 & 1.4000 @xmath217 @xmath220 & 32/9 & 2.8300 & 2.8980 & 2.8736 & 4.5937 & 3.3756 @xmath217 @xmath221 & 40/9 & -17.3400 & -6.6150 & -4.5227 & 29.3125 & 26.1305 @xmath217 @xmath222 & 42/9 & 0.4700 & 1.0260 & 0.8002 & -0.1250 & -0.5855 @xmath217 & 271 & 271 & 267 & 317 & 299 @xmath217 & 0.9893 & 0.9939 & 1.0069 & 0.8537 & 0.9217 @xmath217

this paper reports results of the third - year campaign of monitoring super - eddington accreting massive black holes ( seambhs ) in active galactic nuclei ( agns ) between . ten new targets were selected from quasar sample of sloan digital sky survey ( sdss ) , which are generally more luminous than the seambh candidates in last two years . h lags ( ) in five of the 10 quasars have been successfully measured in this monitoring season . we find that the lags are generally shorter , by large factors , than those of objects with same optical luminosity , in light of the well - known relation . the five quasars have dimensionless accretion rates of . combining measurements of the previous seambhs , we find that the reduction of h lags tightly depends on accretion rates , , where is the h lag from the normal relation . fitting 63 mapped agns , we present a new scaling relation for the broad - line region : $ ] , where is 5100 continuum luminosity , and coefficients oflt - d , , and . this relation is applicable to agns over a wide range of accretion rates , from to . implications of this new relation are briefly discussed .

1604.06218

this is the fifth paper of the series reporting the ongoing large campaign of monitoring super - eddington accreting massive black holes ( seambhs ) in active galaxies and quasars starting from october 2012 . one of the major goals of the campaign is to search for massive black holes with extreme accretion rates through reverberation mapping ( rm ) of broad emission lines and continuum . results from the campaigns in 20122013 and 20132014 have been reported by du et al . ( 2014 , 2015 , hereafter papers i and iv ) , wang et al . ( 2014 , paper ii ) and hu et al . ( 2015 , paper iii ) . this paper carries out the results of seambh2014 sample , which was monitored from september 2014 to june 2015 . with the three monitoring years of observations , we build up a new scaling relation of the broad - line region ( blr ) in this paper . lllllccrr[!ht ] + mrk 335 & 00 06 19.5 & @xmath1520 12 10 & 0.0258 & oct . , 2012 @xmath16 feb . , 2013 & 91 & & @xmath17 & @xmath18 + mrk 1044 & 02 30 05.5 & @xmath1608 59 53 & 0.0165 & oct . , 2012 @xmath16 feb . , 2013 & 77 & & @xmath19 & @xmath20 + iras 04416 + 1215 & 04 44 28.8 & @xmath1512 21 12 & 0.0889 & oct . , 2012 @xmath16 mar . , 2013 & 92 & & @xmath21 & @xmath22 + mrk 382 & 07 55 25.3 & @xmath1539 11 10 & 0.0337 & oct . , 2012 @xmath16 may . , 2013 & 123 & & @xmath23 & @xmath24 + mrk 142 & 10 25 31.3 & @xmath1551 40 35 & 0.0449 & nov . , 2012 @xmath16 apr . , 2013 & 119 & & @xmath25 & @xmath26 + mcg @xmath27 & 11 39 13.9 & @xmath1533 55 51 & 0.0328 & jan . , 2013 @xmath16 jun . , 2013 & 34 & & @xmath28 & @xmath29 + iras f12397 + 3333 & 12 42 10.6 & @xmath1533 17 03 & 0.0435 & jan . , 2013 @xmath16 may . , 2013 & 51 & & @xmath30 & @xmath31 + mrk 486 & 15 36 38.3 & @xmath1554 33 33 & 0.0389 & mar . , 2013 @xmath16 jul . , 2013 & 45 & & @xmath32 & @xmath33 + mrk 493 & 15 59 09.6 & @xmath1535 01 47 & 0.0313 & apr . , 2013 @xmath16 jun . , 2013 & 27 & & @xmath34 & @xmath35 + + sdss j075101.42 + 291419.1 & 07 51 01.4 & @xmath1529 14 19 & 0.1208 & nov . , 2013 @xmath16 may . , 2014 & 38 & & @xmath36 & @xmath37 + sdss j080101.41 + 184840.7 & 08 01 01.4 & @xmath1518 48 40 & 0.1396 & nov . , 2013 @xmath16 apr . , 2014 & 34 & & @xmath38 & @xmath39 + sdss j080131.58 + 354436.4 & 08 01 31.6 & @xmath1535 44 36 & 0.1786 & nov . , 2013 @xmath16 apr . , 2014 & 31 & & @xmath40 & @xmath41 + sdss j081441.91 + 212918.5 & 08 14 41.9 & @xmath1521 29 19 & 0.1628 & nov . , 2013 @xmath16 may . , 2014 & 34 & & @xmath42 & @xmath43 + sdss j081456.10 + 532533.5 & 08 14 56.1 & @xmath1553 25 34 & 0.1197 & nov . , 2013 @xmath16 apr . , 2014 & 27 & & @xmath44 & @xmath45 + sdss j093922.89 + 370943.9 & 09 39 22.9 & @xmath1537 09 44 & 0.1859 & nov . , 2013 @xmath16 jun . , 2014 & 26 & & @xmath46 & @xmath47 + + sdss j075949.54 + 320023.8 & 07 59 49.5 & @xmath1532 00 24 & 0.1880 & sep . , 2014 @xmath16 may . , 2015 & 27 & & @xmath48 & @xmath49 + sdss j080131.58 + 354436.4 & 08 01 31.6 & @xmath1535 44 36 & 0.1786 & oct . , 2014 @xmath16 may . , 2015 & 19 & & @xmath50 & @xmath51 + sdss j084533.28 + 474934.5 & 08 45 33.3 & @xmath1547 49 35 & 0.3018 & sep . , 2014 @xmath16 apr . , 2015 & 18 & & @xmath52 & @xmath53 + sdss j085946.35 + 274534.8 & 08 59 46.4 & @xmath1527 45 35 & 0.2438 & sep . , 2014 @xmath16 jun . , 2015 & 26 & & @xmath54 & @xmath55 + sdss j102339.64 + 523349.6 & 10 23 39.6 & @xmath1552 33 50 & 0.1364 & oct . , 2014 @xmath16 jun . , 2015 & 26 & & @xmath56 & @xmath57 reverberation mapping ( rm ) technique , measuring the delayed echoes of broad lines to the varying ionizing continuum ( bahcall et al . 1972 ; blandford & mckee 1982 ; peterson 1993 ) , is a powerful tool to probe the kinematics and geometry of the blrs in the time domain . countless clouds , which contribute to the smooth profiles of the broad emission lines ( e.g. , arav et al . 1997 ) , are distributed in the vicinity of supermassive black hole ( smbh ) , composing the blr . as an observational consequence of photonionization powered by the accretion disk under the deep gravitational potential of the smbh , the profiles of the lines are broadened , and line emission from the clouds reverberate in response to the varying ionizing continuum . the reverberation is delayed because of light travel difference between h@xmath1 and ionizing photons and is thus expected to deliver information on the kinematics and structure of the blr . the unambiguous reverberation of the lines , detected by monitoring campaigns from ultraviolet to optical bands since the late 1980s , supports this picture of the central engine of agns ( e.g. , clavel et al . 1991 ; peterson et al . 1991 , 1993 ; maoz et al . 1991 ; wanders et al . 1993 ; dietrich et al . 1993 , 1998 , 2012 ; kaspi et al . 2000 ; denney et al . 2006 , 2010 ; bentz et al . 2009 , 2014 ; grier et al . 2012 ; papers i - iv ; barth et al . 2013 , 2015 ; shen et al . 2015a , b ) . the @xmath3 relation was first discussed by koratkar & gaskell ( 1991 ) and peterson ( 1993 ) . robust rm results for 41 agns in the last four decades lead to a simple , highly significant correlation of the form @xmath58 where @xmath59 is the 5100 luminosity in units of @xmath60 ( corrected for host galaxy contamination ) and @xmath61 is the emissivity - weighted radius of the blr ( kaspi et al . 2000 ; bentz et al . 2013 ) . we refer to this type of correlation as the normal @xmath3 relationship . the constants @xmath62 and @xmath63 differ slightly from one study to the next , depending on the number of sources and their exact luminosity range ( e.g. , kilerci eser et al . 2015 ) . for sub - eddington accreting agns , @xmath64 ltd and @xmath65 , but for seambhs they are different ( see paper iv ) . as reported in paper iv , some objects from the seambh2012 and seambh2013 samples have much shorter h@xmath1 lags compared with objects with similar luminosity , and the @xmath3 relation has a large scatter if they are included . in particular , the reduction of the lags increases with the dimensionless accretion rate , defined as @xmath66 , where @xmath67 is the accretion rate , @xmath68 is the eddington luminosity and @xmath69 is the speed of light . furthermore , it has been found , so far in the present campaigns , that seambhs have a range of accretion rates from a few to @xmath70 . this kind of shortened h@xmath1 lags was discovered in the current seambh project ( a comparison with previous campaigns is given in section 6.5 ) . such high accretion rates are characteristic of the regime of slim accretion disks ( abramowicz et al . 1988 ; szuszkiewicz et al . 1996 ; wang & zhou 1999 ; wang et al . 1999 ; mineshige et al . 2000 ; wang & netzer 2003 ; sadowski 2009 ) . these interesting properties needed to be confirmed with observations . we aim to explore whether we can define a new scaling relation , @xmath71 , which links the size of the blr to both the agn luminosity _ and _ accretion rate . we report new results from seambh2014 . we describe target selection , observation details and data reduction in 2 . h@xmath1 lags , bh mass and accretion rates are provided in 3 . properties of h@xmath1 lags are discussed in 4 , and a new scaling relation of h@xmath1 lags is established in 5 . section 6 introduces the fundamental plane , which is used to estimate accretion rates from single - epoch spectra , for application of the new size - luminosity scaling relation of the blr . brief discussions of the shortened lags are presented in 7 . we draw conclusions in 8 . throughout this work we assume a standard @xmath72cdm cosmology with @xmath73 , @xmath74 and @xmath75 ( ade et al . 2014 ) . rlcrllcrlcrll 29.374 & @xmath76 & & 76.365 & @xmath77 & @xmath78 & & 60.324 & @xmath79 & & 112.294 & @xmath80 & @xmath81 + 30.360 & @xmath82 & & 80.319 & @xmath83 & @xmath84 & & 62.314 & @xmath85 & & 116.394 & @xmath86 & @xmath87 + 32.352 & @xmath88 & & 83.314 & @xmath89 & @xmath90 & & 63.299 & @xmath91 & & 119.336 & @xmath92 & @xmath93 + 33.339 & @xmath94 & & 86.422 & @xmath95 & @xmath96 & & 68.397 & @xmath97 & & 135.324 & @xmath98 & @xmath99 + 34.331 & @xmath100 & & 89.378 & @xmath101 & @xmath102 & & 77.318 & @xmath103 & & 139.319 & @xmath104 & @xmath105 + we followed the procedures for selecting seambh candidates described in paper iv . we used the fitting procedures to measure h@xmath1 profile and 5100 luminosity of sdss quasar spectra described by hu et al . ( 2008a , b ) . following the standard assumption that the blr gas is virialized , we estimate the bh mass as @xmath106 where @xmath61 , @xmath2 is the h@xmath1 lag measured in the rest frame , @xmath107days , @xmath108 is the gravitational constant , and @xmath109 is the full - width - half - maximum ( fwhm ) of the h@xmath1 line profile in units of @xmath110 . we take the virial factor @xmath111 in our series of papers ( see some discussions in paper iv ) . in order to select agns with high accretion rates , we employed the formulation of accretion rates derived from the standard disk model of shakura & sunyaev ( 1973 ) . in the standard model it is assumed that the disk gas is rotating with keplerian angular momentum , and thermal equilibrium is localized between viscous dissipation and blackbody cooling . observationally , this model is supported from fits of the so - called big blue bump in quasars ( czerny & elvis 1987 ; wandel & petrosian 1988 ; sun & malkan 1989 ; laor & netzer 1989 ; collin et al . 2002 ; brocksopp et al . 2006 ; kishimoto et al . 2008 ; davis & laor 2011 ; capellupo et al . the dimensionless accretion rate is given by @xmath112 where @xmath113 ( see papers ii and iv ) and @xmath114 is the inclination angle to the line of sight of the disk . we take @xmath115 , which represents a mean disk inclination for a type 1 agns with a torus covering factor of about 0.5 ( it is assumed that the torus axis is co - aligned with the disk axis ) . previous studies estimate @xmath116 [ e.g. , fischer et al . ( 2014 ) find a inclination range of @xmath117 , whereas pancoast et al . ( 2014 ) quote @xmath118 ; see also supplementary materials in shen & ho ( 2014 ) ] , which results in @xmath119 from equation ( 3 ) . this uncertainty is significantly smaller than the average uncertainty on @xmath120 ( @xmath121 dex ) in the present paper , and is thus ignored . equation ( 3 ) applies to agns that have accretion rates @xmath122 , namely excluding the regimes of advection - dominated accretion flows ( adaf ; narayan & yi 1994 ) and of flows with hyperaccretion rates ( @xmath123 ; see appendix a for the validity of equation 3 for seambhs ) . using the normal @xmath3 relation ( bentz et al . 2013 ) , we fitted all the quasar spectra in sdss data release 7 by the procedures in hu et al . ( 2008a , b ) and applied equations ( 2 ) and ( 3 ) to select high@xmath124 targets . we ranked quasars in terms of @xmath120 and chose ones as candidates with the highest @xmath120 . we found that the high@xmath124 quasars are characterized by 1 ) strong optical lines ; 2 ) relatively narrow h@xmath1 lines ( @xmath125 ) ; 3 ) weak lines ; and 4 ) steep 210 kev spectra ( wang et al . these properties are similar to those of nls1s ( osterbrock & pogge 1987 ; boroson & green 1992 ) , but most of the candidates have more extreme accretion rates ( a detailed comparison of seambh properties with normal quasars will be carried out in a separate paper ) . considering that the lags of all targets should be measured within one observing season , and taking into consideration the limitations of the weather of the lijiang station of yunnan observatory ( periods between june and september are raining seasons there ) , we only chose objects with maximum estimated lags of about 100 days or so ( the monitoring periods should be at least a few times the presumed lags ) . also , to ensure adequate signal - to - noise ratio ( s / n ) for measurements of light curves , we restricted the targets to a redshift range of @xmath126 and magnitudes @xmath127 . the fraction of radio - loud objects with @xmath128 is not high . we discarded radio - loud objects based on available first observations , in order to avoid h@xmath1 reverberations potentially affected by nonthermal emission from relativistic jets , or optical continuum emission strongly contaminated by jets . we chose about 20 targets for photometry monitoring , which served as a preselection to trigger follow - up spectroscopic monitoring . the photometric monitoring yielded 10 targets with significant variations ( @xmath129 magnitudes ) , and time lags were successfully measured for 5 objects ( table 1 ) . for an overview of our entire ongoing campaign , table 1 also lists samples from seambh2012 and seambh2013 . to summarize : we have selected about 30 targets for spectroscopic monitoring during the last three years ( 20122014 ) . the successful rate of the monitoring project is about 2/3 . our failure to detect a lag for the remaining 1/3 of the sample are either due to low - amplitude variability or bad weather that leads to poor monitoring cadence . in particular , the seambh2014 observations were seriously affected by the el nio phenomenon . rlcrllcrlcrll[!h ] 29.417 & @xmath130 & & 75.402 & @xmath131 & @xmath132 & & 30.428 & @xmath133 & & 90.382 & @xmath134 & @xmath135 + 30.398 & @xmath136 & & 80.395 & @xmath137 & @xmath138 & & 33.403 & @xmath139 & & 97.433 & @xmath140 & @xmath141 + 33.376 & @xmath142 & & 84.365 & @xmath143 & @xmath144 & & 36.410 & @xmath145 & & 104.293 & @xmath146 & @xmath147 + 36.368 & @xmath148 & & 91.326 & @xmath149 & @xmath150 & & 48.357 & @xmath151 & & 111.261 & @xmath152 & @xmath153 + 38.411 & @xmath154 & & 104.409 & @xmath155 & @xmath156 & & 51.353 & @xmath157 & & 116.448 & @xmath158 & @xmath159 + the seambh project uses the lijiang 2.4 m telescope , which has an alt - azimuth ritchey - chrtien mount with a field de - rotator that enables two objects to be positioned along the same long slit . it is located in lijiang and is operated by yunnan observatories . we adopted the same observational procedures described in detail in paper i , which also introduces the telescope and spectrograph . we employed the yunnan faint object spectrograph and camera ( yfosc ) , which has a back - illuminated 2048@xmath1604608 pixel ccd covering a field of @xmath161 . during the spectroscopic observation , we put the target and a nearby comparison star into the slit simultaneously , which can provide high - precision flux calibration . as in seambh2013 ( paper iv ) , we adopted a @xmath162-wide slit to minimize the influence of atmospheric differential refraction , and used grism 3 with a spectral resolution of 2.9 / pixel and wavelength coverage of 38009000 . to check the accuracy of spectroscopic calibration , we performed differential photometry of the targets using some other stars in the same field . we used an sdss @xmath163-band filter for photometry to avoid the potential contamination by emission lines such as h@xmath1 and h@xmath164 . photometric and spectroscopic exposure times are typically 10 and 60 min , respectively . the reduction of the photometry data was done in a standard way using iraf routines . photometric light curves were produced by comparing the instrumental magnitudes to those of standard stars in the field ( see , e.g. , netzer et al . 1996 , for details ) . the radius for the aperture photometry is typically @xmath165 ( seeing @xmath166 ) , and background is determined from an annulus with radius @xmath167 to @xmath168 . the uncertainties on the photometric measurements include the fluctuations due to photon statistics and the scatter in the measurement of the stars used . the spectroscopic data were also reduced with iraf . the extraction width is fixed to @xmath167 , and the sky regions are set to @xmath169 on both sides of the extracted region . the average s / n of the 5100 continuum of individual spectra are from @xmath17016 to @xmath17022 , except for j085946 , which only has s / n @xmath17112 . the flux of spectroscopic data was calibrated by simultaneously observing a nearby comparsion star along the slit ( see paper i ) . the fiducial spectra of the comparison stars are generated using observations from several nights with the best weather conditions . the absolute fluxes of the fiducial spectra are calibrated using additional spectrophotometric standard stars observed in those nights . then , the in - slit comparison stars are used as standards to calibrate the spectra of targets observed in each night . the sensitivity as a function of wavelength is produced by comparing the observed spectrum of the comparison star to its fiducial spectrum . finally , the sensitivity function is applied to calibrate the observed agn spectrum10% . we multiply the fiducial spectra of the in - slit stars with the bandpass of the sdss r@xmath172 filter and compare their synthesized magnitudes with the magnitudes found in the sdss database . the maximum difference is @xmath17310% . ] . the procedures adopted here resemble the method used by maoz et al . ( 1990 ) and kaspi et al . ( 2000 ) . in order to illustrate the invariance of the comparison stars , we show the light curves from the differential photometry of the comparison stars in appendix b. it is clear that their fluxes are very stable and the variations are less than @xmath1701% . the calibration method of van groningen & wanders ( 1992 ) , based on the emission line and popularily used in many rm campaigns ( e.g. , peterson et al . 1998 ; bentz et al . 2009 ; denney et al . 2010 ; grier et al . 2012 ; barth et al . 2015 ) , is not suitable for seambhs . @xmath174 tends to be weak in seambhs ( especially for the objects in seambh2013 - 2014 ) , and , even worse , is blended with strong around 5016 . applying the calibration method to seambhs results in large statistical ( caused by the weakness of ) and systematic ( caused by the variability of ; see paper iii ) uncertainties . the method based on in - slit comparison stars , used in our campaign , does not rely on and provides accurate flux calibration for the spectra of seambhs . for comparison , in paper i we measured the fluxes in the calibrated spectra of three objects in seambh2012 with relatively strong ; the variation of their flux is on the order of @xmath1703% . this clearly demonstrates the robustness of our flux calibration method based on in - slit comparison stars . the procedures to measure the 5100 and h@xmath1 flux are nearly the same as those given in paper i. the continuum beneath h@xmath1 line is determined by interpolation of two nearby bands ( 47404790 and 5075 -5125 ) in the rest frame . these two bands have minimal contamination from emission lines . the flux of h@xmath1 is measured by integrating the band between 4810 and 4910 after subtraction of the continuum ; the h@xmath1 band is chosen to avoid the influence from lines . the 5100 flux is taken to be the median over the region 5075 - 5125 . detailed information of the observations is provided in table 1 . all the photometry and continuum and h@xmath1 light curves for the five objects with successfully detected lags are listed in tables @xmath175 and shown in figure 1 . we also calculated the mean and rms ( root mean square ) spectra and present them in appendix c. + like the seambh2013 sample , we have no observations that can clearly separate the host galaxies of the agns in seambh2014 . shen et al . ( 2011 ) propose the following empirical relation to estimate the fractional contribution of the host galaxy to the optical continuum emission : @xmath176 , for @xmath177 , where @xmath178 and @xmath179 is the total emission from the agn and its host at 5100 . for @xmath180 , @xmath181 , and the host contamination can be neglected . the host fractions at 5100 for the objects ( j075949 , j080131 , j084533 , j085946 and j102339 ) are ( 27.5% , 37.1% , 14.2% , 19.0% and 31.8% ) . the values of @xmath182 listed in table 5 are the host - subtracted luminosities . we note that this empirical relation is based on sdss spectroscopic observations with a @xmath183 fiber , whereas we used a @xmath184-wide slit . it should apply to our observations reasonably well ( see paper iv for additional discussions on this issue ) . we will revisit this issue in the future using high - resolution images that can more reliably separate the host . as in papers i iv , we used cross - correlation analysis to determine h@xmath1 lags relative to photometric or 5100 continuum light curves . we use the centroid lag for h@xmath1 . the uncertainties on the lags are determined through the flux randomization / random subset sampling " method ( rs / rss ; peterson et al . 1998 , 2004 ) . the cross - correlation centroid distribution ( cccd , described in appendix e ) and cross - correlation peak distribution ( ccpd ) generated by the fr / rss method ( maoz & netzer 1989 ; peterson et al . 1998 , 2004 ; denney et al . 2006 , 2010 , and references therein ) are shown in figure 1 . we used the following criteria to define a successful detection of h@xmath1 lag : 1 ) non - zero lag from the ccf peak and 2 ) a maximum correlation coefficient larger than 0.5 . data for the light curves of the targets are given in tables 24 . all the measurements of the seambh2014 sample are provided in table 5 . the @xmath163-band light curves are generally consistent with the 5100 continuum light curve , but the former usually have small scatter , as shown in figure 1 . we calculated ccfs for the h@xmath1 light curves with both @xmath163-band photometry and with 5100 spectral continuum for all objects . the quality of the h@xmath185 ccfs is usually better than the h@xmath186 ccfs . we show the h@xmath185 ccfs for all objects in figure 1 , except for j080131 . we use the h@xmath185 lags in the following analysis . for j080131 , the @xmath163-band light curve between 200 and 220 days does not match the 5100 continuum light curve , even though h@xmath1 does follow 5100 continuum tightly . notes to individual sources are given in appendix d. there are two ways of calculating bh mass , base either on the rms spectrum ( e.g. , peterson et al . 2004 ; bentz et al . 2009 ; denney et al . 2010 ; grier et al . 2012 ) or on the mean spectrum ( e.g. , kaspi et al . 2005 ; papers i iv ) . different studies also adopt different measures of the line width , typically either the line dispersion @xmath187 ( second moment of the line profile ) or the fwhm . in this study , we choose to parameterize the line width using fwhm , as measured in the mean spectra . the narrow h@xmath1 component may influence the measurement of fwhm . we adopt the same procedure as in paper i to remove the narrow h@xmath1 . we fix narrow h@xmath1/@xmath174 to 0.1 , and measure fwhm from the mean spectra with narrow h@xmath1 subtracted . then we set h@xmath1/@xmath174 to 0 and 0.2 and repeat the process to obtain lower and upper limits to fwhm . the relatively wide slit employed in our campaign ( @xmath184 ) significantly broadens the emission lines by @xmath188 , where @xmath189 is the instrumental broadening that can be estimated from the broadening of selected comparison stars . as in paper iv , we obtain the intrinsic width of the mean spectra from @xmath190 . the fwhm simply obtained here is accurate enough for bh mass estimation . our procedure for bh mass estimation is based on fwhm measured from the mean spectrum ( see explanation in papers i and ii ) . as shown recently in woo et al . ( 2015 ) , the scatter in the scaling parameter ( @xmath191 ) derived in this method is very similar to the scatter in the method based on the rms spectrum . we use equations ( 2 ) and ( 3 ) to calculate accretion rates and bh masses for the five sources listed in table 5 . for convenience and completeness , table 5 also lists h@xmath1 lags , bh masses and accretion rates for the sources from seambh2012 and seambh2013 . our campaign has successfully detected h@xmath1 lags for 18 seambhs since october 2012 . figure 1 _ continued . _ as described in paper ii , there are some theoretical uncertainties in identifying a critical value of @xmath120 to define a seambh ( laor & netzer 1989 ; beloborodov 1998 ; sadowski et al . 2011 ) . following paper ii , we classified seambhs as those objects with @xmath192 . this is based on the idea that beyond this value , the accretion disk becomes slim and the radiation efficiency is reduced mainly due to photon trapping ( sadowski et al . since we currently can not observe the entire spectral energy distribution , we have no direct way to measure @xmath193 , and this criterion is used as an approximate tool to identify seambh candidates . to be on the conservative side , we chose the lowest possible efficiency , @xmath194 ( retrograde disk with @xmath195 ; see bardeen et al . thus , seambhs are objects with @xmath196 . for simplicity , in this paper we use @xmath197 as the required minimum ( papers ii and iv ) . we refer to agns with @xmath198 as seambhs and those with @xmath199 as sub - eddington ones . paper iv clearly shows that the properties of the @xmath3 relation for @xmath198 and @xmath199 are significantly different . rlcrll[!b ] 54.400 & @xmath200 & & 102.431 & @xmath201 & @xmath202 + 56.431 & @xmath203 & & 105.309 & @xmath204 & @xmath205 + 62.333 & @xmath206 & & 111.450 & @xmath207 & @xmath208 + 72.412 & @xmath209 & & 115.406 & @xmath210 & @xmath211 + 76.438 & @xmath212 & & 119.396 & @xmath213 & @xmath214 + figure 2 plots distributions of @xmath182 , ew(h@xmath1 ) , @xmath120 and @xmath215 of all the mapped agns ( 41 from bentz et al . 2013 and the 18 seambhs from our campaign ; see table 7 in paper iv and table 5 here ) . as shown clearly in the diagrams , seambh targets are generally more luminous by a factor of 23 compared to previous rm agns ( figure 2_a _ ) . the bh masses of seambhs are generally less smaller by a factor of 10 compared to previous samples , whereas , as a consequence of our selection , the accretion rates of seambhs are higher by 23 orders of magnitude ( figures 2_c _ and 2_d _ ) . however , ew(h@xmath1 ) of seambhs are not significantly smaller ( figure 2_d _ ) . on average , the high@xmath124 sources have lower mean ew(h@xmath1 ) , consistent with the inverse correlation between ew(h@xmath1 ) and @xmath216 ( e.g. , netzer et al . 2004 ) . lccccccrccc + mrk 335 & @xmath217 & @xmath218 & @xmath219 & @xmath220 & @xmath221 & @xmath222 & @xmath223 & @xmath224 + mrk 1044 & @xmath225 & @xmath226 & @xmath227 & @xmath228 & @xmath229 & @xmath230 & @xmath231 & @xmath232 + mrk 382 & @xmath233 & @xmath234 & @xmath235 & @xmath236 & @xmath237 & @xmath238 & @xmath239 & @xmath240 + mrk 142 & @xmath241 & @xmath242 & @xmath243 & @xmath244 & @xmath245 & @xmath246 & @xmath247 & @xmath248 + iras f12397 & @xmath249 & @xmath250 & @xmath251 & @xmath252 & @xmath253 & @xmath254 & @xmath255 & @xmath256 + mrk 486 & @xmath257 & @xmath258 & @xmath259 & @xmath260 & @xmath261 & @xmath262 & @xmath263 & @xmath264 + mrk 493 & @xmath265 & @xmath266 & @xmath267 & @xmath268 & @xmath269 & @xmath270 & @xmath271 & @xmath272 + iras 04416 & @xmath273 & @xmath274 & @xmath275 & @xmath276 & @xmath277 & @xmath278 & @xmath279 & @xmath280 + + sdss j075101 & @xmath281 & @xmath282 & @xmath283 & @xmath284 & @xmath285 & @xmath286 & @xmath287 & @xmath288 + sdss j080101 & @xmath289 & @xmath290 & @xmath291 & @xmath292 & @xmath293 & @xmath294 & @xmath295 & @xmath296 + sdss j080131 & @xmath297 & @xmath298 & @xmath299 & @xmath300 & @xmath301 & @xmath302 & @xmath303 & @xmath304 + sdss j081441 & @xmath305 & @xmath306 & @xmath307 & @xmath308 & @xmath309 & @xmath310 & @xmath311 & @xmath312 + sdss j081456 & @xmath313 & @xmath314 & @xmath315 & @xmath316 & @xmath317 & @xmath318 & @xmath319 & @xmath320 + sdss j093922 & @xmath321 & @xmath322 & @xmath323 & @xmath324 & @xmath325 & @xmath326 & @xmath327 & @xmath328 + + sdss j075949 & @xmath329 & @xmath330 & @xmath331 & @xmath332 & @xmath333 & @xmath334 & @xmath335 & @xmath336 + sdss j080131 & @xmath337 & @xmath338 & @xmath339 & @xmath340 & @xmath341 & @xmath342 & @xmath343 & @xmath344 + sdss j084533 & @xmath345 & @xmath346 & @xmath347 & @xmath348 & @xmath349 & @xmath350 & @xmath351 & @xmath352 + sdss j085946 & @xmath353 & @xmath354 & @xmath355 & @xmath356 & @xmath357 & @xmath358 & @xmath279 & @xmath359 + sdss j102339 & @xmath360 & @xmath361 & @xmath362 & @xmath363 & @xmath364 & @xmath365 & @xmath366 & @xmath367 the @xmath368 correlation was originally presented by peterson ( 1993 ; his figure 10 , only nine objects ) . it was confirmed by kaspi et al . ( 2000 ) using a sample of 17 low - redshift quasars . bentz et al . ( 2013 ) refined the @xmath3 relation through subtraction of host contamination and found that its intrinsic scatter is only 0.13 dex . paper iv ( table 7 ) provides a complete list of previously mapped agns , based on bentz et al . ( 2013 ) ; we directly use these values lags are consistent , the fwhm of h@xmath1 is very different . we only retain the later observation in the analysis . ] . as in paper iv , for objects with multiple measurements of h@xmath1 lags , we obtain the bh mass from each campaign and then calculate the average bh mass . using the averaged bh mass , we apply it to get accretion rates of the bhs during each monitoring epoch , which are further averaged to obtain the mean accretion rates of those objects ( kaspi et al . 2005 ; bentz et al . 2013 ) . we call this the average scheme . " on the other hand , we may consider each individual measurement of a single object as different objects ( e.g. , bentz et al . 2013 ) . we called this the direct scheme . " although the two approaches are in principle different , we obtain very similar results ( see a comparison in paper iv ) . all correlations of two parameters shown in this paper are calculated with the fitexy method , using the version adopted by tremaine et al . ( 2002 ) , which allows for intrinsic scatter by increasing the uncertainties in small steps until @xmath369 reaches unity ( this is typical for many of our correlations ) . we also emplot the bces method ( akristas & bershady 1996 ) but prefer not to use its results because it is known to give unreliable results in samples containing outliers ( there are a few objects with quite large uncertainties of @xmath120 ) . as shown in paper iv , the h@xmath1 lags of the seambh2013 sample were found to significantly deviate from the normal @xmath3 relation , by a factor of a few . we plot the @xmath3 relation of all samples in figure 3 . for sub - eddington agns ( @xmath370 ) in the direct scheme , @xmath371 with an intrinsic scatter of @xmath372 ( see paper iv ) . using fitexy , we have @xmath373 ( 1.44\pm 0.03)+(0.49\pm 0.03)\log \ell_{44 } & ( { \rm for~ all~\mathdotm } ) , \end{array}\right . \end{split}\ ] ] with intrinsic scatters of @xmath374 . clearly , the intrinsic scatter of seambhs is much larger than the sample of sub - eddington agns . in the averaged scheme , we have @xmath375 for sub - eddington agns , with an intrinsic scatter of @xmath376 ( see paper iv ) , and @xmath377 ( 1.44\pm 0.03)+(0.49\pm 0.03)\log \ell_{44 } & ( { \rm for~ all~\mathdotm } ) , \end{array}\right . \end{split}\ ] ] with intrinsic scatters of @xmath378 . the slope of the correlation for the seambh sample is comparable to that of sub - eddington agns , but the normalization is significantly different . it is clear that the seambh sources increase the scatter considerably , especially over the limited luminosity range occupied by the new sources . as in paper iv , we also tested the correlation between h@xmath1 lag and h@xmath1 luminosity , namely , the @xmath379 relation . the scatter of the @xmath379 correlation is not smaller than that of the @xmath368 correlation , and we do not consider it further . to test the dependence of the blr size on accretion rate , we define a new parameter , @xmath381 , that specifies the deviation of individual objects from the @xmath3 relation of the subsample of @xmath382 sources ( i.e. , @xmath383 as given by equations 4b and 5b for @xmath199 agns in paper iv ) . the scatter of @xmath384 is calculated by @xmath385^{1/2}$ ] , where @xmath386 is the number of objects and @xmath387 is the averaged value . figure 3 provides @xmath388 plots for comparison . figure 4 shows @xmath384 versus @xmath120 , as well as @xmath384 distributions for the @xmath389 and @xmath199 subsamples in the direct ( panels a and b ) and averaged ( panels c and d ) schemes . a kolmogorov - smirnov ( ks ) test shows that the probability that the two subsamples are drawn from the same parent distributions is @xmath390 for the direct scheme and @xmath391 for the averaged scheme . this provides a strong indication that the main cause of deviation from the normal @xmath3 relation is the extreme accretion rate . thus , a single @xmath3 relation for all agns is a poor approximation for a more complex situation in which both the luminosity and the accretion rate determine @xmath392 . from the regression for @xmath198 agns , we obtain the dependence of the deviations of @xmath393 from the @xmath3 relation in figure 4 : @xmath394 ( 0.34\pm0.09)-(0.42\pm0.07)\log \mathdotm & ( { \rm averaged ~scheme } ) , \end{array}\right . \end{split}\ ] ] with @xmath395 , respectively . we have tested the above correlations also for @xmath199 . the fitexy regressions give slopes near 0 , with very large uncertainties : @xmath396 and @xmath397 for figure 4a and 4c , respectively , implying that @xmath388 does not correlate with @xmath120 for the @xmath199 group . all this confirms that @xmath120 is an additional parameter that controls the @xmath3 relation in agns with high accretion rates . we provide evidence h@xmath1 lags depend on luminosity _ and _ accretion rate . there are a total of 28 seambhs ( including those discovered in other studies ) . we now have an opportunity to define a new scaling relation for the blr , one that properly captures the behavior of sub - eddington and super - eddington agns . considering the dependence of @xmath398 ( equation 6 ) , a unified form of the new scaling law can take the form^{\delta_1}$ ] , which is continuous for the transition from sub- to super - eddington sources . the fitting also yields a very rapid transition at @xmath399 , with @xmath400 and @xmath401 ( the present sample is still dominated by sub - eddington agns , with a ratio of 35/63 ) . we prefer the form given by equation ( 7 ) . ] @xmath402,\ ] ] where @xmath403 is to be determined by data . equation ( 7 ) reduces to the normal @xmath3 relation for sub - eddington agns and to @xmath404 for agns with @xmath405 . there are four parameters to describe the new scaling relation , but only two ( @xmath403 and @xmath406 ) are new due to the inclusion of accretion rates ; the other two are mainly determined by sub - eddington agns . the critical value of @xmath403 , which is different from the criterion of seambhs , depends on the sample of seambhs . in order to determine the four parameters simultaneously , we define @xmath407 where @xmath408 is the error bar of @xmath409 . minimizing @xmath369 among all the mapped agns and employing a bootstrap method , we have @xmath410 this new empirical relation has a scatter of @xmath411 , smaller than the scatter ( 0.26 ) of the normal @xmath3 relation for all the mapped agns . the new scaling relation is plotted in figure 5 . equation ( 7 ) shows the dependence of the blr size on accretion rates , but it can not be directly applied to single - epoch spectra for bh mass without knowning @xmath120 . iteration of equation ( 7 ) does not converge . the reason is due to the fact that larger @xmath120 leads to smaller @xmath393 and higher @xmath120 , implying that the iteration from equation ( 1 ) does not converge . du et al . ( 2016b ) devised a new method to determine @xmath120 from single - epoch spectra . beginning with the seminal work of boroson & green ( 1992 ) , it has been well - known that @xmath412 , the flux ratio of broad optical to h@xmath1 , correlates strongly with eddington ratio ( sulentic et al . 2000 ; shen & ho 2014 ) . at the same time , the shape of broad h@xmath1 , as parameterized by @xmath413 , where @xmath414 is the line dispersion , also correlates with eddington ratio ( collin et al . 2006 ) . combining the two produces produces a strong bivariate correlation , which we call the fundamental plane of the blr , of the form @xmath415 where @xmath416 in order to explore the relation between blr size and accretion rate , we define a dimensionless radius for the blr , @xmath417 , where @xmath418 cm is the gravitational radius . as in paper iv , we insert equation ( 3 ) into @xmath419 to replace @xmath420 , to obtain @xmath421 . this relation implies that @xmath419 increases with accretion rates as @xmath422 for sub - eddington agns , whereas in seambhs @xmath423 ( as shown in figure 6_a _ ) and @xmath419 and tends toward a maximum saturated value of @xmath424 , where @xmath425 is the minimum velocity width of h@xmath1 ( see equation 15 in paper iv ) . we note that the minimum observed fwhm values of h@xmath164 ( which is comparable to h@xmath1 ) is @xmath426 among low - mass agns ( greene & ho 2007 ; ho & kim 2016 ) . indeed , this limit is consistent with the saturation trend of @xmath419 ( figure 6_a _ ) . we note that the relatively large scatter in figure 6_a _ is mostly due to the uncertainties in bh mass . in order to better understand the relation between the blr and the central engine , we define , as in paper iv , the parameter @xmath427 , which reduces to @xmath428 we would like to point out that equation ( 12 ) describes the coupled system of the blr and the accretion disks . it is therefore expected that @xmath429 is a synthetic parameter describing the photoionization process including ionizing sources . it is easy to observationally test equation ( 12 ) using rm results . figure 6_b _ plots @xmath429 versus @xmath120 . it is very clear that the observed data for objects with @xmath199 agree well with equation ( 12 ) . furthermore , there is a clear saturation of @xmath429 for objects with @xmath389 objects . all these results strengthen the conclusions drawn in paper iv . as in that work , we define an empirical relation @xmath430,\ ] ] where @xmath431 in fact , we can get @xmath429 by inserting equation ( 3 ) into ( 7 ) , and find that it is in agreement with equation ( 13 ) . from the saturated@xmath432 , we have the maximum value of @xmath433 this result provides a strong constraint on theoretical models of super - eddington accretion onto bhs . the shortened h@xmath1 lags is the strongest distinguishing characteristic so far identified between super- and sub - eddington agns . two factors may lead to shortened lags for seambhs . first , wang et al . ( 2014c ) showed that , in the shakura - sunyaev regime , retrograde accretion onto a bh can lead to shorter h@xmath1 lags . the reason is due to the suppression of ionizing photons in retrograde accretion compared with prograde accretion . the second factor stems from self - shadowing effects of the inner part of slim disks ( e.g. , li et al . 2010 ) , which efficiently lower the ionizing flux received by the blr ( wang et al . when @xmath120 increases , the ratio of the disk height to disk radius increases due to radiation pressure ; the radiation field becomes anisotropic ( much stronger than the factor of @xmath434 ) due to the optically thick funnel of the inner part of the slim disk . in principle , the radiation from a slim disk saturates ( @xmath435 ) , and the total ionizing luminosity slightly increases with accretion rate , but the self - shadowing effects efficiently suppress the ionizing flux to the blr clouds . for face - on disks of type 1 agns , observers receive the intrinsic luminosity . if the ionization parameter is constant , the ionization front will significantly shrink , and hence the h@xmath1 lag is shortened in seambhs compared with sub - eddington agns of the same luminosity . the shortened h@xmath1 lag observed in seambhs can not be caused by retrograde accretion . however , the strong dependence on accretion rate of the deviation from the standard lag - luminosity relation implies that the properties of the ionizing sources are somehow different from those in sub - eddington agns . according to the standard photoionization theory , the observed @xmath436 relation can be explained if @xmath437 and @xmath438 is constant , where @xmath439 , @xmath440 is the ionizing luminosity , @xmath441 is gas density of blr clouds , and @xmath442 is the average energy of the ionizing photons ( bentz et al . 2013 ) . the relation @xmath437 holds for sub - eddington agns , and the constancy of @xmath443 is determined by the clouds themselves . @xmath443 is not expected to vary greatly as a function of eddington ratio . therefore , @xmath444 where the factor @xmath445 describes the anisotropy of the ionizing radiation field , @xmath440 is the shadowed ionizing luminosity received by the blr clouds , and @xmath446 is the blr radius corresponding to @xmath447 , the unshadowed luminosity . based on the classical model of slim disks , wang et al . ( 2014c ) showed that , for a given accretion rate , @xmath448 strongly depends on the orientation of the clouds relative to the disk , and that it range from 1 to a few tens . therefore , the reduction of the h@xmath1 lag can , in principle , reach up to a factor of a few , even 10 , as observed . furthermore , the saturated-@xmath429 implies that the ionizing luminosity received by the blr clouds gets saturated . the theory of super - eddington accretion onto bhs is still controversial . although extensive comparison with models is beyond the scope of this paper , we briefly discuss the implications of the current observations to the theory . two analytical models , which reach diametrically extreme opposite conclusions , have been proposed . abramowicz et al . ( 1988 ) suggested a model characterized by fast radial motion with sub - keplerian rotation and strong photon - trapping . both the shortened lags and saturated@xmath432 may be caused by self - shadowing effects and saturated radiation from a slim disk . both features are expected from the abramowicz et al . model ( wang et al . on the other hand , photon - bubble instabilities may govern the disk structure and lead to very high radiative efficiency ( gammie 1998 ) . if super - eddington accretion can radiate as much as @xmath449 ( equation 14 in begelman 2002 ) , the disk remains geometrically thin . in such an extreme situation , self - shadowing effects are minimal , h@xmath1 lags should not be reduced , and @xmath450saturation disappears . lt - d , @xmath10 , @xmath11 and @xmath12 . the scatter of the blr size is greatly reduced to @xmath451 . , scaledwidth=48.0% ] recent numerical simulations that incorporate outflows ( e.g. , jiang et al . 2014 ) and relativistic jets ( sadowski et al . 2015 ) also suggest that super - eddington accretion flows can maintain a high radiative efficiency . however , most agns with high accretion rates are radio - quiet ( ho 2002 ; greene & ho 2006 ) , in apparent contradiction with the numerical simulation predictions . furthermore , evidence for @xmath429-saturation also does not support the models with high radiative efficiency . recent modifications of the classical slim disk model that include photo - trapping appear promising ( e.g. , cao & gu 2015 ; sadowski et al . 2014 ) , but the situation is far from settled . whatever the outcome , the results from our observations provide crucial empirical constraints on the models . if the blr is flattened , its inclination angle to the observer will influence @xmath215 , and hence @xmath120 ( see equation 3 ) . to zero - order approximation , the observed width of the broad emission lines follow @xmath452^{1/2}v_{\rm k},\ ] ] where @xmath453 is the keplerian velocity and @xmath454 is the height of the flatten blr ( e.g. , collin et al . 2006 ) . for a geometrically thin blr , @xmath455 , @xmath456 , and hence @xmath457 , which is extremely sensitive to the inclination can be severely overestimated for low inclinations . on the other hand , many arguments ( e.g. , goad & korista 2014 ) support @xmath458 , and the inclination angle significantly influences @xmath215 only for @xmath459 . currently , the values of @xmath460 are difficult to estimate , but detailed modelling of rm data suggests @xmath461 ( li et al . 2013 ; pancoast et al . 2014 ) . if true , this implies that the blr is not very flattened , and hence the inclination angle only has a minimal influence on @xmath215 and @xmath120 . the objects in our seambh sample are very similar to nls1s . as previous rm agn samples include nls1s , why have previous studies not noticed that nls1s deviate from the @xmath3 relation ( e.g. , figure 2 in bentz 2011 ) ? we believe that the reason is two - fold . first , the number of nls1s included in previous rm campaigns was quite limited ( denney et al . 2009 , 2010 ; bentz et al . 2008 , 2009 ; see summary in bentz 2011 ) . the level of optical variability in nls1s is generally very low ( klimek et al . 2004 ) , and many previous attempts at rm have proved to be unsuccessful ( e.g. , giannuzzo & stirpe 1996 ; giannuzzo et al . second , not all nls1s are necessarily highly accreting . our seambh sample was selected to have high accretion rates ( see @xmath120 listed in table 7 of paper iv ) , generally higher than that of typical nls1s previously studied successfully through rm . as discussed in wang et al . ( 2014b ) and in paper iv , high accretion rates lead to anisotropic ionizing radiation , which may explain the shortened blr lags . once its discovery , quasars as the brightest celestial objects in the universe had been suggested for cosmology ( sandage 1965 ; hoyle & burbidge 1966 ; longair & scheuer 1967 ; schmidt 1968 ; bahcall & hills 1973 ; burbidge & odell 1973 ; baldwin et al . unfortunately , the diversity of observed quasars made these early attempts elusive . after five decades since its discovery , quasars are much well understood : accretion onto supermassive black holes is powering the giant radiation , in particular , the bh mass can be reliably measured . quasars as the most powerful emitters renewed interests for cosmology in several independent ways : 1 ) the normal @xmath3 relation ( horn et al . 2003 ; watson et al . 2011 ; czerny et al . 2013 ) ; 2 ) the linear relation between bh mass and luminosity in super - eddington quasars ( wang et al . 2013 ; paper - ii ) ; 3 ) eddington agns selected by eigenvector 1 ( marziani & sulentic 2014 ) ; 4 ) x - ray variabilities ( la franca et al . 2014 ) and 5 ) @xmath462 relation ( risaliti & lusso 2015 ) . these parallel methods will be justified for cosmology by their feasibility of experiment periods and measurement accuracy . the strength of seambhs makes its application more convenient for cosmology . selection of seambhs only depends on single epoch spectra through the fundamental plane ( equation 10 ) . bh mass can be estimated by the new scaling relation ( equation 7 ) . we will apply the scheme outlined by wang et al . ( 2014a ) to the sample of selected seambhs for cosmology in a statistical way ( in preparation ) . on the other hand , the shortened h@xmath1 lags greatly reduce monitoring periods if seambhs are applied as standard candles , in particular , the reduction of lags govern by super - eddington accretion can cancel the cosmological dilltion factor of @xmath463 . otherwise , the monitoring periods of sub - eddington agns should be extended by the same factor of @xmath463 for measurements of h@xmath1 lags . such a campaign of using the normal @xmath3 relation for cosmology will last for a couple of years , even 10 years for bright high@xmath464 quasars . similarly to extension of the @xmath3 relation to - and -lines ( vestergaard & peterson 2006 ) , we can extend equation ( 7 ) to and lines for the scaling relations with luminosity as @xmath465 and @xmath466 , respectively , where @xmath467 and @xmath468 are sizes of the and regions , and @xmath469 is the 3000 luminosity . such extended relations conveniently allow us to investigate cosmology by making use of large samples of high@xmath464 quasars without time - consuming rm campaigns . it is urgent for us to make use of kinds of standard candles to test the growing evidence for dynamical dark energy ( e.g. , zhao et al . 2012 ; ade et al . we present the results of the third year of reverberation mapping of super - eddington accreting massive black holes ( seambhs ) . h@xmath1 lags of five new seambhs have been detected . similar to the seambh2012 and seambh2013 samples , we find that the seambh2014 objects generally have shorter h@xmath1 lags than the normal @xmath3 relation , by a factor of a few . in total , we have detected h@xmath1 lags for 18 seambhs from this project , which have accretion rates from @xmath470 to @xmath471 . the entire seambh sample allows us to establish a new scaling relation for the blr size , which depends not only on luminosity but also on accretion rate . the new relation , applicable over a wide range of accretion rates from @xmath472 to @xmath14 , is given by @xmath7 $ ] , where @xmath8 is 5100 continuum luminosity , and coefficients of @xmath9lt - d , @xmath10 , @xmath11 and @xmath12 . . the validity of equation ( 3 ) can be justified for application to seambhs . solutions of slim disks are transonic and usually given only by numerical calculations ( abramowicz et al . when the accretion rate of the disk is high enough , the complicated structure of the disk reduces to a self - similar , analytical form ( wang & zhou 1999 ) . using the self - similar solutions ( wang & zhou 1999 ; wang et al . 1999 ) , we obtained the radius of disk region emitting optical ( 5100 ) photons , @xmath473 and the photon - trapping radius @xmath474 we used the blackbody relation @xmath475 , where @xmath476 is the boltzmann constant , @xmath477 is the effective temperature of the disk surface , @xmath478 is the planck constant , and @xmath479 cm is the schwartzschild radius . equation ( 3 ) holds provided @xmath480 , namely @xmath481 in this regime , optical radiation is not influenced by photon - trapping effects . we would also like to point out that the bh spin only affects emission from the innermost regions of the accretion disk rather than the regions emitting 5100 photons . in the present campaign , no seambh so far has been found to exceed this critical value . beyond this critical value of accretion rate , optical photons are trapped by the accretion flow . we call this the hyper - accretion regime . here the cited @xmath482 below equation ( 3 ) is not a strict value of the adaf threshold since it depends on several factors , such as viscosity and outer boundary conditions . there are a few mapped agns with @xmath483 ( ngc 4151 , ngc 5273 , 3c 390.3 and ngc 5548 ; see table 7 in paper iv ) , but we do not discuss them in this paper because they do not influence our conclusions . recently , reprocessing of x - rays ( e.g. , frank et al . 2002 ; cackett et al . 2007 ) has been found to play an important role in explaining the variability properties of ngc 5548 ( e.g. , fausnaugh et al . 2015 ) . the fraction of x - ray emission to the bolometric luminosity strongly anti - correlates with the eddington ratio ( see figure 1 in wang et al . this result is usually interpreted to mean that the hot corona becomes weaker with increasing accretion rate , as a result of more efficiently cooling of the corona by uv and optical photons from the cold disk . this suggests that agns with high accretion rates will have less reprocessed emission , such that equation ( 3 ) would be more robust in seambhs . in order to avoid selecting variable stars as comparison stars , we examined their variability . to test the invariance of the comparison stars used in our spectroscopic observation , we performed differential photometry by comparing them with other stars in the same field . we typically use six stars for the differential photometry . the light curves of the stars are shown in figure 7 . on average , the standard deviations in the light curves of the comparison stars are 1% . this guarantees that they can be used as standards for spectral calibration . figure 7 shows the light curves of the comparison stars for each seambh targets . the averaged and rms spectra of the seambh2014 sample are provided in this appendix . following the standard way , we calculated the averaged spectrum as @xmath484 and the rms spectrum as @xmath485^{1/2},\ ] ] where @xmath486 is the @xmath487th observed spectrum and @xmath386 is the total number of observed spectra . they are shown in figure 8 . we note that both the averaged and rms spectra are affected by the broadening effects of the @xmath184-slit on the observed profiles . using the richards - lucy iteration , we can correct the observed profiles ( averaged and rms ) for velocity - resolved mapping , which will be carried out in a separate paper ( du et al . we briefly remark on individual objects for which h@xmath1 lags have been successfully measured . we failed in getting lag measurements for the other five objects because either their flux variations are very small or the data sampling rate was inadequate . _ j075949 _ : the detected h@xmath1 lag arises from two major peaks in the light curves . _ j080131 _ : the monitoring observations during @xmath488 did not yield a well - determined h@xmath1 lag because of the lack of h@xmath1 response to the second continuum flare ( see paper iv ) . during @xmath488 , the first reverberation of h@xmath1 , which can be clearly seen during the first 70 days of the light curves , yields a very significant lag , as shown in the ccf with a rest - frame centroid lag of @xmath489 days ( with a very high coefficient of @xmath490 ) . we monitored this object again in this observing season ( figure 1 ) . we successfully measure @xmath491 days , consistent with last season s result . _ j084533 _ : its continuum slightly decreased before being monitored for @xmath492 days , and steadily increased until @xmath493 days and then decreased again . although the ccf has a very flat peak close to @xmath494 , monte carlo simulations show that h@xmath1 lag is around 20 days , which arises from the two peaks in the h@xmath1 and @xmath163-band light curves . _ j085946 _ : the ccf peaks near 0.6 , which results from the two major dips in the h@xmath1 and @xmath163-band light curves . there are two peaks with roughly the same correlation coefficients around @xmath17020 days and @xmath17070 days in the observed frame , respectively . considering the relatively poor data quality of this object , it is difficult to distinguish which is the true response . the centroid lag represents the average of these two peaks ( responses ) , and its uncertainties cover the distribution ( figure 1 ) obtained in fr / rss method . so , we use it in the analysis of main text . _ j102339 _ : the detected lag is from the dip feature around @xmath495 days in light curves . for multiple - peaked ccfs with similar correlation coefficients , it is ambiguous as to which peak should be used to calculate the final lag . in such cases , we use the cccd to determine the lag . however , there are two approaches to calculate the centroid time lag in the ccf , as illustrated in figure 9 . in approach 1 , the cccd tends to be smoother than the ccpd ( see figure 1 ) , whereas in in approach 2 the cccd and ccpd always have a similar distribution . we adopt approach 1 in our analysis . if the ccf has two or even three peaks , the two approaches give different centroid lags . however , when the quality of the data is high and the ccf is unimodal , the two approaches yield the same results . it should be pointed out that approach 2 is often employed in the literature .

we consider gradient fields whose law takes the gibbs boltzmann form , where the sum runs over nearest neighbors . we assume that the potential admits the representation,\ ] ] where is a positive measure with compact support in . hence , the potential is symmetric , but nonconvex in general . while for strictly convex s , the translation - invariant , ergodic gradient gibbs measures are completely characterized by their tilt , a nonconvex potential as above may lead to several ergodic gradient gibbs measures with zero tilt . still , every ergodic , zero - tilt gradient gibbs measure for the potential above scales to a gaussian free field . and .

0704.3086

gradient fields belong to a class of models that arise in equilibrium statistical mechanics , for example , as approximations of critical systems and as effective interface models . although their definition is rather simple and , in fact , quite a lot is known ( see the reviews by funaki @xcite , velenik @xcite or sheffield @xcite ) , there is still much to be learned . in this note , we study gradient fields on a lattice . here , the field is a collection of real - valued random variables @xmath6 and the distribution of @xmath7 on @xmath8 is given by the formal expression @xmath9 where @xmath10 is the lebesgue measure , @xmath11 refers to an unordered nearest - neighbor pair on @xmath12 and @xmath2 is an even , measurable function , called the _ potential _ , which is bounded from below and grows superlinearly at @xmath13 . of course , to define the measure ( [ grad - field ] ) precisely , we have to restrict the above expression to a finite subset of @xmath12 and fix the @xmath7 s on its boundary ; @xmath14 is then the normalizing constant . another way to regularize the expression ( [ grad - field ] ) is to consider directly measures on all of @xmath8 whose conditional probabilities in finite sets take the above form . in our context , this standard definition is hampered by the fact that , due to the unbounded nature of the fields @xmath15 , no such infinite - volume measures may exist at all . however , if one restricts attention to ( the @xmath16-algebra generated by ) the gradient variables @xmath17 then infinite - volume measures exist under the above conditions on @xmath2 . since the measure depends only on gradients , we refer to such measures as _ gradient gibbs measures _ ( ggm ) , in accordance with funaki @xcite and sheffield @xcite . throughout , we will focus on translation - invariant ggms . an important characteristic is their _ tilt_. for a translation - invariant ggm @xmath18 , there exists a unique tilt vector @xmath19 such that @xmath20 for every edge @xmath21 of @xmath12 , which we regard as a vector in this formula . of course , this definition is only really meaningful for the ggms that are ergodic , that is , trivial on the @xmath22-algebra of events invariant under all lattice translations . indeed , in the ergodic case , @xmath23 represents the average incline of typical configurations . for the case of quadratic @xmath2 , the _ massless free field _ , the measure ( [ grad - field ] ) is gaussian and so many desired characteristics are amenable to explicit computations . the challenge for mathematicians has been to develop an equivalent level of understanding for nonquadratic @xmath2 s . a good amount of progress in this direction has been made in the last ten to fifteen years : brydges and yau @xcite ( and also earlier works , e.g. , by gawdzki and kupiainen @xcite and magnen and snor @xcite ) studied the effect of analytic perturbations of the quadratic potentials and concluded that the large - scale behavior is that of the massless free field . naddaf and spencer @xcite proved the same nonperturbatively for strictly convex potentials @xmath2 and ggms with zero tilt . the corresponding extension to nonzero tilt was obtained by giacomin , olla and spohn @xcite . for the same class of potentials , funaki and spohn @xcite proved a bijection between the ergodic ggms and their tilt . sheffield @xcite characterized translation - invariant ggms by means of a gibbs variational principle and extended funaki and spohn s results to fields taking only a discrete set of values . we refer to the reviews by funaki @xcite , velenik @xcite and sheffield @xcite for further results and references . as a unifying feature , all the ( nonperturbative ) results mentioned are based on the strict convexity of the potential @xmath2 , be it for the use of the brascamp lieb inequality @xcite , helffer sjstrand random walk representation @xcite , coupling to langevin dynamics @xcite and the cluster - swapping algorithm @xcite . one would naturally like to have a nonperturbative approach that works even for nonconvex potentials . with this motivation , biskup and koteck @xcite recently studied the ggms for nonconvex @xmath2 that are a log - mixture of centered gaussians , @xmath24 where @xmath4 is a positive measure with compact support in @xmath25 . surprisingly , already for the simplest nontrivial case , @xmath26 with @xmath27 , it was shown that , in @xmath28 , there is a value @xmath29 at which one can construct _ two _ distinct , translation - invariant , gradient gibbs measures of zero tilt . the relevant conclusion from @xcite for the general theory is that the one - to - one correspondence between ergodic ggms and their tilt breaks down once @xmath2 is sufficiently nonconvex . the next question which naturally arises is how to understand what happens to the scaling limit . the purpose of this note is to show that , regardless of the occurrence of phase transitions , for potentials of the form ( [ v - repr ] ) , every translation - invariant , ergodic ggm with zero tilt scales to a gaussian free field ( gff ) . the proof is based on the fact utilized already in @xcite that ( [ v - repr ] ) allows us to represent every ggm as a mixture over gaussian gradient measures with a random coupling constant @xmath30 for each edge @xmath11 . its covariance is simply the inverse of the operator @xmath31,\ ] ] where we take , once and for all , @xmath32 . fluctuations _ in the gaussian measure can be analyzed by invoking a random walk representation ; @xmath33 is the generator of a random walk with symmetric random jump rates , known , equivalently , as a _ random conductance model_. the name arises naturally from the electrostatic interpretation of this problem ( cf . doyle and snell @xcite , in which one views @xmath12 as a resistor network with _ conductance _ @xmath30or resistivity @xmath34assigned to an edge @xmath11 ) . as it turns out ( see lemma [ lemma - ergodic ] ) , if the initial ggm is ergodic , then so is the law of the conductances . this makes homogenization a possible tool . much work has been done in the past two decades on the problem of random walks with random conductances . for our purposes , it suffices to invoke two known results : kipnis and varadhan s @xcite invariance principle ( i.e. , scaling of the random walk to brownian motion ) and delmotte and deuschel s @xcite annealed derivative heat kernel bounds . ( note that in the helffer sjstrand random walk representation , as used in @xcite , one also has to study a random walk in a random environment . however , this random environment fluctuates in time , while , in our case , it is static . ) this takes care of the fluctuations of the field ; an important technical issue is thus the control of the mean . this is where the zero - tilt restriction comes in ( see lemma [ lemma - gaussian ] , corollary [ cor - tilt ] and discussion in section [ sec7 ] ) . while this paper has been in the process of revision , further developments have occurred in the study of gradient models with nonconvex potentials . cotar , deuschel and mller @xcite have shown that for nonconvex perturbations of potentials @xmath2 where the size of the nonconvex region is small compared to typical fluctuations of the field , the conclusions are as in the convex case . ( their precise condition is a bound on the @xmath35-norm of the negative part of the second derivative . ) this is a high - temperature result ; work in progress by adams , koteck and mller @xcite addresses the low - temperature case when nonconvexities are allowed only sufficiently far away from the absolute minimum of @xmath2 . our contribution remains valuable despite these advances as it applies to all potentials of the type ( [ v - repr ] ) , including those for which phase coexistence occurs . this paper is organized as follows . in section [ sec2 ] , we precisely define the concept of the gradient gibbs measure and state our main theorem . in section [ sec3 ] , we introduce the extended gradient gibbs measures and characterize their conditional marginals . this will naturally lead to the aforementioned connections with random walks in reversible random environments . to keep the main line of the argument intact , we first finish proving our main result in section [ sec4 ] and only then expound on the random walk connections in section [ sec5 ] . section [ sec7 ] is devoted to the discussion of the limitations to zero tilt and some open questions concerning gradient gibbs measures . as mentioned above , infinite - volume measures on the field variables @xmath36 may not always exist , particularly in sufficiently low dimensions . to make our statements uniform in dimension , we will focus attention on the gradient variables . however , not even that will be entirely straightforward because the gradient variables satisfy a host of `` hard - core '' constraints which , in a sense , encapsulate most of the interaction . since @xmath37 is gradient , one has @xmath38 whenever @xmath39 are the vertices of a cycle in @xmath12 of length four . we will often write @xmath40 for the _ positively oriented _ edge @xmath21 in @xmath12 . throughout , we will only work with positively oriented edges and will use @xmath41 to denote the set of such edges with both endpoints in the set @xmath42 . the constraints ( [ loop ] ) are implemented at the level of the a priori measure which is defined as follows . fix a configuration @xmath43 that obeys ( [ loop ] ) and , for @xmath42 finite , let @xmath44 be the lebesgue measure on the linear subspace of configurations @xmath45 such that @xmath46 for all @xmath47 and that @xmath48 obeys the constraints ( [ loop ] ) . note that if @xmath49 is a configuration such that @xmath50 for every nearest - neighbor pair @xmath11 , then @xmath51 is , to within a normalization constant , the projection to gradient variables of the lebesgue measure on @xmath52 subject to the boundary condition @xmath49 . next , we will give a precise definition of the notion of gradient gibbs measure . for a finite @xmath42 , consider the _ specification _ @xmath53 , which is a measure in the first coordinate and a function of the boundary condition in the second coordinate , defined by @xmath54\\[-8pt ] & & \qquad:=\frac1{z_\lambda(\eta_{{\mathbb b}(\lambda)^{{\mathrm{c } } } } ) } \exp\biggl\ { - \mathop{\sum_{\langle x , y\rangle}}_{x\in\lambda , y\in\lambda \cup\partial\lambda } v(\eta_{xy } ) \biggr\ } \nu_\lambda^{(\eta_{{\mathbb b}(\lambda)^{{\mathrm{c}}}})}\bigl({{d}}\eta_{{\mathbb b}(\lambda ) } \bigr).\nonumber\end{aligned}\ ] ] here , @xmath55 is the normalizing constant . let @xmath56 . we say that a measure @xmath18 on @xmath57 is a _ gradient gibbs measure _ if the regular conditional probability @xmath58 in any finite @xmath42 satisfies @xmath59 for @xmath18-a.e . @xmath37 . most of this paper is restricted to translation - invariant gradient gibbs measures . to define the required notation , for each @xmath60 , let @xmath61 be the `` translation by @xmath62 '' which acts on configurations @xmath37 by shifting the origin to position @xmath62 , @xmath63 we say that @xmath18 is _ translation - invariant _ if @xmath64 for all @xmath60 and that it is _ ergodic _ if @xmath65 for every event @xmath66 such that @xmath67 for all @xmath60 . as is usual for problems involving random fields , we will interpret samples from gradient gibbs measures as random linear functionals on an appropriate space of functions . let @xmath68 denote the set of all infinitely differentiable functions @xmath69 with compact support . given a configuration @xmath70 of gradients satisfying the conditions ( [ loop ] ) , we can find a configuration of the field @xmath71 such that ( [ eta - phi ] ) holds for every nearest - neighbor pair of sites . the configuration @xmath7 is determined uniquely once we fix the value at one site , for example , @xmath72 . for any function @xmath73 , we introduce the random linear functional @xmath74 which , under the condition @xmath75 does not depend on the choice of the special value @xmath72 . the functional @xmath76 can be naturally extended to a somewhat larger space , defined as follows . let @xmath77 denote the laplace differential operator in @xmath78 and consider the set @xmath79 note that each @xmath80 automatically obeys ( [ zero - sum ] ) . the set @xmath81 is endowed with a natural quadratic form @xmath82 , defined as @xmath83 we thus define the norm @xmath84^{1/2}\ ] ] and let @xmath85 be the completion of @xmath81 in this norm . note that @xmath85 corresponds to the case @xmath86 in the family of sobolev spaces @xmath87 . the condition that @xmath88 is natural once we realize that this quantity will represent the variance of the limiting gaussian field . the extension of @xmath7 to @xmath85 is implied by the following lemma . [ lemma - cont ] suppose that @xmath4 in ( [ v - repr ] ) has support bounded away from zero and let @xmath18 be a translation - invariant , ergodic , zero - tilt gradient gibbs measure for the potential @xmath2 . there then exists a constant @xmath89 such that for each @xmath80 , @xmath90 in particular , @xmath7 extends to a linear functional @xmath91 . note that ( [ 4.1 ] ) means that the map @xmath92 is continuous in @xmath93-norm . if we want to avoid questions about accumulations of null sets , this permits us to work with only a countable number of @xmath94 s at any each time . [ in particular , we do not claim that @xmath92 is continuous in any pointwise sense . ] this will not pose any problems because we will content ourselves with the following ( weaker ) definition of a gaussian free field based on the standard approach via gaussian hilbert spaces ( cf . sheffield @xcite , section 2.4 ) . [ def - gff ] we say that a family @xmath95 of random variables on a probability space @xmath96 is a gaussian free field if the map @xmath97 is linear a.s . and each @xmath98 is gaussian with mean zero and variance @xmath99 our goal is to show that the family of random variables @xmath100 has , asymptotically , in the scaling limit , the law of a linear transformation of a gaussian free field . to pass to this limit , we have to impose the condition that the test functions are slowly varying , which we take to be on the scale @xmath101 . for @xmath102 and a function @xmath69 , let @xmath103 and note that the normalization ensures that @xmath104\\[-8pt ] & = & \varepsilon^2(f , f)+(f,(-\delta)^{-1}f)\mathop{\le}_{\varepsilon\le1}\vert f\vert_{{{\mathcal h}}}^2.\nonumber\end{aligned}\ ] ] let @xmath105 denote the linear functional acting on @xmath106 via @xmath107 the main theorem is the gaussian scaling limit for @xmath108 . [ thm - main ] suppose that @xmath2 is as in ( [ v - repr ] ) with @xmath4 compactly supported in @xmath5 . let @xmath18 be a gradient gibbs measure for the potential @xmath2 which we assume to be ergodic with respect to the translations of @xmath12 and to have zero tilt . then , for every @xmath109 , @xmath110 where @xmath111 is the inverse of the operator @xmath112 with @xmath113 denoting some positive semidefinite , nondegenerate , @xmath114 matrix . in other words , the law of @xmath105 on the linear dual @xmath115 of any finite - dimensional linear subspace @xmath116 converges weakly to that of a gaussian field with mean zero and covariance @xmath117 . there follow some additional observations and remarks concerning the model under consideration and the results above since @xmath118 is dominated by a multiple of @xmath119 from below , the integral in ( [ limit ] ) , interpreted as the quadratic form @xmath120 , is well defined for all @xmath121 . 2 . note that in ( [ phi_epsilon ] ) , the individual @xmath7 s get scaled by @xmath122 , not @xmath123 as one might expect from the conventional central limiting reasoning . this has to do with the fact that the variables @xmath36 are strongly correlated . these correlations are weaker for the gradients @xmath124 which adhere to the `` usual '' central limit scaling . in @xmath125 and for general potentials @xmath2 , the increments @xmath40 are in fact i.i.d . and the scaling limit follows from the standard central limit theorem . 3 . in @xmath126 , the matrix @xmath113 is not necessarily a multiple of unity since , in general , @xmath18 is not guaranteed to be invariant under reflections and rotations of @xmath12 . [ nevertheless , we expect that every zero - tilt , translation - invariant , ergodic measure for the isotropic interaction ( [ spec ] ) will inherit these symmetries . ] to get convergence of @xmath105 to gff in the sense of definition [ def - gff ] , one must thus scale the argument of @xmath7 by the root of the corresponding eigenvalue of @xmath127 in each of its principal directions . 4 . the absence of strict convexity does not permit us to use the general argument of funaki and spohn @xcite for the _ existence _ of an ergodic ggm with zero ( or any other prescribed ) tilt . to show that such ggms do exist and that our theorem [ thm - main ] is not vacuous we note that , by lemma 4.8 of biskup and koteck @xcite , every weak limit of torus measures exhibits exponential concentration of the empirical tilt ; one then just needs to choose any ergodic component . note that this lemma applies only to zero tilt ( cf . @xcite , remark 4.9 ) . the restriction to zero tilt is actually a significant drawback of our analysis . the main reason is our inability to characterize the scaling limit of the so - called corrector for the corresponding random walk problem . see section [ sec7 ] for more details . 6 . in the example studied by biskup and koteck @xcite [ cf . ( [ two - deltas ] ) ] , the two ggms coexisting at the transitional value @xmath128 of @xmath129 were proven to exhibit different characteristic fluctuations . it follows that the corresponding scaling limits will be distinguished by their stiffness coefficients @xmath130 . + moreover , by theorem 2.5 of @xcite , for @xmath131 , the transition in the @xmath28 model with ( [ two - deltas ] ) lies on a self - dual line , that is , @xmath132 the transition presumably stays on this line even as one slides the ratio @xmath133 toward one . however , it disappears before @xmath133 hits one because , for @xmath134 , the potential @xmath2 is convex and so there is only one ggm with zero tilt @xcite . at such a point of disappearance , physicists often expect nontrivial critical fluctuations . notwithstanding , our results show that this is not the case . we avoid the context of the `` stronger '' definition of gff as a random element in an appropriate banach space ( cf . gross @xcite or sheffield @xcite , section 2.2 ) . this definition is appealing in @xmath125 , where the limiting functional @xmath97 actually admits the integral representation @xmath135 with @xmath136 denoting a continuous diffusion with generator @xmath137 , but in @xmath126 , the corresponding field becomes less and less regular with increasing dimension and the appeal is lost . however , this context would be ideal if one wished to discuss the notion of tightness and convergence in law for the limit in theorem [ thm - main ] . both lemma [ lemma - cont ] and theorem [ thm - main ] are proved in section [ sec4 ] . the key idea underlying the representation ( [ v - repr ] ) is that the auxiliary variable @xmath138 in the expression for @xmath2 may be elevated to a genuine degree of freedom associated with the corresponding edge . specifically , given a gradient gibbs measure @xmath18 with potential ( [ v - repr ] ) , for each finite @xmath139 , consider the measure @xmath140 on @xmath141 defined by @xmath142 where @xmath143 and @xmath144 are borel sets . the representation ( [ v - repr ] ) ensures that @xmath145 is a consistent family of measures ; by kolmogorov s extension theorem , these are projections from a unique measure @xmath146 onto configurations @xmath147 . the restriction of @xmath146 to the @xmath37 s gives us back @xmath18 ; we call @xmath146 an _ extension _ of @xmath18 . the measure @xmath146 is gibbs for the hamiltonian @xmath148 , so we will refer to it as an _ extended gradient gibbs measure _ ( see biskup and koteck @xcite for further facts on extended ggms ) . to ease the notation , whenever @xmath21 is an edge between @xmath62 and @xmath149 , we may interchangeably write @xmath150 and @xmath30 for the same quantity . furthermore , for the same reasons , it will even be convenient to assume that @xmath151 and work with the @xmath138 s as symmetric objects . we proceed with a series of lemmas that characterize the properties of @xmath146 . [ lemma - iid ] let @xmath18 be a gradient gibbs measure for the potential @xmath2 and let @xmath146 be its extension to @xmath152 . consider the @xmath22-field @xmath153 . for @xmath146-a.e . @xmath37 , the regular conditional distribution @xmath154 , regarded as a measure on the @xmath138 s , takes the product form @xmath155.\ ] ] recall that @xmath156 . the identity ( [ consistent ] ) implies that @xmath140 coincides with @xmath146 on @xmath157 . however , @xmath158 has the desired product form by definition and so the claim follows by standard approximation arguments . [ lemma - ergodic ] let @xmath18 be a gradient gibbs measure and let @xmath146 be its extension to @xmath152 . if @xmath18 is translation - invariant and ergodic , then so is @xmath146 . the uniqueness of the extension of measures ( [ consistent ] ) implies that @xmath146 is translation - invariant if @xmath18 is translation - invariant and so it remains to prove that ergodicity is also inherited . let @xmath159 be a translation - invariant event , that is , @xmath160 if and only if @xmath161 for all @xmath62 . our task is to show that @xmath162 . first , we invoke the ergodicity of @xmath18 . consider the function @xmath163 since @xmath66 and @xmath146 are translation - invariant , we have @xmath164 but @xmath94 is @xmath165-measurable and the restriction of @xmath146 to @xmath165 is @xmath18 , which we assumed to be ergodic . hence , @xmath94 is constant almost surely . let @xmath166 denote this constant . we will use a standard approximation argument to show that @xmath167 . since @xmath66 is an event from the product @xmath22-algebra , there exists a sequence of events @xmath168 such that @xmath169 the bound @xmath170 then shows that @xmath171 in @xmath172 . since @xmath66 is translation - invariant , we have @xmath173 . each indicator can be approximated by the indicator of the event @xmath174 ; a simple bound gives @xmath175 for @xmath62 with @xmath176 , the fact that @xmath177 is a product measure ( cf . lemma [ lemma - iid ] ) implies that @xmath174 and @xmath178 are independent . hence , @xmath179 rolling the approximations backward , we then conclude that the left - hand side converges to @xmath166 in @xmath172 , while the right - hand side converges to @xmath180 ( note that all expectations are bounded ) . it follows that @xmath181 , that is , @xmath182 . as @xmath183 the proof is finished . our next goal will be to characterize also the conditional measure given the @xmath138 s . this will , in turn , require some facts from the theory of random walks with random conductances . we will frequently borrow facts from an associated potential theory which will be expounded in section [ sec5 ] . let us choose a configuration @xmath184 with @xmath185 and recall the formula ( [ generator ] ) for the generator @xmath33 of the random walk among conductances @xmath138 . we will focus on the action of @xmath33 on functions of both the environment @xmath138 and the position @xmath62 that satisfy the following _ shift covariance _ property : @xmath186 with @xmath60 and @xmath21 a coordinate unit vector in @xmath78 , subject to the condition @xmath187 this makes the function completely determined by its values at the neighbors of the origin . a function of this kind is said to be _ harmonic _ for the above random walk if @xmath188 for ( almost ) every @xmath138 . as it turns out , harmonic , shift - covariant functions are uniquely determined ( a.s . ) by their mean with respect to ergodic measures on the conductances . [ lemma - harmonic ] let @xmath189 be a translation - invariant , ergodic probability measure on configurations @xmath190 such that @xmath191 for some @xmath102 . let @xmath192 be a measurable function which is : 1 . harmonic in the sense of ( [ harmonic ] ) , @xmath189-a.s . shift - covariant in the sense of ( [ shift - invariance ] ) and ( [ zero - cond ] ) , @xmath189-a.s . ; 3 . square integrable in the sense that @xmath193 for all @xmath62 with @xmath194 . if @xmath195 for all @xmath62 with @xmath194 , then @xmath196 a.s . for all @xmath60 . we defer the proof , and further discussion of the consequences of shift covariance and harmonicity , to section [ sec5 ] . returning to the gradient fields , we now characterize the conditional law given the @xmath138 s . [ lemma - gaussian ] let @xmath18 be a translation - invariant , ergodic gradient gibbs measure with zero tilt and let @xmath146 be its extension to @xmath152 . consider the @xmath22-field @xmath197 . for @xmath146-a.e . @xmath138 , the conditional law @xmath198 , regarded as a measure on the set of configurations @xmath199 with the @xmath7 s defined from the @xmath37 s via ( [ eta - phi ] ) , is gaussian with mean zero , @xmath200 and covariance given by @xmath201 . explicitly , for each @xmath202 with finite support and @xmath203 , @xmath204 the fact that the conditional measure is a multivariate gaussian law with covariance @xmath205 can be checked by direct inspection of ( [ consistent ] ) . the only nontrivial task is to identify the mean . first , we note that the loop conditions ( [ loop ] ) ensure that there exists a function @xmath206 such that @xmath207 and @xmath208 for all unit vectors @xmath21 in the coordinate directions . we claim that @xmath209 is harmonic in the sense of ( [ harmonic ] ) . indeed , @xmath210 where we write , thanks to the loop conditions , @xmath211 . using the fact that @xmath146 is gibbs , we can now also condition on @xmath212 ; the conditional measure @xmath213 is gaussian with the explicit form @xmath214 where @xmath14 is an appropriate normalization constant . it is easy to check that the mean of @xmath215 under @xmath216 is exactly @xmath217 , proving that @xmath218 . next , we observe that the translation invariance of @xmath146 implies that @xmath219 and so @xmath209 is shift - covariant , as defined in ( [ shift - invariance ] ) and ( [ zero - cond ] ) . finally , the definition of @xmath209 and the fact that @xmath146 has zero tilt imply that @xmath220 as @xmath209 obeys all conditions of lemma [ lemma - harmonic ] , we have @xmath221 @xmath146-a.s . our reference to the random walk with generator @xmath33 is not limited to lemma [ lemma - harmonic ] ; we will also need to know some specific properties of this random walk . first , we will need to know that the position of the walk satisfies a central limit theorem . let @xmath222 denote the continuous - time random walk with the generator @xmath33 and let @xmath223 denote the law of @xmath224 subject to the initial condition @xmath225 . the following lemma goes back to kipnis and varadhan @xcite . [ lemma - bm ] let @xmath18 be a measure on @xmath57 which is translation - invariant , ergodic and obeys @xmath226 for some @xmath102 . there then exists a positive semidefinite , nondegenerate , @xmath114 matrix @xmath227 such that for every @xmath228 , the annealed distribution @xmath229 converges weakly to the law of the multivariate normal @xmath230 . the main result of @xcite actually shows that the annealed law of the entire _ path _ @xmath231 converges to that of ( a linear transform of ) brownian motion . however , the above is all that will be needed for the purposes of the present paper . apart from a central limit asymptotics , we will also need an estimate on the heat kernel of the above random walk . the following lemma is a consequence of the main result of delmotte and deuschel @xcite . let @xmath18 be a law of the conductances satisfying the ellipticity condition @xmath232 for some @xmath233 . there is then a @xmath234 such that @xmath235 here , @xmath236 is the discrete spatial derivative in the @xmath237th coordinate direction , that is , @xmath238 . by formula ( 1.5b ) in @xcite , theorem 1.1 , @xmath239 where @xmath240 is the probability of the continuous - time simple random walk to be at @xmath62 at time @xmath241 . this probability is bounded from above by a constant times @xmath242 . the goal of this section is to prove theorem [ thm - main ] concerning the scaling limit of @xmath243 . we begin by proving @xmath93-continuity of the random functional @xmath92 on @xmath85 , as stated in lemma [ lemma - cont ] . for convenience of notation , whenever @xmath244 is an operator on @xmath245 , we will extend it to an operator on @xmath246 via the formula @xmath247 where @xmath248 is the kernel of @xmath244 in the canonical basis in @xmath249 . proof of lemma [ lemma - cont ] let @xmath146 be the extended gradient gibbs measure corresponding to @xmath18 . recall the notation @xmath33 for the generator of the random walk among conductances @xmath184 and let @xmath250 denote the generator of the simple random walk ( i.e. , the special case of @xmath33 when all @xmath251 ) . choose @xmath252 . lemma [ lemma - gaussian ] and the fact that @xmath253 is linear in the @xmath37 s imply that @xmath254 where @xmath255 is as defined above . by assumption on the support of @xmath4 , we know that @xmath256 @xmath146-a.s . , by which we immediately have the operator inequalities @xmath257 therefore , it suffices to bound the quadratic form associated with the ( homogeneous ) discrete laplacian @xmath250 by the quadratic form defining the @xmath85-space : @xmath258 for some constant @xmath89 and all @xmath94 in a dense subset of @xmath85 . to this end , we pick @xmath121 in the schwartz class and let @xmath259 be its ( @xmath93-norm - preserving ) fourier transform . a direct calculation now yields @xmath260^d}{{d}}k \frac1{(-\hat{\mathcal l})(k ) } \biggl| \mathop{\sum_{k'\dvtx\exists\ell\in{\mathbb z}^d}}_{k - k'=2\pi\ell } \hat f(k')\prod_{j=1}^d\frac{1-{{e}}^{-{{i}}k_j'}}{{{i}}k_j ' } \biggr|^2,\ ] ] where @xmath261 is the generalized eigenvalue of the lattice laplacian . introducing @xmath262 to denote the corresponding quantity for the continuum laplacian , we invoke the cauchy schwarz inequality for the sum over @xmath263 to get @xmath264^d}{{d}}k\ , \biggl\ { \mathop{\sum _ { k'\dvtx\exists\ell\in{\mathbb z}^d}}_{k - k'=2\pi\ell } @xmath265 is well defined on the set @xmath266^d\dvtx k_j\ne0 , j=1,\ldots , d\}$ ] of full lebesgue measure in @xmath267^d$ ] . we claim that @xmath268 we will show this by proving that the summand in ( [ ck - def ] ) is bounded by a constant times the product @xmath269 . indeed , for the @xmath270 term , we use the fact that the ratio @xmath271 is bounded throughout @xmath272 , and the same for the ratios @xmath273 . when @xmath274 , we set @xmath237 to be the first index @xmath275 such that @xmath276 and bound the @xmath277 term by @xmath278 . we then bound the ratio of @xmath279 terms by unity and the @xmath275th term in the product by a constant times @xmath280 . [ the @xmath281 term is not needed because @xmath282 . ] the product @xmath269 is summable over @xmath283 uniformly in @xmath284 and so ( [ c - sup ] ) is proved . bounding @xmath285 by its supremum in ( [ csbound ] ) , we can merge the sum and the integral to get @xmath286 . the desired bound ( [ discrete - continuous ] ) then follows . the inclusion of @xmath93-norm of @xmath94 in @xmath287 is crucial for the bound ( [ discrete - continuous ] ) . indeed , on the basis of ( [ fourier ] ) , it is not hard to construct functions for which the ratio @xmath288 is arbitrarily large . this is caused by the fact that the spectrum of @xmath289 extends all the way to infinity , while that of @xmath290 is bounded . the continuity established in lemma [ lemma - cont ] allows us to work only with smooth and compactly supported test functions . we will nevertheless need one more regularity bound before we can delve into the proof of our main result . [ lemma - reg ] let @xmath18 be a translation - invariant law on the conductances subject to the ellipticity condition @xmath232 for some @xmath102 . there then exists @xmath89 such that whenever @xmath291 for some @xmath292 , @xmath293 where @xmath294 is the set function on borel subsets of @xmath78 defined by @xmath295 with @xmath296 denoting the set of points in @xmath297^d$ ] with vanishing @xmath237th coordinate . translation invariance of @xmath18 and a simple integration by parts tells us that @xmath298^d}{{d}}z\int_{[0,1]^d}{{d}}z'\ , \delta g(x+z)\delta g(y+z ' ) e_\mu p^x_\kappa(x_t = y ) \nonumber\\[-8pt]\\[-8pt ] & & \qquad= \sum_{x , y\in{\mathbb z}^d}\sum_{i , j=1}^d\int_{r_i}{{d}}z\int_{r_j}{{d}}z'\ , \partial_ig(x+z)\,\partial_j g(y+z')\nonumber\\ & & \qquad\quad\hspace*{94pt}{}\times \nabla_i\nabla_j e_\mu p^x_\kappa(x_t = y),\nonumber\end{aligned}\ ] ] where @xmath299 stands for the partial derivative with respect to the @xmath237th coordinate . restricting the integrations and sums so that the arguments @xmath300 and @xmath301 are in the support of @xmath302 , bounding the partial derivatives by @xmath303 and applying the estimate ( [ hcub ] ) , we obtain the desired bound . the consequence of lemma [ lemma - reg ] that will concern us is as follows . [ cor - tightness ] for @xmath18 as in lemma [ lemma - reg ] and any @xmath304 , @xmath305 choose @xmath94 of the form @xmath291 and note that @xmath306 , where @xmath307 . first , we observe that @xmath308 next , we note that , since the support of @xmath302 is the closure of a nonempty bounded open set , a simple covering argument tells us that @xmath309 where @xmath310 is the lebesgue measure of @xmath311 . as a consequence , there exists a constant @xmath312 such that @xmath313 plugging ( [ 4.13a ] ) and ( [ 4.15a ] ) into ( [ reg - bd ] ) , we get , for @xmath314 , @xmath315 the functions on the left ( indexed by @xmath316 ) are uniformly integrable in @xmath23 in all @xmath317 . having dispensed with regularity considerations , we can now proceed to establish the principal fact underlying the proof of theorem [ thm - main ] . [ lemma - scaling ] let @xmath18 be a translation - invariant , ergodic measure on @xmath318 such that @xmath319 for some @xmath320 . there then exists a positive semidefinite , nondegenerate @xmath114 matrix @xmath321 such that @xmath322 in @xmath18-probability for each @xmath323 , where @xmath137 is defined from @xmath227 by ( [ q - operator ] ) . the key to the proof is the following lemma . for every @xmath228 and any @xmath323 , @xmath324 let @xmath325 be the continuous - time random walk with the generator @xmath33 and let @xmath223 denote the law of the walk started from @xmath62 . by lemma [ lemma - bm ] , the annealed law of @xmath326 tends weakly to that of the multivariate normal @xmath327 for some positive semidefinite , nondegenerate @xmath114 matrix @xmath321 . as a consequence , if @xmath328 is a family of functions that are uniformly equicontinuous and bounded , then we have @xmath329 now , let @xmath137 be the generator of the brownian motion with mean zero and covariance @xmath227 , that is , @xmath137 is the operator in ( [ q - operator ] ) . we then have @xmath330 as @xmath33 is the generator of the random walk @xmath331 , we similarly derive @xmath332\\[-8pt ] & & \qquad= \varepsilon^d \int_{[0,1]^d\times[0,1]^d } { { d}}z_1\,{{d}}z_2 \sum_{x\in { \mathbb z}^d}f(\varepsilon x+\varepsilon z_1)e_\kappa^x\bigl(f(\varepsilon z_2+\varepsilon x_{t\varepsilon^{-2}})\bigr).\nonumber\end{aligned}\ ] ] we thus need to show that the right - hand side of ( [ flf ] ) tends to that of ( [ fqf ] ) . note that if @xmath94 is supported in @xmath333^d$ ] , then the integral in ( [ fqf ] ) can be restricted to @xmath334^d$ ] and the sum over @xmath62 in ( [ flf ] ) to , say , @xmath335 ( once @xmath336 ) . substituting @xmath337 with @xmath60 , @xmath338 and @xmath339^d$ ] in ( [ fqf ] ) allows us to put both terms on the same footing . subtracting ( [ fqf ] ) from ( [ flf ] ) , taking expectation with respect to @xmath18 and applying the cauchy schwarz inequality , we thus get @xmath340^d } { { d}}z \sum_{x\dvtx|x|\le 2m/\varepsilon } e_{\mu } | e_\kappa^xf(\varepsilon z+\varepsilon x_{t\varepsilon ^{-2}})\\ & & \qquad\quad\hspace*{139.2pt}{}-ef(\varepsilon x+\varepsilon z+{\mathcal n}_t ) @xmath18 , we may replace @xmath341 by the expression @xmath342 inside the expectation . for @xmath343 , @xmath344^d\}\ ] ] is an equicontinuous family of uniformly bounded functions . then , ( [ l2limit ] ) tells us that the right - hand side of ( [ long ] ) tends to zero as @xmath345 , which proves the desired claim ( [ 4.9 ] ) . proof of proposition [ lemma - scaling ] to extract ( [ 4.8 ] ) from ( [ 4.9 ] ) , we note that for any @xmath346 , @xmath347 replacing @xmath94 by @xmath348 and scaling @xmath23 by @xmath349 , we find that @xmath350 by ( [ 4.9 ] ) , the function being integrated tends to @xmath351 in probability for each @xmath23 ; the monotonicity in @xmath23 ( and continuity of the limit ) ensures that the convergence is actually uniform ( in probability ) on compact intervals . by corollary [ cor - tightness ] , the integral can be truncated to a finite interval in @xmath35-norm and similarly for the integral of the limit , which is finite since @xmath94 is finite in the domain of @xmath111 . it follows that @xmath352 in @xmath18-probability [ and @xmath353 . this is the desired conclusion ( [ 4.8 ] ) . we note that to control the tail of the integral in ( [ integral - repr ] ) in @xmath354 , it suffices to invoke the diagonal heat kernel estimate @xmath355 which , in the elliptic case , is an immediate consequence of the mixing theory for markov chains based on isoperimetric inequalities . this is sufficient because the finiteness of the green function in @xmath354 permits us to define @xmath356 , even for @xmath357 . this enables us to reduce the general case to positive @xmath94 by decomposing the test function into a positive and a negative part and applying @xmath358 which is uniformly integrable when @xmath354 . however , to include @xmath28 , we can not disregard the cancellations due to the vanishing of @xmath359 and thus the stronger derivative bound ( [ hcub ] ) is necessary . a similar situation occurred in giacomin , olla and spohn @xcite where a stronger nash continuity estimate was required to include @xmath28 . we are now ready to establish the main result of this paper . proof of theorem [ thm - main ] let @xmath18 be a translation - invariant , ergodic , gradient gibbs measure with zero tilt and let @xmath146 be its extension to @xmath152 . we want to prove that @xmath243 tends weakly to a normal random variable with mean zero and variance @xmath360 . by lemma [ lemma - cont ] , it suffices to prove this for @xmath323 . by lemma [ lemma - gaussian ] , we know that @xmath76 is gaussian conditional on @xmath138 . the standard formula for any gaussian random variable @xmath224 , @xmath361 implies , via ( [ mean - zero ] ) and ( [ variance ] ) , that @xmath362 by proposition [ lemma - scaling ] , we have @xmath363 in @xmath146-probability . since the right - hand side of ( [ 4.9a ] ) is a bounded continuous function of this inner product , ( [ limit ] ) follows by means of the bounded convergence theorem . the proof of the key lemma [ lemma - harmonic ] leads us to the study of potential theory for operators depending on a random environment that fall into the class of _ random conductance models_. a good deal of what is to follow exists explicitly , or implicitly , in the literature . we have borrowed some of the notation from the paper of mathieu and piatnitski @xcite , although the formalism draws on earlier works in homogenization theory ; see , for instance , the book by jikov , kozlov and oleinik @xcite . notwithstanding , the content of section [ sec5.2 ] appears to be new . consider a translation - invariant @xmath189 probability measure on @xmath364 ( endowed with the product @xmath22-algebra ) satisfying the ellipticity condition @xmath365 let @xmath366 denote the closure of the set of all local functions in the topology induced by the inner product @xmath367 let @xmath368 denote the set of coordinate vectors in @xmath12 . the translations by the vectors in @xmath369 induce natural unitary maps @xmath370 on @xmath366 defined via @xmath371 apart from square integrable functions , we will also need to work with vector fields , by which we will generally mean measurable functions @xmath372 or @xmath373 , depending on the context . we will sometimes write @xmath374 for @xmath375note that these may still be vector - valued . while we index vector fields only by the _ positive _ coordinate vectors , in certain situations , it is convenient to have them also defined for the negative coordinate directions via @xmath376 as we will see , this definition will automatically ensure that the cycle condition ( see lemma [ lemma - cycle ] below ) holds for the trivial cycles crossing only a single edge . let @xmath377 be the set of all vector fields with @xmath378 , where @xmath379 denotes the inner product @xmath380 examples of such functions are the gradients @xmath381 of local functions @xmath382 defined component - wise via the formula @xmath383 we denote by @xmath384 the closure of the set of gradients of local functions in the topology induced by the above inner product . [ lemma - cycle ] let @xmath385 . then , @xmath209 satisfies the cycle condition @xmath386 for any finite ( nearest - neighbor ) cycle @xmath387 on @xmath12 . in particular , there exists a shift - covariant function @xmath388 such that @xmath389 for every @xmath390 . the cycle condition ( [ cycle ] ) holds trivially for all gradients of local functions . indeed , if @xmath391 , then , in light of ( [ minus - b ] ) , we have @xmath392 a corresponding limit extends this to all of @xmath393 . to define @xmath394 , we integrate properly shifted values of @xmath209 along a path from zero to @xmath62 ; the cycle condition guarantees that the result is independent of the choice of path and that @xmath395 is shift - covariant . we will henceforth use the convention of writing @xmath395 for the extension of a shift - covariant vector field @xmath396 to a function on @xmath12 . notice that the shift @xmath397 extends naturally via @xmath398 next , let us characterize the functions in @xmath399 . [ lemma - divergence ] for @xmath400 , let @xmath401 be the function in @xmath366 defined by @xmath402,\ ] ] where @xmath403 is the coordinate vector opposite to @xmath21 . we then have @xmath404 if @xmath209 satisfies the cycle condition and @xmath395 is its extension , then @xmath405 . these are direct consequences of the definitions , the translation invariance of @xmath189 and a simple calculation . note that @xmath401 plays the role of the _ divergence_that is , the total flow out of a given vertex of vector field @xmath209 . however , we prefer to denote it by @xmath250 to emphasize its connection with the operator @xmath33 . clearly , all @xmath385 are shift - covariant and have zero mean . a question which naturally arises is whether every shift - covariant zero - mean @xmath209 is in @xmath393 . ( note that this is analogous to asking whether every closed differential form is exact . ) our answer to this is in the affirmative . [ thm - complement ] suppose @xmath189 is ergodic . then , every @xmath400 which obeys the cycle condition ( [ cycle ] ) and @xmath406 satisfies @xmath407 . again , recall that ( [ cycle ] ) and zero expectation are necessary for @xmath385 . the above implies that these conditions are also sufficient . to prove the theorem , we will need the following lemma . [ lemma - projection ] let @xmath408 denote the orthogonal projection onto @xmath409 in @xmath377 . if @xmath189 is ergodic and @xmath400 satisfies ( [ cycle ] ) , then @xmath410 , @xmath411-almost surely . fix @xmath400 that obeys ( [ cycle ] ) and let @xmath395 be the corresponding shift - covariant function . we will prove the claim only for the component @xmath412 ; the other cases follow analogously . by translation covariance and the @xmath93 ergodic theorem , we have @xmath413 if @xmath189 were separately ergodic ( i.e. , ergodic with respect to @xmath414 alone ) , then the claim would immediately follow by the fact that every @xmath414-invariant function must be constant . to make up for the potential lack of separate ergodicity , we note that translation covariance of @xmath209 and the fact that @xmath395 obeys the cycle conditions together yield @xmath415\\[-8pt ] & = & \bar u(\cdot , n{\hat{{e}}}_1)-u(\cdot,{\hat{{e}}}_j)+t_{n{\hat{{e}}}_1}u(\cdot,{\hat{{e}}}_j).\nonumber \ ] ] it follows that @xmath416 also converges to @xmath417 ( in @xmath93 ) and so , by the continuity of @xmath397 , @xmath418 hence , @xmath419 is invariant under all shifts and is therefore constant @xmath189-a.s . proof of theorem [ thm - complement ] suppose @xmath189 is ergodic and let @xmath400 obey ( [ cycle ] ) and @xmath406 . the boundedness of the @xmath150 s away from zero and infinity ensures that @xmath400 if and only if all of its components are in @xmath366 . our task is to construct functions @xmath420 such that @xmath421 in @xmath377 . we define @xmath422 and note that this is the unique solution of the equation @xmath423 . this observation implies that @xmath424 and so the first component of @xmath425 converges to that of @xmath209 , provided @xmath426 in @xmath366 . to see what happens with the other components of @xmath425 , we note that the cycle condition ( [ cycle ] ) translates into @xmath427 applying this to the definition of @xmath428 , we conclude that @xmath429 where @xmath430 is defined as @xmath428 , but with @xmath431 replaced by @xmath432 . again , it suffices to show that @xmath433 in @xmath366 , which will boil down to the same argument as for @xmath434 . to prove that @xmath426 , we note that , for the inner product in ( [ bracket - product ] ) , @xmath435 introducing the notation @xmath436 for the average @xmath437 inserting this into the above sum and reordering the terms , we get @xmath438 by lemma [ lemma - projection ] , the @xmath93 ergodic theorem and the fact that @xmath209 has zero expectation in @xmath189 , we have @xmath439 and so @xmath440 as @xmath441 . a straightforward estimate now shows that the sum in ( [ hh ] ) is @xmath442 and so @xmath443 , as desired . not every function in @xmath377 necessarily belongs to @xmath444 . a prime example is the _ position _ vector field @xmath445 . indeed , let @xmath446 be the projection @xmath447 since @xmath448 , it satisfies ( [ cycle ] ) and we may extend it to a function @xmath449 mapping @xmath450 by setting @xmath451 and integrating the gradients along oriented paths . lemma [ lemma - divergence ] implies that @xmath452 is harmonic , @xmath453 . lemma [ lemma - projection ] and ( [ on - axis ] ) show that @xmath454 and so @xmath455 . it follows that @xmath456 and so @xmath457 . the function @xmath449 is generally referred to as the _ corrector _ because it corrects for the nonharmonicity of the position function . the corrector can be defined by appealing to spectral theory ( kipnis and varadhan @xcite , also berger and biskup @xcite ) ; the above `` projection '' definition is inspired by those in giacomin , olla and spohn @xcite and mathieu and piatnitski @xcite . as a side remark , we note that the function @xmath452 actually allows us to characterize the space of all square integrable shift - covariant functions . [ cor - characterize ] suppose @xmath189 is ergodic . then , every shift - covariant @xmath458-valued @xmath459 can be obtained from @xmath460 , where @xmath62 is the position function and @xmath461 is its orthogonal projection onto @xmath393 , by means of the linear transformation @xmath462e_\nu u_k,\qquad j=1,\ldots , d.\ ] ] here , @xmath374 stand for @xmath375 and similarly for ( @xmath78-valued objects ) @xmath463 and @xmath464 . in particular , for @xmath458-valued vector fields , we have @xmath465 the observation ( [ l2-decompose])which , in the language borrowed from differential geometry , implies that linear transforms of @xmath460 are the only closed forms that are not exact was previously discovered in the context of multicolored exclusion processes ( quastel @xcite , theorem 9.1 ) . however , while our proof is based only on a soft , poisson equation - based argument ( [ hepsilon])([au1-constant ] ) , that of @xcite requires an explicit bound on the spectral gap for the corresponding dynamics . the reason may be the reliance of @xcite on the _ spatial _ ergodic theorem , which , naturally , leads to bounds involving the poincar inequality and/or spectral gap . proof of corollary [ cor - characterize ] this is a simple consequence of theorem [ thm - complement ] . let @xmath466 be the @xmath467-component vector field whose ( @xmath458-valued ) components are defined by @xmath468e_\nu u_k,\qquad j=1,\ldots , d.\ ] ] we will show that @xmath469 . first , both @xmath209 and @xmath470 obey the cycle condition , so @xmath471 also does . as @xmath448 implies @xmath472 , the fact that @xmath473 shows that @xmath474 , that is , @xmath475 . theorem [ thm - complement ] implies that @xmath476 . on the other hand , @xmath477 implies that @xmath478 and so , by lemma [ lemma - divergence ] , @xmath479 . thus , @xmath480 also holds . it follows that @xmath481 , as claimed . as for ( [ l2-decompose ] ) , the argument we just used ensures that a shift - covariant field @xmath400 can be written as @xmath482 , where @xmath483 is a vector with components @xmath484 , plus a shift - covariant vector field with zero expectation . by theorem [ thm - complement ] , the latter is in @xmath485 . we still have to supply the proof of lemma [ lemma - harmonic ] . proof of lemma [ lemma - harmonic ] since @xmath302 has square integrable components , we have @xmath486 . as @xmath302 is shift - covariant and has zero expectation , theorem [ thm - complement ] implies that @xmath487 . however , @xmath302 is also harmonic and so , in turn , lemma [ lemma - divergence ] forces @xmath488 . thus , @xmath489 , as desired . recall that @xmath449 is the extension of @xmath490 subject to the condition @xmath451 . with this object at hand , we may even remove the restriction to zero slope in lemma [ lemma - gaussian ] . [ cor - tilt ] let @xmath18 be a translation - invariant , ergodic , gradient gibbs measure for the potential ( [ v - repr ] ) and let @xmath146 be its extension to @xmath152 . let @xmath19 be the tilt of @xmath18 and let @xmath491 . then , @xmath198 is gaussian with mean @xmath492\ ] ] and covariance given by @xmath201 . let @xmath395 denote the extension of the shift - covariant vector field @xmath493 , @xmath390 . clearly , we have @xmath494 inspecting the proof of lemma [ lemma - gaussian ] , all formulae carry over , except ( [ u - exp ] ) , which becomes @xmath495 thus , @xmath496 is harmonic , shift - covariant and of zero mean , so @xmath497 by lemma [ lemma - divergence ] and theorem [ thm - complement ] . the proofs in the present note often rely on the fact that the random walk with generator @xmath33 is uniformly elliptic . this enters via the assumption that @xmath4 , defining the potential @xmath2 , has support bounded away from zero . while we believe that the general picture carries over , even if we let the support extend all the way to zero , a number of steps in the proof become quite subtle . for instance , the pointwise heat kernel asymptotic for this walk may take a radically different form ( berger et al . @xcite ) and it is not known under what conditions on the environment the walk scales to brownian motion . progress in this direction for i.i.d . ( or i.i.d . dominated ) environments has been made only recently ( mathieu @xcite , biskup and prescott @xcite , barlow and deuschel @xcite ) . another interesting open question concerns the annoying restriction to _ zero - tilt _ gradient gibbs measures . as is seen from corollary [ cor - tilt ] , once the tilt is nonzero , ( [ 4.9a ] ) has to be modified to @xmath498}|{\mathscr{f}}\bigr)(\kappa)={{e}}^{-{{i}}(t\cdot\bar\chi)(f)-1/2(f,(-{\mathcal l}_\kappa)^{-1}f)}.\ ] ] when we plug in @xmath348 for @xmath94 , the second term in the exponent still converges to @xmath360 in probability as @xmath499 . to show convergence of @xmath500 to gaussian free field , we thus have to show that @xmath501 converges in law to a normal random variable . if @xmath2 is strictly convex and of the form ( [ v - repr ] ) , then we know this to be true by the main result of giacomin , olla and spohn @xcite . however , the general case seems to be a hard open problem ( cf . @xcite , conjecture 5 ) . already in @xmath125 , where the corrector can be written explicitly as the sum @xmath502 with @xmath503 , the desired result requires that the resistivities @xmath504 exhibit a specific correlation decay @xcite , theorem 7.6 . very little information concerning this is known in @xmath505 ; see mourrat @xcite for some recent progress in this direction . our final remark concerns a generalization to potentials @xmath2 , where ( [ v - repr ] ) has been modified to @xmath506 with @xmath507 denoting a family of strictly convex , even , measurable functions with uniformly superlinear growth at @xmath13 and a uniform lower bound . in this case , we may still consider the extended gradient gibbs measures ; however , the conditional law given the @xmath138 s is no longer gaussian . notwithstanding , given @xmath138 , the helffer sjstrand representation still applies and leads to a random walk in a dynamic random environment . its annealed central limit theorem would imply the gaussian scaling limit for the @xmath7-field . this project was initiated at the kavli institute for theoretical physics during the program `` stochastic geometry and field theory : from growth phenomena to disordered systems . '' the kitp program was supported by the nsf grant phy99 - 07949 . we wish to thank jean - dominique deuschel for a number of interesting suggestions , and scott sheffield and jeremy quastel for discussions on related subjects . we are also grateful to the anonymous referees for careful reading of the first version of this paper and for a number of constructive suggestions .

interference effects in quantum dots between different transport channels can lead to a strong suppression of conductance , which cuts like a canyon through the common conductance plot [ phys . rev . lett . * 104 * , 186804 ( 2010 ) ] . in the present work , we consider the thermoelectric transport properties of the canyon of conductance suppression using the second - order von neumann approach . we observe a characteristic signal for the zeros of the thermopower . this demonstrates that thermoelectric measurements are an interesting complimentary tool to study complex phenomena for transport through confined systems .

1504.04048

thermoelectric effects in nanoscale structures have been frequently studied with the aim to improve the efficiency of devices . @xcite in addition , it has been recently demonstrated that thermopower measurements can serve as an interesting tool to characterize complex scenarios due to coherences and interactions in nanoscale systems . @xcite here the intrinsic advantage of thermoelectric measurements is that they probe asymmetries around the fermi level and therefore can easily provide information about excited states . @xcite close to a degeneracy of energy levels in a quantum dot , interference and correlation effects can play an important role for transport and lead to pronounced quantum mechanical phenomena . for a spin - degenerate quantum dot ( qd ) the conductance experiences an enhancement for low temperatures due to the kondo effect . @xcite the thermopower of the single and multiple quantum dots in the presence of the kondo effect was examined both theoretically @xcite and experimentally . @xcite on the other hand , in the case of two degenerate levels with equal spin , at degeneracy , conductance can be suppressed due to electron correlation and interference effects . @xcite such a spin - polarized two - level model was widely used to interpret the phenomena of phase lapses of transmission in aharonov - bohm interferometer containing a qd . @xcite additionally , due to their small size the quantum dots also can exhibit coulomb blockade effect . @xcite the thermopower of the usual coulomb blockade sequential tunneling peaks and the cotunneling signal were addressed in refs . . in ref . a system of two spin - polarized degenerate levels was realized in an insb nanowire qd , where different @xmath0-factors allow to control level crossings for the same spin by a magnetic field . the experiment showed , in good agreement with supporting calculations , that the suppression cuts as a canyon through the standard conductance plot for different parities of the level couplings . a more detailed analysis of the conductance spectrum can be found in ref . . a related two - level system was optimized for achieving high thermoelectric performance in ref . . in this work we further elaborate on thermoelectric properties and focus on the fingerprint of conductance suppression in the thermopower signal . ( color online ) schematic of the system : a quantum dot with single particle levels @xmath1 and @xmath2 coupled to the leads via tunneling barriers . the leads have a temperature difference @xmath3 which can give rise to a current flowing through the dot . ] the paper is organized as follows . in section [ sec : model ] the model for the spin - polarized two - level quantum dot is introduced . results for conductance and thermopower are presented in section [ sec : results ] . here we focus on the zeros of the thermopower , which are relatively easy to extract experimentally . we start in section [ subsec : uzero ] with a discussion of the non - interacting case , where results are obtained using transmission formalism . here we show that up to five zeros in the thermopower can be found if the gate voltage is varied . furthermore , we establish the role of temperature and level broadening for the existence of multiple zeros . the impact of finite qd charging energy is addressed in section [ subsec : ufinite ] , where calculations are performed by the second order von neumann ( 2vn ) approach . this reveals the full scenario for the canyon of conductance suppression . concluding remarks are given in section [ sec : concl ] . the two - level quantum dot system under investigation is depicted in fig . its energy levels @xmath4 are assumed to be individually tunable . in practice this is achieved by a gate voltage @xmath5 ( where @xmath6 is the electron charge , @xmath7 is the chemical potential in the left and right lead , respectively ) ; and by controlling the detuning @xmath8 , which , for example , can be physically realised by a magnetic field if the two levels have different @xmath0-factors.@xcite the total hamiltonian for the system consisting of the qd coupled to two leads via tunneling barriers can be written as @xmath9 the two - level spinless qd hamiltonian @xmath10 , in a single particle basis , is @xmath11 where @xmath12 is the charging energy due to coulomb repulsion between electrons in the dot . in our calculations and corresponding plots we shift the zero value of the gate voltage to the particle - hole symmetric point and use @xmath13 the lead hamiltonian @xmath14 is @xmath15 where @xmath16 denotes the lead state and @xmath17 denotes the lead . we assume , that the leads are thermalized resulting in an occupation function @xmath18 with different temperatures @xmath19 , for the left and right lead , respectively . the tunneling between the dot and the lead is governed by the hamiltonian @xmath20 that reads @xmath21 where , for simplicity , the tunneling amplitudes @xmath22 are assumed to be the same for all values of @xmath16 and the coupling strengths are defined as @xmath23 . here we performed the @xmath16-sum using the flat density of states approximation , i. e. , @xmath24 , with @xmath25 denoting the density of states at the fermi level and @xmath26 denoting the bandwidth of the leads . we also assume @xmath26 to be the largest energy scale in the problem . in a nonequilibrium situation a current @xmath27 through the dot can be generated both by an applied bias @xmath28 and a temperature difference @xmath29 between the leads . close to equilibrium this establishes a linear relation between @xmath30 and the bias @xmath31 for which this current vanishes . this defines the seebeck coefficient ( or thermopower ) @xmath32 via @xmath33 this coefficient can be numerically obtained for a given transport model . for the units of the thermopower we use @xmath34 and for the units of conductance we use @xmath35 . lastly , in linear response to applied bias @xmath31 or temperature difference @xmath3 the hamiltonian eq . can be mapped to generalized anderson model , which was discussed in ref . . this mapping for model parameters considered in this paper is discussed in [ sec : mapanderson ] . for the non - interacting case , when there is no charging energy @xmath36 , the thermopower @xmath32 and conductance @xmath37 can be calculated exactly from the electronic transmission @xmath38 as @xcite [ gstransform ] @xmath39 with @xmath40 where the limit of vanishing temperature difference @xmath41 and vanishing bias @xmath42 between the leads was taken . the transmission is calculated using the caroli formula @xcite @xmath43,\ ] ] where @xmath44 is the retarded / advanced green s function of the quantum dot electrons and @xmath45 is the coupling strength given by @xmath46 the retarded / advanced green s function can be obtained from @xmath47 by replacing @xmath48 for the retarded function @xmath49 and @xmath50 for the advanced function @xmath51 , where @xmath52 is positive infinitesimal . here @xmath53 is the self - energy given by @xmath54 in the case of infinite bandwidth @xmath55 . if the following choice of tunneling amplitudes ( corresponding to the canyon of conductance suppression ) is made @xmath56 we get the coupling strength matrices @xmath57 with @xmath58 , and the transmission function takes a simple and intuitive form as a sum of two breit - wigner resonances : @xcite @xmath59 here the minus sign between both terms relates to the different parities of the tunnel couplings for both levels as defined in eq . ( [ amplitudes ] ) . for small @xmath60 both terms cancel at @xmath61 resulting in zero transmission . this provides a line of conductance suppression in the @xmath62 plane . in order to see whether the canyon of conductance suppression gives qualitatively different results for the thermopower , we will make a comparison to the case when all the tunneling amplitudes have the same sign : @xmath63 this choice leads to such coupling strength matrices @xmath64 the transmission function acquires a more complicated structure than a sum of two resonances : @xcite @xmath65 for zero energy @xmath66 the above transmission is zero at @xmath67 . for the interacting case , we use the second order von neumann ( 2vn ) approach , @xcite which contains both coherent effects and cotunneling . however , the method fails around and below the kondo temperature where strong correlations between the dot and the lead electrons appear . for our considered spinless system the kondo temperature can be defined through the mapping to generalized anderson model as discussed by kashcheyevs _ et al . _ @xcite numerically , the thermopower is found in an iterative procedure by varying the voltage until the current vanishes . the governing equations of the method and the solution procedure of those equations is presented in [ sec:2vn ] . in the numerical calculations we use @xmath68 and a finite temperature difference @xmath69 ( about an order of magnitude lower ) in order to achieve convergent and reliable results . furthermore , we define the average temperature @xmath70 as a reference point . all the calculations in figs . [ fig3]-[fig5 ] are done using the 2vn approach . for the non - interacting case @xmath36 the 2vn approach reproduces the exact transmission formalism results . @xcite we explicitly checked that for present calculations the values @xmath71 give the linear conductance @xmath37 and the linear thermopower @xmath32 on the scale of figs . [ fig3]-[fig5 ] within the 2vn approach . additionally , the value of bandwidth @xmath72 is chosen large enough that the results in the considered parameter regime do not depend on @xmath73 . our main focus is on the zeros of the thermopower @xmath32 , which is a rather distinct spectroscopic feature , especially when there is a sign change of @xmath32 . to emphasize the parameter values where the zeros appear we plot the square root of the absolute value of the thermopower @xmath74 and use a grayscale scheme where the zero value corresponds to white . ( color online ) the thermopower @xmath32 ( left column ) and conductance @xmath37 ( right column ) as a function of the detuning @xmath75 and gate voltage @xmath76 in the non - interacting case @xmath36 . the asymmetry values are @xmath77 for ( a ) to ( d ) , respectively . for the contour plots the temperature is set to @xmath78 and @xmath79 denotes the sign of the thermopower for these contour plots . the dashed ( blue ) curves depict zeros of thermopower , given by eqs . , for the case of a small temperature @xmath80 . panel ( e ) gives a cut of the contour plots at the detuning @xmath81 . ] ( color online ) same as fig . [ fig2 ] just for the case where the tunneling amplitudes have the same sign . the dashed ( blue ) curves depict zeros of thermopower , given by eqs . . ] we start by discussing the thermopower for the non - interacting @xmath36 case.@xcite if the temperature is the lowest energy scale in the system , i. e. , @xmath82 , the zeros of the thermopower can be obtained directly from the transmission function by using the mott approximation , @xcite which yields the following expression @xmath83 then we see that in this limit the zeros of thermopower are given by the zeros of the derivative of the transmission , i. e. , @xmath84 , when @xmath85 . using eq . , this condition defines the following curves in the @xmath86 space : [ anlzerostherm ] @xmath87 ^ 2}.\end{aligned}\ ] ] from the above expressions we see that for given detuning @xmath75 between the levels there can be up to five zeros , when the gate voltage @xmath76 is scanned . three zeros appear once the distinct peaks , related to the energies of the single particle levels , become resolvable and this happens when the separation between the levels is larger than the energy scale determined by @xmath60 . these zero s curves are given by the two solutions ( [ anlzerostherm]a ) , and by solution in ( [ anlzerostherm]c ) with the minus sign for @xmath88 or the plus sign for @xmath89 . additional two zeros appear due to destructive interference , and for finite detuning @xmath90 this happens only when the couplings to the levels are asymmetric , @xmath91 . for symmetric couplings , @xmath92 , destructive interference leads to zero thermopower for @xmath93 , but without the sign change , because it is a solution of double multiplicity . figure [ fig2 ] depicts the zeros given by analytical expressions eqs . by blue dashed curves for different asymmetries @xmath94 . these lines can be compared to the thermopower @xmath95 ( in grayscale ) where the temperature is comparable to the coupling @xmath60 . we see that for larger values of @xmath75 or @xmath76 the zeros , corresponding to single particle levels , for the finite temperature case match eqs . , but there are some discrepancies when @xmath75 and @xmath76 do not exceed @xmath96 . for the interference features we see that they are sensitive to finite temperature , i. e. , depending on the coupling asymmetry they appear for larger values of @xmath97 . [ fig2]e shows the calculated data for a fixed @xmath98 . for @xmath99 ( dashed blue curve ) we clearly see the five zeros in thermopower for the finite temperature calculation . for the case of the same sign tunneling amplitudes from eqs . and we find that there can be up to three zeros in the thermopower for @xmath100 for given @xmath75 : [ samesignzeros ] @xmath101 for low energies in the transmission eq . the destructive interference corresponds to a zero given by eq . , i.e. , the gate voltage has to be within the window of energies given by the detuning , i.e. , @xmath102 . this is in contrast to the canyon of conductance suppression situation , where the destructive interference appears for @xmath103 . additionally , the canyon of conductance suppression exhibits a thermopower zero for @xmath102 just because there is a minimum of conductance , and there is no necessity for destructive interference for this zero to appear . on the other hand , for the same sign amplitudes the large values of the thermopower close to a zero give a hint for the presence of the destructive interference ( see fig . [ fig2s ] ) . @xcite the role of finite temperature can be understood as follows . in the different limit , when the tunneling coupling is the smallest energy scale , i. e. , @xmath104 , we obtain the following expression for transmission to lowest order in @xmath60 @xmath105,\ ] ] which is valid for both eq . and eq . . applying eq . yields the condition for zeros of thermopower @xmath106 the above eq . can have only up to three gate voltage @xmath76 solutions for given @xmath75 , i. e. , the features arising due to destructive interference for the canyon of conductance suppression get washed out due to high temperature [ as also seen from eq . ] . additionally , for symmetric coupling , @xmath92 , eq . yields the following condition for resolution of two levels , @xmath107 which has a solution @xmath108 . @xcite from fig . [ fig2]a we see that both the temperature @xmath109 and the coupling @xmath60 contribute to the effective broadening and the two levels become resolvable in the thermopower for @xmath110 in the case of the canyon of conductance suppression . for the same sign tunneling amplitudes we obtain that the resolution of the thermopower zeros around @xmath93 is determined just by the temperature broadening and not by @xmath60 ( see fig . [ fig2s]a ) . ( color online ) the thermopower @xmath32 and conductance @xmath37 evolution as a function of increasing charging energy for symmetric coupling configuration @xmath92 . the charging energy values are @xmath111 for ( a ) to ( c ) , respectively . for the 2vn thermopower calculations we have set @xmath112 , @xmath113 and for conductance calculations we have set @xmath114 , @xmath113 , @xmath115 . also a finite bandwidth @xmath72 is used . the dashed ( green ) lines denote the bare resonances given by eq . . the dashed ( red ) lines show a region of detuning , where the thermopower is not defined because of the zero conductance . ] ( color online ) the thermopower @xmath32 and conductance @xmath37 evolution as a function of increasing charging energy for asymmetric coupling configuration @xmath99 . the charging energy values are @xmath116 for ( a ) to ( e ) , respectively . other parameter values are as in fig . [ fig3 ] . ] ( color online ) the thermopower @xmath32 and conductance @xmath37 evolution as a function of increasing charging energy for the same sign tunneling amplitudes eq . and asymmetric coupling configuration @xmath99 . the charging energy values are @xmath117 for ( a ) to ( c ) , respectively . other parameter values are as in fig . the ( red ) horseshoe patches denote the region of parameters , where our numerical procedure for solving the 2vn equations does not converge . ] the thermopower @xmath32 ( first column ) calculated for canyon of conductance @xmath37 suppression ( second column ) discussed in ref . . the charging energy values are @xmath118 and @xmath119 for ( a ) and ( b ) , respectively . note that here the parametrization @xmath120 , @xmath121 @xmath122 , @xmath123 , with @xmath124 , for tunneling amplitudes is used . other parameter values are as in fig . [ fig3 ] . ] in order to study the impact of the finite interaction @xmath126 , we first examine a case of equal coupling strengths for the two levels , i. e. , @xmath92 . the corresponding thermopower and conductance are shown in fig . [ fig3 ] . changing the interaction strength @xmath12 causes a minor change in the conductance , while the changes are more visible in the thermopower . we note that for @xmath126 the 2vn method can give negative diagonal reduced matrix elements @xmath127}_{bb}$ ] defined in eq . and also the linear conductance @xmath37 can become negative , which is unphysical . @xcite this is the reason that we do not address the interference features in the thermopower appearing at large gate voltages @xmath76 . the conductance plots in the right column of fig . [ fig3 ] show , that the conductance peaks for large @xmath128 appear at particular resonances , where the quantum dot ground states differing by a single charge cross in energy . these are the common sequential tunneling peaks occurring at @xmath129 which are shown by dashed green lines in fig . [ fig3 ] . for zero detuning @xmath75 these _ sequential tunneling _ lines are broken by a canyon of conductance suppression . now we consider the thermopower in the left column of fig . [ fig3 ] . around the sequential tunneling lines there is an equal amount of electron and hole tunneling for the given resonance and the average energy of tunneling particles becomes zero.@xcite this provides a zero in the thermopower . as can be seen for large @xmath128 there is actually a zero close to the dashed green lines , as expected . a further zero is occurring in between ( at @xmath130 for the case @xmath92 considered here ) , where electron and hole tunneling for different resonances compensate each other . for the charging energy , @xmath131 , smaller than the effective broadening of the levels , there is a region of detuning , where the level crossing of empty with singly occupied and singly with doubly occupied states in the dot can not be resolved ( see fig . [ fig3]a , b ) . for the corresponding thermopower , there is only one sign shift for all gate voltages around zero detuning . for larger detunings @xmath132 the levels can be resolved and again three sign shifts appear in the thermopower : at the two - level crossings and at the electron - hole symmetry point @xmath133 . when the charging energy becomes larger than the effective broadening there are always three sign shifts in the thermopower as can bee seen from fig . [ fig3]c . the behavior of thermopower and conductance for different @xmath12 when the coupling is asymmetric @xmath99 is shown in fig . [ fig4 ] . here we focus on the thermopower zeros , which appear for low values of gate voltage and detuning , i.e. , we do not address zeros , which appear due to interference ( as discussed for @xmath36 above ) . the canyon of conductance suppression is now tilted due to the asymmetric coupling as discussed already in ref . . more interestingly , the appearance of the thermopower zeros is changed quite drastically : for @xmath134 , the curve of zero thermopower becomes strongly tilted around @xmath135 . it takes a characteristic s - shaped form , around its center denoted by a box in figs . [ fig4]a - c . increasing @xmath12 pushes two further thermopower zeros ( their onset is denoted by two asterisks in figs . [ fig4]b , c ) towards the central s - shaped curve . around critical @xmath136 there is a qualitative change ( compare fig . [ fig4]c and fig . [ fig4]d ) , where the s - shaped curve merges with the other ones . finally , for large @xmath12 ( fig . [ fig4]e ) the situation resembles the case for symmetric coupling , @xmath92 , addressed in fig . [ fig3]c . the situation when all the tunneling amplitudes have the same sign is depicted in fig . [ fig4s ] . for @xmath137 we see that the evolution of zeros is qualitatively the same as in fig . however , for the canyon of conductance suppression the zero lines have a jump around @xmath138 ( see figs . [ fig4]d and [ fig4]e ) . additionally , for @xmath139 the s - shaped zero thermopower curve does not appear for the same sign configuration . lastly , in fig . [ fig5 ] we present the thermopower behavior for a system , whose conductance was studied experimentally in ref . . these results may be relevant for future thermoelectric experiments on similar type devices , and shows the situation when the couplings have more general structure than eq . . note that for given parameters in fig . [ fig5 ] we have the level @xmath2 more strongly coupled than the level @xmath140 , which is opposite from the cases studied in figs . [ fig2]-[fig4 ] . this results to a mirrored behavior of thermopower and conductance along zero detuning @xmath93 . otherwise , the introduced additional asymmetries in the couplings between the left and the right lead does not alter the qualitative behavior of the asymmetric coupling @xmath99 thermopower seen in fig . the canyon of conductance suppression known from refs . has been further investigated with thermopower @xmath32 acting as a probing tool . zeros of the thermopower are a telling spectroscopic feature , which for the system studied in this paper yields information about the level coupling asymmetry @xmath94 ( compare fig . [ fig3 ] vs. fig . [ fig4 ] ) and charging energy @xmath12 compared to the effective broadening of the levels given by @xmath141 . additionally , for the non - interacting case it was shown that up to five zeros can be observed when scanning the gate voltage for given detuning . this shows that the thermopower measurement could be useful to resolve features appearing due to destructive interference [ see fig . [ fig2 ] and eq . ] . additionally , by comparing the same sign tunneling amplitude thermopower [ eq . ] to canyon of conductance suppression thermopower it was shown that the canyon of conductance suppression has a unique signature around zero detuning @xmath93 ( compare fig . [ fig4 ] vs. fig . [ fig4s ] ) . we thank k. g. l. pedersen , a. donarini , and j. knig for useful discussions . this work was supported by the swedish research council ( vr ) and nanolund . in the 2vn approach @xcite we approximately solve the equation @xmath142,\ ] ] by considering the density matrix @xmath143 elements , which connect the states differing by up to two electron or hole excitations . such a treatment corresponds to the so - called resonant tunneling approximation in real - time diagrammatic approach , @xcite and yields an exact current for non - interacting systems ( @xmath36 for our considered system ) . @xcite we note that the 2vn method equations were originally derived in ref . , and here we present this derivation for convenience with a slightly different notation . we write the governing equations in the many - particle eigenbasis @xmath144 of the dot hamiltonian @xmath145 . expressed in this many - particle basis the tunneling hamiltonian @xmath20 becomes @xmath146 here we used the _ letter convention _ : if more than one state enters an equation , then the position of the letter in the alphabet follows the particle number ( for example @xmath147 , @xmath148 , @xmath149 ) . in such a way the sum @xmath150 restricts to those combinations , where @xmath151 . for our considered system we have four many - particle eigenstates @xmath152 and such many - particle tunneling amplitudes @xmath153 the density matrix elements are defined as @xmath154}={\langleag\rvert}\rho{\lvertbg'\rangle},\ ] ] where @xmath155 , with @xmath156 denoting the eigenstate of the dot hamiltonian @xmath145 and @xmath157 denoting the eigenstate of the lead hamiltonian @xmath14 . here the label @xmath158 provides the number of electron or hole excitations needed to transform @xmath157 into @xmath159 . for example , we consider the matrix elements of the type @xmath160}={\langlebg\rvert}\rho{\lvertb'g\rangle},\\ & \rho_{bg-\kappa , ag}^{[1]}={\langlebg-\kappa\rvert}\rho{\lvertag\rangle},\\ & \rho_{dg-\kappa-\kappa',bg}^{[2]}={\langledg-\kappa-\kappa'\rvert}\rho{\lvertbg\rangle},\\ & \rho_{bg-\kappa+\kappa',b'g}^{[2]}={\langlebg-\kappa+\kappa'\rvert}\rho{\lvertb'g\rangle}. \end{aligned}\ ] ] here we have introduced the following notation @xmath161 @xmath162 by neglecting all the density matrix elements with more than two electron or hole excitation @xmath163 from eq . we obtain the equations @xmath164}&=(e_{b}-e_{b'})\rho_{bg , b'g}^{[0]}\\ & + { } t_{ba_1,\kappa_1}\rho_{a_1g+\kappa_1,b'g}^{[1]}(-1)^{n_{a_1}}\\ & + { } t_{bc_1,\kappa_1}\rho_{c_1g-\kappa_1,b'g}^{[1]}(-1)^{n_{b}}\\ & -{}\rho_{bg , c_1g-\kappa_1}^{[1]}(-1)^{n_{b'}}t_{c_1b',\kappa_1}\\ & -{}\rho_{bg , a_1g+\kappa_1}^{[1]}(-1)^{n_{a_1}}t_{a_1b',\kappa_1 } , \end{aligned}\ ] ] @xmath165 } & = ( e_{c}-e_{\kappa}-e_{b})\rho_{cg-\kappa , bg}^{[1]}\\ & + { } t_{cb_1,\kappa_1}\rho_{b_1g-\kappa+\kappa_1,bg}^{[2]}(-1)^{n_{b_1}}\\ & + { } t_{cd_1,\kappa_1}\rho_{d_1g-\kappa-\kappa_1,bg}^{[2]}(-1)^{n_{c}}\\ & -{}\rho_{cg-\kappa , c_1g-\kappa_1}^{[2]}(-1)^{n_{b}}t_{c_1b,\kappa_1}\\ & -{}\rho_{cg-\kappa , a_1g+\kappa_1}^{[2]}(-1)^{n_{a_1}}t_{a_1b,\kappa_1 } , \end{aligned}\ ] ] @xmath166 } & \approx(e_{b}-e_{\kappa}+e_{\kappa'}-e_{b'})\rho_{bg-\kappa+\kappa',b'g}^{[2]}\\ & + { } t_{ba_1,\kappa}\rho_{a_1g-\kappa+\kappa'+\kappa , b'g}^{[1]}(-1)^{n_{a_1}}\\&+{}t_{bc_1,\kappa'}\rho_{c_1g-\kappa+\kappa'-\kappa',b'g}^{[1]}(-1)^{n_{b}}\\&-{}\rho_{bg-\kappa+\kappa',c_1g-\kappa}^{[1]}(-1)^{n_{b'}}t_{c_1b',\kappa}\\ & -{}\rho_{bg-\kappa+\kappa',a_1g+\kappa'}^{[1]}(-1)^{n_{a_1}}t_{a_1b',\kappa ' } , \end{aligned}\ ] ] @xmath167 } & \approx(e_{d}-e_{\kappa}-e_{\kappa'}-e_{b})\rho_{dg-\kappa-\kappa',bg}^{[2]}\\ & + { } t_{dc_1,\kappa}\rho_{c_1g-\kappa-\kappa'+\kappa , bg}^{[1]}(-1)^{n_{c_1}}\\&+{}t_{dc_1,\kappa'}\rho_{c_1g-\kappa-\kappa'+\kappa',bg}^{[1]}(-1)^{n_{c_1}}\\&-{}\rho_{dg-\kappa-\kappa',c_1g-\kappa}^{[1]}(-1)^{n_{b}}t_{c_1b,\kappa}\\ & -{}\rho_{dg-\kappa-\kappa',c_1g-\kappa'}^{[1]}(-1)^{n_{b}}t_{c_1b,\kappa'}. \end{aligned}\ ] ] note that all indices with subscript 1 like @xmath168 are summed over . additionally , phase factors like @xmath169 appear due to order exchange of the lead operators with the dot operators , i.e. , @xmath170 . summing eqs . and over all the lead states @xmath157 we get @xmath171}&=(e_{b}-e_{b'})\phi_{bb'}^{[0]}\\ & + { } t_{ba_1,\kappa_1}\phi_{a_1b',\kappa_1}^{[1 ] } + { } t_{bc_1,\kappa_1}\phi_{c_1b',\kappa_1}^{[1]}\\ & -{}\phi_{bc_1,\kappa_1}^{[1]}t_{c_1b',\kappa_1 } -{}\phi_{ba_1,\kappa_1}^{[1]}t_{a_1b',\kappa_1 } , \end{aligned}\ ] ] @xmath172}&\approx(e_{c}-e_{\kappa}-e_{b})\phi_{cb,\kappa}^{[1]}\\ & + { } t_{cb_1,\kappa}\phi_{b_1b}^{[0]}f_{\kappa } -{}\phi_{cc_1}^{[0]}t_{c_1b,\kappa}f_{-\kappa},\\ & + { } t_{cb_1,\kappa_1}\phi_{b_1b,-\kappa+\kappa_1}^{[2 ] } + { } t_{cd_1,\kappa_1}\phi_{d_1b,-\kappa-\kappa_1}^{[2]}\\ & + { } \phi_{cc_1,-\kappa+\kappa_1}^{[2]}t_{c_1b,\kappa_1 } + { } \phi_{ca_1,-\kappa-\kappa_1}^{[2]}t_{a_1b,\kappa_1 } , \end{aligned}\ ] ] @xmath173}&\approx(e_{b}-e_{\kappa}+e_{\kappa'}-e_{b'})\phi_{bb',-\kappa+\kappa'}^{[2]}\\ & -{}t_{ba_1,\kappa}\phi_{a_1b',\kappa'}^{[1]}f_{\kappa } + { } t_{bc_1,\kappa'}\phi_{c_1b',\kappa}^{[1]}f_{-\kappa'}\\ & -{}\phi_{bc_1,\kappa'}^{[1]}t_{c_1b',\kappa}f_{-\kappa } + { } \phi_{ba_1,\kappa}^{[1]}t_{a_1b',\kappa'}f_{\kappa ' } , \end{aligned}\ ] ] @xmath174}&\approx(e_{d}-e_{\kappa}-e_{\kappa'}-e_{b})\phi_{db,-\kappa-\kappa'}^{[2]}\\ & -{}t_{dc_1,\kappa}\phi_{c_1b,\kappa'}^{[1]}f_{\kappa } + { } t_{dc_1,\kappa'}\phi_{c_1b,\kappa}^{[1]}f_{\kappa'}\\ & -{}\phi_{dc_1,\kappa'}^{[1]}t_{c_1b,\kappa}f_{-\kappa } + { } \phi_{dc_1,\kappa}^{[1]}t_{c_1b,\kappa'}f_{-\kappa'}. \end{aligned}\ ] ] where we introduced the following notation @xmath175}=\sum_{g}\rho_{bg , b'g}^{[0]},\\ & \phi_{cb,\kappa}^{[1]}=\sum_{g}\rho_{cg-\kappa , bg}^{[1]}(-1)^{n_{b}},\quad \phi_{bc,\kappa}^{[1]}=\big[\phi_{cb,\kappa}^{[1]}\big]^{*},\nonumber\\ & \phi_{ca,-\kappa-\kappa'}^{[2]}=-\sum_{g}\rho_{cg-\kappa-\kappa',ag}^{[2]},\nonumber\\ & \phi_{bb',-\kappa+\kappa'}^{[2]}=+\sum_{g}(1-\delta_{\kappa\kappa'})\rho_{bg-\kappa+\kappa',b'g}^{[2]},\nonumber\\ & f_{\kappa}\equiv f_{k\ell}=(\exp[(e_{k}-\mu_{\ell})/k_{{\mathrm}{b}}t_{\ell}]+1)^{-1},\nonumber\\ & f_{-\kappa}\equiv 1-f_{k\ell},\nonumber\end{aligned}\ ] ] and when going from eq . to eq . we have used @xmath176}=\rho_{b_1g-\kappa+\kappa , bg}^{[0]}+(1-\delta_{\kappa\kappa_1})\rho_{b_1g-\kappa+\kappa_1,bg}^{[2]},\\ & \rho_{cg-\kappa , c_1g-\kappa_1}^{[2]}=\rho_{cg-\kappa , c_1g-\kappa}^{[0]}+(1-\delta_{\kappa\kappa_1})\rho_{cg-\kappa , c_1g-\kappa_1}^{[2]}. \end{aligned}\ ] ] here we have also assumed that the electrons in the leads are _ thermally distributed _ according to the fermi - dirac distribution @xmath177 and that this distribution is not affected by the coupling to the quantum dots . this assumption leads to the following relations for eq . @xmath178}\approx f_{\kappa}\phi_{b_1b}^{[0]},\\ & \sum_{g}\rho_{cg-\kappa , c_1g-\kappa}^{[0]}\approx f_{-\kappa}\phi_{cc_1}^{[0 ] } , \end{aligned}\ ] ] for eq . @xmath179}(-1)^{n_{a_1}}\approx - f_{\kappa}\phi_{a_1b',\kappa'}^{[1]},\nonumber\\ & \sum_{g}\rho_{c_1g-\kappa+\kappa'-\kappa',b'g}^{[1]}(-1)^{n_{b}}\approx f_{-\kappa'}\phi_{c_1b',\kappa}^{[1]},\\ & \sum_{g}\rho_{bg-\kappa+\kappa',c_1g-\kappa}^{[1]}(-1)^{n_{b'}}\approx f_{-\kappa}\phi_{bc_1,\kappa'}^{[1]},\nonumber\\ & \sum_{g}\rho_{bg-\kappa+\kappa',a_1g+\kappa'}^{[1]}(-1)^{n_{a_1}}\approx - f_{\kappa'}\phi_{ba_1,\kappa}^{[1]},\nonumber\end{aligned}\ ] ] and for eq . @xmath180}(-1)^{n_{c_1}}\approx-\phi_{c_1b,\kappa'}^{[1]}f_{\kappa},\nonumber\\ & \sum_{g}\rho_{c_1g-\kappa-\kappa'+\kappa',bg}^{[1]}(-1)^{n_{c_1}}\approx\phi_{c_1b,\kappa}^{[1]}f_{\kappa'},\\ & \sum_{g}\rho_{dg-\kappa-\kappa',c_1g-\kappa}^{[1]}(-1)^{n_{b}}\approx\phi_{dc_1,\kappa'}^{[1]}f_{-\kappa},\nonumber\\ & \sum_{g}\rho_{dg-\kappa-\kappa',c_1g-\kappa'}^{[1]}(-1)^{n_{b}}\approx-\phi_{dc_1,\kappa}^{[1]}f_{-\kappa'}.\nonumber\end{aligned}\ ] ] for the stationary state we assume the conditions @xmath181}=0,\quad { i}\hbar\frac{{\partial}}{{\partial}{t}}\phi_{cb,\kappa}^{[1]}=0,\quad { i}\hbar\frac{{\partial}}{{\partial}{t}}\phi_{bb',-\kappa+\kappa'}^{[2]}=0,\quad { i}\hbar\frac{{\partial}}{{\partial}{t}}\phi_{db,-\kappa-\kappa'}^{[2]}=0 , \end{aligned}\ ] ] which allow to write @xmath182}$ ] in terms of @xmath183}$ ] as @xmath184}= \frac{-{}t_{ba_1,\kappa}\phi_{a_1b',\kappa'}^{[1]}f_{\kappa } + { } t_{bc_1,\kappa'}\phi_{c_1b',\kappa}^{[1]}f_{-\kappa ' } -{}\phi_{bc_1,\kappa'}^{[1]}t_{c_1b',\kappa}f_{-\kappa } + { } \phi_{ba_1,\kappa}^{[1]}t_{a_1b',\kappa'}f_{\kappa'}}{e_{\kappa}-e_{\kappa'}-e_{b}+e_{b'}+{i}\eta},\\ & \phi_{db,-\kappa-\kappa'}^{[2]}= \frac{-{}t_{dc_1,\kappa}\phi_{c_1b,\kappa'}^{[1]}f_{\kappa } + { } t_{dc_1,\kappa'}\phi_{c_1b,\kappa}^{[1]}f_{\kappa ' } -{}\phi_{dc_1,\kappa'}^{[1]}t_{c_1b,\kappa}f_{-\kappa } + { } \phi_{dc_1,\kappa}^{[1]}t_{c_1b,\kappa'}f_{-\kappa'}}{e_{\kappa}+e_{\kappa'}-e_{d}+e_{b}+{i}\eta}. \end{aligned}\ ] ] here we have added a positive infinitesimal @xmath185 to ensure a proper decay of initial conditions . after inserting the above expressions into eq . we obtain the integral equations of the 2vn method for the stationary state @xmath186}_{cb,{\kappa } } + t_{cb_1,{\kappa}}f_{{\kappa}}\phi^{[0]}_{b_1b } -\phi^{[0]}_{cc_1}f_{-{\kappa}}t_{c_1b,{\kappa}}\\ & + \frac{t_{cb_1,{\kappa}_1}\left [ \color{black } { t_{b_1c_1,{\kappa}_1}f_{-{\kappa}_1}\phi^{[1]}_{c_1b,{\kappa } } + \phi^{[1]}_{b_1a_1,{\kappa}}f_{{\kappa}_1}t_{a_1b,{\kappa}_1 } } \color{black } { -t_{b_1a_1,{\kappa}}f_{{\kappa}}\phi^{[1]}_{a_1b,{\kappa}_1 } -\phi^{[1]}_{b_1c_1,{\kappa}_1}f_{-{\kappa}}t_{c_1b,{\kappa } } } \right]}{e_{{\kappa}}-e_{{\kappa}_1}-e_{b_1}+e_{b}+{i}\eta}\\ & + \frac{t_{cd_1,{\kappa}_1}\left [ \color{black } { t_{d_1c_1,{\kappa}_1}f_{{\kappa}_1}\phi^{[1]}_{c_1b,{\kappa } } + \phi^{[1]}_{d_1c_1,{\kappa}}f_{-{\kappa}_1}t_{c_1b,{\kappa}_1 } } \color{black } { -t_{d_1c_1,{\kappa}}f_{{\kappa}}\phi^{[1]}_{c_1b,{\kappa}_1 } -\phi^{[1]}_{d_1c_1,{\kappa}_1}f_{-{\kappa}}t_{c_1b,{\kappa } } } \right]}{e_{{\kappa}}+e_{{\kappa}_1}-e_{d_1}+e_{b}+{i}\eta}\\ & + \frac{\left [ \color{black } { t_{cd_1,{\kappa}_1}f_{-{\kappa}_1}\phi^{[1]}_{d_1c_1,{\kappa } } + \phi^{[1]}_{cb_1,{\kappa}}f_{{\kappa}_1}t_{b_1c_1,{\kappa}_1 } } \color{black } { -t_{cb_1,{\kappa}}f_{{\kappa}}\phi^{[1]}_{b_1c_1,{\kappa}_1 } -\phi^{[1]}_{cd_1,{\kappa}_1}f_{-{\kappa}}t_{d_1c_1,{\kappa } } } \right]t_{c_1b,{\kappa}_1}}{e_{{\kappa}}-e_{{\kappa}_1}-e_{c}+e_{c_1}+{i}\eta}\\ & + \frac{\left [ \color{black } { t_{cb_1,{\kappa}_1}f_{{\kappa}_1}\phi^{[1]}_{b_1a_1,{\kappa } } + \phi^{[1]}_{cb_1,{\kappa}}f_{-{\kappa}_1}t_{b_1a_1,{\kappa}_1 } } \color{black } { -t_{cb_1,{\kappa}}f_{{\kappa}}\phi^{[1]}_{b_1a_1,{\kappa}_1 } -\phi^{[1]}_{cb_1,{\kappa}_1}f_{-{\kappa}}t_{b_1a_1,{\kappa } } } \right]t_{a_1b,{\kappa}_1}}{e_{{\kappa}}+e_{{\kappa}_1}-e_{c}+e_{a_1}+{i}\eta } , \end{aligned}\ ] ] @xmath187 } + { } t_{ba_1,\kappa_1}\phi_{a_1b',\kappa_1}^{[1 ] } + { } t_{bc_1,\kappa_1}\phi_{c_1b',\kappa_1}^{[1 ] } -{}\phi_{bc_1,\kappa_1}^{[1]}t_{c_1b',\kappa_1 } -{}\phi_{ba_1,\kappa_1}^{[1]}t_{a_1b',\kappa_1}.\quad\quad\ ] ] additionally , we impose the normalisation condition for the diagonal reduced - density matrix elements : @xmath188}=1.\ ] ] the integral equation under interest has the structure @xmath189}_{\kappa}=f_{\kappa}+\sum_{\kappa_1}k_{\kappa,\kappa_1}\phi^{[1]}_{\kappa_1}.\ ] ] it is solved iteratively on an equidistant energy grid @xmath190 by having @xmath191 discretization points for our considered calculations . the zeroth iteration of @xmath183}_{\kappa,0}=f_{\kappa}$ ] is determined by making a local approximation , i.e. , terms of the form @xmath183}_{ba,\kappa_1}$ ] which have integrated momentum label @xmath192 are neglected . then the first correction is determined as @xmath193}_{\kappa,1}=\sum_{\kappa_1}k_{\kappa,\kappa_1}f_{\kappa_1}$ ] . the higher order corrections are given by @xmath193}_{\kappa , n}=\sum_{\kappa_1}k_{\kappa,\kappa_1}\delta\phi^{[1]}_{\kappa',n-1}$ ] and the solution is expressed as @xmath183}_{\kappa}=\phi^{[1]}_{\kappa,0}+\sum_{n}\delta\phi^{[1]}_{\kappa',n}$ ] . we make iterations up to @xmath194 , which yields good convergence for most parameter values . in these iterations we need to evaluate a lot hilbert transform of the form : @xmath195}_{\kappa})=\frac{1}{\pi}\int_{-d}^{d}\frac{\phi^{[1]}_{\kappa'}{\mathrm{d}}{e_{\kappa'}}}{e_{\kappa}-e_{\kappa'}\pm{i}\eta}\\ & \phantom { ....... } = \frac{1}{\pi}{\mathcal}{p}\int_{-d}^{d}\frac{\phi^{[1]}_{\kappa'}{\mathrm{d}}{e_{\kappa'}}}{e_{\kappa}-e_{\kappa'}}\mp{i}\phi^{[1]}_{\kappa}\theta(d-{\lverte_{\kappa}\rvert } ) . \end{aligned}\ ] ] the principal value integrals are efficiently evaluated on equidistant grid with @xmath196 points using fast fourier transform , which has complexity @xmath197 . @xcite finally , we are interested in the current going from the lead @xmath198 into the quantum dots , which is given by @xmath199}],\ ] ] and which shows that @xmath200}$ ] are the energy resolved current amplitudes . in linear response to applied bias @xmath31 or temperature difference @xmath3 the hamiltonian eq . can be mapped to generalized anderson model by mixing the left lead electrons with the right lead electrons , and the dot orbital @xmath201 with the orbital @xmath202 . the generalized anderson model has the form : @xcite @xmath203 where @xmath204 , @xmath205 represents a pseudo - spin and @xmath206 is an effective magnetic field . the hamiltonian is expressed in the form of by performing a singular value decomposition ( svd ) on the tunneling hamiltonian as @xmath207 @xmath208 using the rotations @xmath209 and @xmath210 the pseudo - spin operators are expressed in terms of original basis operators as : @xmath211 for the tunneling amplitudes of the form eq . we obtain @xmath212 which corresponds to parallel field configuration . for the same sign tunneling amplitudes we obtain @xmath213 @xmath214 lastly , for tunneling amplitudes examined in fig . [ fig5 ] we get : @xmath215 @xmath216 in all above cases we have @xmath217 , @xmath218 . 80ifxundefined [ 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 10.1002/adma.200600527 [ * * , ( ) ] link:\doibase 10.1038/nmat2090 [ * * , ( ) ] link:\doibase 10.1103/physrevb.75.041301 [ * * , ( ) ] link:\doibase 10.1088/1367 - 2630/10/8/083016 [ * * , ( ) ] link:\doibase 10.1103/physrevb.80.195409 [ * * , ( ) ] link:\doibase 10.1103/physrevb.81.235127 [ * * , ( ) ] link:\doibase 10.1103/physrevb.84.241107 [ * * , ( ) ] link:\doibase 10.1103/physrevb.85.085117 [ * * , ( ) ] link:\doibase 10.1088/1367 - 2630/15/10/105011 [ * * , ( ) ] link:\doibase 10.1088/1367 - 2630/16/1/015029 [ * * , ( ) ] link:\doibase 10.1103/physrevb.90.165428 [ * * , ( ) ] link:\doibase 10.1063/1.4893928 [ * * , ( ) ] link:\doibase 10.1038/34373 [ * * , ( ) ] link:\doibase 10.1126/science.281.5376.540 [ * * , ( ) ] link:\doibase 10.1209/epl / i2001 - 00559 - 8 [ * * , ( ) ] link:\doibase 10.1088/0953 - 8984/14/45/316 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.88.136601 [ * * , ( ) ] link:\doibase 10.1103/physrevb.73.075307 [ * * , ( ) ] link:\doibase 10.1143/jpsj.76.074709 [ * * , ( ) ] link:\doibase 10.1088/0953 - 8984/19/8/086214 [ * * , ( ) ] link:\doibase 10.1063/1.2836331 [ * * , ( ) ] link:\doibase 10.1016/j.physb.2009.07.118 [ * * , ( ) ] link:\doibase 10.1103/physrevb.82.113306 [ * * , ( ) ] link:\doibase 10.1103/physrevb.86.075303 [ * * , ( ) ] link:\doibase 10.1103/physrevb.86.165106 [ * * , ( ) ] link:\doibase 10.1088/0953 - 8984/25/50/505305 [ * * , ( ) ] link:\doibase 10.1103/physrevb.88.085313 [ * * , ( ) ] link:\doibase 10.1103/physrevb.90.165116 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.95.176602 [ * * , ( ) ] link:\doibase 10.1103/physrevb.49.17123 [ * * , ( ) ] link:\doibase 10.1103/physrevb.65.245301 [ * * , ( ) ] link:\doibase 10.1103/physrevb.67.195335 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.96.146801 [ * * , ( ) ] link:\doibase 10.1103/physrevb.75.115313 [ * * , ( ) ] link:\doibase 10.1088/1367 - 2630/9/5/113 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.74.4047 [ * * , ( ) ] link:\doibase 10.1103/physrevb.55.13726 [ * * , ( ) ] link:\doibase 10.1038/385417a0 [ * * , ( ) ] link:\doibase 10.1038/nature03899 [ * * , ( ) ] link:\doibase 10.1088/1367 - 2630/9/5/123 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.98.186802 [ * * , ( ) ] link:\doibase 10.1007/bf00683469 [ * * , ( ) ] @noop _ _ ( , , ) link:\doibase 10.1103/physrevb.46.9667 [ * * , ( ) ] link:\doibase 10.1016/0378 - 4371(93)90505-x [ * * , ( ) ] link:\doibase 10.1016/0038 - 1098(93)90819 - 9 [ * * , ( ) ] link:\doibase 10.1209/0295 - 5075/22/1/011 [ * * , ( ) ] link:\doibase 10.1103/physrevb.55.r10197 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.81.5197 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.86.280 [ * * , ( ) ] link:\doibase 10.1103/physrevb.66.045301 [ * * , ( ) ] link:\doibase 10.1103/physrevb.65.115332 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.104.186804 [ * * , ( ) ] link:\doibase 10.1103/physrevb.83.205412 [ * * , ( ) ] link:\doibase 10.1103/physrevb.84.113415 [ * * , ( ) ] link:\doibase 10.1103/physrevb.33.551 [ * * , ( ) ] link:\doibase 10.1088/0953 - 8984/17/25/014 [ * * , ( ) ] link:\doibase 10.1103/physrevb.73.085406 [ * * , ( ) ] link:\doibase 10.1088/0022 - 3719/4/8/018 [ * * , ( ) ] link:\doibase 10.1103/physrev.49.519 [ * * , ( ) ] link:\doibase 10.1143/jpsj.76.034715 [ * * , ( ) ] link:\doibase 10.1088/0256 - 307x/26/4/047301 [ * * , ( ) ] link:\doibase 10.1143/jpsj.81.104601 [ * * , ( ) ] link:\doibase 10.1103/physrevb.71.045317 [ * * , ( ) ] link:\doibase 10.1007/s10909 - 011 - 0417 - 2 [ * * , ( ) ] link:\doibase 10.1103/physrevb.72.195330 [ * * , ( ) ] @noop link:\doibase 10.1063/1.2938087 [ * * , ( ) ] link:\doibase 10.1088/1751 - 8113/46/6/065301 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.40.416 [ * * , ( ) ] @noop _ _ ( , ) link:\doibase 10.1103/physrev.181.1336 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.81.3491 [ * * , ( ) ] link:\doibase 10.1103/physrevb.62.r16275 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.65.771 [ * * , ( ) ] link:\doibase 10.1103/physrevb.75.235314 [ * * , ( ) ] link:\doibase 10.1016/j.physe.2009.06.069 [ * * , ( ) ] link:\doibase 10.1103/physrevb.54.16820 [ * * , ( ) ] _ _ , @noop master s thesis , ( ) @noop _ _ ( , , )

we deal with the equivariant estimation of scatter and location for-dimensional data , giving emphasis to scatter . it is important that the estimators possess both a high efficiency for normal data and a high resistance to outliers , that is , a low bias under contamination . the most frequently employed estimators are not quite satisfactory in this respect . the minimum volume ellipsoid ( mve ) and minimum covariance determinant ( mcd ) estimators are known to have a very low efficiency . s - estimators with a monotonic weight function like the bisquare have a low efficiency for small and their efficiency tends to one with increasing . unfortunately , this advantage is paid for by a serious loss of robustness for large . we consider four families of estimators with controllable efficiencies whose performance for moderate to large has not been explored to date : s - estimators with a non - monotonic weight function ( rocke 1996 ) , mm - estimators ,-estimators , and the stahel - donoho estimator . two types of starting estimators are employed : the mve computed through subsampling , and a semi - deterministic procedure proposed by pea and prieto ( 2007 ) for outlier detection . a simulation study shows that the rocke estimator starting from the pea - prieto estimator and with an adequate tuning , can simultaneously attain high efficiency and high robustness for and the mm estimator can be recommended for . keywords : mm - estimator , tau - estimator , s - estimator , stahel - donoho estimator , kullback - leibler divergence .

1504.03389

consider a sample @xmath5 we look for substitutes @xmath6 * * and * * @xmath7 of the sample mean vector and covariance matrix , that are resistant to atypical observations . we also want estimators that have a high efficiency for normal samples@xmath8 as a measure of robustness we consider not only the breakdown point but also the maximum expected kullback - leibler divergence between the estimator and the true value . under contamination . the most frequently employed estimators are not quite satisfactory in this respect . the minimum volume ellipsoid ( mve ) and the minimum covariance determinant ( mcd ) estimators ( rousseeuw 1985 ) are known to have a low efficiency . this efficiency can be increased by means of a one - step reweighting . croux and haesbroeck ( 1999 , tables vii and viii ) computed the finite - sample efficiencies of the reweighted mcd ; although they are much higher than for the raw estimator , they are still low if one wants a high breakdown point . s - estimators ( davies 1987 ) with a monotonic weight function like the bisquare have a low efficiency for small @xmath9 rocke ( 1996 ) showed that their efficiency tends to one with increasing @xmath0 ; unfortunately , this advantage is paid for with a serious loss of robustness for large @xmath0 . we restrict ourselves to equivariant estimators . there exist many non - equivariant proposals ; but the comparison between equivariant and non - equivariant estimators is difficult . in particular , a non - equivariant estimator is more difficult to tune for a given efficiency , since the latter depends on the correlations . among the published equivariant proposals , there are four families of estimators with controllable efficiencies : non - monotonic s - estimators ( rocke 1996 ) , mm - estimators ( tatsuoka and tyler 2000 ) , @xmath2-estimators ( lopuhaa 1991 ) and the estimator proposed independently by stahel ( 1981 ) and donoho ( 1982 ) but their behavior for large dimensions has not been explored to date . we compare their behaviors employing different weight functions . a simulation study shows that the rocke and mm estimators , with an adequate weight function and an adequate tuning , can simultaneously attain high efficiency and high robustness . it will be seen below that if we have a good @xmath10 it is easy to find a good equivariant @xmath11 but the converse is not true . for this reason we shall put more emphasis on the estimation of the scatter matrix . since all the considered estimators are based on the iterative minimization of a non - convex function , the starting values are crucial . subsampling is the standard way to compute starting values ; but we shall see that a semi - deterministic equivariant procedure proposed by pea and prieto ( 2007 ) may yield both shorter computing times and better statistical performances . in section [ secmesti ] we describe monotonic m - estimators ; section [ secminscale ] deals with estimators based on the minimization of a robust scale of mahalanobis distances . sections [ secmm ] and [ sec_stadono ] deal with mm and stahel - donoho estimators respectively . in section [ secrho ] we discuss the choice of the @xmath12function for mm- and @xmath13estimators . section [ seccomputing ] deals with computational details . in section [ secsimula ] the estimators are compared through a simulation study . in section [ secreal ] the estimators are applied to a real data set . finally section [ secconclu ] summarizes the results . section [ secappend ] is an appendix containing the full results of the simulations , the approximations for the tuning constants and some details on the rocke and the pea - prieto procedures . for @xmath14 and @xmath15 define the ( squared ) mahalanobis distance as @xmath16 let @xmath17 be a bounded nonincreasing weight function . then monotonic m - estimators ( maronna 1976 ) are defined as solutions of @xmath18 where for brevity we put @xmath19 the uniqueness of the solutions requires that @xmath20 be nondecreasing . unfortunately , this implies ( maronna , 1976 ) that the breakdown point is @xmath21 which makes these estimators unreliable except for small @xmath9 besides , this fact holds even if @xmath22 is known , while the asymptotic breakdown point of @xmath23 with known @xmath24 is 0.5 with an adequate @xmath25 this shows that the main problem to attain high robustness is the scatter matrix . for @xmath26 let @xmath27 be a robust scale . put @xmath28 a general family of estimators can be defined by @xmath29 where the condition @xmath30 rules out trivial solutions with @xmath31 . if @xmath32 we have the minimum volume ellipsoid ( mve ) estimator , and if @xmath33 is a trimmed mean , we have the minimum covariance determinant ( mcd ) estimator , both proposed by rousseeuw ( 1985 ) . the first one is very robust , but has a null asymptotic efficiency ; the second is very popular , but its asymptotic efficiency is very low ; see ( paindaveine and van bever , 2014 ) and references therein , and its maximum contamination bias increases rapidly with @xmath0 ( agostinelli et al , 2015 , table 1 ) . the condition @xmath30 means that we estimate the shape of the scatter . given the shape , the size can easily estimated to yield consistency at the normal model ( maronna et al . , section 6.3.2 ) . a simple way is to put@xmath34 instead of the median , one could use more efficient scales , such as an m - scale , but exploratory simulations indicate that they do not yield better results . let @xmath35 be a scale m - estimator defined as solution of @xmath36 where @xmath37 controls the breakdown point , and @xmath38 $ ] is smooth and nondecreasing in @xmath39 with @xmath40 and @xmath41 then s - estimators ( davies 1987 ) are defined as solutions of ( [ defescamini ] ) with @xmath33 given by ( [ defmscale ] ) . the maximum finite - sample replacement breakdown point is attained when @xmath42 and its value is equal to this @xmath43 see ( maronna et al . , 2006 , section 6.4.2 ) . a popular @xmath44 is the bisquare given by @xmath45{ccc}1-\left ( 1-d\right ) ^{3 } & \mathrm{if } & d\leq1\\ 1 & \mathrm{if } & d>1 . \end{array } \right . \label{defibis}\ ] ] note that the usual bisquare @xmath44 employed for regression is actually @xmath46 however , since we are dealing with the _ squared _ distances , we employ in ( [ defibis ] ) @xmath47 it is easy to show that s - estimators satisfy the estimating equations@xmath48 with @xmath49 that is , they satisfy the equations ( [ defmesti])-([defmesti2 ] ) which define monotonic m - estimators , with weight function @xmath49 here , since @xmath44 is bounded @xmath20 is not a nondecreasing function , and therefore this case is different from monotonic m - estimators . in particular , the breakdown point is not bounded by @xmath50 as shown by ( [ bdp - s ] ) . for the bisquare , the weight function is@xmath51 ( where @xmath52 denotes the indicator ) , which is decreasing . it seems intuitive that the weights of the observations should decrease with their outlyingness . however it will be seen in the next section that monotonicity is not necessarily favorable . rocke ( 1996 ) showed that if @xmath17 is nonincreasing , the efficiency of the estimator tends to one when @xmath53 a similar result was derived by kent and tyler ( 1996 , page 1363 ) for their constrained m - estimators . table [ tabefisbis ] shows the efficiencies ( to be defined later ) of the bisquare s - estimator of scatter for normal @xmath0-dimensional data . [ c]cccccccc@xmath0 & 2 & 5 & 10 & 20 & 30 & 40 & 50 + efficiency & 0.427 & 0.793 & 0.930 & 0.976 & 0.984 & 0.990 & 0.992 + however , it will be seen that the price for this increase in efficiency is a decrease in robustness . more precisely , although the breakdown point does not tend to zero with increasing @xmath1 the bias caused by contamination grows rapidly with @xmath9 this fact suggests that we need estimators with a controllable efficiency . but while in regression the efficiency has to be controlled to make it higher , here we need to prevent it from becoming too high . based on the fact that for large @xmath0 the @xmath0-variate standard normal distribution @xmath54 is concentrated near the spherical shell with radius @xmath55 rocke ( 1996 ) proposed estimators with non - monotonic weight functions . maronna et al . ( 2006 ) proposed a modification of rocke s biflat function , namely@xmath56 \mathrm{i}\left ( 1-\gamma\leq d\leq1+\gamma\right ) \label{rockeweight}\ ] ] with@xmath57 where @xmath58 is the @xmath59-quantile of the @xmath60 distribution with @xmath0 degrees of freedom , and @xmath61 is small to control the efficiency . maronna et al ( 2006 , sec . 6.8 ) dealt only with location . the performance of the respective scatter matrix will be studied below . @xmath62estimators were proposed by yohai and zamar ( 1988 ) to obtain robust regression estimators with controllable efficiency , and later lopuha ( 1991 ) employed the same approach for multivariate estimation . this approach requires two functions @xmath63 and @xmath64 for given @xmath65 call @xmath66 the solution of @xmath67 then the estimator minimizes the @xmath68-scale@xmath69 here @xmath70 where @xmath71 is chosen to regulate the efficiency . originally , @xmath2-estimators were proposed to obtain estimators with higher efficiency than s - estimators for small @xmath1 which required @xmath72 but for large @xmath0 we need @xmath73 in order to decrease the efficiency . mm - estimators were initially proposed by yohai ( 1987 ) to obtain regression estimators with a controllable efficiency . this approach has been used in the multivariate setting by lopuha ( 1992 ) and tatsuoka and tyler ( 2000 ) . here we give a simplified version of the latter . let @xmath74 be an initial very robust although possibly inefficient estimator . put@xmath75 and call @xmath33 the respective m - scale@xmath76 the estimator is defined by @xmath77 with @xmath78 such that@xmath79 where @xmath80 and the constant @xmath71 is chosen to control efficiency . it can be shown that the solution satisfies the equations @xmath81 with @xmath82 actually , it is not necessary to obtain the absolute minimum in ( [ defmm_min ] ) . as with regression mm - estimators ( yohai 1987 ) it is possible to show that any solution of ( [ defmm ] ) for with the objective function ( [ defmm_min ] ) is lower that for the initial estimator , has the same asymptotic behavior as the absolute minimum and has a breakdown point at least as high as the initial estimator . like @xmath2-estimators , mm estimators were originally proposed to obtain estimators with higher efficiency than s - estimators for small @xmath83 but here for large @xmath0 the constant has to chosen to prevent the efficiency becoming too high . let @xmath84 and @xmath85 be univariate location and dispersion statistics , e.g. , the median and mad . define for any @xmath86 the _ outlyingness _ @xmath87 : @xmath88 where the supremum is over @xmath89 with @xmath90 or equivalently over the spherical surface @xmath91 . here @xmath92 denotes @xmath93 . let @xmath17 ( the _ weight function _ ) be a positive function . the stahel donoho estimator of location and scatter , @xmath94 is a weighted mean and covariance matrix , with weights @xmath95 . if @xmath17 is continuous , and @xmath96 and @xmath97 are bounded for @xmath98 , the estimators have asymptotic breakdown point @xmath99 for all @xmath0 at continuous multivariate models , if @xmath100 and @xmath33 have asymptotic breakdown point @xmath99 ( see hampel et al.1986 ) . the finite - sample breakdown point was derived by tyler ( 1994 ) . maronna and yohai ( 1995 ) showed that these estimators have order @xmath101-consistency . their asymptotic distribution was given by zuo et al . ( 2004 ) . maronna and yohai ( 1995 ) recommended a huber - type @xmath102 however , further exploratory simulations indicate that better results are obtained with the weight function described in the next section . the numerical computation of these estimators is difficult . stahel ( 1981 ) proposed an approximate algorithm based on subsampling , the cost of which increases rapidly with @xmath0 . pea and prieto ( 2007 ) proposed a fast algorithm for outlier detection which combines the projections on a set of @xmath103 deterministic directions that are extrema of the kurtosis , and a set of random directions . although this method was originally meant for data analysis , it offers two further uses . first , the resulting projections can be employed to compute the stahel - donoho estimator ; second , the method yields a robust ( but probably inefficient ) estimator that can be used as a starting point for the iterative computing of the estimators described above . further details about this procedure are given in section [ seccomputing ] the most popular @xmath44 in robust methods seems to be the bisquare . yohai and zamar ( 1997 ) proposed a @xmath44 for regression with certain optimality properties . a simplified variant of this function is given by muler et al ( 2002 ) . its version for multivariate estimation has weight function @xmath104{ccc}1 & \mathrm{if } & d\leq4\\ q\left ( d\right ) & \mathrm{if } & 4<d\leq9\\ 0 & \mathrm{if } & d>9 \end{array } \right . , \label{optiweight}\ ] ] where@xmath105 is such that @xmath17 is continuous and differentiable at @xmath106 and @xmath107 the respective @xmath44 function is@xmath108{ccc}d & \mathrm{if } & d\leq4\\ s\left ( d\right ) & \mathrm{if } & 4<d\leq9\\ 6.494 & \mathrm{if } & d>9 \end{array } \right . , \ ] ] where@xmath109 figure [ figweigths ] shows the bisquare and optimal weight functions , scaled with their respective tuning constants for the mm - estimator with 90% efficiency and @xmath110 . it is seen that the optimal @xmath44 yields a smaller cutoff point . [ tbh ] weights.eps all estimators described above are computed as iterative reweighted means and covariances , starting from an initial estimator . for s- , @xmath62 and mm estimators this algorithm ensures that the objective function descends at each iteration . this need not happen with the rocke estimator , which has a non - monotonic weight function . maronna et al . ( 2006 , section 6.4.4 ) describe an algorithm which ensures attaining a local minimum . the ( approximate ) mve is computed with 1000 subsamples and using the improvement described in ( maronna et al . , 2006 , section 6.7.3 ) . since in all cases we attempt to minimize a non - convex function , the initial estimator is an essential part of the procedure . the standard way to obtain a robust and equivariant starting point is subsampling . however , ensuring a high enough breakdown point with large @xmath0 may require an impractically large number of subsamples . besides , our experiments indicate that the breakdown point may be much lower than expected when @xmath111 is small ( say , @xmath112 which is not uncommon with high - dimensional data sets . for these reasons we need a faster and more reliable starting point . pea and prieto ( 2007 ) proposed an equivariant and semi - deterministic procedure for outlier detection , based on finding directions that maximize or minimize the kurtosis of the respective projections , plus a set of random specific directions aimed at detecting outliers . here we employ this procedure ( which they call kurtosis plus specific directions , henceforth abbreviated as ksd ) as an estimator by itself . in the present setting it would not be competitive with the other estimators because its efficiency can not be tuned ( see table [ tabefikurto ] below ) , but we shall use it as an initial estimator competing with the sampling - based mve . there are no theoretical results on the breakdown point of ksd . however , the simulations in ( pea and prieto 2007 , table 4 ) suggest that it can yield reliable results even with 40% of outliers . a limited theoretical result is given in section [ secappend ] . given a starting point , the rocke estimator is computed iteratively as described in section 9.6.3 of ( maronna et al . , 2006 ) . the form of the weight function ensures that for normal data sets , most of the data have positive weights . since real data are seldom normal , it may happen that for data sets with large @xmath0 and low ratio @xmath111 the proportion of data with positive weights is small . if the data set is nearly collinear , this may cause @xmath113 to be ill - conditioned , which affects the computation of mahalanobis distances . for this reason , if at the first iteration the number of data with positive weights is less than @xmath114 the tuning constant is enlarged until this number is @xmath115 as a reference distribution we take the @xmath0-variate normal @xmath116 in order to measure the performance of a given estimator @xmath117 we need a measure of distance between an estimator and the true value . recall that the kullback - leibler divergence between densities @xmath118 and @xmath119 is@xmath120 if both densities belong to the same parametric model with parameter vector @xmath121 : @xmath122 then @xmath123 induces a distance between parameters : @xmath124 in the normal family , for @xmath22 with known @xmath125 we have@xmath126 and for @xmath125 with known @xmath127 we have@xmath128 since all estimators are equivariant we may in the simulations take without loss of generality @xmath129 . each estimator is evaluated by @xmath130 monte carlo average of the kullback - leibler divergences @xmath131 given in ( [ dkl(mu)])-([dkl(sigma ) ] ) . we generate @xmath132 samples @xmath133}$ ] of size @xmath134 from @xmath135 the estimators compared are : * rocke with tuning constant @xmath136 see ( [ gama - alfa ] ) * mm with bisquare and optimal @xmath137 with tuning constant @xmath71 ; see ( [ defmm ] ) * @xmath2 with bisquare and optimal @xmath137 with tuning constant @xmath71 ; see([rho1 - 2 ] ) * stahel - donoho with weight function @xmath138 where @xmath139 is defined in ( [ optiweight ] ) for all estimators we employed both the mve and ksd estimators as starting values . the tuning constants were chosen to attain an efficiency of 0.9 ( see below ) . we add for completeness four other estimators with uncontrollable efficiency : * the s - estimator ( s - e ) with @xmath140 in ( [ defmscale ] ) and biweight @xmath141 the optimal @xmath44 yielded similar results . * the mve and ksd estimators . * the mcd with breakdown point ( [ bdp - s ] ) and one - step reweighting , computed with the code in the library libra ( verboven and hubert 2005 ) . all scatter estimators are corrected for size by means of ( [ corretama ] ) , except for the mcd for which the code applies a consistency correction . call @xmath142 the sample covariance matrix . for each estimator @xmath113 we define@xmath143 the constants for each estimator are chosen to attain finite - sample efficiencies of 0.90 . to this end we computed for each estimator its tuning constants for @xmath144 with @xmath145 10 and 20 and @xmath0 between 5 and 50 , and then fitted the constants as functions of @xmath134 and @xmath0 . the simulation showed the efficiency can not be controlled in all cases , namely * for @xmath146 the maximum efficiency of the rocke estimator is 0.876 for all @xmath61s , and is still lower for smaller @xmath0 . the explanation is that when @xmath61 tends to zero , the estimator does not tend to the covariance matrix unless @xmath0 is large enough . * the minimum efficiency of the @xmath2-estimators over all constants @xmath71 tends to one with increasing @xmath1 for both @xmath12functions . in particular , it is @xmath1470.95 for @xmath148 the reason is that when @xmath71 is small , the @xmath62scale approaches the m - scale , and therefore the @xmath62estimators approaches the s - estimator . table [ tabefikurto ] shows the efficiencies of the ksd estimator . [ c]llll@xmath0 & @xmath134 & scatter & location + & & & + & & & + & & & + & & & + & & & + & & & + & & & + & & & + & & & + it seen that the efficiency depends heavily on the ratio @xmath111 and can be rather low for @xmath149 we deal first with shift contamination . for contamination rate @xmath150 let @xmath151.$ ] given @xmath152 we replace the first coordinate:@xmath153 the outlier size @xmath154 is varied between 1 and 12 in order to find the maximum @xmath155 the constant @xmath156 determines the scatter of the outliers . we employed the values @xmath157 and 0.2 , and @xmath158 and 0.5 . the simulations were run for @xmath159 5 , 10 , 15 , 20 and 30 , and @xmath160 with @xmath161 10 and 20 . since the complete results are rather bulky , they are given in section [ sectablas ] . here we give the most important conclusions from them . examination of the tables shows that * the price paid for the high efficiency of s - e is a large loss of robustness . * ksd is always better than mve as a starting estimator for mm and @xmath2 . * ksd is generally better than subsampling for s - d . * the optimal @xmath44 is always better than the bisquare @xmath44 for both mm and @xmath2 * in all situations , the best estimators are mm and @xmath2 with optimal @xmath137 rocke , and s - d , all starting from ksd . * although the results for @xmath158 and 0.5 are different , the comparisons among estimators are almost the same . * the relative performances of the estimators for location and scatter are similar . * the relative performances of the estimators for @xmath162 @xmath163 and @xmath164 are similar . for these reasons we give in table [ tabresumen ] a reduced version of the results , for @xmath165 and @xmath166 and the maximum @xmath167s of the scatter estimators corresponding to mm and @xmath2 ( both with optimal @xmath168 rocke and s - d , all starting from ksd . for completion we add s - e with ksd start , and the reweighted mcd . the results for estimators with efficiency less than 0.9 are shown in italics . [ c]cccccccc@xmath0 & @xmath169 & mm & @xmath2 & rocke & s - d & s - e & mcd + 5 & 0.1 & & & & & _ 1.09 _ & _ 1.99 _ + & 0.2 & & & & & _ 4.38 _ & _ 17.58 _ + & & & & & & _ 3.54 _ & _ 6.66 _ + & & & & & & _ 11.26 _ & _ 21.89 _ + & & & & & & _ 6.68 _ & _ 12.53 _ + & & & & & & _ 19.82 _ & _ 28.33 _ + & & & & & & _ 10.03 _ & _ 16.46 _ + & & & & & & _ 25.41 _ & _ 32.04 _ + & & & & & & _ 18.39 _ & _ 17.66 _ + & & & & & & _ 49.14 _ & _ 34.02 _ + it is seen that * the performance of s - d is competitive for @xmath170 but is poor for @xmath171 * for @xmath172 mm has the best overall performance . * for @xmath3 rocke has the best overall performance . * the mcd has a poor performance . figure [ figconta10_20_0 ] shows the values of @xmath167 as a function of the outlier size @xmath154 for some of the estimators in the case @xmath173 @xmath174 and @xmath175 here mm - opt stands for mm with optimal @xmath44 . all estimators in the second panel start from ksd . [ tbh ] conta_10_20_0.eps the plot confirms the superiority of rocke+ksd . recently hubert et al , ( 2015 ) proposed two deterministic estimators , called dets and detmm , of which the latter has a tuneable efficiency . we compare it with rocke+sd . the nominal efficiency of detmm is chosen as 0.90 . the scenario is the same as above . however , since detmm is not equivariant , the model is now n@xmath176 where @xmath177 has unit diagonal elements and all non - diagonal elements equal to @xmath141 we chose the extreme cases @xmath178 and @xmath179 since both yield qualitatively similar results , we show in table [ tabsimudeter ] only the results from the first case . [ c]cccccccccc@xmath169 & @xmath156 & & & & + & & @xmath180 & 100 & 200 & 400 & & 100 & 200 & 400 + 0.1 & 0 & & & & & & & & + & & & & & & & & & + & 0.5 & rocke+ksd & & & & & & & + & & detmm & & & & & & & + 0.2 & 0 & rocke+ksd & & & & & & & + & & detmm & & & & & & & + & 0.5 & rocke+ksd & & & & & & & + & & detmm & & & & & & & + the performance of detmm is clearly poor . we have not been able to find an explanation for this disappointing behavior . we compare the computing times of the rocke estimator with mve and ksd starts , and of detmm . the results are the average of 10 runs with normal samples , on a pc with intel tm12 duo cpu and 3.01 ghz . the values of @xmath134 were @xmath181 @xmath163 and @xmath182 with @xmath0 between 10 and 100 . the number of subsamples for the mve was made to increase slowly as @xmath183 table [ tabtiempos ] displays the results , where for brevity we show only the values for @xmath173 50 , 80 and 100 . [ c]ccccc@xmath0 & @xmath134 & rocke+mve & rocke+ksd & detmm + & & & & + & & & & + & & & & + & & & & + & & & & + & & & & + & & & & + & & & & + & & & & + & & & & + & & & & + & & & & + it is seen that rocke+ksd is faster than detmm for @xmath184 and rocke+mve . however it is slower than detmm for @xmath185 this rapid increase in computing time is probably due to the optimization procedure employed by ksd , and may be improved upon by choosing a more efficient optimizer . we deal with the well - known wine data set , available at the uci machine learning repository : https://archive.ics.uci.edu/ml/datasets/wine , which has been employed as a benchmark data set for pattern recognition ; see e. g. ( aeberhard et al , 1994 ) , and consists of three classes with 13 variables . the estimators were applied to the data of class 3 , with @xmath186 and @xmath187 since ksd and mve yielded similar results as initial estimators , we show only the results corresponding to the former . figures [ figwine1 ] and [ figwine2 ] contain the qq - plots of the ( squared ) mahalanobis distances for the different estimators . [ tbh ] figwine1.eps [ tbh ] figwine2.eps rocke , mm and detmm pinpoint respectively 8 , 6 and 5 possible outliers ; s - d seems to pinpoint an excessive number of possible outliers ; while s - e and @xmath2 behave like the classical estimator , showing no suspicious points . some subject - matter knowledge would be necessary to decide how atypical the suspicious points are . the rocke estimator has a controllable efficiency for @xmath188 with equal efficiencies , the rocke estimator with ksd start outperforms all its competitors for shift contamination its computing time is competitive for @xmath189 and can probably be improved upon . it can therefore be recommended for estimation with @xmath190 . for @xmath191 we can recommend mm with optimal @xmath44 and ksd start . tables [ tabsim05_10 ] to [ tabsim_3020 ] in this section contain the detailed results of the simulation . in each scenario , the smallest and next - to - smallest values are marked as bold and italic , respectively . this is done only for estimators with controllable efficiency ; in particular , rocke is not considered for @xmath172 for its efficiency in this case is less than 0.90 . one would expect the values for a given estimator to decrease when @xmath134 increases . however , in many cases this does not hold for the estimators based on the mve . we have re - run the simulations with a different seed , and also employed medians instead of means to rule out atypical cases , but this pattern appears nevertheless . we have not been able to find an explanation for this phenomenon . since it always affects the largest values , it does not influence the conclusions . [ c]cccccccccc@xmath156 & & start & & & + 0 & @xmath180 & & 25 & 50 & 100 & & 25 & 50 & 100 + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + [ c]rccccccccc@xmath156 & & start & & & + 0 & @xmath180 & & 25 & 50 & 100 & & 25 & 50 & 100 + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + 0.5 & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + & & & & & & & & & + [ c]cccccccccc@xmath156 & & start & & & + 0 & @xmath180 & & 50 & 100 & 200 & & 50 & 100 & 200 + & s - e & mve & & & & & & & + & s - e & ksd & & & & & & & + & mm - bisq . & mve & & & & & & & + & mm - bisq . & ksd & & & & & & & + & mm - opt . & mve & & & & & & & + & mm - opt . & ksd & & & & & & & + & @xmath2-bisq . & mve & & & & & & & + & @xmath2-bisq . & ksd & & & & & & & + & @xmath2-opt . & mve & & & & & & & + & @xmath2-opt . & ksd & & & & & & & + & rocke & mve & & & & & & & + & rocke & ksd & & & & & & & + & s - d & subs & & & & & & & + & s - d & ksd & & & & & & & + & ksd & & & & & & & & + & mcd & & 10.95 & 6.66 & 5.28 & & 1.87 & 1.20 & 0.94 + & mve & & & & & & & & + 0.5 & s - e & mve & & & & & & & + & s - e & ksd & & & & & & & + & mm - bisq & mve & & & & & & & + & mm - bisq . & ksd & & & & & & & + & mm - opt . & mve & & & & & & & + & mm - opt . & ksd & & & & & & & + & @xmath2-bisq . & mve & & & & & & & + & @xmath2-bisq . & ksd & & & & & & & + & @xmath2-opt . & mve & & & & & & & + & @xmath2-opt . & ksd & & & & & & & + & rocke & mve & & & & & & & + & rocke & ksd & & & & & & & + & s - d & subs & & & & & & & + & s - d & ksd & & & & & & & + & ksd & & & & & & & & + & mcd & & 5.43 & 2.70 & 1.72 & & 0.79 & 0.51 & 0.37 + & mve & & & & & & & & + [ c]cccccccccc@xmath156 & & start & & & + 0 & @xmath180 & & 50 & 100 & 200 & & 50 & 100 & 200 + & s - e & mve & & & & & & & + & s - e & ksd & & & & & & & + & mm - bisq . & mve & & & & & & & + & mm - bisq . & ksd & & & & & & & + & mm - opt . & mve & & & & & & & + & mm - opt . & ksd & & & & & & & + & @xmath2-bisq . & mve & & & & & & & + & @xmath2-bisq . & ksd & & & & & & & + & @xmath2-opt . & mve & & & & & & & + & @xmath2-opt . & ksd & & & & & & & + & rocke & mve & & & & & & & + & rocke & ksd & & & & & & & + & s - d & subs & 12.20 & 10.07 & 9.08 & & 3.13 & 2.88 & 2.89 + & s - d & ksd & 34.59 & 7.94 & 7.09 & & 6.02 & 2.35 & 2.18 + & ksd & & & & & & & & + & mcd & & 26.45 & 21.89 & 17.84 & & 10.50 & 9.21 & 6.82 + & mve & & & & & & & & + 0.5 & s - e & mve & & & & & & & + & s - e & ksd & & & & & & & + & mm - bisq . & mve & & & & & & & + & mm - bisq . & ksd & & & & & & & + & mm - opt . & mve & & & & & & & + & mm - opt . & ksd & & & & & & & + & @xmath2-bisq . & mve & & & & & & & + & @xmath2-bisq . & ksd & & & & & & & + & @xmath2-opt . & mve & & & & & & & + & @xmath2-opt . & ksd & & & & & & & + & rocke & mve & & & & & & & + & rocke & ksd & & & & & & & + & s - d & subs & & & & & & & + & s - d & ksd & & & & & & & + & ksd & & & & & & & & + & mcd & & 22.33 & 18.35 & 15.85 & & 9.24 & 7.19 & 7.09 + & mve & & & & & & & & + [ c]cccccccccc@xmath156 & & start & & & + & @xmath180 & & 75 & 150 & 300 & & 75 & 150 & 300 + 0 & s - e & mve & 6.93 & 6.68 & 6.35 & & 1.03 & 1.17 & 1.22 + & s - e & ksd & 7.15 & 6.68 & 6.43 & & 0.99 & 1.16 & 1.21 + & mm - bisq . & mve & 5.47 & 5.65 & 5.65 & & 0.88 & 1.13 & 1.23 + & mm - bisq . & ksd & 5.19 & 5.17 & 5.15 & & 0.81 & 1.04 & 1.08 + & mm - opt . & mve & 3.83 & 2.92 & 2.55 & & 0.50 & 0.48 & 0.46 + & mm - opt . & ksd & * 3.33 * & 2.38 & 2.25 & & 0.46 & 0.38 & 0.41 + & @xmath2-bisq . & mve & 7.42 & 7.91 & 7.98 & & 1.20 & 1.56 & 1.68 + & @xmath2-bisq . & ksd & 7.93 & 7.85 & 7.73 & & 1.68 & 1.52 & 1.47 + & @xmath2-opt . & mve & 4.11 & 3.23 & 2.73 & & 0.55 & 0.59 & 0.48 + & @xmath2-opt . & ksd & 3.69 & 2.98 & 2.63 & & 0.51 & 0.49 & 0.48 + & rocke & mve & 3.80 & 2.58 & 2.04 & & _ 0.43 _ & _ 0.34 _ & 0.31 + & rocke & ksd & _ 3.67 _ & * 1.95 * & _ 1.44 _ & & * 0.36 * & * 0.27 * & * 0.23 * + & s - d & subs . & & & & & & & + & s - d & ksd & & & & & & & + & ksd & & 8.08 & 4.19 & 1.66 & & 0.69 & 0.53 & 0.31 + & mcd & & 13.71 & 12.53 & 10.51 & & 2.78 & 2.06 & 1.99 + & mve & & 8.44 & 5.09 & 3.19 & & 0.73 & 0.55 & 0.32 + 0.5 & s - e & mve & 7.83 & 6.38 & 5.73 & & 0.98 & 1.02 & 0.93 + & s - e & ksd & 6.59 & 5.68 & 5.26 & & 0.86 & 0.93 & 0.92 + & mm - bisq . & mve & 8.61 & 5.29 & 4.71 & & 1.09 & 0.89 & 0.9 + & mm - bisq . & ksd & 4.84 & 4.58 & 4.29 & & 0.71 & 0.81 & 0.8 + & mm - opt . & mve & 7.55 & 4.16 & 3.33 & & 0.91 & 0.62 & 0.55 + & mm - opt . & ksd & _ 3.55 _ & 2.79 & 2.48 & & 0.50 & 0.46 & 0.45 + & @xmath2-bisq . & mve & 8.37 & 6.36 & 6.23 & & 1.06 & 1.09 & 1.12 + & @xmath2-bisq . & ksd & 6.05 & 6.17 & 6.17 & & 0.87 & 1.05 & 1.11 + & @xmath2-opt . & mve & 7.36 & 4.08 & 3.28 & & 0.89 & 0.62 & 0.55 + & @xmath2-opt . & ksd & 3.87 & 2.93 & 2.56 & & 0.53 & 0.51 & 0.47 + & rocke & mve & 7.69 & 4.07 & 3.21 & & 0.92 & 0.60 & 0.52 + & rocke & ksd & 4.03 & _ 2.16 _ & _ 1.68 _ & & 0.47 & _ 0.35 _ & _ 0.31 _ + & s - d & subs . & & & & & & & + & s - d & ksd & & & & & & & + & ksd & & 7.32 & 2.66 & 1.66 & & 0.69 & 0.43 & 0.32 + & mcd & & 9.10 & 7.06 & 5.95 & & 1.30 & 1,19 & 0.99 + & mve & & 15.05 & 8.54 & 5.92 & & 1.53 & 1.09 & 0.86 + [ c]cccccccccc@xmath156 & & start & & & + & @xmath180 & & 75 & 150 & 300 & & 75 & 150 & 300 + 0 & s - e & mve & & & & & & & + & s - e & ksd & & & & & & & + & mm - bisq . & mve & & & & & & & + & mm - bisq . & ksd & & & & & & & + & mm - opt & mve & & & & & & & + & mm - opt . & ksd & & & & & & & + & @xmath2-bisq . & mve & & & & & & & + & @xmath2-bisq . & ksd & & & & & & & + & @xmath2-opt . & mve & & & & & & & + & @xmath2-opt . & ksd & & & & & & & + & rocke & mve & & & & & & & + & rocke & ksd & & & & & & & + & s - d & subs . & 23.84 & 21.68 & 21.09 & & 6.90 & 6.68 & 6.99 + & s - d & ksd & 98.01 & 12.31 & 10.80 & & 12.54 & 3.97 & 3.74 + & ksd & & & & & & & & + & mcd & & 35.14 & 28.33 & 20.25 & & 13.28 & 10.30 & 8.06 + & mve & & & & & & & & + 0.5 & s - e & mve & & & & & & & + & s - e & ksd & & & & & & & + & mm - bisq . & mve & & & & & & & + & mm - bisq . & ksd & & & & & & & + & mm - opt . & mve & & & & & & & + & mm - opt . & ksd & & & & & & & + & @xmath2-bisq . & mve & & & & & & & + & @xmath2-bisq . & ksd & & & & & & & + & @xmath2-opt . & mve & & & & & & & + & @xmath2-opt . & ksd & & & & & & & + & rocke & mve & & & & & & & + & rocke & ksd & & & & & & & + & s - d & subs . & & & & & & & + & s - d & ksd & & & & & & & + & ksd & & & & & & & & + & mcd & & 30.80 & 22.85 & 17.43 & & 11.54 & 6.03 & 5.13 + & mve & & & & & & & & + [ c]cccccccccc@xmath156 & & start & & & + & @xmath180 & & 100 & 200 & 400 & & 100 & 200 & 400 + 0 & s - e & mve & & & & & & & + & s - e & ksd & & & & & & & + & mm - bisq . & mve & & & & & & & + & mm - bisq . & ksd & & & & & & & + & mm - opt . & mve & & & & & & & + & mm - opt . & ksd & & & & & & & + & @xmath2-bisq . & mve & & & & & & & + & @xmath2-bisq . & ksd & & & & & & & + & @xmath2-opt . & mve & & & & & & & + & @xmath2-opt . & ksd & & & & & & & + & rocke & mve & & & & & & & + & rocke & ksd & & & & & & & + & s - d & subs & & & & & & & + & s - d & ksd & & & & & & & + & ksd & & & & & & & & + & mcd & & 23.11 & 16.46 & 12.85 & & 3.70 & 2.64 & 1.93 + & mve & & & & & & & & + 0.5 & s - e & mve & & & & & & & + & s - e & ksd & & & & & & & + & mm - bisq . & mve & & & & & & & + & mm - bisq . & ksd & & & & & & & + & mm - opt . & mve & & & & & & & + & mm - opt . & ksd & & & & & & & + & @xmath2-bisq . & mve & & & & & & & + & @xmath2-bisq . & ksd & & & & & & & + & @xmath2-opt . & mve & & & & & & & + & @xmath2-opt . & ksd & & & & & & & + & rocke & mve & & & & & & & + & rocke & ksd & & & & & & & + & s - d & subs & & & & & & & + & s - d & ksd & & & & & & & + & ksd & & & & & & & & + & mcd & & 20.94 & 15.20 & 12.45 & & 3.41 & 2.42 & 1.86 + & mve & & & & & & & & + [ c]cccccccccc@xmath156 & & start & & & + & @xmath180 & & 100 & 200 & 400 & & 100 & 200 & 400 + 0 & s - e & mve & & & & & & & + & s - e & ksd & & & & & & & + & mm - bisq . & mve & & & & & & & + & mm - bisq . & ksd & & & & & & & + & mm - opt . & mve & & & & & & & + & mm - opt . & ksd & & & & & & & + & @xmath2-bisq . & mve & & & & & & & + & @xmath2-bisq . & ksd & & & & & & & + & @xmath2-opt . & mve & & & & & & & + & @xmath2-opt . & ksd & & & & & & & + & rocke & mve & & & & & & & + & rocke & ksd & & & & & & & + & s - d & subs & 34.50 & 32.90 & 31.81 & & 13.23 & 12.30 & 13.43 + & s - d & ksd & 92.02 & 17.09 & 15.55 & & 18.31 & 5.76 & 5.33 + & ksd & & & & & & & & + & mcd & & 39.35 & 32.04 & 25.62 & & 15.00 & 12.50 & 8.73 + & mve & & & & & & & & + 0.5 & s - e & mve & & & & & & & + & s - e & ksd & & & & & & & + & mm - bisq . & mve & & & & & & & + & mm - bisq . & ksd & & & & & & & + & mm - opt . & mve & & & & & & & + & mm - opt . & ksd & & & & & & & + & @xmath2-bisq . & mve & & & & & & & + & @xmath2-bisq . & ksd & & & & & & & + & @xmath2-opt . & mve & & & & & & & + & @xmath2-opt . & ksd & & & & & & & + & rocke & mve & & & & & & & + & rocke & ksd & & & & & & & + & s - d & subs & & & & & & & + & s - d & ksd & 17.34 & 13.67 & 12.44 & & & & + & ksd & & & & & & & & + & mcd & & 35.13 & 27.90 & 23.71 & & 13.99 & 10.24 & 7.70 + & mve & & & & & & & & + [ c]cccccccccc@xmath156 & & start & & & + & @xmath180 & & 150 & 300 & 600 & & 150 & 300 & 600 + 0 & s - e & mve & & & & & & & + & s - e & ksd & & & & & & & + & mm - bisq . & mve & & & & & & & + & mm - bisq . & ksd & & & & & & & + & mm - opt . & mve & & & & & & & + & mm - opt . & ksd & & & & & & & + & @xmath2-bisq . & mve & & & & & & & + & @xmath2-bisq . & ksd & & & & & & & + & @xmath2-opt . & mve & & & & & & & + & @xmath2-opt . & ksd & & & & & & & + & rocke & mve & & & & & & & + & rocke & ksd & & & & & & & + & s - d & subs & & & & & & & + & s - d & ksd & & & & & & & + & ksd & & & & & & & & + & mcd & & 22.47 & 17.66 & 12.42 & & 3.41 & 2.68 & 2.17 + & mve & & & & & & & & + 0.5 & s - e & mve & & & & & & & + & s - e & ksd & & & & & & & + & mm - bisq . & mve & & & & & & & + & mm - bisq . & ksd & & & & & & & + & mm - opt . & mve & & & & & & & + & mm - opt . & ksd & & & & & & & + & @xmath2-bisq . & mve & & & & & & & + & @xmath2-bisq . & ksd & & & & & & & + & @xmath2-opt . & mve & & & & & & & + & @xmath2-opt . & ksd & & & & & & & + & rocke & mve & & & & & & & + & rocke & ksd & & & & & & & + & s - d & subs . & & & & & & & + & s - d & ksd & & & & & & & + & ksd & & & & & & & & + & mcd & & 19.46 & 16.35 & 12.34 & & 2.99 & 2.47 & 1.80 + & mve & & & & & & & & + [ c]cccccccccc@xmath156 & & start & & & + & @xmath180 & & 150 & 300 & 600 & & 150 & 300 & 600 + 0 & s - e & mve & & & & & & & + & s - e & ksd & & & & & & & + & mm - bisq . & mve & & & & & & & + & mm - bisq . & ksd & & & & & & & + & mm - opt . & mve & & & & & & & + & mm - opt . & ksd & & & & & & & + & @xmath2-bisq . & mve & & & & & & & + & @xmath2-bisq . & ksd & & & & & & & + & @xmath2-opt . & mve & & & & & & & + & @xmath2-opt . & ksd & & & & & & & + & rocke & mve & & & & & & & + & rocke & ksd & & & & & & & + & s - d & subs . & & & & & & & + & s - d & ksd & & & & & & & + & ksd & & & & & & & & + & mcd & & & & & & & & + & mve & & & & & & & & + 0.5 & s - e & mve & & & & & & & + & s - e & ksd & & & & & & & + & mm - bisq . & mve & & & & & & & + & mm - bisq . & ksd & & & & & & & + & mm - opt . & mve & & & & & & & + & mm - opt . & ksd & & & & & & & + & @xmath2-bisq . & mve & & & & & & & + & @xmath2-bisq . & ksd & & & & & & & + & @xmath2-opt . & mve & & & & & & & + & @xmath2-opt . & ksd & & & & & & & + & rocke & mve & & & & & & & + & rocke & ksd & & & & & & & + & s - d & subs . & & & & & & & + & s - d & ksd & & & & & & & + & ksd & & & & & & & & + & mcd & & 42.37 & 29.51 & 22.62 & & 13.27 & 10.58 & 9.43 + & mve & & & & & & & & + for the rocke estimator , the value of @xmath61 in ( [ gama - alfa ] ) which yields 90% efficiency is approximated for @xmath190 by @xmath198 with@xmath199{cccc}\hline start & $ a$ & $ b$ & $ c$\\\hline mve & 0.00436 & -0.5030 & 0.4214\\ ksd & 0.00216 & -1.0078 & 0.8156\\\hline \end{tabular } \ \ \ ] ] finally , for the stahel - donoho estimator the weight function is @xmath201 where the values of @xmath71 that yield 90% efficiency are given by @xmath202 with@xmath199{lccc}\hline & $ a$ & $ b$ & $ c$\\\hline \multicolumn{1}{c}{subsampling } & \multicolumn{1}{r}{5.116 } & 63.820 & \multicolumn{1}{r}{2.213}\\ \multicolumn{1}{c}{ksd } & \multicolumn{1}{r}{6.564 } & 0.211 & \multicolumn{1}{r}{24.286}\\\hline \end{tabular } \ \ \ \ ] ] the ksd procedure is defined in the same way as the stahel - donoho estimator , but with a different set of directions . the population version is as follows . let @xmath204 be a random vector with distribution @xmath205 let @xmath206 be a set of directions @xmath207 with @xmath208 let @xmath127 and @xmath209 be univariate robust location and scale estimators . the outlyingness of a point @xmath210 is defined as @xmath211 the location and scale estimators are defined as weighted means and covariance matrix with weights @xmath212 where @xmath213 is a nonincreasing function for @xmath214 for a sample , the estimator is defined as above with @xmath215 the empirical distribution . in the ( theoretical ) stahel - donoho estimator , @xmath216 is the set of all directions , and @xmath17 is a smooth function ; in actual practice , a finite set of directions obtained by subsampling is employed . the ksd procedure employs two sets of directions : @xmath217 . the first one is deterministic , and consists of a set of @xmath0 orthogonal directions maximizing the kurtosis of @xmath218 and @xmath219 directions minimizing it . the other is a set of random specific directions obtained through a stratified sampling . we shall deal only with the first one . besides , @xmath17 is of hard rejection type : @xmath220 where @xmath59 depends on @xmath9 theoretical calculations with ksd seem extremely difficult , and for this reason we will limit ourselves to a very simplified case . we consider only the population case with point - mass contamination ; furthermore we assume that the uncontaminated data are elliptically distributed . it will be shown that if @xmath127 and @xmath209 have breakdown 0.5 , so has the ksd estimator . let @xmath221 be an elliptical distribution with fourth moments and consider the contaminated distribution @xmath222 with @xmath223 because of the estimator s equivariance it may be assumed that @xmath221 is radial , with zero means and identity covariance matrix , and that @xmath224 where @xmath225 are the elements of the canonical base and @xmath226 . put @xmath227 where @xmath228 is the first coordinate of @xmath229 the rotational symmetry implies that @xmath230 for all @xmath231 with @xmath208 we will show that the direction of the contamination , i.e. , @xmath232 is always included in the set . it is straightforward to show that the kurtosis of a projection @xmath218 under @xmath215 is @xmath233 where @xmath234 and@xmath235 it follows that @xmath236 depends on @xmath207 only through @xmath237.$ ] a laborious but straightforward calculation shows that the derivative of @xmath238 has the form @xmath239 where @xmath240 does not depend on @xmath152 and @xmath241 the location of the extrema depends only on the sign of @xmath242 although the result holds in general , to simplify the analysis we consider only the case @xmath243 . and we assume @xmath244 there are two cases . if @xmath245 then @xmath246 for @xmath247,$ ] and therefore @xmath248 is a minimizer of @xmath249 if @xmath250 there are maxima at @xmath251 and @xmath252 and therefore the set of maximizing directions contains @xmath253 and a set of orthogonal @xmath207 s which are orthogonal to @xmath253 it follows that @xmath254 note that @xmath255 and @xmath256 depend on @xmath152 but since @xmath257 they are bounded . therefore for @xmath154 large enough , @xmath258 will be larger than the cutoff value @xmath59 and will therefore have null weight . this finishes the proof . agostinelli , c. leung , a. , yohai , v.j . and zamar , r.h .. robust estimation of multivariate location and scatter in the presence of cellwise and casewise contamination doi 10.1007/s11749 - 015 - 0453 - 3 ( 2015 ) . rousseeuw p.j . multivariate estimation with high breakdown point . in : grossmann w. , pflug g. , vincze i. , and wertz w. ( eds . ) , mathematical statistics and applications , vol b , 283297 . reidel , dordrecht ( 1985 ) .

for a quantum system , a density matrix that is not pure can arise , via averaging , from a distribution of its wave function , a normalized vector belonging to its hilbert space . while itself does not determine a unique , additional facts , such as that the system has come to thermal equilibrium , might . it is thus not unreasonable to ask , which , if any , corresponds to a given thermodynamic ensemble ? to answer this question we construct , for any given density matrix , a natural measure on the unit sphere in , denoted . we do this using a suitable projection of the gaussian measure on with covariance . we establish some nice properties of and show that this measure arises naturally when considering macroscopic systems . in particular , we argue that it is the most appropriate choice for systems in thermal equilibrium , described by the canonical ensemble density matrix . may also be relevant to quantum chaos and to the stochastic evolution of open quantum systems , where distributions on are often used . key words : canonical ensemble in quantum theory ; probability measures on hilbert space ; gaussian measures ; density matrices .

quant-ph0309021

in classical mechanics , ensembles , such as the microcanonical and canonical ensembles , are represented by probability distributions on the phase space . in quantum mechanics , ensembles are usually represented by density matrices . it is natural to regard these density matrices as arising from probability distributions on the ( normalized ) wave functions associated with the thermodynamical ensembles , so that members of the ensemble are represented by a random state vector . there are , however , as is well known , many probability distributions which give rise to the same density matrix , and thus to the same predictions for experimental outcomes @xcite .. the measure that gives equal weight to these two states corresponds to the same density matrix as the one giving equal weight to @xmath5 and @xmath6 . however the physical situation corresponding to the former measure , a mixture of two grotesque superpositions , seems dramatically different from the one corresponding to the latter , a routine mixture . it is thus not easy to regard these two measures as physically equivalent . ] moreover , as emphasized by landau and lifshitz @xcite , the energy levels for macroscopic systems are so closely spaced ( exponentially small in the number of particles in the system ) that `` the concept of stationary states [ energy eigenfunctions ] becomes in a certain sense unrealistic '' because of the difficulty of preparing a system with such a sharp energy and keeping it isolated . landau and lifshitz are therefore wary of , and warn against , regarding the density matrix for such a system as arising solely from our lack of knowledge about the wave function of the system . we shall argue , however , that despite these caveats such distributions can be both useful and physically meaningful . in particular we describe here a novel probability distribution , to be associated with any thermal ensemble such as the canonical ensemble . while probability distributions on wave functions are natural objects of study in many contexts , from quantum chaos @xcite to open quantum systems @xcite , our main motivation for considering them is to exploit the analogy between classical and quantum statistical mechanics @xcite . this analogy suggests that some relevant classical reasonings can be transferred to quantum mechanics by formally replacing the classical phase space by the unit sphere @xmath7 of the quantum system s hilbert space @xmath2 . in particular , with a natural measure @xmath8 on @xmath7 one can utilize the notion of typicality , i.e. , consider properties of a system common to `` almost all '' members of an ensemble . this is a notion frequently used in equilibrium statistical mechanics , as in , e.g. , boltzmann s recognition that typical phase points on the energy surface of a macroscopic system are such that the empirical distribution of velocities is approximately maxwellian . once one has such a measure for quantum systems , one could attempt an analysis of the second law of thermodynamics in quantum mechanics along the lines of boltzmann s analysis of the second law in classical mechanics , involving an argument to the effect that the behavior described in the second law ( such as entropy increase ) occurs for typical states of an isolated macroscopic system , i.e. for the overwhelming majority of points on @xmath7 with respect to @xmath9 . probability distributions on wave functions of a composite system , with hilbert space @xmath2 , have in fact been used to establish the typical properties of the reduced density matrix of a subsystem arising from the wave function of the composite . for example , page @xcite considers the uniform distribution on @xmath7 for a finite - dimensional hilbert space @xmath2 , in terms of which he shows that the von neumann entropy of the reduced density matrix is typically nearly maximal under appropriate conditions on the dimensions of the relevant hilbert spaces . given a probability distribution @xmath1 on the unit sphere @xmath7 of the hilbert space @xmath2 there is always an associated density matrix @xmath10 @xcite : it is the density matrix of the mixture , or the statistical ensemble of systems , defined by the distribution @xmath1 , given by @xmath11 for any projection operator @xmath12 , @xmath13 is the probability of obtaining in an experiment a result corresponding to @xmath12 for a system with a @xmath1-distributed wave function . it is evident from that @xmath10 is the second moment , or covariance matrix , of @xmath1 , provided @xmath1 has mean 0 ( which may , and will , be assumed without loss of generality since @xmath14 and @xmath15 are equivalent physically ) . while a probability measure @xmath1 on @xmath7 determines a unique density matrix @xmath0 on @xmath2 via , the converse is not true : the association @xmath16 given by is many - to - one .- dimensional hilbert space the uniform probability distribution @xmath17 over the unit sphere has density matrix @xmath18 with @xmath19 the identity operator on @xmath2 ; at the same time , for every orthonormal basis of @xmath2 the uniform distribution over the basis ( which is a measure concentrated on just @xmath20 points ) has the same density matrix , @xmath21 . an exceptional case is the density matrix corresponding to a pure state , @xmath22 , as the measure @xmath1 with this density matrix is almost unique : it must be concentrated on the ray through @xmath14 , and thus the only non - uniqueness corresponds to the distribution of the phase . ] there is furthermore no unique `` physically correct '' choice of @xmath1 for a given @xmath0 since for any @xmath1 corresponding to @xmath0 one could , in principle , prepare an ensemble of systems with wave functions distributed according to this @xmath1 . however , while @xmath0 itself need not determine a unique probability measure , additional facts about a system , such as that it has come to thermal equilibrium , might . it is thus not unreasonable to ask : which measure on @xmath7 corresponds to a given thermodynamic ensemble ? let us start with the _ microcanonical _ ensemble , corresponding to the energy interval @xmath23 $ ] , where @xmath24 is small on the macroscopic scale but large enough for the interval to contain many eigenvalues . to this there is associated the spectral subspace @xmath25 , the span of the eigenstates @xmath26 of the hamiltonian @xmath27 corresponding to eigenvalues @xmath28 between @xmath29 and @xmath30 . since @xmath25 is finite dimensional , one can form the _ microcanonical density matrix _ @xmath31 with @xmath32}(h)$ ] the projection to @xmath25 . this density matrix is diagonal in the energy representation and gives equal weight to all energy eigenstates in the interval @xmath23 $ ] . but what is the corresponding _ microcanonical measure _ ? the most plausible answer , given long ago by schrdinger @xcite and bloch @xcite , is the ( normalized ) uniform measure @xmath33 on the unit sphere in this subspace . @xmath34 is associated with @xmath35 via . note that a wave function @xmath36 chosen at random from this distribution is almost certainly a nontrivial superposition of the eigenstates @xmath37 with random coefficients @xmath38 that are identically distributed , but not independent . the measure @xmath35 is clearly stationary , i.e. , invariant under the unitary time evolution generated by @xmath27 , and it is as spread out as it could be over the set @xmath39 of allowed wave functions . this measure provides us with a notion of a `` typical wave function '' from @xmath25 which is very different from the one arising from the measure @xmath40 that , when @xmath27 is nondegenerate , gives equal probability @xmath41 to every eigenstate @xmath42 with eigenvalue @xmath43 $ ] . the measure @xmath40 , which is concentrated on these eigenstates , is , however , less robust to small perturbations in @xmath27 than is the smoother measure @xmath35 . our proposal for the canonical ensemble is in the spirit of the uniform microcanonical measure @xmath35 and reduces to it in the appropriate cases . it is based on a mathematically natural family of probability measures @xmath1 on @xmath7 . for every density matrix @xmath0 on @xmath2 , there is a unique member @xmath1 of this family , satisfying for @xmath44 , namely the _ gaussian adjusted projected measure _ @xmath3 , constructed roughly as follows : eq . ( i.e. , the fact that @xmath10 is the covariance of @xmath1 ) suggests that we start by considering the gaussian measure @xmath45 with covariance @xmath0 ( and mean 0 ) , which could , in finitely many dimensions , be expressed by @xmath46 ( where @xmath47 is the obvious lebesgue measure on @xmath2 ) . this is not adequate , however , since the measure that we seek must live on the sphere @xmath7 whereas @xmath45 is spread out over all of @xmath2 . we thus adjust and then project @xmath45 to @xmath7 , in the manner described in section [ sec : defmeasure ] , in order to obtain the measure @xmath3 , having the prescribed covariance @xmath0 as well as other desirable properties . it is our contention that _ a quantum system in thermal equilibrium at inverse temperature @xmath48 should be described by a random state vector whose distribution is given by the measure @xmath49 associated with the density matrix for the canonical ensemble , _ @xmath50 in order to convey the significance of @xmath3 as well as the plausibility of our proposal that @xmath49 describes thermal equilibrium , we recall that a system described by a canonical ensemble is usually regarded as a subsystem of a larger system . it is therefore important to consider the notion of the distribution of the wave function of a subsystem . consider a composite system in a pure state @xmath51 , and ask what might be meant by the wave function of the subsystem with hilbert space @xmath52 . for this we propose the following . let @xmath53 be a ( generalized ) orthonormal basis of @xmath54 ( playing the role , say , of the eigenbasis of the position representation ) . for each choice of @xmath55 , the ( partial ) scalar product @xmath56 , taken in @xmath54 , is a vector belonging to @xmath52 . regarding @xmath55 as random , we are led to consider the random vector @xmath57 given by @xmath58 where @xmath59 is the normalizing factor and @xmath60 is a random element of the basis @xmath53 , chosen with the quantum distribution @xmath61 we refer to @xmath62 as the _ conditional wave function _ @xcite of system 1 . note that @xmath62 becomes doubly random when we start with a random wave function in @xmath63 instead of a fixed one . the distribution of @xmath62 corresponding to ( [ psi1def ] ) and ( [ marg ] ) is given by the probability measure on @xmath64 , @xmath65 where @xmath66 denotes the `` delta '' measure concentrated at @xmath67 . while the density matrix @xmath68 associated with @xmath69 always equals the reduced density matrix @xmath70 of system 1 , given by @xmath71 the measure @xmath72 itself usually depends on the choice of the basis @xmath73 . it turns out , nevertheless , as we point out in section [ sec : typicality ] , that @xmath74 is a universal function of @xmath70 in the special case that system 2 is large and @xmath14 is typical ( with respect to the uniform distribution on all wave functions with the same reduced density matrix ) , namely @xmath75 . thus @xmath3 has a distinguished , universal status among all probability measures on @xmath7 with density matrix @xmath0 . to further support our claim that @xmath49 is the right measure for @xmath76 , we shall regard , as is usually done , the system described by @xmath76 as coupled to a ( very large ) heat bath . the interaction between the heat bath and the system is assumed to be ( in some suitable sense ) negligible . we will argue that if the wave function @xmath14 of the combined `` system plus bath '' has microcanonical distribution @xmath35 , then the distribution of the conditional wave function of the ( small ) system is approximately @xmath77 ; see section [ sec : hb1 ] . indeed , a stronger statement is true . as we argue in section [ sec : hb2 ] , even for a typical _ fixed _ microcanonical wave function @xmath14 of the composite , i.e. , one typical for @xmath35 , the conditional wave function of the system , defined in , is then approximately @xmath49-distributed , for a typical basis @xmath53 . this is related to the fact that for a typical microcanonical wave function @xmath14 of the composite the reduced density matrix for the system is approximately @xmath76 @xcite . note that the analogous statement in classical mechanics would be wrong : for a fixed phase point @xmath78 of the composite , be it typical or atypical , the phase point of the system could never be random , but rather would merely be the part of @xmath78 belonging to the system . the remainder of this paper is organized as follows . in section [ sec : defmeasure ] we define the measure @xmath3 and obtain several ways of writing it . in section [ sec : prop ] we describe some natural mathematical properties of these measures , and suggest that these properties uniquely characterize the measures . in section [ sec : hb1 ] we argue that @xmath49 represents the canonical ensemble . in section [ sec : typicality ] we outline the proof that @xmath3 is the distribution of the conditional wave function for _ most _ wave functions in @xmath63 with reduced density matrix @xmath0 if system 2 is large , and show that @xmath49 is the typical distribution of the conditional wave function arising from a fixed microcanonical wave function of a system in contact with a heat bath . in section [ sec : rem ] we discuss other measures that have been or might be considered as the thermal equilibrium distribution of the wave function . finally , in section [ sec : two ] we compute explicitly the distribution of the coefficients of a @xmath49-distributed state vector in the simplest possible example , the two - level system . in this section , we define , for any given density matrix @xmath0 on a ( separable ) hilbert space @xmath79 the gaussian adjusted projected measure @xmath3 on @xmath7 . this definition makes use of two auxiliary measures , @xmath45 and @xmath80 , defined as follows . @xmath45 is the gaussian measure on @xmath2 with covariance matrix @xmath0 ( and mean 0 ) . more explicitly , let @xmath81 be an orthonormal basis of eigenvectors of @xmath0 and @xmath82 the corresponding eigenvalues , @xmath83 such a basis exists because @xmath0 has finite trace . let @xmath84 be a sequence of independent complex - valued random variables having a ( rotationally symmetric ) gaussian distribution in @xmath85 with mean @xmath86 and variance @xmath87 ( where @xmath88 means expectation ) , i.e. , @xmath89 and @xmath90 are independent real gaussian variables with mean zero and variance @xmath91 . we define @xmath45 to be the distribution of the random vector @xmath92 note that @xmath93 is not normalized , i.e. , it does not lie in @xmath7 . in order that @xmath93 lie in @xmath2 at all , we need that the sequence @xmath84 be square - summable , @xmath94 is finite . in fact , @xmath95 more generally , we observe that for any measure @xmath1 on @xmath2 with ( mean 0 and ) covariance given by the trace class operator @xmath96 , @xmath97 we have that , for a random vector @xmath36 with distribution @xmath1 , @xmath98 . it also follows that @xmath93 almost surely lies in the positive spectral subspace of @xmath0 , the closed subspace spanned by those @xmath26 with @xmath99 , or , equivalently , the orthogonal complement of the kernel of @xmath0 ; we shall call this subspace @xmath100 . note further that , since @xmath45 is the gaussian measure with covariance @xmath0 , it does not depend ( in the case of degenerate @xmath0 ) on the choice of the basis @xmath101 among the eigenbases of @xmath0 , but only on @xmath0 . since we want a measure on @xmath7 while @xmath45 is not concentrated on @xmath7 but rather is spread out , it would be natural to project @xmath45 to @xmath7 . however , since projecting to @xmath7 changes the covariance of a measure , as we will point out in detail in section [ sec : densitymatrix ] , we introduce an adjustment factor that exactly compensates for the change of covariance due to projection . we thus define the adjusted gaussian measure @xmath80 on @xmath2 by @xmath102 since @xmath103 by , @xmath80 is a probability measure . let @xmath104 be a @xmath80-distributed random vector . we define @xmath3 to be the distribution of @xmath105 with @xmath106 the projection to the unit sphere ( i.e. , the normalization of a vector ) , @xmath107 putting differently , for a subset @xmath108 , @xmath109 where @xmath110 denotes the cone through @xmath111 . more succinctly , @xmath112 where @xmath113 denotes the action of @xmath106 on measures . more generally , one can define for any measure @xmath1 on @xmath2 the `` adjust - and - project '' procedure : let @xmath114 be the adjusted measure @xmath115 ; then the adjusted - and - projected measure is @xmath116 , thus defining a mapping @xmath117 from the measures on @xmath2 with @xmath118 to the probability measures on @xmath7 . we then have that @xmath119 . we remark that @xmath120 , too , lies in @xmath100 almost surely , and that @xmath121 does _ not _ have distribution @xmath3nor covariance @xmath0 ( see sect . [ sec : densitymatrix ] ) . we can be more explicit in the case that @xmath0 has finite rank @xmath122 , e.g. for finite - dimensional @xmath2 : then there exists a lebesgue volume measure @xmath123 on @xmath124 , and we can specify the densities of @xmath45 and @xmath80 , [ mundensity ] @xmath125 with @xmath126 the restriction of @xmath0 to @xmath100 . similarly , we can express @xmath3 relative to the @xmath127dimensional surface measure @xmath128 on @xmath129 , [ mudensity ] @xmath130 we note that @xmath131 where @xmath132 is the microcanonical density matrix given in and @xmath133 is the microcanonical measure . in this section we prove the following properties of @xmath3 : * property 1 * _ the density matrix associated with @xmath3 in the sense of is @xmath0 , i.e. , @xmath134 . _ * property 2 * _ the association @xmath135 is _ covariant _ : for any unitary operator @xmath136 on @xmath2 , @xmath137 where @xmath138 is the adjoint of @xmath136 and @xmath139 is the action of @xmath136 on measures , @xmath140 . in particular , @xmath3 is stationary under any unitary evolution that preserves @xmath0 . _ * property 3 * _ if @xmath141 has distribution @xmath142 then , for any basis @xmath53 of @xmath54 , the conditional wave function @xmath62 has distribution @xmath143 . ( `` gap of a product density matrix has gap marginal . '' ) _ we will refer to the property expressed in property 3 by saying that the family of gap measures is _ we note that when @xmath141 has distribution @xmath3 and @xmath0 is not a tensor product , the distribution of @xmath62 need not be @xmath75 ( as we will show after the proof of property 3 ) . before establishing these properties let us formulate what they say about our candidate @xmath49 for the canonical distribution . as a consequence of property 1 , the density matrix arising from @xmath144 in the sense of is the density matrix @xmath145 . as a consequence of property 2 , @xmath49 is stationary , i.e. , invariant under the unitary time evolution generated by @xmath27 . as a consequence of property 3 , if @xmath146 has distribution @xmath147 and systems 1 and 2 are decoupled , @xmath148 , where @xmath149 is the identity on @xmath150 , then the conditional wave function @xmath62 of system @xmath151 has a distribution ( in @xmath52 ) of the same kind with the same inverse temperature @xmath48 , namely @xmath152 . this fits well with our claim that @xmath49 is the thermal equilibrium distribution since one would expect that if a system is in thermal equilibrium at inverse temperature @xmath48 then so are its subsystems . we conjecture that the family of gap measures is the only family of measures satisfying properties 13 . this conjecture is formulated in detail , and established for suitably continuous families of measures , in section [ sec : uniqueness ] . the following lemma , proven in section [ gmgm ] , is convenient for showing that a random wave function is gap - distributed : [ gaussgap ] let @xmath153 be a measurable space , @xmath1 a probability measure on @xmath153 , and @xmath154 a hilbert - space - valued function . if @xmath155 is @xmath45-distributed with respect to @xmath156 , then @xmath157 is @xmath3-distributed with respect to @xmath158 . in this subsection we establish property 1 . we then add a remark on the covariance matrix . of property 1 from we find that @xmath159 because @xmath160 is the covariance matrix of @xmath45 , which is @xmath0 . ( a number above an equal sign refers to the equation used to obtain the equality . ) * remark on the covariance matrix . * the equation @xmath134 can be understood as expressing that @xmath3 and @xmath45 have the same covariance . for a probability measure @xmath1 on @xmath2 with mean 0 that need not be concentrated on @xmath7 , the covariance matrix @xmath161 is given by @xmath162 suppose we want to obtain from @xmath1 a probability measure on @xmath7 having the same covariance . the projection @xmath163 of @xmath1 to @xmath7 , defined by @xmath164 for @xmath108 , is not what we want , as it has covariance @xmath165 however , @xmath166 does the job : it has the same covariance . as a consequence , a naturally distinguished measure on @xmath7 with given covariance is the gaussian adjusted projected measure , the gap measure , with the given covariance . we establish property 2 and then discuss in more general terms under which conditions a measure on @xmath7 is stationary . of property 2 under a unitary transformation @xmath136 , a gaussian measure with covariance matrix @xmath96 transforms into one with covariance matrix @xmath167 . since @xmath168 , @xmath169 transforms into @xmath170 ; that is , @xmath171 and @xmath172 are equal in distribution , and since @xmath173 , we have that @xmath174 and @xmath175 are equal in distribution . in other words , @xmath176 transforms into @xmath177 , which is what we claimed in . in this subsection we discuss a criterion for stationarity under the evolution generated by @xmath178 . consider the following property of a sequence of complex random variables @xmath84 : @xmath179 ( the phase @xmath180 exists when @xmath181 . ) condition implies that the distribution of the random vector @xmath182 is stationary , since @xmath183 . note also that implies that the distribution has mean 0 . we show that the @xmath184 have property . to begin with , the @xmath185 obviously have this property since they are independent gaussian variables . since the density of @xmath80 relative to @xmath45 is a function of the moduli alone , also the @xmath186 satisfy . finally , since the @xmath187 are functions of the @xmath188 while the phases of the @xmath189 equal the phases of the @xmath190 , also the @xmath184 satisfy . we would like to add that is not merely a sufficient , but also almost a necessary condition ( and _ morally _ a necessary condition ) for stationarity . since for any @xmath36 , the moduli latexmath:[$|z_n| = of @xmath36 takes place in the ( possibly infinite - dimensional ) torus @xmath192 contained in @xmath7 . independent uniform phases correspond to the uniform measure @xmath123 on @xmath193 . @xmath123 is the only stationary measure if the motion on @xmath194 is uniquely ergodic , and this is the case whenever the spectrum @xmath195 of @xmath27 is linearly independent over the rationals @xmath196 , i.e. , when every finite linear combination @xmath197 of eigenvalues with rational coefficients @xmath198 , not all of which vanish , is nonzero , see @xcite . this is true of generic hamiltonians , so that @xmath123 is generically the unique stationary distribution on the torus . but even when the spectrum of @xmath27 is linearly dependent , e.g. when there are degenerate eigenvalues , and thus further stationary measures on the torus exist , these further measures should not be relevant to thermal equilibrium measures , because of their instability against perturbations of @xmath27 @xcite . the stationary measure @xmath123 on @xmath193 corresponds , for given moduli @xmath199 or , equivalently , by setting @xmath200 for a given probability measure @xmath201 on the spectrum of @xmath27 , to a stationary measure @xmath202 on @xmath7 that is concentrated on the embedded torus . the measures @xmath202 are ( for generic @xmath27 ) the extremal stationary measures , i.e. , the extremal elements of the convex set of stationary measures , of which all other stationary measures are mixtures . lemma [ gaussgap ] is more or less immediate from the definition of @xmath3 . a more detailed proof looks like this : of lemma [ gaussgap ] by assumption the distribution @xmath203 of @xmath36 with respect to @xmath1 is @xmath45 . thus for the distribution of @xmath36 with respect to @xmath204 , we have @xmath205 . thus , @xmath206 has distribution @xmath207 . we have already remarked in the introduction that the orthonormal basis @xmath53 of @xmath54 , used in the definition of the conditional wave function , could be a _ generalized _ basis , such as a `` continuous '' basis , for which it is appropriate to write @xmath208 instead of the `` discrete '' notation @xmath209 we used in . we wish to elucidate this further . a generalized orthonormal basis @xmath210 indexed by the set @xmath211 is mathematically defined by a unitary isomorphism @xmath212 , where @xmath213 denotes a measure on @xmath211 . we can think of @xmath211 as the configuration space of system 2 ; as a typical example , system 2 may consist of @xmath214 particles in a box @xmath215 , so that its configuration space is @xmath216 with @xmath213 the lebesgue measure ( which can be regarded as obtained by combining @xmath214 copies of the volume measure on @xmath217 ) . was supposed to be the _ configuration _ , corresponding to the positions of the particles belonging to system 2 . for our purposes here , however , the physical meaning of the @xmath218 is irrelevant , so that any generalized orthonormal basis of @xmath219 can be used . ] the formal ket @xmath220 then means the delta function centered at @xmath221 ; it is to be treated as ( though strictly speaking it is not ) an element of @xmath54 . the definition of the conditional wave function @xmath62 then reads as follows : the vector @xmath222 can be regarded , using the isomorphism @xmath223 , as a function @xmath224 . is to be understood as meaning @xmath225 where @xmath226 is the normalizing factor and @xmath227 is a random point in @xmath211 , chosen with the quantum distribution @xmath228 which is how is to be understood in this setting . as @xmath14 is defined only up to changes on a null set in @xmath211 , @xmath62 may not be defined for a particular @xmath227 . its distribution in @xmath52 , however , is defined unambiguously by . in the most familiar setting with @xmath229 , we have that @xmath230 . in the following , we will allow generalized bases and use continuous instead of discrete notation , and set @xmath231 . of property 3 the proof is divided into four steps . _ we can assume that @xmath232 where @xmath104 is a @xmath80-distributed random vector in @xmath233 . we then have that @xmath234 where @xmath235 is the normalization in @xmath52 , and where the distribution of @xmath227 , given @xmath104 , is @xmath236 @xmath104 and @xmath227 have a joint distribution given by the following measure @xmath237 on @xmath238 : @xmath239 thus , what needs to be shown is that with respect to @xmath237 , @xmath240 is @xmath143-distributed . _ _ if @xmath141 is @xmath241-distributed and @xmath242 is fixed , then the random vector @xmath243 with @xmath244 is @xmath245-distributed . _ this follows , more or less , from the fact that a subset of a set of jointly gaussian random variables is also jointly gaussian , together with the observation that the covariance of @xmath246 is @xmath247 more explicitly , pick an orthonormal basis @xmath248 of @xmath150 consisting of eigenvectors of @xmath249 with eigenvalues @xmath250 , and note that the vectors @xmath251 form an orthonormal basis of @xmath233 consisting of eigenvectors of @xmath252 with eigenvalues @xmath253 . since the random variables @xmath254 are independent gaussian random variables with mean zero and variances @xmath255 , so are their linear combinations @xmath256 with variances ( because variances add when adding independent gaussian random variables ) @xmath257 thus @xmath258 is @xmath245-distributed , which completes step 2 . _ _ if @xmath141 is @xmath241-distributed and @xmath259 is random with any distribution , then the random vector @xmath260 is @xmath245-distributed . _ this is a trivial consequence of step 2 . _ step 4 . _ apply lemma [ gaussgap ] as follows . let @xmath261 , @xmath262 , and @xmath263 ( which means that @xmath221 and @xmath14 are independent ) . by step 3 , the hypothesis of lemma [ gaussgap ] ( for @xmath264 ) is satisfied , and thus @xmath265 is @xmath143-distributed with respect to @xmath266 where we have used that @xmath267 . but this is , according to step 1 , what we needed to show . to verify the statement after property 3 , consider the density matrix @xmath268 for a pure state @xmath269 of the form @xmath270 , where @xmath271 and @xmath272 are respectively orthonormal bases for @xmath52 and @xmath54 and the @xmath82 are nonnegative with @xmath273 . then a @xmath3-distributed random vector @xmath36 coincides with @xmath269 up to a random phase , and so @xmath274 . choosing for @xmath275 the basis @xmath272 , the distribution of @xmath62 is not @xmath75 but rather is concentrated on the eigenvectors of @xmath70 . when the @xmath82 are pairwise - distinct this measure is the measure @xmath276 we define in section [ sec : nu ] . in this section we use property 3 , i.e. , the fact that gap measures are hereditary , to show that @xmath49 is the distribution of the conditional wave function of a system coupled to a heat bath when the wave function of the composite is distributed microcanonically , i.e. , according to @xmath35 . consider a system with hilbert space @xmath52 coupled to a heat bath with hilbert space @xmath54 . suppose the composite system has a random wave function @xmath277 whose distribution is microcanonical , @xmath35 . assume further that the coupling is negligibly small , so that we can write for the hamiltonian @xmath278 and that the heat bath is large ( so that the energy levels of @xmath279 are very close ) . it is a well known fact that for macroscopic systems different equilibrium ensembles , for example the microcanonical and the canonical , give approximately the same answer for appropriate quantities . by this equivalence of ensembles @xcite , we should have that @xmath280 for suitable @xmath281 . then , since @xmath3 depends continuously on @xmath0 , we have that @xmath282 . thus we should have that the distribution of the conditional wave function @xmath62 of the system is approximately the same as would be obtained when @xmath36 is @xmath49-distributed . but since , by , the canonical density matrix is then of the form @xmath283 we have by property 3 that @xmath62 is approximately @xmath152-distributed , which is what we wanted to show . the previous section concerns the distribution of the conditional wave function @xmath62 arising from the microcanonical distribution of the wave function of the composite . it concerns , in other words , a _ random _ wave function of the composite . the result there is the analogue , on the level of measures on hilbert space , of the well known result that if a microcanonical density matrix is assumed for the composite , the reduced density matrix @xmath70 of the system , defined as the partial trace @xmath284 , is canonical if the heat bath is large @xcite . as indicated in the introduction , a stronger statement about the canonical density matrix is in fact true , namely that for a _ fixed _ ( nonrandom ) wave function @xmath14 of the composite which is typical with respect to @xmath35 , @xmath285 when the heat bath is large ( see @xcite ; for a rigorous study of special cases of a similar question , see @xcite ) . , the reduced density matrix becomes proportional to the identity on @xmath52 for typical wave functions relative to the uniform distribution on @xmath7 ( corresponding to @xmath35 for @xmath286 and @xmath287 ) . ] this stronger statement will be used in section [ sec : hb2 ] to show that a similar statement holds for the distribution of @xmath62 as well , namely that it is approximately @xmath152-distributed for a typical fixed @xmath288 and basis @xmath289 of @xmath54 . but we must first consider the distribution of @xmath62 for a typical @xmath290 . in this section we argue that for a typical wave function of a big system the conditional wave function of a small subsystem is approximately gap - distributed , first giving a precise formulation of this result and then sketching its proof . we give the detailed proof in @xcite . let @xmath291 , where @xmath52 and @xmath54 have respective dimensions @xmath20 and @xmath292 , with @xmath293 . for any given density matrix @xmath294 on @xmath52 , consider the set @xmath295 where @xmath296 is the reduced density matrix for the wave function @xmath14 . there is a natural notion of ( normalized ) uniform measure @xmath297 on @xmath298 ; we give its precise definition in section [ sec : typoutline ] . we claim that for fixed @xmath20 and large @xmath292 , the distribution @xmath299 of the conditional wave function @xmath62 of system 1 , defined by and for a basis @xmath289 of @xmath54 , is close to @xmath143 for the overwhelming majority , relative to @xmath297 , of vectors @xmath300 with reduced density matrix @xmath294 . more precisely : _ for every @xmath301 and every bounded continuous function @xmath302 , @xmath303 regardless of how the basis @xmath289 is chosen . _ here we use the notation @xmath304 it is important to resist the temptation to translate @xmath297 into a density matrix in @xmath2 . as mentioned in the introduction , to every probability measure @xmath1 on @xmath7 there corresponds a density matrix @xmath10 in @xmath2 , given by , which contains all the empirically accessible information about an ensemble with distribution @xmath1 . it may therefore seem a natural step to consider , instead of the measure @xmath305 , directly its density matrix @xmath306 , where @xmath307 is the identity on @xmath54 . but since our result concerns properties of most wave functions relative to @xmath1 , it can not be formulated in terms of the density matrix @xmath10 . in particular , the corresponding statement relative to another measure @xmath308 on @xmath7 with the same density matrix @xmath309 could be false . noting that @xmath10 has a basis of eigenstates that are product vectors , we could , for example , take @xmath310 to be a measure concentrated on these eigenstates . for any such state @xmath14 , @xmath299 is a delta - measure . the result follows , by , lemma [ gaussgap ] , and the continuity of @xmath117 , from the corresponding statement about the gaussian measure @xmath245 on @xmath52 with covariance @xmath294 : _ for every @xmath301 and every bounded continuous @xmath311 , _ @xmath312 _ where @xmath313 is the distribution of @xmath314 ( not normalized ) with respect to the uniform distribution of @xmath315 . _ we sketch the proof of and give the definition of @xmath297 . according to the schmidt decomposition , every @xmath300 can be written in the form @xmath316 where @xmath317 is an orthonormal basis of @xmath318 , @xmath319 an orthonormal system in @xmath54 , and the @xmath320 are coefficients which can be assumed real and nonnegative . from one reads off the reduced density matrix of system 1 , @xmath321 as the reduced density matrix is given , @xmath322 , the orthonormal basis @xmath323 and the coefficients @xmath320 are determined ( when @xmath294 is nondegenerate ) as the eigenvectors and the square - roots of the eigenvalues of @xmath294 , and , @xmath298 is in a natural one - to - one correspondence with the set @xmath324 of all orthonormal systems @xmath325 in @xmath54 of cardinality @xmath20 . ( if some of the eigenvalues of @xmath294 vanish , the one - to - one correspondence is with @xmath326 where @xmath327 . ) the haar measure on the unitary group of @xmath54 defines the uniform distribution on the set of orthonormal bases of @xmath54 , of which the uniform distribution on @xmath324 is a marginal , and thus defines the uniform distribution @xmath297 on @xmath298 . ( when @xmath294 is degenerate , @xmath297 does not depend upon how the eigenvectors @xmath328 of @xmath294 are chosen . ) the key idea for establishing from the schmidt decomposition is this : @xmath329 is the average of @xmath292 delta measures with equal weights , @xmath330 , located at the points @xmath331 now regard @xmath14 as random with distribution @xmath297 ; then the @xmath332 are @xmath292 random vectors , and @xmath333 is their empirical distribution . if the @xmath334 coefficients @xmath335 were _ independent gaussian ( complex ) random variables with ( mean zero and ) variance @xmath336 , then the @xmath332 would be @xmath292 independent drawings of a @xmath245-distributed random vector ; by the weak law of large numbers , their empirical distribution would usually be close to @xmath245 ; in fact , the probability that @xmath337 would converge to 1 , as @xmath338 . however , when @xmath339 is a random orthonormal system with uniform distribution as described above , the expansion coefficients @xmath335 in the decomposition of the @xmath340 s @xmath341 will not be independent since the @xmath340 s must be orthogonal and since @xmath342 . nonetheless , replacing the coefficients @xmath335 in by independent gaussian coefficients @xmath343 as described above , we obtain a system of vectors @xmath344 that , in the limit @xmath345 , form a uniformly distributed orthonormal system : @xmath346 ( by the law of large numbers ) and @xmath347 for @xmath348 ( since a pair of randomly chosen vectors in a high - dimensional hilbert space will typically be almost orthogonal ) . this completes the proof . while this result _ suggests _ that @xmath49 is the distribution of the conditional wave function of a system coupled to a heat bath when the wave function of the composite is a typical _ fixed _ microcanonical wave function , belonging to @xmath25 , it does not quite _ imply _ it . the reason for this is that @xmath25 has measure 0 with respect to the uniform distribution on @xmath79 even when the latter is finite - dimensional . nonetheless , there is a simple corollary , or reformulation , of the result that will allow us to cope with microcanonical wave functions . we have indicated that for our result the choice of basis @xmath289 of @xmath54 does not matter . in fact , while @xmath299 , the distribution of the conditional wave function @xmath62 of system 1 , depends upon both @xmath349 and the choice of basis @xmath289 of @xmath54 , the distribution of @xmath299 itself , when @xmath14 is @xmath297-distributed , does not depend upon the choice of basis . this follows from the fact that for any unitary @xmath136 on @xmath54 @xmath350 ( and the invariance of the haar measure of the unitary group of @xmath54 under left multiplication ) . it similarly follows from that for fixed @xmath349 , the distribution of @xmath299 arising from the uniform distribution @xmath237 of the basis @xmath289 , in the set @xmath351 of all orthonormal bases of @xmath54 , is the same as the distribution of @xmath299 arising from the uniform distribution @xmath297 of @xmath14 with a fixed basis ( and the fact that the haar measure is invariant under @xmath352 ) . we thus have the following corollary : _ let @xmath349 and let @xmath353 be the corresponding reduced density matrix for system 1 . then for a typical basis @xmath289 of @xmath54 , the conditional wave function @xmath62 of system 1 is approximately @xmath143-distributed when @xmath292 is large : for every @xmath301 and every bounded continuous function @xmath302 , @xmath354 _ it is an immediate consequence of the result of section [ sec : ref ] that for any fixed microcanonical wave function @xmath14 for a system coupled to a ( large ) heat bath , the conditional wave function @xmath62 of the system will be approximately gap - distributed . when this is combined with the `` canonical typicality '' described near the beginning of section [ sec : typicality ] , we obtain the following result : _ consider a system with finite - dimensional hilbert space @xmath52 coupled to a heat bath with finite - dimensional hilbert space @xmath54 . suppose that the coupling is weak , so that we can write @xmath355 on @xmath356 , and that the heat bath is large , so that the eigenvalues of @xmath279 are close . then for any wave function @xmath14 that is typical relative to the microcanonical measure @xmath35 , the distribution @xmath299 of the conditional wave function @xmath62 , defined by and for a typical basis @xmath289 of the heat bath , is close to @xmath49 for suitable @xmath357 , where @xmath358 . in other words , in the thermodynamic limit , in which the volume @xmath359 of the heat bath and @xmath360 go to infinity and @xmath361 is constant , we have that for all @xmath362 , and for all bounded continuous functions @xmath363 , @xmath364 where @xmath365 . _ we note that if @xmath289 were an energy eigenbasis rather than a typical basis , the result would be false . we review in this section other distributions that have been , or may be , considered as possible candidates for the distribution of the wave function of a system from a canonical ensemble . one possibility , which goes back to von neumann @xcite , is to consider @xmath9 as concentrated on the eigenvectors of @xmath0 ; we denote this distribution @xmath366 after the first letters of `` eigenvector '' ; it is defined as follows . suppose first that @xmath0 is nondegenerate . to select an @xmath366-distributed vector , pick a unit eigenvector @xmath42 , so that @xmath367 , with probability @xmath82 and randomize its phase . this definition can be extended in a natural way to degenerate @xmath0 : @xmath368 where @xmath369 denotes the eigenspace of @xmath0 associated with eigenvalue @xmath201 . the measure @xmath366 is concentrated on the set @xmath370 of eigenvectors of @xmath0 , which for the canonical @xmath371 coincides with the set of eigenvectors of @xmath27 ; it is a mixture of the microcanonical distributions @xmath372 on the eigenspaces of @xmath27 in the same way as in classical mechanics the canonical distribution on phase space is a mixture of the microcanonical distributions . note that @xmath373 , and that in particular @xmath374 is not , when @xmath27 is nondegenerate , the uniform distribution @xmath40 on the energy eigenstates with energies in @xmath375 $ ] , against which we have argued in the introduction . the distribution @xmath366 has the same properties as those of @xmath3 described in properties 13 , except when @xmath0 is degenerate : _ the measures @xmath366 are such that ( a ) they have the right density matrix : @xmath376 ; ( b ) they are covariant : @xmath377 ; ( c ) they are hereditary at nondegenerate @xmath0 : when @xmath378 and @xmath0 is nondegenerate and uncorrelated , @xmath379 , then @xmath366 has marginal ( i.e. , distribution of the conditional wave function ) @xmath380 . _ \(a ) and ( b ) are obvious . for ( c ) let , for @xmath381 , @xmath382 be a basis consisting of eigenvectors of @xmath249 with eigenvalues @xmath250 . note that the tensor products @xmath383 are eigenvectors of @xmath0 with eigenvalues @xmath384 , and by nondegeneracy all eigenvectors of @xmath0 are of this form up to a phase factor . since an @xmath366-distributed random vector @xmath36 is almost surely an eigenvector of @xmath0 , we have latexmath:[$\psi = e^{i\theta } @xmath214 , and @xmath386 . the conditional wave function @xmath62 is , up to the phase , the eigenvector @xmath387 of @xmath294 occurring as the first factor in @xmath36 . the probability of obtaining @xmath388 is @xmath389 . and @xmath390 are multiplicatively independent , in the sense that @xmath391 can occur only trivially , i.e. , when @xmath392 and @xmath393 . in particular , the nondegeneracy of @xmath294 and @xmath390 is irrelevant . ] in contrast , for a _ degenerate _ @xmath379 the conditional wave function need not be @xmath380-distributed , as the following example shows . suppose @xmath294 and @xmath390 are nondegenerate but @xmath394 for some @xmath395 ; then an @xmath366-distributed @xmath36 , whenever it happens to be an eigenvector associated with eigenvalue @xmath384 , is of the form @xmath396 , almost surely with nonvanishing coefficients @xmath397 and @xmath398 ; as a consequence , the conditional wave function is a multiple of @xmath399 , which is , for typical @xmath227 and unless @xmath400 and @xmath401 have disjoint supports , a nontrivial superposition of eigenvectors @xmath402 , @xmath403 with different eigenvalues and thus can not arise from the @xmath380 distribution . also in the case of the degeneracy of @xmath404 : if the orthonormal basis @xmath289 used in the definition of conditional wave function consists of eigenvectors of @xmath390 , then the distribution of the conditional wave function is @xmath380 . ] note also that @xmath366 is discontinuous as a function of @xmath0 at every degenerate @xmath0 ; in other words , @xmath405 is , like @xmath40 , unstable against small perturbations of the hamiltonian . ( and , as with @xmath40 , this fact , quite independently of the considerations on behalf of gap - measures in sections [ sec : hb1 ] and [ sec : typicality ] , suggests against using @xmath406 as a thermal equilibrium distribution . ) moreover , @xmath366 is highly concentrated , generically on a one - dimensional subset of @xmath7 , and in the case of a finite - dimensional hilbert space @xmath2 fails to be absolutely continuous relative to the uniform distribution @xmath407 on the unit sphere . for further discussion of families @xmath408 of measures satisfying the analogues of properties 13 , see section [ sec : uniqueness ] . here is another distribution on @xmath2 associated with the density matrix @xmath0 . let the random vector @xmath36 be @xmath409 the @xmath410 being independent random vectors with distributions @xmath372 . in case all eigenvalues are nondegenerate , this means the coefficients @xmath84 of @xmath36 , @xmath411in sharp contrast with the moduli when @xmath36 is @xmath3-distributed . and in contrast to the measure @xmath366 considered in the previous subsection , the weights @xmath82 in the density matrix now come from the fixed size of the coefficients of @xmath36 when it is decomposed into the eigenvectors of @xmath0 , rather than from the probability with which these eigenvectors are chosen . this measure , too , is stationary under any unitary evolution that leaves @xmath0 invariant . in particular , it is stationary in the thermal case @xmath371 , and for generic @xmath27 it is an extremal stationary measure as characterized in section [ sec : stationarity ] ; in fact it is , in the notation of the last paragraph of section [ sec : stationarity ] , @xmath202 with @xmath412 . this measure , too , is highly concentrated : for a hilbert space @xmath2 of finite dimension @xmath20 , it is supported by a submanifold of real dimension @xmath413 where @xmath292 is the number of distinct eigenvalues of @xmath27 , hence generically it is supported by a submanifold of just half the dimension of @xmath2 . in @xcite , guerra and loffredo consider the canonical density matrix @xmath76 for the one - dimensional harmonic oscillator and want to associate with it a diffusion process on the real line , using stochastic mechanics @xcite . since stochastic mechanics associates a process with every wave function , they achieve this by finding a measure @xmath414 on @xmath415 whose density matrix is @xmath76 . they propose the following measure @xmath414 , supported by coherent states . with every point @xmath416 in the classical phase space @xmath417 of the harmonic oscillator there is associated a coherent state @xmath418 with @xmath419 , thus defining a mapping @xmath420 , @xmath421 . let @xmath422 be the classical hamiltonian function , and consider the classical canonical distribution at inverse temperature @xmath423 , @xmath424 let @xmath425 then @xmath426 is the distribution on coherent states arising from @xmath427 . the density matrix of @xmath414 is @xmath76 @xcite . this measure is concentrated on a 2-dimensional submanifold of @xmath415 , namely on the set of coherent states ( the image of @xmath96 ) . note also that not every density matrix @xmath0 on @xmath428 can arise as the density matrix of a distribution on the set of coherent states ; for example , a pure state @xmath429 can arise in this way if and only if @xmath14 is a coherent state . in a similar spirit , one may consider , on a finite - dimensional hilbert space @xmath2 , the distribution @xmath430 that maximizes the gibbs entropy functional @xmath431 = - \int\limits_{{\mathscr{s}}({\mathscr{h } } ) } \!\!\ ! u(d\psi ) \:\ : f(\psi ) \ , \log f(\psi)\ ] ] under the constraints that @xmath432 be a probability distribution with mean 0 and covariance @xmath433 : @xmath434 a standard calculation using lagrange multipliers leads to @xmath435 with @xmath436 a self - adjoint matrix determined by and ; comparison with shows that @xmath432 is not a gap measure . ( we remark , however , that another gibbs entropy functional , @xmath437 = -\int_{\mathscr{h}}\lambda(d\psi ) \ : f(\psi ) \ , \log f(\psi)$ ] , based on the lebesgue measure @xmath123 on @xmath2 instead of @xmath407 , is maximized , under the constraints that the mean be 0 and the covariance be @xmath0 , by the gaussian measure , @xmath438 . ) there is no apparent reason why the family of @xmath432 measures should be hereditary . the situation is different for the microcanonical ensemble : here , the distribution @xmath439 that we propose is in fact the maximizer of the appropriate gibbs entropy functional @xmath440 . which functional is that ? since any measure @xmath441 on @xmath7 whose covariance matrix is the projection @xmath442}(h)$ ] must be concentrated on the subspace @xmath25 and thus can not be absolutely continuous ( possess a density ) relative to @xmath407 , we consider instead its density relative to @xmath443 , that is , we consider @xmath444 and set @xmath445 = - \int\limits_{{\mathscr{s}}({\mathscr{h}}_{e , \delta } ) } u_{e , \delta } ( d\psi ) \ , f(\psi ) \ , \log f(\psi ) \,.\ ] ] under the constraints that the probability measure @xmath432 have mean 0 and covariance @xmath446 , @xmath447 $ ] is maximized by @xmath448 , or @xmath449 ; in fact even without the constraints on @xmath432 , @xmath447 $ ] is maximized by @xmath450 . brody and hughston @xcite have proposed the following distribution @xmath1 to describe thermal equilibrium . they observe that the projective space arising from a finite - dimensional hilbert space , endowed with the dynamics arising from the unitary dynamics on hilbert space , can be regarded as a classical hamiltonian system with hamiltonian function @xmath451 ( and symplectic form arising from the hilbert space structure ) . they then define @xmath1 to be the classical canonical distribution of this hamiltonian system , i.e. , to have density proportional to @xmath452 relative to the uniform volume measure on the projective space ( which can be obtained from the symplectic form or , alternatively , from @xmath407 by projection from the sphere to the projective space ) . however , this distribution leads to a density matrix , different from the usual one @xmath76 given by , that does not describe the canonical ensemble . as @xmath366 is a family of measures satisfying properties 13 for _ most _ density matrices @xmath0 , the question arises whether there is any family of measures , besides @xmath3 , satisfying these properties for _ all _ density matrices . we expect that the answer is no , and formulate the following uniqueness conjecture : _ given , for every hilbert space @xmath2 and every density matrix @xmath0 on @xmath2 , a probability measure @xmath408 on @xmath7 such that properties 13 remain true when @xmath3 is replaced by @xmath408 , then @xmath453 . _ in other words , we conjecture that @xmath454 is the only hereditary covariant inverse of . this is in fact true when we assume in addition that the mapping @xmath455 is suitably continuous . here is the argument : when @xmath0 is a multiple of a projection , @xmath456 for a subspace @xmath457 , then @xmath408 must be , by covariance @xmath458 , the uniform distribution on @xmath459 , and thus @xmath460 in this case . consider now a composite of a system ( system 1 ) and a large heat bath ( system 2 ) with hilbert space @xmath461 and hamiltonian @xmath462 , and consider the microcanonical density matrix @xmath34 for this system . by equivalence of ensembles , we have for suitable @xmath463 that @xmath464 where @xmath465 . by the continuity of @xmath1 and @xmath466 , @xmath467 now consider , for a wave function @xmath36 with distribution @xmath468 respectively @xmath469 , the distribution of the conditional wave function @xmath62 : by heredity , this is @xmath470 respectively @xmath471 . since the distribution of @xmath62 is a continuous function of the distribution of @xmath36 , we thus have that @xmath472 . since we can make the degree of approximation arbitrarily good by making the heat bath sufficiently large , we must have that @xmath473 . for any density matrix @xmath0 on @xmath52 that does not have zero among its eigenvalues , there is an @xmath474 such that @xmath475 for @xmath476 , and thus we have that @xmath453 for such a @xmath0 ; since these are dense , we have that @xmath453 for all density matrices @xmath0 on @xmath52 . since @xmath52 is arbitrary we are done . markov processes in hilbert space have long been considered ( see @xcite for an overview ) , particularly diffusion processes and piecewise deterministic ( jump ) processes . this is often done for the purpose of numerical simulation of a master equation for the density matrix , or as a model of continuous measurement or of spontaneous wave function collapse . such processes could arise as follows . since the conditional wave function @xmath62 arises from the wave function @xmath56 by inserting a random coordinate @xmath227 for the second variable ( and normalizing ) , any dynamics ( i.e. , time evolution ) for @xmath227 , described by a curve @xmath477 and preserving the quantum probability distribution of @xmath227 , for example , as given by bohmian mechanics @xcite , gives rise to a dynamics for the conditional wave function , @xmath478 , where @xmath479 evolves according to schrdinger s equation and @xmath480 is the normalizing factor . in this way one obtains a stochastic process ( a random path ) in @xmath64 . in the case considered in section [ sec : hb1 ] , in which @xmath54 corresponds to a large heat bath , this process must have @xmath481 as an invariant measure . it would be interesting to know whether this process is approximately a simple process in @xmath64 , perhaps a diffusion process , perhaps one of the markov processes on hilbert space considered already in the literature . in this last section , we consider a two - level system , with @xmath482 and @xmath483 and calculate the joint distribution of the energy coefficients @xmath484 and @xmath485 for a @xmath486-distributed @xmath36 as explicitly as possible . we begin with a general finite - dimensional system , @xmath487 , and specialize to @xmath488 later . one way of describing the distribution of @xmath36 is to give its density relative to the hypersurface area measure @xmath128 on @xmath489 ; this we did in . another way of describing the joint distribution of the @xmath84 is to describe the joint distribution of their moduli @xmath199 , or of @xmath490 , as the phases of the @xmath84 are independent ( of each other and of the moduli ) and uniformly distributed , see . for greater clarity , from now on we write @xmath495 instead of @xmath84 . a relation similar to that between @xmath3 , @xmath80 , and @xmath45 holds between the joint distributions of the @xmath496 , of the @xmath497 , and of the @xmath498 . the joint distribution of the @xmath498 is very simple : they are independent and exponentially distributed with means @xmath499 . since the density of @xmath500 relative to @xmath501 , @xmath502 , is a function of the moduli alone , and since , according to , @xmath503 , we have that @xmath504 thus , @xmath505 where each @xmath506 . finally , the @xmath496 arise by normalization , @xmath507 we now specialize to the two - level system , @xmath488 . since @xmath508 , it suffices to determine the distribution of @xmath509 , for which we give an explicit formula in below . we want to obtain the marginal distribution of from the joint distribution of the @xmath497 in @xmath510 , the first quadrant of the plane , as given by . to this end , we introduce new coordinates in the first quadrant : @xmath511where @xmath512 and @xmath513 . conversely , we have @xmath514 and @xmath515 , and the area element transforms according to @xmath516 with @xmath519 and @xmath520 for @xmath521 . the density @xmath522 of the distribution of @xmath509 is depicted in figure 1 for various values of @xmath24 . for @xmath523 , @xmath522 is identically 1 . for @xmath524 , we have @xmath525 , so that @xmath526 is decreasing monotonically from @xmath527 at @xmath528 to @xmath529 at @xmath530 ; hence , @xmath522 is increasing monotonically from @xmath531 to @xmath532 . for @xmath533 , we have @xmath534 , and hence @xmath522 is decreasing monotonically from @xmath531 to @xmath532 . in all cases @xmath522 is convex since @xmath535 . _ acknowledgments . _ we thank andrea viale ( universit di genova , italy ) for preparing the figure , eugene speer ( rutgers university , usa ) for comments on an earlier version , james hartle ( uc santa barbara , usa ) and hal tasaki ( gakushuin university , tokyo , japan ) for helpful comments and suggestions , eric carlen ( georgia institute of technology , usa ) , detlef drr ( lmu mnchen , germany ) , raffaele esposito ( universit di laquila , italy ) , rossana marra ( universit di roma `` tor vergata '' , italy ) , and herbert spohn ( tu mnchen , germany ) for suggesting references , and juan diego urbina ( the weizmann institute , rehovot , israel ) for bringing the connection between gaussian random wave models and quantum chaos to our attention . we are grateful for the hospitality of the institut des hautes tudes scientifiques ( bures - sur - yvette , france ) , where part of the work on this paper was done . the work of s. goldstein was supported in part by nsf grant dms-0504504 , and that of j. lebowitz by nsf grant dmr 01 - 279 - 26 and afosr grant af 49620 - 01 - 1 - 0154 . the work of r. tumulka was supported by infn and by the european commission through its 6th framework programme `` structuring the european research area '' and the contract nr . rita - ct-2004 - 505493 for the provision of transnational access implemented as specific support action . the work of n. zangh was supported by infn . lebowitz , j. l. : microscopic reversibility and macroscopic behavior : physical explanations and mathematical derivations . in _ 25 years of non - equilibrium statistical mechanics _ , proceedings , sitges conference , barcelona , spain , 1994 , in _ lecture notes in physics _ , j.j . brey , j. marro , j.m . rub , and m. san miguel ( eds . ) ( springer - verlag , berlin , 1995 ) . tasaki , h. : from quantum dynamics to the canonical distribution : general picture and a rigorous example . lett . _ * 80 * , 13731376 ( 1998 ) . urbina , j. d. and richter , k. : random wave models . _ elsevier encyclopedia of mathematical physics _ ( 2006 ) . to appear . von neumann , j. : _ mathematical foundations of quantum mechanics _ ( princeton university press , princeton , 1955 ) . translation of _ mathematische grundlagen der quantenmechanik _ ( springer - verlag , berlin , 1932 ) .

an effective - mass theory with a deformation - induced ( an axial ) gauge field is proposed as a theoretical framework to study graphene edge . though the gauge field is singular at edge , it can represent the boundary condition and this framework is adopted to solve the scattering problems for the zigzag and armchair edges . furthermore , we solve the scattering problem in the presence of a mass term and an electromagnetic field . it is shown that the mass term makes the standing wave at the dirac point avoid the zigzag edge , by which the local density of states disappears , and the lowest and first landau states are special near the zigzag edge . the ( chiral ) gauge theory framework provides a useful description of graphene edge .

1003.5036

the graphene edge has attracted much attention , @xcite because it is the source of a wide variety of notable phenomena . for example , the zigzag edge possesses localized edge states . @xcite the edge states enhance the local density of states near the fermi energy . @xcite as a result , the spins of the edge states may be polarized by coulombic interaction . @xcite another type of edge , the armchair edge , does not support edge states . the zigzag edge is the source of intravalley scattering , while the armchair edge gives rise to intervalley scattering . the transport properties near the armchair edge may differ significantly from that near the zigzag edge ; @xcite however , the reason for this variety is unclear . the schrdinger equation is a differential equation ; therefore , an appropriate boundary condition should be imposed on the equation . the boundary condition is sensitive to the situation of the edge , while the local dynamics , as described by the schrdinger equation , are the same everywhere in a graphene sample . the wave function and energy spectrum are dependent on the boundary condition . in this sense , the boundary condition is the origin of the variety . @xcite in this paper , we attempt to construct a theoretical framework in which the edge is taken into account as a gauge field , and not as a boundary condition for the wave function . we show that the framework is useful to obtain and understand the standing wave and edge states . this paper is organized as follows . in sec . [ sec : pspin ] , the qualitative features of the reflections from the zigzag and armchair edges are shown using the kinematics for elastic scattering . in sec . [ sec : model ] a general form of the electronic hamiltonian is given for a graphene sheet with edges . in secs . [ sec : zig ] and [ sec : arm ] , the scattering problem is solved for both the zigzag and armchair edges , and the standing wave solution is obtained . a discussion and summary are given in sec . [ sec : dis ] . in the inset of fig . [ fig : pspin ] , we consider the zigzag edge parallel to the @xmath0-axis , by which translational symmetry along the @xmath1-axis is broken . thus , the incident state with wave vector @xmath2 is elastically scattered by the zigzag edge , and the wave vector of the reflected state becomes @xmath3 . in contrast , the armchair edge parallel to the @xmath1-axis breaks the translational symmetry along the @xmath0-axis , so that the wave vector of the reflected state is @xmath4 . the brillouin zone ( bz ) is given by 90@xmath5 rotation of the hexagonal lattice , so that for the incident state near the k point in fig . [ fig : pspin ] , the zigzag edge reflected state is also near the k point , while the armchair edge reflected state is near the k@xmath6 point . therefore , scattering by the zigzag edge is intravalley scattering , while that by the armchair edge is intervalley scattering . points . appendix [ app : rot - pspin ] outlines why the pseudospins at the ( three ) equivalent k ( k@xmath6 ) points are not identical . [ inset - top ] the hexagonal unit cell of graphene consists of a ( solid circle ) and b ( open circle ) atoms . the @xmath7 coordinate system is fixed as shown . the vectors @xmath8 and @xmath9 are primitive translations . the length of each of these is @xmath10 ( @xmath11 , where @xmath12 is the c - c bond length ) . [ inset - middle ] the vectors @xmath13 and @xmath14 are reciprocal lattice vectors defined by @xmath15 . [ inset - bottom ] the vectors @xmath16 are expressed as @xmath17 , @xmath18 , and @xmath19 , where @xmath20 ( @xmath21 ) is the dimensionless unit vector for the @xmath0-axis ( @xmath1-axis ) . ] pseudospin is defined as the expected value of the pauli matrices @xmath22 with respect to the two component bloch function . the pseudospin provides information concerning the relative phase and the relative amplitude between the two components of the bloch function , and it can be used to characterize scattering at the edges . @xcite the bloch function of the conduction state with wave vector @xmath23 is given by @xmath24 where @xmath25 , @xmath26 denotes the complex conjugate of @xmath27 , and @xmath16 ( @xmath28 ) are the vectors pointing to the nearest - neighbor b atoms from an a atom [ see the inset of fig . [ fig : pspin ] ] . the pseudospin is then given by @xmath29}{|f_{\bf k}| } , \ \ \langle \sigma_y \rangle_{\bf k } = \frac{{\rm im}[f_{\bf k}]}{|f_{\bf k}| } , \ \ \langle \sigma_z \rangle_{\bf k } = 0 , \label{eq : pspin2d}\end{aligned}\ ] ] where @xmath30 = \cos \left(\frac{k_y a}{\sqrt{3 } } \right ) + 2\cos\left(\frac{k_y a}{2\sqrt{3}}\right ) \cos\left(\frac{k_x a}{2}\right ) , \\ & { \rm im}[f_{\bf k } ] = \sin \left(\frac{k_y a}{\sqrt{3 } } \right ) - 2\sin\left(\frac{k_y a}{2\sqrt{3}}\right ) \cos\left(\frac{k_x a}{2}\right ) . \end{split } \label{eq : ref - imf}\end{aligned}\ ] ] the pseudospin , @xmath31 , may be regarded as a two - dimensional vector field , because @xmath32 . the arrows in fig . [ fig : pspin ] show the pseudospin field , @xmath33 . @xmath34 is proportional to @xmath35 $ ] ; therefore , the angle of each arrow with respect to the @xmath36-axis represents the relative phase of the bloch function between a and b atoms . for example , at the @xmath37 point @xmath38 in fig . [ fig : pspin ] , the arrow is pointing toward the negative @xmath36-axis . this implies that the wave function forms an antisymmetric combination with respect to the a and b atoms , which can be checked by setting @xmath38 in eq . ( [ eq : bloch ] ) . since @xmath34 is an odd function of @xmath39 as shown in eq . ( [ eq : ref - imf ] ) , the pseudospin in fig . [ fig : pspin ] at @xmath2 and that at @xmath3 point to different orientations with respect to @xmath40 . in contrast , the pseudospin at @xmath2 and that at @xmath4 point toward the same orientation . thus , the pseudospin component perpendicular to the zigzag edge flips , while the pseudospin is invariant for the armchair edge . we have seen for the zigzag edge that the reflection is intravalley scattering and that the pseudospin component perpendicular to the edge flips . for the armchair edge , the reflection is intervalley scattering and the pseudospin is invariant . more details concerning the scattering , for example , the relative phase between the incident and reflected waves and the edge states are difficult to obtain within the above argument . in subsequent sections we will explore an effective hamiltonian to obtain the standing wave and the edge states . the fact that the pseudospin flips at the zigzag edge leads us to consider a gauge field for the edge that couples with the pseudospin in a manner similar to that an electromagnetic gauge field couples with the real spin . here , we show the formulation , in which the effect of the edge is included into the hamiltonian as a deformation - induced gauge field . @xcite to begin with , we consider a change of the nearest - neighbor hopping integral from the average value , @xmath41 , as @xmath42 , where @xmath10 @xmath43 denotes the direction of a bond parallel to @xmath16 in the inset of fig . [ fig : pspin ] . the deviation @xmath44 represents a lattice deformation in a graphene sheet . the low energy effective - mass equation for deformed graphene is written as @xmath45 where @xmath46 and @xmath47 are two - component wavefunctions that represent the electrons near the k and k@xmath6 points , respectively . the hamiltonian for deformed graphene is written as @xcite @xmath48 where @xmath49 is the momentum operator , @xmath50 , and @xmath51 . a lattice deformation @xmath44 enters the hamiltonian through the deformation - induced gauge field @xmath52 , where @xmath53 is expressed by a linear combination of @xmath54 as @xcite @xmath55 the @xmath53 field causes intravalley scattering , while the perturbation that is relevant to intervalley scattering is given by a linear combination of @xmath56 and @xmath57 as @xcite @xmath58 in fig . [ fig : graphene](a ) , we consider cutting the c - c bonds located on the @xmath0-axis at @xmath59 in order to introduce the zigzag edge in a flat graphene sheet . after cutting the bonds , the graphene sheet splits into two semi - infinite parts : @xmath60 and @xmath61 . the cutting is represented as @xmath62 , @xmath63 and @xmath64 . from eq . ( [ eq : gauge ] ) , the corresponding deformation - induced gauge field is then written as @xmath65 , where @xmath66 is not vanishing only for the c - c bonds located on the @xmath0-axis at @xmath59 as @xmath67 . since @xmath66 is defined for the c - c bond , @xmath66 is meaningful when it is integrated from @xmath68 to @xmath69 , where @xmath69 is of the same order as the c - c bond length and will be taken to be zero at the end of calculation in the continuum limit . note also that the vector direction of @xmath53 is perpendicular to that of the bond with a modified hopping integral . since the zigzag edge is not the source of intervalley scattering , intervalley scattering can be ignored . hereafter , we consider the electrons near the k point for the zigzag edge . moreover , separation of variables can be employed , due to translational symmetry along the @xmath0-axis . as a result , @xmath46 and @xmath70 in eq . ( [ eq : h ] ) can be replaced with @xmath71 and @xmath72 . the energy eigenequation can then be simplified as @xmath73 , where the hamiltonian is @xmath74 . \label{eq : hk_zigzag}\end{aligned}\ ] ] this hamiltonian is solved in sec . [ sec : zig ] . the cutting which produces the klein edges @xcite is represented as @xmath75 , @xmath76 and @xmath77 . from eq . ( [ eq : gauge ] ) , the corresponding deformation - induced gauge field is then written as @xmath78 , where @xmath66 is the gauge field for the zigzag edge . note that the direction of the @xmath53 field for the klein edge is opposite that of the zigzag edge . are cut to introduce the zigzag edge ( klein edge ) . the cutting is represented as a deformation - induced gauge field @xmath65 . ( b ) the deformation - induced gauge field for the armchair edge is given by @xmath79 . ] the armchair edge can be introduced by cutting the bonds located on @xmath80 , as shown in fig . [ fig : graphene](b ) . by setting @xmath75 , @xmath81 and @xmath82 in eq . ( [ eq : gauge ] ) , the deformation - induced gauge field for the armchair edge is written as @xmath83 with the limit of @xmath84 . due to translational symmetry along the @xmath1-axis , @xmath85 is replaced to @xmath86 in eq . ( [ eq : h ] ) . thus , the hamiltonian is given by @xmath87 with @xmath88 . this hamiltonian can be reduced further by means of a gauge symmetry in the following manner . since @xmath89 does not depend on @xmath1 , it can be represented in terms of a scalar function @xmath90 , as @xmath91 . using the gauge transformation : @xmath92 and @xmath93 , @xmath89 can be erased from the hamiltonian for each valley . however , note that as a result of this gauge transformation , @xmath94 must be changed into @xmath95 . to minimize notation , let us use @xmath94 to denote this gauge transformed field , so that @xmath96}$ ] . the hamiltonian for the armchair edge is then written as @xmath97 this will be solved in sec . [ sec : arm ] . note that by introducing @xmath98 ( @xmath99 ) matrices defined by @xmath100 the unperturbed hamiltonian is represented in a compact fashion as @xmath101 , where @xmath102 is a @xmath103 identity matrix . in eqs . ( [ eq : gauge ] ) and ( [ eq : gauge2 ] ) , we assume @xmath104 , and ignore the higher order term of @xmath44 . as a result of this simplification , the relationship between @xmath53 and @xmath44 may deviate from eq . ( [ eq : gauge ] ) when @xmath105 . however , note that the direction and not the strength of the @xmath53 field can be determined by eq . ( [ eq : gauge ] ) , even for the case where @xmath106 . consideration of this point is given in appendix [ app : rot - pspin ] . the scattering problem for the zigzag edge is solved in this section . standing wave solutions are constructed in sec . [ sec : exs ] , and the properties of the solutions are examined in detail . localized edge states are constructed in sec . [ ssec : es ] . the behavior of the standing wave in the presence of a mass term and an external magnetic field is examined in secs . [ ssec : mass ] and [ ssec : emf ] , respectively . the local density of states near the zigzag edge is calculated analytically in sec . [ ssec : ldos ] . to begin with , solutions are constructed for the case of @xmath107 in eq . ( [ eq : hk_zigzag ] ) . let @xmath108 be the eigenstate of the unperturbed hamiltonian @xmath109 . @xmath110 satisfies @xmath111 ; therefore , a general solution may be constructed from the basis function @xmath108 to satisfy the constraint equation , @xmath112 where @xmath113 is a real number phase . the phase @xmath113 can not be an arbitrary value . the successive operation of eq . ( [ eq : sym ] ) on @xmath114 gives @xmath115 , and hence @xmath113 should be @xmath116 or @xmath117 . note that a set of functions satisfying eq . ( [ eq : sym ] ) is useful for construction of solutions in the case @xmath118 , because @xmath119 also satisfies @xmath120 . this constraint comes from the inversion symmetry of the gauge field with respect to @xmath59 , @xmath121 . from eq . ( [ eq : sym ] ) , we have @xmath122 . thus , @xmath108 can be rewritten as @xmath123 by substituting eq . ( [ eq : wf - zig ] ) into @xmath124 , we obtain simultaneous differential equations : @xmath125 where @xmath126 and @xmath127 are defined as @xmath128 for the case @xmath129 , eq . ( [ eq : phiab - def ] ) implies that @xmath126 is an even function [ @xmath130 , while @xmath127 is an odd function [ @xmath131 . thus , they can be parameterized as follows : @xmath132 where the parameters @xmath133 and @xmath134 can be determined from eq . ( [ eq : hsym ] ) . by substituting eq . ( [ eq : func0 ] ) into eq . ( [ eq : hsym ] ) , we obtain the secular equation @xmath135 the solution of this secular equation satisfies @xmath136 let @xmath137 be the polar angle between vector @xmath138 and the @xmath36-axis . then , @xmath139 and @xmath140 where @xmath141 , and the second equation of eq . ( [ eq : enea ] ) can be rewritten as @xmath142 $ ] for the eigenstate with positive energy @xmath143 . assuming that @xmath144 $ ] , we have @xmath145 $ ] . substituting these into eq . ( [ eq : func0 ] ) gives @xmath146 then , eq . ( [ eq : phisa ] ) is substituted into eq . ( [ eq : phiab - def ] ) with @xmath129 to give @xmath147 similarly , for the case where @xmath148 , we have @xmath149 the energies of the eigenstates @xmath150 and @xmath151 are equal , and therefore a general solution can be expressed as a superposition of the degenerate eigenstates , as @xmath152 where @xmath153 is a real number . the value of @xmath153 is determined as follows . the hamiltonian , @xmath154 , is identical to the unperturbed hamiltonian @xmath110 for @xmath1 to satisfy @xmath155 , so that @xmath156 satisfies the eigenequation @xmath157 for @xmath155 . we need to solve @xmath73 locally for @xmath158 . by parameterizing the eigenstate of @xmath119 as @xmath159 , we obtain the constraint equation for @xmath160 and @xmath156 as @xmath161 + \sigma_x a_x^{\rm q}(y)n(y ) \right\ } \phi^{f}(y ) = 0 . \label{eq : constz}\end{aligned}\ ] ] to obtain eq . ( [ eq : constz ] ) we must place @xmath159 to @xmath73 , and use @xmath162 . here , we have assumed that the energy eigenvalues of the standing wave @xmath163 and of @xmath156 are the same . this assumption is valid for the standing wave , because the energy eigenvalue is determined by the bulk hamiltonian @xmath110 and the energy does not change through elastic scattering . however , note that this assumption is not valid for the edge states , which @xmath73 must be solved directly ( see sec . [ ssec : es ] for more details ) . now , eq . ( [ eq : constz ] ) is equivalent to the two successive equations : @xmath164 the following two cases can be considered for this successive equation . one case is that the solution satisfies @xmath165 the first ( second ) equation of eq . ( [ eq : case1 ] ) ensures the second ( first ) equation of eq . ( [ eq : caseeq ] ) . the other case is that the solution satisfies @xmath166 the two conditions , eqs . ( [ eq : case1 ] ) and ( [ eq : case2 ] ) , correspond to the standing wave in the upper semi - infinite graphene plane for @xmath60 and that in the lower plane for @xmath61 in the limit of @xmath167 , as shown in the following . for the case of eq . ( [ eq : case1 ] ) , the first equation is integrated with respect to @xmath1 , to obtain @xmath168 hence , when @xmath169 , @xmath170 is negligible compared with @xmath171 , and therefore the standing wave appears only for @xmath61 . in contrast , when @xmath172 , the standing wave appears only for @xmath60 . the other condition in eq . ( [ eq : case1 ] ) holds for the limit of @xmath173 by setting @xmath174 in eq . ( [ eq : phify ] ) , because @xmath175 this condition leads to @xmath176 , which represents the boundary conditions for the zigzag and klein edges shown in fig . [ fig : graphene](a ) . thus , eq . ( [ eq : case1 ] ) covers two situations , depending on the direction of the gauge field ; @xmath177 or @xmath178 . that is , when @xmath177 , eq . ( [ eq : case1 ] ) corresponds to the upper semi - infinite graphene plane with the zigzag edge , while when @xmath178 , eq . ( [ eq : case1 ] ) corresponds to the lower semi - infinite graphene plane with the klein edge . similarly , when @xmath177 , eq . ( [ eq : case2 ] ) corresponds to the lower semi - infinite graphene plane with the zigzag edge , while when @xmath178 , eq . ( [ eq : case2 ] ) corresponds to the upper semi - infinite graphene plane with the klein edge . from eq . ( [ eq : weight ] ) , it follows that the gauge field for the edge should be large , @xmath179 . in ref . , the following was obtained analytically @xmath180 where @xmath181 is the parameter that specifies the deformation as @xmath182 [ see fig . [ fig : graphene ] ] . the right - hand side gives logarithmic singularities for @xmath183 and @xmath184 . the limit @xmath185 corresponds to the zigzag edge , while the limit @xmath186 represents the klein edge . note that when @xmath186 , the electron is unable to have a finite amplitude on the a and b atoms located at @xmath59 , which effectively represents the klein edge . because of the singularity , @xmath160 that satisfies eq . ( [ eq : weight ] ) is similar to the step function ; @xmath187 for @xmath188 , and otherwise @xmath189 . now , by setting @xmath174 in eq . ( [ eq : phify ] ) , the standing wave in the conduction band is expressed as @xmath190 where the plane wave parallel to the edge with the length @xmath191 is included . the standing wave in the valence band is obtained by using the particle - hole symmetry of the hamiltonian , @xmath192 , as @xmath193 : @xmath194 here , we consider the pseudospin of the standing wave . the pseudospin for an eigenstate @xmath195 is defined by the expected value of the pauli matrices as @xmath196 ( @xmath197 ) , where @xmath198 is a pseudospin density defined by @xmath199 . note that the @xmath1-component of the pseudospin is proportional to the imaginary part of the bloch function , such as @xmath200 $ ] . the bloch function of the standing wave is real , so that the @xmath1-component of the pseudospin for the standing wave vanishes , that is , @xmath201 . note also that @xmath201 means that the current normal to the zigzag edge vanishes . it is interesting to note that @xmath201 holds whenever @xmath202 and @xmath203 can be taken as real numbers . this indicates that the result @xmath201 is not sensitive to the value of @xmath153 , but depends only on the fact that @xmath204 does not have a relative phase between the two components . the condition of eq . ( [ eq : phib0 ] ) means that the pseudospin density is locally polarized into the positive @xmath205-axis near the zigzag edge , that is , @xmath206 and @xmath207 . actually , by substituting @xmath208 into eq . ( [ eq : s - sol ] ) , the standing wave near the zigzag edge has amplitude only at a - atoms . this polarization of the pseudospin is consistent with the fact that the gauge field @xmath66 has a non - vanishing deformation - induced magnetic field , @xmath209 at the zigzag edge . the presence of the @xmath210 field at the zigzag edge causes local polarization of the standing wave pseudospin near the zigzag edge , similar to the polarization of a real spin by a magnetic field . we will show in sec . [ ssec : ldos ] that this polarization of the pseudospin causes anomalous behavior to appear in the local density of states ( ldos ) near the zigzag edge . a zigzag nanoribbon is given by introducing another zigzag edge at @xmath211 , in addition to the zigzag edge at @xmath59 . suppose that the edge atoms at @xmath211 are b - atoms , which imposes the boundary condition on the wave function at @xmath212 as @xmath213 . this leads to the constraint equation for @xmath2 , @xmath214 where @xmath215 is an integer . it is noted that this equation reproduces @xmath216 which was obtained by brey and fertig in ref . [ the negative sign in front of @xmath36 is a matter of notation ] . note that @xmath215 should be a nonzero integer , because the equation does not possess a solution when @xmath217 . for the case where the edge at @xmath211 is the klein edge , the boundary condition on the wave function at @xmath212 becomes @xmath218 . this leads to @xmath219 , where @xmath215 is a positive integer . in addition to the standing wave derived in the previous subsection , @xmath119 possesses localized edge states . @xcite here , we show how to construct the edge states . the following observation is useful in order to obtain the edge states . instead of eq . ( [ eq : func0 ] ) , we assume @xmath220 by substituting eq . ( [ eq : func1 ] ) into eq . ( [ eq : hsym ] ) , the secular equation is obtained : @xmath221 the solution of this secular equation satisfies @xmath222 by introducing the @xmath223 variable , which satisfies @xmath224 we have @xmath225 . for the case we have @xmath227 . by inserting this into eq . ( [ eq : func1 ] ) and setting @xmath228 , we obtain @xmath229 by substituting eq . ( [ eq : phisa - e ] ) into eq . ( [ eq : phiab - def ] ) with @xmath129 , we have @xmath230 similarly , for the case @xmath148 , we have @xmath231 the energies of @xmath151 and @xmath150 are equal ; therefore , the basis function may be chosen as @xmath232 the functions @xmath233 and @xmath234 are exponentially increasing and decreasing functions of @xmath1 , respectively . thus , neither @xmath233 nor @xmath234 is a normalized wave function all over the space , @xmath235 . however , note that @xmath233 and @xmath234 can be normalizable wave functions for @xmath61 and @xmath60 , respectively . we also note that the pseudospin of @xmath233 is given by @xmath236 , while that of @xmath234 is @xmath237 . from the above observation , we parameterized the localized eigenstate as @xmath238 where @xmath239 is a normalization constant , and the modulation of the pseudospin is represented by a function @xmath240 . substituting eq . ( [ eq : phir ] ) into @xmath73 gives simultaneous differential equations for @xmath240 , @xmath241 by summing and subtracting both sides of eq . ( [ eq : w+g - d ] ) , the energy eigenequation can be rewritten as @xmath242 the solution of the second equation is given by @xmath243 the sign of @xmath240 changes across the zigzag edge , and this sign change indicates that the @xmath205-component of the pseudospin flips at the edge . the flip is induced by the gauge field @xmath244 . to represent this , we integrate the first equation of eq . ( [ eq : weyl+g ] ) from @xmath245 to @xmath69 , and acquire @xmath246 we have neglected other terms , because they are proportional to @xmath69 and become zero in the limit of @xmath247 . by substituting eq . ( [ eq : const - g(y ) ] ) into eq . ( [ eq : w - s - integ ] ) , we find @xmath248 hence , eq . ( [ eq : const - g(y ) ] ) becomes @xmath249 having described the wave function of the localized state , let us now calculate @xmath250 and @xmath251 . to this end , we use the first equation of eq . ( [ eq : weyl+g ] ) for @xmath155 and obtain @xmath252 moreover , using eq.([eq : g - ene ] ) , we find @xmath253 in addition to this localized state , there is another localized state for the same @xmath36 with the same @xmath251 , but with the opposite sign of @xmath250 . this results from the particle - hole symmetry of the hamiltonian , and the wave function is given by @xmath254 . in the following , we will show that the solutions can reproduce all the properties of the edge states known in the tight - binding lattice ( tb ) model , @xcite such as the asymmetric energy band structure with respect to the k ( k@xmath6 ) point , the flat energy band , and the pseudospin structure . the asymmetric energy band structure with respect to the k ( k@xmath6 ) point originates from the normalization condition of the wave function , which requires that @xmath251 should be positive . this requirement restricts the value of @xmath36 in eq . ( [ eq : xi - n ] ) . when @xmath244 is positive , eq . ( [ eq : xi - n ] ) indicates that the localized states appear only at @xmath255 around the k point . this is the reason why the localized states appear in the energy spectrum only at one side around the k point . a similar argument can be used for the k@xmath6 point , which concludes that the localized state appears at @xmath256 around the k@xmath6 point . the hamiltonian around the k@xmath6 point is expressed by @xmath257 therefore , we obtain different signs in front of @xmath258 in @xmath119 and eq . ( [ eq : weyl - k ] ) , which causes the negative sign in front of the right - hand side of eq . ( [ eq : xi - n ] ) to disappear for the k@xmath6 point . thus , when @xmath244 is negative ( klein edges ) , edge states appear on the opposite side ; @xmath256 around the k point and @xmath259 around the k@xmath6 point . calculations on the tb model with klein edges also agree with the results obtained here . a singularity of the gauge field , @xmath260 , is the origin of the flat energy dispersion and the pseudospin polarization of the edge states . when @xmath261 , @xmath250 in eq . ( [ eq : ene ] ) becomes zero . the zero energy eigenvalue between the k and k@xmath6 points in the band structure corresponds to the flat energy band of the edge state . @xcite moreover , from eq . ( [ eq : asol ] ) , @xmath262 for @xmath188 and @xmath263 for @xmath264 are obtained . in this case , the localized state is a pseudospin - up state @xmath265 for @xmath188 and a pseudospin - down state @xmath266 for @xmath264 . hence , a singular gauge field at the zigzag edge causes polarization of the pseudospin of the localized states . polarization of the pseudospin means that the wave function has amplitude only at the a ( or b ) atom , so that this result agrees with the result from the tb model for the edge state . @xcite comparing eqs . ( [ eq : xi - n ] ) and ( [ eq : ene ] ) with eqs . ( [ eq : kycond ] ) and ( [ eq : solu1 ] ) , the relation between the variable @xmath223 and the field @xmath66 is observed as @xmath267 . here , we note that @xmath239 in eq . ( [ eq : phir ] ) is not a function of @xmath1 , but is a constant for @xmath268 . therefore , the edge states appear on both sides of the zigzag edge , @xmath60 and @xmath61 , while the standing waves appear on only one side of the edge under the limit @xmath269 . with this limit , the edge states can be confined to one side of the edge , because the energy of the localized state becomes @xmath270 , and therefore , the superposition of an edge state , @xmath163 , and its electron - hole pair state , @xmath254 , is a solution . it is easy to see that @xmath271 has amplitude only for @xmath272 , while @xmath273 has amplitude only for @xmath274 . the wave function of the edge state for @xmath61 is then given by @xmath275 where the normalization constant has been fixed , @xmath276 , by assuming that the system is a semi - infinite graphene plane . we note that the mass term , @xmath277 , is proportional to the particle - hole symmetry operator , @xmath278 . thus , the mass term automatically restricts the region where the edge states can appear ( @xmath60 or @xmath61 ) , and this is shown in appendix [ app : edge ] . finally , we consider the edge states in nanoribbons . note first that the exact localization length for the case of a zigzag nanoribbon with width @xmath279 satisfies @xmath280 , \end{split } \label{eq : xiexact}\end{aligned}\ ] ] which is obtained by analytical continuation @xmath281 for eq . ( [ eq : bery ] ) . comparing this equation with eq . ( [ eq : xi - n ] ) shows that the large value of the gauge field in eq . ( [ eq : xi - n ] ) corresponds to the case of @xmath282 in eq . ( [ eq : xiexact ] ) . this is consistent with having solved the hamiltonian locally near the edge , in which it was implicitly assumed that the condition @xmath282 is satisfied . except the edge states whose localization length is in the order of @xmath279 , eqs . ( [ eq : xi - n ] ) and ( [ eq : xiexact ] ) give almost identical values of @xmath283 , which justifies the description using the gauge field . note also that the condition @xmath282 also represents the condition @xmath284 , which is clear from the second equation in eq . ( [ eq : xiexact ] ) . to solve the hamiltonian for the edge states with @xmath285 , eq . ( [ eq : weyl+g ] ) must be solved globally , for example , on a circle , which is a challenging issue . let us reconsider the scattering problem for the case where the hamiltonian includes a mass term . the total hamiltonian is given by @xmath286 , where the mass , @xmath287 , is a constant over the space @xmath288 . the solutions of @xmath289 can be constructed from the solutions of @xmath119 as follows . for @xmath119 , the standing wave solutions , eqs . ( [ eq : s - sol ] ) and ( [ eq : s - solv ] ) , satisfy @xmath290 for the mass term , because @xmath291 , we obtain @xmath292 thus , by changing the basis state from @xmath293 and @xmath294 into @xmath295 @xmath296 $ ] , the hamiltonian is represented as @xmath297 here , the angle @xmath298 is defined as @xmath299 where @xmath300 . the normalized eigenvectors of the matrix in eq . ( [ eq : hkmat ] ) are then @xmath301 for the @xmath302 and @xmath303 eigenvalues , respectively . @xmath304 and @xmath305 are obtained from eqs . ( [ eq : s - sol ] ) and ( [ eq : s - solv ] ) ; therefore , the standing wave near the zigzag edge is given by @xmath306 where we have omitted to write the plane wave parallel to the edge . the factors @xmath307 and @xmath308 appear in a manner similar to the eigenvalue problem of the spin magnetic moment in a magnetic field . in this subsection , solutions are constructed for a magnetic field applied perpendicular to the graphene plane . @xcite a magnetic field @xmath309 can be represented by the electromagnetic gauge field as @xmath310 . this gauge field is included in the hamiltonian by substituting the momentum operator @xmath311 with @xmath312 . for the case @xmath313 , the eigenequation becomes @xmath314 \phi({\bf r } ) = e \phi({\bf r}),\end{aligned}\ ] ] and the solutions are given by the landau states , which are specified by an integer @xmath215 and a center coordinate @xmath315 as @xmath316 where @xmath317 is a normalization constant , @xmath318 , and @xmath319 is a hermite polynomial defined by @xmath320 ( @xmath321 ) . the energy eigenvalue of @xmath322 is given by @xmath323 . a method similar to that in sec . [ sec : exs ] is used to solve the scattering problem in the presence of a magnetic field . the energy eigenstate of @xmath324 is parameterized as @xmath325 . substituting this into @xmath326 , and using @xmath327 , we obtain the constraint equation for @xmath160 and @xmath322 , @xmath161 + \sigma_x a_x^{\rm q}(y)n(y ) \right\ } \phi^{\rm ll}_{ny}({\bf r } ) = 0.\end{aligned}\ ] ] two cases can be considered as a solution for this successive equation [ see eqs . ( [ eq : case1 ] ) and ( [ eq : case2 ] ) ] . here , we choose the case where @xmath328 from eq . ( [ eq : ll ] ) , the second equation leads to @xmath329 with the limit @xmath173 ( @xmath330 ) . the number of zeros of @xmath331 is @xmath332 , so that there are @xmath332 solutions of eq . ( [ eq : constb ] ) , which are denoted as @xmath333 ( @xmath334 ) . the solutions can then be written as @xmath335 note that @xmath336 with a large value of @xmath315 that satisfies @xmath337 , can be an approximate solution , due to the exponential factor in eq . ( [ eq : constb ] ) . the solution with a large value of @xmath315 represents the wave function in the bulk , and is not sensitive to the details of the edge . the solutions given in eq . ( [ eq : llz ] ) concern the landau states near the zigzag edge , and these are examined in the following . for the case that @xmath215 is an odd integer , @xmath338 satisfies eq . ( [ eq : constb ] ) , because @xmath339 . the wave function with @xmath338 decays according to @xmath340 , and the amplitude has a maximum at the zigzag edge . note that the localization length is in the order of @xmath341 ( @xmath342/\sqrt{b[{\rm t}]}$ ] ) , which is larger than the localization length of the edge state @xmath343 where @xmath344 takes a value of the same order as the lattice constant . the lowest landau level ( @xmath217 ) can not satisfy the condition of eq . ( [ eq : constb ] ) because @xmath345 and the amplitudes of b - atoms do not vanish at the edge . thus , the lowest landau level is absent for the k point . on the other hand , the lowest landau level appears for the k@xmath6 point . the hamiltonian for the k@xmath6 point is given by @xmath346.\end{aligned}\ ] ] for the case of @xmath313 , @xmath324 and @xmath347 are related as @xmath348 , and therefore the solutions for the k@xmath6 point are given by @xmath349 . the constraint equation for the k@xmath6 point is @xmath161 + \sigma_x a_x^{\rm q}(y)n(y ) \right\ } \sigma_y \phi^{\rm ll}_{ny'}({\bf r } ) = 0,\end{aligned}\ ] ] which reduces to the condition @xmath350 . the solution is then given by @xmath351 where @xmath352 denotes the solution of the constraint equation , @xmath353 this condition is satisfied for @xmath217 , so that the lowest landau level appears for the k@xmath6 point . there is no constraint for the value of @xmath315 . for the case of the first landau levels ( @xmath354 ) , the landau level for the k point appears , while that for the k@xmath6 point disappears . therefore , near the zigzag edge , the lowest and first landau levels are not symmetric with respect to the k and k@xmath6 points . several groups have conducted scanning tunneling spectroscopy ( sts ) measurements to determine the ldos near the step edge of graphite . @xcite a peak structure in the ldos due to the edge states has been extensively discussed by many authors . here , we calculate the ldos near the zigzag edge . we show that some characteristic features that originate from the pseudospin polarization , the edge states , and the mass appear in the ldos . let us first review the ldos for graphene without an edge . assuming that electrons are non - interacting , the bulk ldos is given by @xmath355 where @xmath356 is proportional to @xmath357 , which results from the dirac cone spectrum . note that the actual ldos is given by @xmath358 , where @xmath359 ( @xmath360 ) accounts for the spin ( valley ) degrees of freedom . next , the ldos near the zigzag edge is calculated using the solutions given in eq . ( [ eq : mass - wave ] ) . the ldos has the form , @xmath361 where @xmath362 is defined as @xmath363 by performing the integral with respect to the angle @xmath364 in eq . ( [ eq : rfac ] ) , we obtain an analytical result for @xmath362 as @xmath365 where @xmath366 is a function of @xmath250 according to @xmath367 , and the functions @xmath368 and @xmath369 are defined as @xmath370 here , @xmath371 is a bessel function of order @xmath372 . the number located on each solid line represents the distance ( corresponding to @xmath373 in the inset ) from the zigzag edge . the ldos at @xmath374 vanishes near the zigzag edge , which is emphasized by the dashed circle . the dashed line denotes the ldos in the bulk which is defined by the ldos at @xmath375 . a peak structure due to the edge states is plotted for comparison . note that there are several intrinsic perturbations @xcite that can change the position of the peak . ] because the case of @xmath376 has been considered elsewhere , @xcite we consider the case @xmath377 here . ( [ eq : ldos_bulk ] ) holds for @xmath378 . the bulk ldos vanishes for the case @xmath379 , as shown by the dashed line in fig . [ fig : ldos ] . note that the ldos disappears suddenly at @xmath380 , and the bulk ldos has a step like structure at @xmath381 . the bulk ldos is symmetric with respect to @xmath382 , even for the case @xmath383 . however , note that the ldos near the edge is not symmetric for the case of @xmath383 , which is clear from the different signs in front of the function @xmath369 in eq . ( [ eq : ldos ] ) . in fig . [ fig : ldos ] , the ldos are plotted at @xmath384 , 1 , 2 , and 3 [ nm ] for the case of @xmath385 ev . note that for the case @xmath386 ev , the corresponding ldos curve is given by interchanging the conduction and valence bands in fig . [ fig : ldos ] . the asymmetry in the ldos near the edge appears at the following points . first , a step structure appears only at @xmath387 ev . at @xmath388 ev , the ldos vanishes , and the step structure is absent , as indicated by the dashed circle in fig . [ fig : ldos ] . the absence of the ldos at @xmath388 ev can be explained by the zigzag edge consisting of a - atoms makes the standing wave polarized into a - atoms near the zigzag edge . however , eigenstates with energy @xmath374 should be polarized into b - atoms by the factors in eq . ( [ eq : bfieldp ] ) , and the amplitude of a - atoms are strongly suppressed by the mass term . therefore , electrons with energy @xmath374 can not approach the zigzag edge , and therefore the ldos disappears . secondly , the ldos peak of the edge states appears only at @xmath387 ev . this is a straightforward consequence of the edge state amplitude appearing only for a - atoms . the absence of the ldos at @xmath374 and the presence of the peak at @xmath389 due to the edge states occurs at different sides of the band edge . to plot the ldos of the edge states in fig . [ fig : ldos ] , we have used @xmath390 where @xmath391 is a phenomenological parameter that represents the energy uncertainty of the edge states , for which we assume @xmath392 mev . this result has been derived in ref . for the case of @xmath376 . note that @xmath393 decreases as @xmath394 , which is a slowly decreasing function compared with the exponential decay wave function of the edge state . in this section , the scattering problem for the armchair edge is solved using a method similar to that used in sec . [ sec : zig ] . the standing wave shows that the pseudospin does not change its direction through the reflection at the armchair edge . solutions for the case of @xmath395 in eq . ( [ eq : armchairh ] ) are constructed first , and then used as the basis functions to construct the standing wave near the armchair edge . let @xmath396 represent the solution of the unperturbed hamiltonian , @xmath397 . the perturbed hamiltonian satisfies @xmath398 , so that the functions @xmath396 that satisfy the constraint equation @xmath399 are useful for construction of solutions in the case of @xmath400 . from eq . ( [ eq : xmirror ] ) , we may write @xmath401 by using eq . ( [ eq : wf_arm ] ) , the energy eigenequation becomes @xmath402 where @xmath403 and @xmath404 are defined as @xmath405 for the case @xmath129 , @xmath403 is an even function , while @xmath404 is an odd function . for the case of @xmath148 , @xmath403 is an odd function , while @xmath404 is an even function . for the case @xmath129 , we can set @xmath406 substituting these into eq . ( [ eq : dirac3 ] ) , we obtain the secular equation : @xmath407 the solutions of this secular equation satisfy @xmath408 , and the eigenfunction in the conduction band is given by @xmath409 , which is defined as @xmath410 by substituting eq . ( [ eq : cond_sa-1 ] ) into eq . ( [ eq : psisa_def ] ) , we obtain @xmath411 . using eq . ( [ eq : wf_arm ] ) , it can be seen that @xmath412 similarly , for the case @xmath148 , we have @xmath413 new basis functions are defined using eqs . ( [ eq : aphi0 ] ) and ( [ eq : aphipi ] ) , as @xmath414 the eigenstate @xmath415 represents a free propagating state with momentum @xmath138 near the k point , while @xmath416 represents a state with momentum @xmath417 near the k@xmath6 point . it is clear that these are eigenstates in the absence of the edge . in the presence of the armchair edge , neither @xmath415 nor @xmath416 is an eigenstate , but a true eigenstate is the standing wave that is given by a superposition between @xmath415 and @xmath416 as @xmath418 to find @xmath419 , it is useful to rewrite the total hamiltonian as @xmath420 , \label{eq : hdec}\end{aligned}\ ] ] where @xmath94 is expressed in terms of real and imaginary parts , as @xmath421 . in sec . [ sec : model ] , we have shown that @xmath96}$ ] , where @xmath91 . @xmath422 is an even function with respect to @xmath0 ; therefore , @xmath90 can be taken as an odd function , so that the field satisfies @xmath423 . from this condition , it follows that @xmath424 is an even function , while @xmath425 is an odd function . next , we construct solutions for the case @xmath426 . let us define @xmath427 and @xmath428 using a real function @xmath429 as @xmath430 since @xmath415 and @xmath416 are the solutions of @xmath431 , we obtain the following equations for @xmath429 from @xmath432 , @xmath433 all four components of this matrix are reduced into the same differential equation : @xmath434 . @xmath435 is an odd function , so that we have @xmath436 by using @xmath437 . because @xmath438 when @xmath439 , the constant of integration can be taken as zero . as a result , we have @xmath440 . therefore , @xmath429 can take only a non zero value for @xmath441 , and the mixing between @xmath415 and @xmath416 is negligible in the bulk . finally , we assume that the solution of the total hamiltonian of eq . ( [ eq : hdec ] ) has the form of @xmath442.\end{aligned}\ ] ] the constraint equation for @xmath443 is then given by @xmath444 this constraint equation has the same form as eqs . ( [ eq : case1 ] ) and ( [ eq : case2 ] ) . performing the integral for @xmath0 from @xmath68 to @xmath69 in eq . ( [ eq : difn ] ) gives @xmath445 as we have shown in fig . [ fig : graphene ] , @xmath446 is a negative large quantity . thus , @xmath447 has an amplitude only for @xmath448 , while @xmath449 has an amplitude only for @xmath450 . @xmath451 and @xmath452 for @xmath453 ; therefore , the standing wave near the armchair edge is written as @xmath454 note that the bloch functions for the k and k@xmath6 points are the same , which indicates that the pseudospins of the incident and reflected waves are equal , as shown in fig . [ fig : pseudospin - arm ] . thus , the berry s phase of the standing wave near the armchair edge is given by @xmath455 , which is in contrast to the case of the zigzag edge . @xcite the boundary condition for the armchair edge does not forbid an electronic state to cross the dirac singularity point , and therefore the electron can pick up a nontrivial berry s phase . of one valley into @xmath417 of another valley , and the two wave functions of the different valleys form a standing wave . the pseudospin is unchanged by the armchair edge . note that the pseudospin for states near the k@xmath6 point is not parallel to the vector @xmath456 , while the pseudospin for states near the k point is parallel to the vector @xmath138 . ] to understand the behavior of the standing wave in more detail , the density of @xmath98 was examined . the density for an eigenstate @xmath195 is defined by the expected value of @xmath98 as @xmath457 . it is then straightforward to check from eq . ( [ eq : armwf ] ) that @xmath458 , @xmath459 , and @xmath460 . @xmath461 vanishes near the armchair edge ( at @xmath462 ) , and @xmath463 takes a maximum value at the edge . this behavior can be understood from eq . ( [ eq : hdec ] ) , in which @xmath464 couples with @xmath446 . since @xmath446 is singular at @xmath462 , @xmath461 can not have a non - zero value at @xmath462 . the result @xmath465 indicates that time - reversal symmetry is preserved . let us examine the landau states near the armchair edge . the electromagnetic gauge field @xmath466 for an external magnetic field @xmath309 is included in the hamiltonian @xmath467 of eq . ( [ eq : h ] ) by the substitution @xmath468 . the hamiltonian satisfies @xmath469 , and therefore the solution can be written as @xmath470 let @xmath471 be the solution for the case @xmath395 . then @xmath471 satisfies the following energy eigenequation : @xmath472 \phi_{\rm k}({\bf r } ) = e \phi_{\rm k}({\bf r}).\end{aligned}\ ] ] the solutions are the landau states specified by integer @xmath215 and a center coordinate @xmath473 as [ see eq . ( [ eq : ll ] ) ] @xmath474 applying the parity transformation @xmath475 to @xmath476 , we obtain @xmath477 the matrix @xmath278 on the right - hand side can be understood by applying the parity transformation @xmath475 to this energy eigenequation : @xmath472 \phi_{\rm k}(-{\bf r } ) = -e \phi_{\rm k}(-{\bf r}).\end{aligned}\ ] ] the negative sign in front of the right - hand side shows that the energy eigenvalue of @xmath478 is opposite to that of @xmath471 . by substituting eq . ( [ eq : llp ] ) into eq . ( [ eq : armcy ] ) , we obtain @xmath479 by repeating the same argument given in the previous subsection , the following standing wave solutions are obtained : @xmath480 there are no constraints for the value of @xmath473 . it is then a straightforward calculation to check that @xmath461 vanishes at the armchair edge . a realistic graphene edge may be a mixture of zigzag and armchair edges . @xcite the construction of the standing wave near the general edge is one of the interesting applications for our framework . we believe that the hamiltonian in eq . ( [ eq : h ] ) can describe the low - energy electrons in a graphene plane with a general edge . however , note that this issue is related to the coherence length of the standing wave . in the present paper , we have not considered perturbations that break coherence , such as electron - phonon interaction . interestingly , the electron - phonon interaction can also be represented as a deformation - induced gauge field . @xcite thus , the gauge field description for the graphene edge may be useful when we consider such issues . the effective - mass model of eq . ( [ eq : h ] ) is equivalent to a chiral gauge theory for graphene that has been proposed by jackiw and pi . @xcite indeed , by applying @xmath481 to @xmath47 in eq.([eq : hwave ] ) , the hamiltonian in eq . ( [ eq : h ] ) may be rewritten as @xmath482 which is the electronic hamiltonian of the chiral gauge theory . they have investigated zero - mode solutions of the hamiltonian with a topological vortex for @xmath53 on the background of kekul distortion for @xmath483 , in the context of fractionalization of quantum number . @xcite our trial is then to study the graphene edge as a chiral gauge theory , although our results in this paper do not clarify fully the topological features of the graphene edge . it is interesting to note that one may find an advantage of a chiral gauge theory when we consider the real spins of the electrons . for example , the magnetism of the edge states may be understood as a parity anomaly phenomenon . @xcite the various field - theoretical techniques may be utilized to explore the electronic properties near the edge . note also that the perturbation which mixes the electrons in the two valleys has been examined in the studies on the topological defect in graphene . @xcite we have taken into account the edge as a part of the hamiltonian . this strategy stems from the tight - binding lattice model , in which the edge is automatically included as a part of the hamiltonian . a similar concept is found in the article by berry _ et al . _ @xcite , in which the authors modeled the edge using a mass term , @xmath484 . they considered that a singularity of the mass @xmath485 outside of the edge is necessary , in order to uniquely specify the pseudospin . we have observed a similar situation for the deformation - induced gauge field for the edge , that is , the field is singular at the edge . it is also interesting to note that @xmath483 in eq . ( [ eq : deq ] ) corresponds to the mass of a dirac fermion , and that the armchair edge is a singular point as for the mass . in summary , we have proposed a framework in which the edge is represented as the deformation - induced gauge field . we have used the framework to investigate the standing waves and edge states in the presence of a mass term and a magnetic field . the description of the edge using the deformation - induced gauge field is one attempt to better understand the edge . if we can describe the variety of edge structures as different configurations of a single gauge field , it provides a basis to further explore the properties near the edge . this work was financially supported by a grant - in - aid for specially promoted research ( no . 20001006 ) from the ministry of education , culture , sports , science and technology ( mext ) . the configurations of the pseudospin field for three equivalent corners of the graphene bz are not the same , as shown in fig . [ fig : pspin ] . consideration of this pseudospin behavior is given in this appendix . the tight - binding hamiltonian can be written as @xcite @xmath486 where @xmath487 ( @xmath28 ) . note that @xmath488 satisfies @xmath489 where @xmath215 and @xmath287 are integers . hence , the representations of @xmath490 , @xmath491 , and @xmath492 are different from each other , and are related via @xmath493 and @xmath494 , where @xmath495 for the solution @xmath496 of @xmath490 , we have the corresponding solution of the effective hamiltonian at @xmath497 as @xmath498 . @xmath499 is a rotational matrix for the pseudospin around the @xmath205-axis , so that the pseudospin of @xmath496 and that of @xmath500 are related by rotation around the @xmath205-axis by an angle of @xmath501 . this explains why the configurations of the pseudospin field around the three equivalent k ( k@xmath6 ) points are different from each other , as shown in fig . [ fig : pspin ] . next , we consider the effective hamiltonians for three equivalent k ( k@xmath6 ) points . by expanding @xmath488 around the wave vector of the k point , @xmath502 , we obtain @xmath503 . using @xmath504 , @xmath505 , and @xmath506 , we have @xmath507 , where @xmath508 and @xmath509 . then @xmath510 , and @xmath511 are obtained . the same argument can be applied to the k@xmath6 points . for the k@xmath6 point at @xmath512 , we obtain the effective hamiltonian @xmath513 . it is then straightforward to obtain @xmath514 , and @xmath515 . this difference in the representations of the effective hamiltonians does not cause a problem , because a coordinate transformation can be used to eliminate the @xmath499 matrix from one effective hamiltonian ( see also appendices in ref . ) . @xcite here , we imply the coordinate transformation as the rotation of the @xmath0 and @xmath1-axes by @xmath516 . a coordinate transformation can not alter the physics , and therefore the physical result derived from the effective hamiltonians are the same . rather , by using the change of the effective hamiltonians under a translation given by the reciprocal lattice vectors , a constraint for the form of the effective hamiltonians can be obtained . for example , the deformation hamiltonian , @xmath517 , should transform in the same way as @xmath518 . therefore , we must have @xmath519 , @xmath520 , and @xmath521 . otherwise , there would be three physically distinct effective hamiltonians for the same k point . the deformation - induced gauge field @xmath53 satisfies this constraint , because @xmath522 note that this equation is equivalent to eq . ( [ eq : gauge ] ) . the phase factor of @xmath523 appears when we change @xmath524 to @xmath525 and to @xmath526 , due to the factor of @xmath527 on the right - hand side . a notable feature is that the constraint must be satisfied for a strong lattice deformation that corresponds to a large value of @xmath53 . therefore , the direction of the gauge field does not change , although the values of @xmath528 are renormalized for a strong deformation . the edge states in the presence of a mass term is of interesting , because the magnetism of the edge states is related to the generation of a local spin - dependent mass term due to the coulombic interaction . @xcite here , we show how to obtain the edge states in the presence of a uniform mass term . by substituting eq . ( [ eq : phir ] ) into @xmath529 , we obtain instead of eq . ( [ eq : weyl+g ] ) @xmath530 where the variables @xmath531 and @xmath153 are respectively defined as @xmath532 the solution of the second equation in ( [ eq : g ] ) is @xmath533 the first equation in eq . ( [ eq : g ] ) is integrated with respect to @xmath1 from @xmath68 to @xmath69 . considering the limit @xmath173 , only singular functions of @xmath66 and @xmath240 at @xmath59 can survive after the integration , so that we obtain @xmath534 using eqs . ( [ eq : caseg ] ) and ( [ eq : gauge_con ] ) , we see from the first equation in ( [ eq : g ] ) that @xmath535 holds except very close to the edge . from eqs . ( [ eq : gauge_con ] ) and ( [ eq : xs ] ) , we see that @xmath251 is given by @xmath536 note that @xmath251 in the presence of a mass term is identical to @xmath251 in the absence of the mass given in eq . ( [ eq : xi - n ] ) . the mass term would affect @xmath251 , but this is not the case . from eqs . ( [ eq : x ] ) and ( [ eq : xs ] ) we obtain the energy eigenvalue @xmath537 when @xmath538 , we obtain @xmath539 and @xmath540 . according to the definition of @xmath153 in eq . ( [ eq : x ] ) , the sign of @xmath153 depends on the signs of both @xmath287 and @xmath250 . let us first consider the case of @xmath541 , by which we have @xmath542 . using this expression for @xmath153 in eq . ( [ eq : caseg ] ) , we obtain @xmath543 to determine @xmath544 , we substitute eq . ( [ eq : ene ] ) into eq . ( [ eq : x ] ) , and considering that @xmath545 can be approximated as @xmath546 for @xmath547 , we then have @xmath548 for @xmath549 . since @xmath550 for the zigzag edge , eq . ( [ eq : caseg(y ) ] ) becomes @xmath551 when @xmath552 . the wave function of this eigenstate for @xmath60 , which has unpolarized pseudospin , is negligible due to the normalization . thus , the localized state with energy @xmath553 in the valence energy band can appear near the edge only for @xmath61 , and the wave function is given by @xmath554 . similarly , for the case of @xmath555 , we have @xmath556 the corresponding wavefunction is has the pseudospin down state , which appears only for @xmath60 near the edge . it is noted that the mass term automatically selects the region where the edge state can appear , @xmath60 or @xmath61 . this is reasonable , because we have used the particle - hole symmetry operator @xmath278 to restrict the edge state only for @xmath60 or @xmath61 in sec . [ ssec : es ] . the particle - hole symmetry operator is nothing but the mass term .

astrometric measurements are presented for seven nearby stars with previously detected planets : six m dwarfs ( gj 317 , gj 667c , gj 581 , gj 849 , gj 876 , and gj 1214 ) and one k dwarf ( bd 3166 ) . measurements are also presented for six additional nearby m dwarfs without known planets , but which are more favorable to astrometric detections of low mass companions , as well as three binary systems for which we provide astrometric orbit solutions . observations have baselines of three to thirteen years , and were made as part of the recons long - term astrometry and photometry program at the ctio / smarts 0.9 m telescope . we provide trigonometric parallaxes and proper motions for all 16 systems , and perform an extensive analysis of the astrometric residuals to determine the minimum detectable companion mass for the 12 m dwarfs not having close stellar secondaries . for the six m dwarfs with known planets , we are not sensitive to planets , but can rule out the presence of all but the least massive brown dwarfs at periods of 2 12 years . for the six more astrometrically favorable m dwarfs , we conclude that none have brown dwarf companions , and are sensitive to companions with masses as low as 1 for periods longer than two years . in particular , we conclude that proxima centauri has no jovian companions at orbital periods of 2 12 years . these results complement previously published m dwarf planet occurrence rates by providing astrometrically determined upper mass limits on potential super - jupiter companions at orbits of two years and longer . as part of a continuing survey , these results are consistent with the paucity of super - jupiter and brown dwarf companions we find among the over 250 red dwarfs within 25 pc observed longer than five years in our astrometric program .

1407.4820

during the brief history of extrasolar planet investigations , our understanding of the relative populations of different types of planets has been limited by the observational biases of the techniques employed . with the advent of sophisticated transit searches and hypersensitive radial velocity measurements , significant progress has been made discovering various types of planets that orbit stars with periods up to a few years . less progress has been made in discovering planets in longer orbits , and particularly around nearby m dwarfs , which account for at least 74% of the stellar population within 10 pc @xcite . m dwarfs offer fertile ground for companion searches , as @xcite have inferred that a high fraction of m dwarfs host terrestrial planets at short orbital periods . less is known about the populations of jupiter - mass planets and brown dwarfs around m dwarfs , particularly at orbital periods longer than a few years . to understand how m dwarf planetary systems form and evolve , we must probe the full regime of companion masses and orbital periods . transit techniques are geometrically biased towards companions with small orbits , while radial velocity techniques are biased towards massive companions with short periods that exert large gravitational accelerations on their host stars . direct imaging techniques are limited to young , giant planets at large separations . astrometric techniques , which measure the positions of stars on the plane of the sky , are most sensitive to jovian - type planets in jovian - type orbits . while radial velocity observing programs are now becoming sensitive to such companions @xcite , the astrometric results presented here have longer observational baselines , of up to 13 years . furthermore , astrometry can detect companions with a large range of inclinations and orientations , and allow for the determination of orbit inclinations and accurate companion masses . to date the majority of nearby extrasolar planets around m dwarfs have been discovered by radial velocity searches , which tend to select the brightest m dwarfs . as discussed in more detail in @xmath2[sec : analysis ] , in ground - based imaging programs the brightest targets generally have the noisiest astrometric residuals due to the short exposures required and the lack of comparably bright reference stars . with the exception of gj 1214 , five m dwarfs in our sample were found to have planets using radial velocity techniques , and are among the brightest targets in our astrometric program . an extreme case is the k dwarf bd @xmath010 3166 , for which we are not sensitive to sub - stellar companions , but for which we provide the first accurate parallax . for comparison , we have included six additional m dwarfs that are less bright , less massive , and closer , and therefore more favorable to companion detection via astrometry . to calibrate our analysis , we have also included three confirmed stellar binaries with clear photocentric perturbations for which we have characterized the orbits . these binaries were chosen from the roughly two dozen binaries in our observing program with clear astrometric perturbations because we have observed multiple orbital periods , and can most accurately characterize the orbits . astrometric solutions for proper motion and parallax are given for each of the 16 systems targeted , plus orbital solutions for three binaries . a detailed analysis of the astrometric residuals is given to search for companions to the 12 m dwarf systems without close stellar companions . periodograms of the astrometric residuals have been generated , along with detection limits based on simulations of 10 million hypothetical companions to each star . these are the first results of a larger recons survey for companions orbiting more than 250 red dwarfs within 25 pc for which we have at least five years of coverage . as observations continue , this sample will grow , further constraining the population of brown dwarf and super - jupiter companions in long period orbits around m dwarfs . finally , to provide context for these results we provide a comprehensive list of the 17 m dwarfs within 25 pc having exoplanets as of 1 july 2014 , including the six targeted in this work . the 0.9 m telescope at ctio is equipped with a 2048 @xmath3 2048 tektronix ccd camera with 0.401 pixel@xmath4 plate scale @xcite . only the center quarter of the chip is used for astrometric and photometric observations , yielding a 6.8 square field of view . astrometric exposures are taken through one of four filters , @xmath5 ( old ) , @xmath5 ( new ) , @xmath6 , or @xmath7 ( old ) , @xmath5 ( new ) , @xmath6 , and @xmath7 filters are 5438 , 5475 , 6425 , and 8075 , respectively . ] ( hereafter without subscripts , and the @xmath8 filters combined ) . depending on the brightnesses of the science targets , reference stars , and sky conditions , exposure times vary from 20 to 1200 s for targets with 9 @xmath9 . for optimal centroiding , exposure times are set so that either science or reference stars have maximum peak adu of @xmath1050,000 ( digital saturation occurs at 65,537 adu ) . observations are almost always made within @xmath11 30 minutes of a science target s transit to minimize the corrections required for differential color refraction , as described in @xcite . three to five frames are typically taken each night , depending primarily on the exposure time required . to enable routine calibration of the science images , bias and dome flat frames are taken nightly . instrument setups for most stars have been kept constant during the 13 years of observations . however , we have used two @xmath8 filters , dubbed the `` old '' tek#2 @xmath8 filter ( @xmath12 = 5438 , @xmath13 = 1026 ) and `` new '' tek#1 v filter ( @xmath12 = 5475 , @xmath13 = 1000 ) , because the `` old '' filter cracked in 2005 february . the `` new '' @xmath8 filter was used between 2005 and 2009 . the `` old '' @xmath8 filter was reinstated in 2009 july after confirming that the crack in the corner did not significantly affect astrometric residuals . as discussed in @xcite , a reliable parallax can be obtained using data from both filters as long as at least 1 - 2 years of data ( depending on observing frequency ) have been taken in each filter . reductions containing both `` old '' and `` new '' @xmath8 frames can exhibit offsets of a several milliarcseconds ( mas ) in residuals on both axes . this has been mitigated by choosing close - in reference stars , and only using frames taken near the meridian . in total , 7 of the 16 systems discussed in this paper were observed astrometrically in the @xmath8 filter . further details about the filters and their effects on the astrometry can be found in @xcite and @xcite . the paths traced on the sky by science stars result from the combinations of proper motions and parallactic shifts . details of the data reduction process used to separate these motions are given in @xcite and @xcite . briefly , we ( 1 ) use sextractor @xcite to measure centroids , ( 2 ) define a six - constant plate model to find plate constants ( given in equation ( 4 ) of @xcite ) , ( 3 ) assume that ensembles of reference stars have zero mean parallax and proper motion , ( 4 ) solve least - square equations for multi - epoch images ( given in equation ( 5 ) of @xcite ) , and ( 5 ) convert from relative parallax to absolute parallax by estimating the distances of the reference stars photometrically . our typical centering precision is 2.1 - 3.5 mas , or 0.5 - 0.9% of a pixel , depending on the filter , with @xmath14 being the best and @xmath8 being the worst . to correct the relative parallax to an absolute parallax , photometric distances are estimated by comparing @xmath15 colors to @xmath16 for single , main - sequence stars in the recons 10 pc sample @xcite . a distance is estimated for each reference star , and the correction to absolute parallax is then computed using the weighted mean distance of the entire reference field . the uncertainty on the correction is determined using equation ( 6 ) in @xcite . in the case of a binary with a given combination of magnitude and mass differences , where @xmath17 is the photocentric semi - major axis of the orbit of the primary , @xmath18 is the fractional mass ( @xmath19 ) , @xmath20 is the relative luminosity ( @xmath21 ) , and @xmath22 is the semi - major axis of the relative orbit of the two components . the perturbation we detect here is @xmath17 . ] , we detect its photocenter orbit around its barycenter , in addition to the motions due to parallax and proper motion . hence , the residuals of our typical binary star s parallax reduction are significantly offset from zero . in order to get a better parallax result and calculate the photocenter s orbital elements , we first carry out a standard reduction for proper motion and parallax . we then fit a photocentric orbit to the residuals , i.e , we treat these residuals as a binary orbit , using the techniques described in @xcite . based on the orbital elements we calculate , this photocentric orbit in components of r.a . and decl . is then removed from the centroids of the science star at each epoch . finally , we re - calculate proper motion and parallax using these corrected centroids . the final `` cleaned '' proper motions and parallaxes are the values given for the binaries in table [ tab1 ] . after one iteration , the parallax errors are reduced to those typical of similar program stars , and the residuals are significantly reduced and consistent with the mean errors found for our overall program . @xmath15 photometry was obtained at the ctio 0.9 m using the same instrumental setup used for the astrometry frames . as for astrometry observations , bias and dome flat frames are taken nightly for basic image calibration . all science stars were observed at airmass @xmath23 . exposure times were chosen to reach a signal - to - noise ratio ( s / n ) @xmath24 100 for science stars in each of the @xmath15 filters . combinations of fields that provided 10 or more standard stars from @xcite and/or e - regions from @xcite were observed several times each night to derive transformation equations and extinction curves . further details of photometric data reductions , the definition of transformation equations , errors , etc . , can be found in @xcite and @xcite . spectroscopic observations used to the provide spectral types in table 2 were made at the ctio 1.5 m using the r - c spectrograph and loral 1200 @xmath3 800 ccd camera between 2003 and 2006 . grating # 32 was used in first order with a tilt of 15.@xmath251 , and observations were made using a 2 slit . the order - blocking filter og570 was utilized to provide spectra covering the range of 6000 - 9000 with a resolution of 8.6 . for calibration , bias frames , dome flats , and sky flats were taken at the beginning of each night . further details regarding the 1.5 m spectroscopy program and associated data reduction , including assignment of spectral types , can be found in @xcite . astrometric residuals represent the deviation in a target s measured position from the solution for proper motion and parallax , given for all 16 target stars in table [ tab1 ] . the residuals for each star are plotted in figure [ fig1 ] . each filled circle represents the mean of typically three to five frames taken in a single night , with a corresponding estimate of the nightly mean error . open circles represent nights with only one frame , which are included in the parallax measurement but are not included in the following analysis of the residuals . the three binaries ( gj 748ab , lhs 1582ab , and lhs 3738ab ) exhibit large , periodic perturbations in both r.a . and decl . , indicating the presence of a companion in each case . because the systems are unresolved in our images , we calculate a photocentric semi - major axis , as previously discussed . the remaining targets generally have flat residuals , although some are more scattered ( gj 849 ) than others ( gj 1128 ) . targets with the flattest residuals , such as denis j1048 - 3956 , generally have at least 8 reference stars that closely and evenly surround the target , and are bright enough to have a least 1,000 peak counts when the target has 50,000 . fainter targets tend to have the flattest residuals because they have a larger number of suitable reference stars and require longer exposures , which smooth out short - term seeing effects that produce psfs with poorly - defined centroids . the residuals were analyzed for the presence of companions using lomb - scargle periodograms , shown in figure [ fig2 ] , generated by the method given in @xcite . this method weights each data point by its estimated nightly mean error , and produces a normalized periodogram for the data sequence with power ranging from 0 to 1 . we only attempt to detect companions with periods greater than two years but less than the length of the observations , and have only generated periodograms for those ranges . the data were oversampled in increments of one day to create smooth periodograms . as @xcite have shown , oversampling the periodogram does not significantly increase the risk of false periodogram peaks . as the data are irregularly spaced , it was not possible to use an analytical formula for the false alarm probabilities ( fap ) . instead , empirical fap distribution functions were calculated as outlined in @xcite by generating periodograms for 10,000 sets of random noise . each of the 10,000 data sets has the same number of observations as the actual data , with the same observation times and errors as the actual data . the randomized values of each point within a given data set are assigned from a normal distribution with a standard deviation representing the distribution of offsets from zero of the actual data points . each of these 10,000 periodograms was then sampled in a grid of periods with five day increments between two years and the length of observations to find the highest power occurring in each periodogram . the fap for an observed periodogram peak of a given power is equal to the fraction of the 10,000 data sets with peaks greater than or equal to that power . the normalized powers corresponding to fap values of 0.1 , 0.01 , and 0.001 are shown as dashed lines in figure [ fig2 ] . this corresponds to 10% , 1% , and 0.1% probabilities that the peak is due solely to random noise and the cadence of observations . as a calibration , periodograms and fap functions were also generated for the three confirmed stellar binaries for which astrometric perturbations are clearly evident . these binaries have significant periodogram peaks ( fap @xmath26 0.001 ) in both the r.a . and decl . axes at periods corresponding to their orbital periods , while the remaining 13 targets have no significant peaks . this indicates that our periodograms are sensitive to astrometric perturbations , and that we have not detected any significant periodicity in the residuals of the remaining 12 m dwarfs and one k dwarf without companions . given that there are no companions evident in the residuals of the non - binary targets , we aim to establish lower limits for the companion masses and periods to which we are sensitive , and would have detected were they present . for each target we ran a simulation of 10 million hypothetical companions orbiting each star with masses chosen randomly from a uniform distribution between 0.5 and 80 @xmath1 , and periods chosen randomly from a uniform distribution between 2 years up to the length of observations . although we have observed perturbations as short as 1.2 years , our practical lower limit is 2 years . all geometric parameters ( inclination , eccentricity , time of periastron , longitude of ascending node , and longitude of periastron ) were assigned randomly from a uniform distribution of all possible values , including eccentricity , which was allowed to be as high as 0.99 . the program then calculates the astrometric perturbation that each simulated companion would induce on the primary , using a primary mass calculated from the mass - luminosity relations in @xcite ( equations 5a and 5b ) and the revised relation for the lowest mass stars in @xcite . for den j1048 - 3956 and lp 944 - 020 , which are too faint for the relations , we assume masses of 0.08 @xmath27 . we assume that the companion does not contribute significantly to the overall flux in each system , so that the photocenter of the system is concentric with the primary star . this assumption proves problematic when considering brown dwarf companions to the two latest m dwarfs , den j1048 - 3956 and lp 944 - 020 . however , the trigonometric parallaxes of these targets agree well with their photometric distance estimates ( columns 12 and 13 of table [ tab2 ] ) , and we conclude that neither have nearly equal luminosity companions . the goodness - of - fit between the observed data @xmath28 with errors @xmath29 , and a flat line with @xmath30 was determined using the reduced chi squared statistic @xmath31 where @xmath32 is the number of degrees of freedom , given by @xmath33 for @xmath34 data points and @xmath35 fitted parameters . because we did not attempt to fit a model to the data , but only to analyze how well a simulation fits the data , we set @xmath35 equal to zero . as data points with small error bars are more heavily weighted , we discarded epochs that have unrealistic errors smaller than 1 mas representing less than 5% of the epochs because such points overconstrain the orbits that can be fit . simulated orbits for which @xmath36 was greater than 4 in at least one of either r. a. or decl . we consider to be orbits that we would have detected . this threshold of 4 is based empirically on the values plotted in figure [ fig3 ] , where the lower panel is an inset of the upper . the plotted @xmath36 values were calculated for ( 1 ) the stars included in this paper , ( 2 ) calibration stars with flat residuals we use to monitor potential fluctuations in equipment and our data reduction pipeline , and ( 3 ) additional known binaries with perturbations . we compare these targets residuals to the case of flat residuals , i.e. , an exact astrometric solution with no perturbation . as expected , the binaries with perturbations have large @xmath36 values , indicating that a flat line is a poor fit to the data . the two solid points inside the @xmath37 box are long term perturbations with large gaps in the astrometric observations . as more data are collected , those two points will move to larger @xmath36 values outside the box , but we include the points here for completeness . for those stars with no perturbations , all have @xmath36 less than 4 in both axes , and are centered around 1 . this indicates that we are accurately calculating our measurement errors , and provides an empirical @xmath36 value of 4 , above which we are sensitive to perturbations . table [ tab1 ] gives the parallax and proper motion results for the 16 systems , with details about the astrometric observations ( filters used , number of seasons observed , number of frames used in reductions , time coverage , span of time , and the number of reference stars ) and results ( relative parallaxes , parallax corrections , absolute parallaxes , proper motions , position angles of the proper motions , and the derived tangential velocities based on relative proper motions and parallaxes ) . all but two of the sixteen systems have parallax errors of @xmath102 mas or less . bd @xmath010 3166 and gj 876 have larger errors due to combinations of faint reference stars and short exposures . corrections to absolute parallax are generally less than @xmath102 mas , so systematics in the corrections should not significantly affect the results . three targets ( gj 317 , gj 667c , and bd @xmath010 3166 , ) have corrections of @xmath103@xmath38-@xmath384 mas due to reddening of the reference stars , which skews their photometric distance estimates . in these cases , we adopt a generic correction of 1.50 @xmath11 0.50 mas . the per observation precision for each target is listed in column 16 , representing the mean of the observation errors in r.a . and decl . the percentage of companions eliminated listed in column 17 is discussed in @xmath2[sec : discuss ] . lccccccccrrrrrrccc + gj 317 & 08 40 59.21 & @xmath023 27 22.6 & r & 5c & 75 & 2009.04 - 2013.38 & 4.35 & 7 & 64.04 @xmath11 1.45 & 1.50 @xmath11 0.50 & 65.54 @xmath11 1.53 & 930.7 @xmath11 1.1 & 330.5 @xmath11 0.13 & 65.5 & 4.99 & & + bd @xmath010 3166 & 10 58 28.79 & @xmath010 46 13.4 & i & 7s & 71 & 2004.43 - 2011.50 & 7.07 & 6 & 13.84 @xmath11 3.04 & 1.50 @xmath11 0.50 & 15.34 @xmath11 3.08 & 185.9 @xmath11 1.5 & 269.1 @xmath11 0.67 & 52.4 & 10.85 & & + gj 581 & 15 19 26.83 & @xmath007 43 20.1 & v & 14s & 267 & 2000.58 - 2013.38 & 12.80 & 11 & 157.67 @xmath11 1.57 & 1.12 @xmath11 0.17 & 158.79 @xmath11 1.58 & 1224.3 @xmath11 0.4 & 266.0 @xmath11 0.03 & 36.5 & 7.93 & 96% & a + gj 1214 & 17 15 18.92 & @xmath3904 57 50.1 & i & 4c & 80 & 2010.39 - 2013.38 & 3.00 & 9 & 68.20 @xmath11 1.26 & 1.88 @xmath11 0.18 & 70.08 @xmath11 1.27 & 945.5 @xmath11 1.4 & 142.0 @xmath11 0.17 & 63.9 & 5.02 & & + gj 667c & 17 18 58.82 & @xmath034 59 48.6 & v & 11s & 140 & 2003.52 - 2013.38 & 9.86 & 5 & 139.38 @xmath11 1.98 & 1.50 @xmath11 0.50 & 140.88 @xmath11 2.04 & 1154.1 @xmath11 0.6 & 101.0 @xmath11 0.05 & 38.8 & 7.66 & 93% & + gj 849 & 22 09 40.34 & @xmath004 38 26.8 & v & 11s & 135 & 2003.52 - 2013.39 & 9.86 & 5 & 113.78 @xmath11 1.97 & 2.27 @xmath11 0.30 & 116.05 @xmath11 1.99 & 1118.0 @xmath11 0.5 & 90.8 @xmath11 0.04 & 45.7 & 9.53 & 81% & + gj 876 & 22 53 16.75 & @xmath014 15 49.2 & v & 11s & 85 & 2003.52 - 2013.39 & 9.87 & 6 & 210.97 @xmath11 3.99 & 2.14 @xmath11 0.57 & 213.11 @xmath11 4.03 & 1149.4 @xmath11 1.1 & 125.7 @xmath11 0.11 & 25.6 & 8.43 & 99% & + + gj 1061 & 03 35 59.72 & @xmath044 30 45.5 & r & 13s & 194 & 1999.62 - 2012.95 & 13.32 & 7 & 269.92 @xmath11 1.29 & 0.94 @xmath11 0.08 & 270.86 @xmath11 1.29 & 827.7 @xmath11 0.3 & 117.7 @xmath11 0.04 & 14.5 & 7.59 & 79% & a + lp 944 - 020 & 03 39 35.25 & @xmath035 25 43.8 & i & 8s & 59 & 2003.95 - 2012.94 & 8.99 & 10 & 154.53 @xmath11 1.03 & 1.36 @xmath11 0.10 & 155.89 @xmath11 1.03 & 408.3 @xmath11 0.3 & 48.5 @xmath11 0.07 & 12.4 & 2.13 & 94% & a + gj 1128 & 09 42 46.36 & @xmath068 53 06.1 & v & 13s & 167 & 2000.23 - 2013.12 & 12.89 & 8 & 153.54 @xmath11 0.75 & 0.73 @xmath11 0.11 & 154.27 @xmath11 0.76 & 1123.0 @xmath11 0.2 & 356.1 @xmath11 0.02 & 34.5 & 3.17 & 80% & a + denis j1048 - 3956 & 10 48 14.56 & @xmath039 56 07.0 & i & 13s & 200 & 2001.15 - 2013.27 & 12.13 & 11 & 247.23 @xmath11 0.60 & 0.85 @xmath11 0.10 & 248.08 @xmath11 0.61 & 1531.6 @xmath11 0.2 & 229.5 @xmath11 0.01 & 29.3 & 2.92 & 97% & a + scr 1138 - 7721 & 11 38 16.76 & @xmath077 21 48.5 & i & 11s & 134 & 2003.25 - 2013.27 & 10.03 & 12 & 119.60 @xmath11 1.01 & 0.81 @xmath11 0.07 & 120.41 @xmath11 1.01 & 2143.3 @xmath11 0.4 & 287.8 @xmath11 0.02 & 84.4 & 4.20 & 69% & a + proxima cen & 14 29 43.02 & @xmath062 40 46.7 & v & 14s & 205 & 2000.57 - 2013.25 & 12.68 & 5 & 766.41 @xmath11 0.91 & 1.72 @xmath11 0.50 & 768.13 @xmath11 1.04 & 3850.8 @xmath11 0.6 & 282.4 @xmath11 0.02 & 23.8 & 4.83 & 99% & a + + lhs 1582ab & 03 43 22.08 & @xmath009 33 50.9 & r & 11s & 102 & 2000.87 - 2012.94 & 12.06 & 7 & 48.84 @xmath11 1.18 & 2.00 @xmath11 0.26 & 50.84 @xmath11 1.21 & 509.4 @xmath11 0.3 & 52.7 @xmath11 0.06 & 47.5 & 3.88 & & a + gj 748ab & 19 12 14.60 & @xmath3902 53 11.0 & v & 10s & 154 & 2004.45 - 2013.39 & 8.95 & 11 & 97.77 @xmath11 1.15 & 2.22 @xmath11 0.41 & 99.99 @xmath11 1.22 & 1857.8 @xmath11 0.5 & 107.4 @xmath11 0.02 & 88.1 & 5.80 & & + lhs 3738ab & 21 58 49.13 & @xmath032 26 25.5 & r & 12s & 151 & 1999.64 - 2012.81 & 13.17 & 12 & 50.82 @xmath11 1.01 & 1.40 @xmath11 0.21 & 52.22 @xmath11 1.03 & 535.2 @xmath11 0.3 & 229.1 @xmath11 0.06 & 48.6 & 2.50 & & a + [ tab1 ] lrrrcrrrcccrrcc + gj 317 & 12.01 & 10.84 & 9.37 & 3 & 7.934 & 7.321 & 7.028 & m3.5 v & 1 & 0.35 & 15.26 @xmath11 0.36 & 9.70 @xmath11 1.53 & 12 & + bd @xmath010 3166 & 10.03 & 9.58 & 9.19 & 3 & 8.611 & 8.300 & 8.124 & k3.0 v & 2 & 0.85 & 65.19 @xmath11 13.64 & & & a + gj 581 & 10.56 & 9.44 & 8.03 & 3 & 6.706 & 6.095 & 5.837 & m3.0 v & 2 & 0.30 & 6.30 @xmath11 0.06 & 6.60 @xmath11 1.03 & 12 & + gj 1214 & 14.71 & 13.27 & 11.50 & 3 & 9.750 & 9.094 & 8.782 & m4.5 v & 1 & 0.14 & 14.27 @xmath11 0.24 & 12.42 @xmath11 2.00 & 12 & + gj 667c & 10.34 & 9.29 & 8.09 & 3 & 6.848 & 6.322 & 6.036 & m1.5 v & 2 & 0.36 & 7.10 @xmath11 0.10 & 9.41 @xmath11 1.49 & 12 & + gj 849 & 10.38 & 9.27 & 7.87 & 3 & 6.510 & 5.899 & 5.594 & m3.0 v & 2 & 0.42 & 8.62 @xmath11 0.15 & 5.73 @xmath11 0.92 & 12 & + gj 876 & 10.18 & 8.97 & 7.40 & 3 & 5.934 & 5.349 & 5.010 & m3.5 v & 2 & 0.27 & 4.69 @xmath11 0.09 & 3.46 @xmath11 0.54 & 12 & + + gj 1061 & 13.09 & 11.45 & 9.47 & 6 & 7.523 & 7.015 & 6.610 & m5.0 v & 3 & 0.11 & 3.69 @xmath11 0.02 & 3.55 @xmath11 0.60 & 12 & + lp 944 - 020 & 18.69 & 16.39 & 13.98 & 3 & 10.725 & 10.017 & 9.548 & m9.0 v & 4 & 0.08 & 6.42 @xmath11 0.04 & 7.04 @xmath11 1.32 & 11 & + gj 1128 & 12.74 & 11.36 & 9.62 & 3 & 7.953 & 7.385 & 7.037 & m4.0 v & 2 & 0.15 & 6.48 @xmath11 0.03 & 6.33 @xmath11 1.00 & 12 & + denis j1048 - 3956 & 17.37 & 14.98 & 12.47 & 4 & 9.538 & 8.905 & 8.447 & m8.0 v & 2 & 0.08 & 4.03 @xmath11 0.01 & 4.48 @xmath11 0.73 & 10 & + scr 1138 - 7721 & 14.78 & 13.20 & 11.24 & 4 & 9.399 & 8.890 & 8.521 & m5.0 v & 3 & 0.11 & 8.31 @xmath11 0.07 & 9.45 @xmath11 1.71 & 12 & + proxima cen & 11.13 & 9.45 & 7.41 & 3 & 5.357 & 4.835 & 4.384 & m5.0 v & 2 & 0.11 & 1.30 @xmath11 0.01 & 1.15 @xmath11 0.18 & 12 & b + + lhs 1582ab & 14.69j & 13.33j & 11.60j & 4 & 9.799j & 9.177j & 8.854j & m4.5 vj & 1 & & 19.67 @xmath11 0.47 & 13.27 @xmath11 2.25 & 12 & c + gj 748ab & 11.10j & 9.95j & 8.47j & 3 & 7.087j & 6.572j & 6.294j & m3.5 vj & 2 & & 10.00 @xmath11 0.12 & 7.69 @xmath11 1.26 & 12 & c + lhs 3738ab & 15.78j & 14.29j & 12.46j & 3 & 10.654j & 10.091j & 9.761j & m4.5 vj & 5 & & 19.15 @xmath11 0.38 & 18.50 @xmath11 2.96 & 12 & c + [ tab2 ] nine of the 16 targets in this paper have parallaxes previously published by recons , and are noted in column 18 . the results presented here supersede those published previously by recons because additional data and improved reduction techniques have been used , as discussed in detail in @xcite . the identical parallax of lp 944 - 020 is also presented in @xcite as part of a study of the stellar hydrogen burning limit . for bd @xmath010 3166 , we did not run simulations because it is too massive and far away for us to detect any type of substellar companion . however , we do provide the first accurate parallax , and conclude that bd @xmath010 3166 is not physically related to the star with a similar proper motion , lp 731 - 076 , that is @xmath40 away @xcite . photometric and spectroscopic results are provided in table [ tab2 ] . @xmath15 photometry was taken using the ctio 0.9 m ( number of nights of photometry in column 5 ) , with errors in @xmath41 mag @xcite . @xmath42 photometry was retrieved from the two micron all sky survey ( 2mass , @xcite ) catalog . spectral types are given in column 9 with references in column 10 . mass estimates were calculated as discussed in @xmath2[sec : analysis ] . the photometric distances are calculated using the @xmath43 distance relations ( number of relations in column 14 ) detailed in @xcite . for systems with photometric and trigonometric distances that agree within the errors , we conclude that they lack nearly equal luminosity companions . the trigonometric distances of gj 748ab and lhs 1582ab are greater than their photometric distances due to companion contributions to the systems total flux . the two distances of lhs 3738ab agree well , indicating that the companion is significantly fainter than the primary . gj 317 and gj 849 have discrepant ( at 3.5@xmath44 and 3.1@xmath44 , respectively ) photometric and trigonometric distances , which does not necessarily mean that these stars have stellar companions , as main sequence stars within the same spectral type can vary somewhat in luminosity . @xcite note a radial velocity drift in their observations of gj 849 . @xcite also note this drift , and constrain the minimum companion mass to @xmath45 . our astrometry would show a photocenter shift for unequal mass components , as discussed in @xmath2[sec : observe ] . only components of roughly equal luminosity and mass would provide the additional flux with no perturbation . such a companion would have been observed to separations as close as 1 in our images , which corresponds to @xmath109 au . at a semimajor axis of 9 au , the orbital period is 29.5 years for an equal mass companion . this results in velocities for each component 9.1 km / s for edge - orbits . thus , for most orbital inclinations , such a stellar companion is ruled out by the radial velocity data . therefore , it is unlikely that a stellar companion similar to the primary is contributing to the overluminosity we observe . figure [ fig4 ] shows the range of periods and masses for which 90% of simulated companions would have been detected , based on the simulations for objects with masses from 0.5 to 80 @xmath1 . as discussed in @xmath2[sec : analysis ] , the masses , periods , and orbital parameters of the simulated companions were assigned randomly from a uniform distribution . for this discussion we set the dividing line between planets and brown dwarfs at 13 @xmath1 , and the dividing line between brown dwarfs and stars at 80 @xmath1 . the bottom panel of figure [ fig4 ] is an inset of the top , showing the best case targets in more detail . the noisy nature of the lines is due to the sizes of the bins used in the simulations . the bin sizes were chosen to achieve a reasonably high resolution , while still having enough simulated companions in each bin . the minimum detectable companion mass is smallest for companions with long periods , which produce the largest amplitude perturbations in the astrometric data . the lower panel indicates that for the best case targets ( stars at close distances and of low mass ) . we are most sensitive to jovian - type planets in jovian - type orbits . for a companion at a given mass and orbital period , the amplitude of the resulting astrometric perturbation depends on the orbital parameters of the system , and the mass and distance of the primary . the detection limits we report are based on simulations of companions with a wide range of masses , periods , and orbital parameters . therefore we give a few representative examples of how the results in figure [ fig4 ] translate into astrometric perturbations in mas . in the case of a face - on , circular orbit , a 20 @xmath1 companion in a 4 year orbit around gj 581 would cause a 16 mas perturbation , while a 15 @xmath1 companion in an 8 year orbit would cause a 20 mas perturbation , and a 10 @xmath1 companion in a 12 yr orbit would cause a 17 mas perturbation . for circular , face - on orbits around proxima centauri , companions of 1.5 , 1 , and 0.5 @xmath1 in orbits of 4 , 8 , and 12 years would cause perturbations of 12 , 13 , and 8 mas , respectively . these values are significantly greater than the per observation precisions listed table [ tab1 ] 7.93 mas for gj 581 and 4.83 mas for proxima . thus , the 90% detection thresholds given in figure [ fig4 ] are reasonable for the four planet hosts observed longer than eight years , column 17 of table [ tab1 ] gives the percentages of simulated _ brown dwarf _ companions , ranging from 81 99% , eliminated with orbits between two and eight years , and masses between 13 and 80 @xmath1 . approximately 92% of all simulated brown dwarfs have been eliminated as companions to those stars known to host exoplanets . for the six more astrometrically favorable targets , we calculate the percentages of simulated _ planetary _ companions eliminated with orbits between two and eight years and masses between 1 and 13 @xmath1 , with results ranging from 69 99% . we have eliminated @xmath1086% of all simulated planets with masses of 1 13 @xmath1 around these six astrometrically favorable stars , and effectively all brown dwarf companions in orbital periods of 2 8 years . photocentric orbital solutions for the three binaries are shown in figure [ fig5 ] with the corresponding orbital parameters given in table [ tab3 ] . from our astrometric data for gj 748 ab , we find an orbital period of 2.504 @xmath11 0.025 years , which is consistent with the two detailed studies of the system by @xcite , who found p = 2.466 @xmath11 0.008 years , and @xcite , who found p = 2.469 @xmath11 0.001 years using hst fine guidance sensor data . however , we determine an eccentricity of 0.06 , which is inconsistent with the value of 0.45 found in both of the hst studies . we utilized the orbit - fitting code described in @xcite and set starting eccentricities of 0.05 to 0.95 in increments of 0.05 ; regardless , our data converged to the e = 0.06 value each time . lrrrrrrr gj 748ab & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + lhs 1582ab & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] + lhs 3738ab & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & [email protected] & 130.5@xmath11 5.1 + [ tab3 ] the discrepancy between our eccentricity and that of the hst studies is likely due to our observations of gj 748 ab having been taken at the two different @xmath8 filters discussed in @xmath2 2.1 . while the two filters are photometrically identical within measurable errors @xcite , they are not astrometrically identical . we have analyzed the astrometric residuals for over 500 targets without detectable perturbations in the three different filters ( @xmath46 and @xmath14 ) over the length of our observing program . the r and i filters are stable , but astrometric offsets in the v filters are evident over the time period when the problematic `` new '' @xmath8 filter was used . these offsets have been mitigated as discussed in @xmath2 2.1 , allowing data from both @xmath8 filters to be used to produce reliable parallax results . in the case of gj 748 ab , we are able to recover the correct period , but the offsets in the residuals are likely contributing to the errant eccentricity . we presently do not possess enough observations to perform a reduction of gj 748 ab without the problematic @xmath8 filter data . we include the current solution because it is our only system for which an accurate period has been published , against which to compare our results . in contrast to our photocenter data , the fgs observations resolve the system into two components at 15 epochs over 1.8 years . they are exquisitely sensitive to the separation and position angle of the secondary from the primary , and are to be preferred to our ground - based results for the eccentricity . a clever suggestion by hugh harris of usno has been suggested to solve this dilemma . because in an unresolved system the center of mass location is unknown , the zero points for the residuals in r.a . and are unknown . by shifting the zero points and fitting the residuals , a different eccentricity may be derived . we await the acquisition of resolved data for several more systems before exploring this technique so that a robust analysis can be accomplished . at present errors on the eccentricities in table [ tab3 ] should be treated with caution . the orbital solutions for lhs 1582ab and lhs 3738ab are updated and improved over those presented in @xcite , which were the first orbits presented for each system . in addition to demonstrating the astrometric detection and characterization of unresolved companions , these results can provide additional dynamically determined masses for m dwarfs , once the systems have been resolved . in narrow - angle field astrometry programs such as the one outlined here , the lowest mass companions detectable by astrometry are dependent on a number of factors . these include the apparent brightness of the host star and availability of suitable reference stars , which affect the precision of the astrometric measurements . additionally , the target star mass , companion mass , the luminosity ratio , orbital separation , and system distance determine the size of the astrometric perturbation . the m dwarfs with planets that have been discovered so far , including the six in this paper , are not the most favorable to our astrometric observations . they are among the brighter red dwarfs , with the latest type stars at m3.5v for the radial velocity detections included in this paper . we are rarely sensitive to planets around these stars . however , we are able to rule out the presence of a large fraction of potential brown dwarf companions with masses of @xmath47 @xmath1 . as observations continue , we will be able to rule out companions at longer periods , particularly in the cases of gj 317 and gj 1214 , which we have so far only observed for 3 4 years . in addition , we have demonstrated that we are sensitive to planets with masses of @xmath48 @xmath1 around m dwarfs that are more favorable to astrometric observations . of particular interest , we find that proxima centauri has no companions more massive than 2 @xmath1 with periods of 2 5 years and more massive than 1 @xmath1 for 5 12 years . this is to be contrasted to the results of other companion searches of proxima centauri . using hst fine guidance sensor data , @xcite report a companion detection sensitivity of 1 @xmath1 at a 60 day period . based on 7 years of rv observations , @xcite preclude the presence of companions with @xmath49 at periods @xmath50 years . together these studies eliminate all jupiter mass planets around proxima centauri for orbital periods out to 12 years . gj 163 b & 04 09 15.6 & @xmath053 22 25 & 66.59 @xmath11 1.79 & 2 & 1,2 & 0.03 & 0.024 & 9 & + gj 163 c & & & & & & 0.02 & 0.070 & 9 & + gj 163 d & & & & & & 0.09 & 1.654 & 9 & + gj 176 b & 04 42 55.7 & @xmath3918 57 29 & 110.00 @xmath11 2.00 & 2 & 1,2 & 0.03 & 0.024 & 10 & + gj 179 b & 04 52 05.7 & @xmath3906 28 35 & 80.82 @xmath11 3.78 & 2 & 1,2 & 0.82 & 6.264 & 11 & + gj 317 b & 08 40 55.7 & @xmath023 28 00 & 65.34 @xmath11 0.39 & 3 & 1,3,4 & 1.81 & 1.895 & 4 & a + lhs 2335 b & 10 58 35.0 & @xmath031 08 38 & 50.55 @xmath11 1.55 & 1 & 5 & 0.02 & 0.007 & 12 & + gj 433 b & 11 35 26.9 & @xmath032 32 23 & 112.09 @xmath11 1.43 & 2 & 1,2 & 0.02 & 0.020 & 13 & a + gj 1148 b & 11 41 44.6 & @xmath3942 45 07 & 88.81 @xmath11 2.14 & 2 & 1,2 & 0.30 & 0.113 & 14 & + gj 436 b & 11 42 11.0 & @xmath3926 42 23 & 98.95 @xmath11 2.07 & 2 & 1,2 & 0.07 & 0.007 & 15 & + gj 581 b & 15 19 26.0 & @xmath007 43 20 & 159.28 @xmath11 1.32 & 3 & 1,2,3 & 0.05 & 0.015 & 16 & + gj 581 c & & & & & & 0.02 & 0.035 & 16 & + gj 581 d & & & & & & 0.02 & 0.183 & 16 & b + gj 581 e & & & & & & 0.01 & 0.009 & 16 & + lp 804 - 027 b & 16 12 41.7 & @xmath018 52 31 & 69.46 @xmath11 3.12 & 1 & 2 & 2.10 & 0.306 & 17 & + gj 649 b & 16 58 08.8 & @xmath3925 44 39 & 97.28 @xmath11 1.32 & 2 & 1,2 & 0.33 & 1.638 & 18 & a + gj 1214 b & 17 15 18.9 & @xmath3904 57 50 & 69.04 @xmath11 0.54 & 3 & 1,3,6 & 0.02 & 0.004 & 19 & c + gj 667c b & 17 18 57.1 & @xmath034 59 23 & 138.24 @xmath11 0.57 & 3 & 1,3,7 & 0.02 & 0.020 & 13 & a + gj 667c c & & & & & & 0.01 & 0.077 & 13 & + gj 674 b & 17 28 39.9 & @xmath046 53 42 & 220.11 @xmath11 1.39 & 2 & 1,2 & 0.03 & 0.013 & 20 & + gj 832 b & 21 33 33.9 & @xmath049 00 32 & 202.03 @xmath11 1.00 & 2 & 1,2 & 0.64 & 9.353 & 21 & + gj 849 b & 22 09 40.3 & @xmath004 38 26 & 113.46 @xmath11 1.32 & 3 & 1,2,3 & 0.90 & 5.241 & 22 & a + gj 876 b & 22 53 16.7 & @xmath014 15 49 & 214.45 @xmath11 0.57 & 4 & 1,2,3,8 & 1.95 & 0.167 & 23 & + gj 876 c & & & & & & 0.61 & 0.082 & 23 & + gj 876 d & & & & & & 0.02 & 0.005 & 23 & + gj 876 e & & & & & & 0.04 & 0.342 & 23 & + in the broader context , these results are consistent with recently published searches for jovian companions to m dwarfs at shorter orbital periods . transit searches are unlikely to detect companions at jovian orbits , due to the narrow range of detectable inclinations . @xcite found no jupiter - sized planets in their transit search , which is most sensitive to companions at orbital periods less than 10 days , and conclude that such planets rarely orbit m dwarfs . based on the first two years of their astrometric search , @xcite find no planetary mass companions to the 20 m and l dwarfs they observed . they determine the occurrence rate of planets more massive than @xmath105 @xmath1 to have an upper limit of 9% . the longer time coverage of the astrometric results presented in this work overlap most closely with radial velocity results . @xcite report detection limits based on radial velocity measurements for 102 nearby m dwarfs , of which six are featured in this paper . among the 102 m dwarfs searched , they confirm only two planets with orbital periods longer than 100 days . at an orbital period of 1000 days ( 2.7 years ) , they report a detection sensitivity of @xmath51 around 90% of stars observed , and @xmath52 at 10000 days ( 27 years ) . @xcite report that 6.5% @xmath11 3.0% of m dwarfs host a 1 to 13 @xmath1 planet at a separation less than 20 au , based on their radial velocity survey and high resolution imaging . the results presented here are among the first astrometric searches for jovian companions at jovian orbits , and fill in relatively unexplored mass and period parameter space models of how m dwarf planetary systems form and evolve must now explain the lack of massive companions with long period orbits like those in our solar system . looking forward , the _ gaia _ mission may detect up to 2600 planets within 100 pc @xcite . however , its ability to determine accurate masses and orbits will be limited to orbital periods less than 6 years , a fraction of the time coverage of our ground - based astrometric observations . finding nearby m dwarfs with planets remains an important challenge , as the closest planets are the brightest and most easily studied , and m dwarfs dominate the stellar population . as part of its mission to characterize the solar neighborhood , recons is developing a database of all objects with accurate trigonometric parallaxes placing them within 25 pc ( @xmath53 mas with an error @xmath54 mas ) . for an extrasolar planet to be included in the recons database it must orbit a star that meets the above criteria , or be a free - floating object with a comparable parallax , and be listed in _ both _ the extrasolar planets encyclopaedia ( _ exoplanet.eu _ ) and the exoplanet orbit database ( _ exoplanets.org _ , @xcite ) . the recons database currently contains 1074 systems having m dwarf primaries ( @xmath55 @xmath56 ) within 25 pc , only 17 of which have detected exoplanets as of 2014 july 1 , listed in table [ tab4 ] . the error - weighted mean parallax for each system is given in column 4 , including the parallaxes in this work and published values . minimum masses and orbital periods for planets with references are listed in columns 7 9 . note that only 3 of the 26 reported planets have masses greater than jupiter s . based on the 36 m dwarf primaries within 5 pc @xcite , we anticipate that there are 4500 m dwarf primaries within 25 parsecs , yet only 17 so far have been found to host planets . clearly , many planets lurk undetected in the solar neighborhood . discovering these planets will require a wide variety of survey techniques , and as sensitivities are improved , astrometry will continue to play an important role . we wish to thank cassy davison for her assistance developing the simulations , and sergio dieterich for helpful conversations regarding brown dwarf luminosities , and his assistance in surveying the existing literature . we also wish to thank hugh harris for his helpful suggestion regarding the eccentricities of unresolved binaries . this effort has been supported by the national science foundation via grants ast 05 - 07711 and ast 09 - 08402 , and the long - term cooperative efforts of the national optical astronomy observatories and the members of the smarts consortium . this research has made use of the exoplanet orbit database and the exoplanet data explorer at _ exoplanets.org_ , as well as the extrasolar planet encyclopaedia at _ exoplanet.eu_. this research has also made use of aladin and the simbad database , operated at cds , strasbourg , france . anglada - escud , g. , boss , a. p. , weinberger , a. j. , et al . 2012 , apj , 746 , 37 anglada - escud , g. , rojas - ayala , b. , boss , a. p. , weinberger , a. j. , & lloyd , j. p. 2013 , , 551 , a48 anglada - escud , g. , arriagada , p. , tuomi , m. , et al . 2014 , arxiv:1406.0818 apps , k. , clubb , k. i. , fischer , d. a. , et al . 2010 , , 122 , 156 bailey , j. , butler , r. p. , tinney , c. g. , et al . 2009 , , 690 , 743 bartlett , j. l. , et al . in preparation benedict , g. f. , mcarthur , b. , chappell , d. w. , et al . 1999 , aj , 118 , 1086 benedict , g. f. , mcarthur , b. e. , franz , o. g. , et al . 2001 , aj , 121 , 1607 benedict , g. f. , mcarthur , b. e. , forveille , t. , et al . 2002 , apj , 581 , l115 berta , z. k. , irwin , j. , & charbonneau , d. 2013 , arxiv:1307.3178 bertin , e. , & arnouts , s. 1996 , a&as , 117 , 393 bonfils , x. , mayor , m. , delfosse , x. , et al . 2007 , , 474 , 293 bonfils , x. , gillon , m. , forveille , t. , et al . 2011 , a&a , 528 , a111 bonfils , x. , delfosse , x. , udry , s. , et al . 2013 , a&a , 549 , a109 bonfils , x. , lo curto , g. , correia , a. c. m. , et al . 2013 , , 556 , a110 burt , j. , vogt , s. s. , butler , r. p. , et al . 2014 , , 789 , 114 delfosse , x. , bonfils , x. , forveille , t. , et al . 2013 , a&a , 553 , a8 dieterich , s. b. , henry , t. j. , jao , w .- c . , et al . 2013 , arxiv:1312.1736 dressing , c. d. & charbonneau , d. 2013 , apj , 767 , 95 endl , m. & krster , m. 2008 , , 488 , 1149 fabricius , c. , & makarov , v. v. 2000 , aa&s , 144 , 45 forveille , t. , bonfils , x. , delfosse , x. , et al . 2009 , a&a , 493 , 645 franz , o. g. , wasserman , l. h. , benedict , g. f. , et al . 1999 , working on the fringe : optical and ir interferometry from ground and space , 194 , 28 frescura , f. a. m. , engelbrecht , c. a. , & frank , b. s. 2007 , arxiv : 0706.2225 graham , j. a. 1982 , pasp , 94 , 244 haghighipour , n. , vogt , s. s. , butler , r. p. , et al . 2010 , apj , 715 , 271 harpse , k. b. w. , hardis , s. , hinse , t. c. , et al . 2013 , , 549 , a10 hartkopf , w. i. , mcalister , h. a. , & franz , o. g. 1989 , , 98 , 1014 hawley , s. l. , gizis , j. e. , & reid , i. n. 1996 , aj , 112 , 2799 henry , t. j. , & mccarthy , d. w. , jr . 1993 , aj , 106 , 773 henry , t. j. , ianna , p. a. , kirkpatrick , j. d. , & jahrei , h. 1997 , aj , 114 , 388 henry , t. j. , franz , o. g. , wasserman , l. h. , et al . 1999 , apj , 512 , 864 henry , t. j. , subasavage , j. p. , brown , m. a. , beaulieu , t. d. , jao , w. , & hambly , n. c. 2004 , aj , 128 , 2460 henry , t. j. , jao , w .- c . , subasavage , j. p. , beaulieu , t. d. , ianna , p. a. , costa , e. , & mndez , r. a. 2006 , aj , 132 , 2360 henry , t.j . 2013 , in royal astronomical society of canada observer s handbook 2013 , ed . david m.f . chapman . toronto : royal astronomical society of canada howard , a. w. , johnson , j. a. , marcy , g. w. , et al . 2010 , apj , 721 , 1467 jao , w .- c . , henry , t. j. , subasavage , j. p. , bean , j. l. , costa , e. , ianna , p. a. , & mndez , r. a. 2003 , aj , 125 , 332 jao , w .- c . , henry , t. j. , subasavage , j. p. , brown , m. a. , ianna , p. a. , bartlett , j. l. , costa , e. , & mndez , r. a. 2005 , aj , 129 , 1954 jao , w .- c . , henry , t. j. , subasavage , j. p. , et al . 2011 , , 141 , 117 johnson , j. a. , howard , a. w. , marcy , g. w. , et al . 2010 , , 122 , 149 landolt , a. u. 1992 , aj , 104 , 340 landolt , a. u. 2007 , , 133 , 2502 maness , h. l. , marcy , g. w. , ford , e. b. , et al . 2007 , , 119 , 90 mayor , m. , bonfils , x. , forveille , t. , et al . 2009 , a&a , 507 , 487 montet , b. t. , crepp , j. r. , johnson , j. a. , howard , a. w. , & marcy , g. w. 2013 , arxiv:1307.5849 reid , i. n. , hawley , s. l. , & gizis , j. e. 1995 , aj , 110 , 1838 riedel , a. r. , subasavage , j. p. , finch , c. t. , et al . 2010 , , 140 , 897 rivera , e. j. , laughlin , g. , butler , r. p. , et al . 2010 , , 719 , 890 robertson , p. , mahadevan , s. , endl , m. , & roy , a. 2014 , arxiv:1407.1049 sahlmann , j. , lazorenko , p. f. , segransan , d. , et al . 2014 , arxiv:1403.1275 skrutskie , m. f. , cutri , r. m. , stiening , r. , et al . 2006 , , 131 , 1163 sozzetti , a. , giacobbe , p. , lattanzi , m. g. , et al . 2013 , arxiv:1310.1405 subasavage , j. p. , jao , w .- c . , henry , t. j. , bergeron , p. , dufour , p. , ianna , p. a. , costa , e. , & mndez , r. a. 2009 , aj , 137 , 4547 van altena , w. f. , lee , j. t. , & hoffleit , d. 1995 , in the general catalogue of trigonometric stellar parallaxes ( 4th ed . ; new haven , ct : yale univ . obs . ) van de kamp , p. , 1967 , principles of astrometry ( san francisco : w. h. freeman and company ) van leeuwen , f. 2007 , a&a , 474 , 653 winters , j. g.,henry , t. j. , jao , w .- c . , subasavage , j. p. , finch , c. t. , & hambly , n. c. 2011 , aj , 141 , 21 wright , j. t. , fakhouri , o. , marcy , g. w. , et al . 2011 , pasp , 123 , 412 zechmeister , m. , krster , m. 2009 , a&a , 496 , 577

we study fubini instantons of a self - gravitating scalar field . the fubini instanton describes the decay of a vacuum state under tunneling instead of rolling in the presence of a tachyonic potential . the tunneling occurs from the maximum of the potential , which is a vacuum state , to any arbitrary state , belonging to the tunneling without any barrier . we consider two different types of the tachyonic potential . one has only a quartic term . the other has both the quartic and quadratic terms . we show that , there exist several kinds of new o(4)-symmetric fubini instanton solution , which are possible only if gravity is taken into account . one type of them has the structure with symmetry . this type of the solution is possible only in the de sitter background . we discuss on the interpretation of the solutions with symmetry . pacs numbers : 04.62.+v , 98.80.cq

1204.1521

the very first picture of an inflationary multiverse scenario was proposed in ref . @xcite , in which it would seem that the author wanted to suggest a universe without the cosmological singularity problem using an interesting feature of self - reproducting or regenerating exponential expansion of the universe . a major development in this scenario was triggered by the discovery of the eternal inflationary scenario @xcite and a paradigm for string theory landscape @xcite . the eternal inflation is related to the expanding false vacuum solution with a positive cosmological constant , which in turn means that the inflation is eternal into the future . if the theory has multiple minima then the false vacuum state decays into the true vacuum state , i.e. the phase transition proceeded _ via _ the nucleation of a vacuum bubble . in this scenario the universe is situated within some bubble called a pocket universe @xcite having a certain value of the cosmological constant and the whole universes are referred to as multiverse . the description of self - reproduction including tunneling process and random walk was combined into a scenario called recycling universe @xcite . these scenarios seem to provide an escape from the question of the initial conditions of the universe , i.e. it seems to be eternal into the past . unfortunately , inflationary spacetimes can not be made complete in the past direction @xcite , even though the universe is eternal into the future . there are still interesting arguments on the beginning of the universe @xcite . the string theory landscape is a setting that involves a huge number of different metastable and stable vacua @xcite , originated from different choices of calabi - yau manifolds and generalized magnetic fluxes . the huge number of different vacua can be approximated by the potential of a scalar field . the important thing is the fact that , once the de sitter vacuum can exist , the inflationary expansion is eternal into the future and has the possibility of self - reproduction . on the other hand , there are theories of gauged @xmath1 , @xmath2 supergravity having de sitter(ds ) solution , in which all susys are spontaneously broken . it is well known that the ds solution corresponds to a m / sting theory solution with a non - compact @xmath3- or @xmath4-dimensional internal space , in which a small value of the cosmological constant stems from the @xmath5-form flux . the simplest representative of these kind of theories has a tachyonic potential with the ds maxima @xcite . the potential in the vicinity of the maximum reduces to a form having a quadratic term , that is not metastable but unstable . however , according to some authors , the time for collapse giving rise to the tachyonic potential can be much greater than the age of the universe for anthropic reasoning . if the curvature radius of the potential in the vicinity of the maximum is greater than that used in the above theory , then that will be all together different story . the supergravity analogue of the tachyonic potential could be constructed also by using an exact supergravity solution representing the @xmath6-@xmath7 system @xcite . from the above scenarios , the study of the possibility of the tunneling process for the potential with stable and metastable vacua , or even tachyonic behavior has acquired renewed interest . in the present paper , we will study the tunneling process under a simple tachyonic potential governed by a quartic term both without the quadratic term and with the term as a toy model . to obtain the general solution including the effect of the backreaction , we solve the coupled equations for the gravity and the scalar field simultaneously . although the model has a tachyonic potential , it might still be an useful example to show how the tunneling process occurs in various shapes of the potential provided by the above scenarios . a quantum particle can tunnel through a finite potential barrier _ via _ the so - called barrier penetration . this process can be described by the euclidean solution obeying appropriate boundary conditions . there exist two kinds of euclidean solutions describing quantum tunneling phenomena . one corresponds to an instanton solution representing a stable pseudoparticle configuration characterized by the existence of a nontrivial topological charge . it does not change even if we continuously deform the field , as long as the boundary conditions remain the same . the instanton solution corresponds to the minimum of the euclidean action to pass from the initial to final state @xcite . the solution , in case of a double well potential , describes a general shift in the ground state energy of the classical vacuum due to the presence of an additional potential well , then lifting the so - called classical degeneracy . the other is a bounce solution representing an unstable nontopological configuration that corresponds to a saddle point rather than a minimum of the euclidean action . the second derivative of the euclidean action around the bounce has one negative eigenvalue which leads to the imaginary part of the energy . the existence of the negative eigenvalue implies that the vacuum state is unstable , i.e. the state decays into other states @xcite . the euclidean solutions can also mediate phase transitions . the phase transition describes the sudden change of a physical system from one state to another . the transition are of two different types transition accompanied by temperature or zero temperature . the competition between the entropy and the energy terms in the thermodynamic potential cause thermal phase transitions in which dynamics is irrelevant . in the modern classification scheme , thermal phase transitions are divided into two broad categories either with a discontinuous jump in the first - order derivatives of the free energy or without it . a first - order phase transition is characterized by the discontinuity in the first derivative of the free energy and is associated with the existence of latent heat , whereas a @xmath8th - order phase transition is characterized by the continuity in the first derivative while there is a discontinuity in the @xmath8th - order derivative . a quantum phase transition describes a transition between different phases by quantum fluctuation , which occurs at zero temperature , unlike the case of a thermal phase transition which is governed by a thermal fluctuation @xcite . to simplify things , we consider an asymmetric double well potential to distinguish two different phase transitions at zero temperature . if the initial state is the metastable vacuum state and the tunneling occurs from that state to the other vacuum state , then the transition corresponds to a tunneling process @xcite . on the other hand , if the initial state is the local maxima of the potential and the field is rolling down to one vacuum state continuously rather than any discontinuous jump , then the transition corresponds to the rolling . however , one more channel exists as tunneling and that corresponds to the one without a barrier . in this kind of transition , the initial state on the top of a potential can tunnel to the other state rather than rolling down the potential @xcite . there are two different kind of transitions in this case . one is the tunneling without a barrier representing the tunneling from the local maximum of the potential to the vacuum state @xcite . recently , an analytic study on this type of solution was performed in @xcite . the other is a tunneling without a barrier representing the tunneling from the maximum of the potential to any arbitrary state . this case corresponds to the fubini instanton @xcite , where the tachyonic potential is employed . can we describe the rolling corresponding to the transition between the initial metastable vacuum state and the other final vacuum state ? this may look similar to a superfluid motion by the liquid helium . although to establish the phase transition corresponding to the superfluid motion is itself a very challenging problem , we concentrate on the fubini instantons in this work . the fubini instanton @xcite describes the decay of a vacuum state by the quantum phase transition instead of rolling down the tachyonic potential consisted of a quartic term only . on the other hand , one can consider a tachyonic potential consisted of a quadratic term only , the point @xmath9 is unstable . a small perturbation will cause it to roll down the hill of the potential . originally , it was fubini who introduced a fundamental scale of hadron phenomena by means of the dilatation noninvariant vacuum state in the framework of a scale invariant lagrangian field theory @xcite . however , the solution is a one - parameter family of instanton solutions representing a tunneling without a barrier as an interpolating solution from the maximum of the potential to any arbitrary state . the instanton solution was studied in a conformally invariant model , i.e. a fixed background was used and the effect of the backreaction by instantons was neglected @xcite . this is a good approximation , when the variation of the potential during the transition is much smaller than the maximum of the potential . the instanton has gained much interest now - a - days in the context of anti - de sitter(ads)/conformal field theory correspondence @xcite . the paper is organized as follows : in sec . 2 , we review the fubini instanton in the absence of gravity . we present numerical solutions including the euclidean energy density as an example and analyze the structure of the solution in the theory with a potential having only the quartic self - interaction term . we stress the fact that there is no such solutions with the potential containing both the quartic and the quadratic terms . in sec . 3 , we show that the instanton solutions exist in the curved space . we perform a numerical study to solve the coupled equations for the gravity and the scalar field simultaneously . we show that there exist numerical solutions without oscillation in the initial ads space in the potential with only the quartic term . we also show that there exist numerical solutions in the potential both with the quartic and the quadratic terms irrespective of the value of the cosmological constant , which is possible only when the gravity is switched on . in order to estimate the decay rate of the background state , we compute the action difference between that of the solution and the background obtained by numerical means . we present an oscillating numerical solutions in the potential with only the quartic term with various values of the cosmological constant . one type of these solutions has the structure with @xmath0 symmetry . we will discuss on the interpretation of the solutions with @xmath0 symmetry in the final section . we analyze the behavior of the solutions using the phase diagram method . in sec . 4 , to observe the dynamics of the solutions , we briefly sketch the causal structures of the solutions in the lorentzian spacetime . finally in sec . 5 , we summarize and discuss our results . one can consider the following action in the absence of gravity @xmath10 , \label{min - action}\ ] ] where @xmath11 , @xmath12 is the minkowski metric , and the tachyonic potential has a quartic self - interaction term and also a quadratic term as follows : @xmath13 where @xmath14 and @xmath15 . the plots of potentials ( a ) without the quadratic term and ( b ) with the quadratic term are shown in fig . [ fig : fig01 ] . the potential has a metastable vacuum state at @xmath9 and no other stationary state in fig.[fig : fig01](a ) , while fig . [ fig : fig01](b ) illustrates that the potential has a local minimum at @xmath9 and two maxima @xcite . in both the cases , the potential is not bounded from below . before going to the tunneling problem in four dimensions , we briefly describe the problem in one dimension . one can consider the simplest quantum tunneling problem in one dimension . quantum field theory in one dimension is nothing but ordinary quantum mechanics . in case of @xmath16 , the amplitude for transmission obeys the wkb formula in the semiclassical approximation , in which @xmath17 are the classical turning points . on the other hand , the double - hump potential with @xmath14 and @xmath18 can be considered as an inverted double - well potential for a bounce solution representing the tunneling from @xmath9 to @xmath19 . the solution is given by @xmath20 $ ] , where @xmath21 is an integration constant . the bounce solutions can be easily understood in the euclidean space . the particle can only reach the point @xmath9 at @xmath22 and it bounces off @xmath19 at @xmath23 with a vanishing velocity @xcite . we now turn to the tunneling problem in four dimensions . it is an well - known fact that the massless theory has an instanton @xcite . actually , the instanton corresponds to the bounce solution representing the decay of the background vacuum state . the equation of motion with @xmath24 symmetry , obtained by varying the euclidean action , is then ; @xmath25 where @xmath26 plays the role of the evolution parameter in euclidean space and the second term in the left - hand side plays the role of a damping term . the boundary conditions are @xmath27 the particle in the classical mechanics problem starts at @xmath28 with zero velocity in the inverted potential , and stops at @xmath29 without any oscillation . for the potential with @xmath16 , the analytic solution of the fubini instanton has the form @xmath30 where @xmath31 is the radial length in the euclidean space , @xmath32 is any arbitrary length scale which characterizes the size of the instanton and is related to the initial value @xmath33 . in addition , the value of the scalar field of the center of the solution depends on @xmath32 as @xmath34 . this solution was used in the related perturbation theory @xcite . the characteristic behavior of the analytic solution is ploted in fig . [ fig : fig02 ] in terms of the value of the parameter @xmath32 . we take @xmath35 for all the cases . the solid line denotes the solution with @xmath36 , the dashed line with @xmath37 , and the dotted line with @xmath38 . the corresponding euclidean action is given by @xmath39 where the action does not depend on the parameter @xmath32 due to the consequence of the conformal invariance of the potential and we take that value to be @xmath40 . the action has the same value irrespective of the starting point @xmath33 . in other words , the tunneling from the maximum of the potential to any arbitrary state always happens with same probability . the numerical solutions for @xmath41 and @xmath42 and the euclidean energy including the density variation with @xmath31 are as shown in fig . [ fig : fig03 ] . figure [ fig : fig03](a ) illustrates the numerical solution for @xmath41 , in which the initial value set as @xmath43 and the solution asymptotically approaches the value @xmath44 . figure [ fig : fig03](b ) illustrates @xmath42 with respect to @xmath31 . there is a peak of @xmath42 near @xmath45 . figure [ fig : fig03](c ) depicts the volume energy density , when the density has got the form @xmath46 $ ] . the lower right box in the same figure shows the magnification of the small region clearly representing the existence of a smooth hill . the smooth peak of the volume energy density exists at @xmath47 . there is a disagreement between the location of the peak for the energy density and that for @xmath42 . it clearly reveals the fact that the position with the maximum value for @xmath42 is still not the same as the maximum of the energy density due to non - trivial contribution coming from the potential , @xmath48 . figure [ fig : fig03](d ) shows the euclidean energy @xmath49 for each slice of constant @xmath31 . the euclidean energy signifies the value of energy after the full integration of variables except for @xmath31 in the present work , @xmath50 . there are one minimum and one maximum point for @xmath49 . the location of the minimum of @xmath49 is around @xmath51 , whereas that of the maximum is near @xmath52 . ironically , the location of the minimum of @xmath49 coincides with that of the maximum of @xmath42 . these solutions can be considered as a ball consisting of only a thick wall except for one point at the center of the solution with a lower arbitrary state than the outer vacuum state unlike a vacuum bubble that consists of an inside part with a lower vacuum state and a wall . for a theory with @xmath14 , the conformal invariance is broken and any solution with a finite action is forbidden by scaling argument . in other worlds , the particle can not have enough energy to reach the hill overcome the barrier near @xmath9 since the damping term has got a large value due to a large value of @xmath42 near the initial point @xcite . let us consider the following action : @xmath53 + \oint_{\partial \mathcal m } \sqrt{-h } d^3 x \frac{k - k_o}{\kappa}\ , , \label{f - action}\ ] ] where @xmath54 , @xmath55 , @xmath56 denotes the scalar curvature of the spacetime @xmath57 , @xmath58 is the induced boundary metric , @xmath59 and @xmath60 are the traces of the extrinsic curvatures of @xmath61 for the metric @xmath62 and @xmath63 , respectively . the second term on the right - hand side is the boundary term @xcite . it is necessary to have a well - posed variational problem including the einstein - hilbert term . here we adopt the notations and sign conventions of misner , thorne and wheeler @xcite . we study the creation process of fubini instantons in curved spacetime . in the first place , we consider the massless case , and then we will also consider generalized fubini instantons , the so - called massive case ( see the form of the potential in eq.([min - po ] ) ) . the cosmological constant is given by @xmath64 , such that background space will be ds , flat or ads depending on the signs of @xmath65 . in order to solve the coupled equations , we assume an @xmath24 symmetry for the geometry and the scalar field similar to ref.@xcite @xmath66 \,.\ ] ] and then , @xmath41 and @xmath67 depends only on @xmath31 , and the euclidean equation can be written respectively as follows : @xmath68 and the hamiltonian constraint is then given by @xmath69 in order to yield a meaningful solution , the constraint requires a delicate balance among all the different terms . otherwise the solution can yield qualitatively incorrect behavior @xcite . to solve the eqs . ( [ erho ] ) , we have to impose suitable boundary conditions . when the gravity is switched off , boundary conditions for the fubini instanton are @xmath70 and @xmath29 as in ref . @xcite . while gravity is taken into account , we can write boundary conditions as follows : @xmath71 where @xmath72 is the maximum value of @xmath31 . for the flat and ads background @xmath73 , while @xmath72 is finite for the ds background . the first condition is to obtain a geodesically complete spacetime . the second condition is nothing but eq . ( [ eqcon ] ) . the third condition is the regularity condition as can be seen from the first equation in eq . ( [ erho ] ) . one should find the undetermined initial value of @xmath41 , i.e. @xmath74 , using the undershoot - overshoot procedure @xcite , to satisfy the fourth condition @xmath75 . we employ these conditions for fubini instantons in sec . iii a , b. if the background space is ds , we can impose conditions specified at @xmath76 and @xmath77 . for this purpose , we choose the values of the field @xmath67 and derivatives of the field @xmath41 as follows : @xmath78 the first two conditions are for the background space . the last two conditions are for the scalar field . in general , the solutions satisfying eq . ( [ ourbc-3 ] ) do not guaranty @xmath79 to be zero . for the solution having @xmath80 , the conditions eq . ( [ ourbc-3 ] ) are equivalent to the conditions eq . ( [ ourbc-2 ] ) . if @xmath81 , they represent completely new type of solutions with @xmath0 symmetry . we will discuss this case more in detail in sec . iii c. in order to solve the euclidean field eqs . ( [ erho ] ) and ( [ eqcon ] ) numerically , we rewrite the equations in terms of dimensionless variables as in ref . @xcite . in the present work , we employ the shooting method using the adaptive step size runge - kutta as the integrator similar to the treatment in ref . @xcite . for this procedure we choose the initial values of @xmath82 , @xmath83 , @xmath84 , and @xmath85 at @xmath86 as follows : @xmath87 where @xmath88 for @xmath89 . the minus sign in front of the second formula is due to the negative value of the @xmath90 determined by the sign of @xmath91 at @xmath92 . however , the initial value of @xmath93 is taking to be positive in the present work . once we specify the initial value @xmath94 , the remaining conditions can be exactly determined from eqs . ( [ nbcon2 ] ) . furthermore we impose additional conditions implicitly . to avoid a singular solution at @xmath95 for the euclidean field equations and to demand a @xmath0 symmetry , the conditions @xmath96 and @xmath97 as @xmath98 are needed in the next section . in this work , we require that the value of @xmath99 goes to a value smaller than @xmath100 as @xmath101 , as the exact value of @xmath102 is not known @xcite . the parameter @xmath103 is the ratio between the gravitational constant or planck mass and the mass scale in the theory , @xmath104 , and the parameter @xmath105 is related to the rescaled cosmological constant @xmath106 . to find the probability of the instanton solution , we only consider the euclidean action for the bulk part in eq . ( [ f - action ] ) to get , @xmath107 = 2\pi^2 \int \rho^3 d\eta [ -u]\ , , \label{euclac}\ ] ] where @xmath108 $ ] . we used eqs . ( [ erho ] ) and ( [ eqcon ] ) to arrive at this . the volume energy density has the form : @xmath109 , which has a different sign compared to the sign of the density used in ref . the euclidean energy signifies the energy value after the full integration of variables except for @xmath31 in the present case as @xmath110 . in the beginning , we obtain the numerical solution for @xmath111 . and then we obtain the numerical solution for @xmath14 . we call the space ds when the initial vacuum state has a positive cosmological constant , @xmath112 , flat when @xmath113 and ads when @xmath114 . the rate of decay of a metastable state can be evaluated in terms of the classical configuration and represented as @xmath115 in this approximation , in which the leading semiclassical exponent @xmath116 is the difference between the euclidean action corresponding to the classical solution @xmath117 and the background action @xmath118 . the prefactor @xmath119 is evaluated from the gaussian integral over fluctuations around the background classical solution @xcite . cccccc @xmath120 & @xmath94 & color of plot & @xmath117 & @xmath118 & b + @xmath121 & @xmath122 & red & @xmath123 & @xmath124 & @xmath125 + @xmath126 & @xmath127 & green & @xmath128 & @xmath129 & @xmath130 + @xmath131 & @xmath132 & blue & @xmath133 & @xmath134 & @xmath135 + we perform the numerical work with @xmath111 and take @xmath136 . the solutions without oscillation are only possible in the initial ads background as shown in fig . [ fig : fig04 ] . we guess that there is no solution without any oscillation for the initial flat and ds background . in the given @xmath137 , there may exist the phase space of solutions having the region of an arbitrary @xmath33 . if @xmath137 is increased , the oscillating behavior is appearing in the phase space of solutions @xcite . figure 4(a ) illustrates the solution for @xmath138 , in which the right box in the same figure shows the magnification of a small region representing the initial values of @xmath138 and the behavior of the curves . the curves move upwards with increasing value of @xmath139 , then overlap near @xmath140 , and more downwards with increasing value of @xmath139 . figure 4(b ) shows the solutions of @xmath141 . the curves move downwards with increasing @xmath139 . the shape of the numerical solution @xmath141 can be easily understood if one thinks of the shape of the solution in a fixed ads space as @xmath142 . table 1 shows the dimensionless variables and the color of plot used , and also the actions obtained from fig . [ fig : fig04 ] . from the numerical data , one can easily see that the magnitude of @xmath143 approaches the vacuum state @xmath144 as @xmath139 approaches a vanishing value . the vanishing of @xmath139 means the background geometry which serves as the initial vacuum state is flat . the action difference @xmath145 between the action of the solution @xmath146 and that of the background @xmath147 has positive values . we carry out the action integral in the range @xmath148 numerically as the action difference @xmath145 diverges to infinity if we perform the integration for an infinite @xmath149 value . this divergence is due to the fact that the size of the solution including the outside part in the evolution parameter space decrease compared to the size of the initial background similar to what happens for the case of the nucleation of a vacuum bubble . in the analytic computation , the outside part and the background are simply canceled at the same radius . in the present numerical work , it is difficult to decide the exact size of the solution . thus we straightforwardly compute the action difference and then the difference @xmath145 has got an approximate behavior @xmath150 which cause the divergence at infinity . if this minor error is cured , the action difference has a finite value . cccccc @xmath120 & @xmath94 & color of plot & @xmath117 & @xmath118 & b + @xmath131 & @xmath151 & black & @xmath152 & @xmath153 & @xmath154 + @xmath126 & @xmath155 & red & @xmath156 & @xmath157 & @xmath158 + @xmath121 & @xmath159 & green & @xmath160 & @xmath161 & @xmath162 + @xmath163 & @xmath164 & blue & @xmath165 & @xmath166 & @xmath165 + @xmath167 & @xmath168 & sky blue & @xmath169 & @xmath170 & @xmath171 + @xmath172 & @xmath173 & pink & @xmath174 & @xmath175 & @xmath176 + @xmath177 & @xmath178 & yellow & @xmath179 & @xmath180 & @xmath181 + now we perform the numerical work with @xmath14 and take @xmath182 . this type of solutions belongs to usual tunneling with a barrier . we obtained the numerical solutions with an arbitrary cosmological constant as shown in fig . [ fig : fig05 ] . the solutions are only possible for specific @xmath143s . the figures represent the vary fact that the solution is only possible in curved spacetime irrespective of the value of the cosmological constant . figure 5(a ) illustrates the solution of @xmath138 . the upper right box in the same figure shows the magnification of the small region representing behavior of the curves which move to the left with an increase in @xmath139 . figure [ fig : fig05](b ) shows @xmath93 with respect to @xmath149 . the upper right box in the same figure shows the magnification of the small region representing behavior of the curves moving below with increasing value of @xmath139 . figure [ fig : fig05](c ) illustrates the solutions of @xmath141 . the curves move downwards with increasing value of @xmath139 . the shape of the numerical solutions of @xmath141 can be easily understood if one consider a fixed space . in the fixed flat space , @xmath183 . in the ds space , @xmath184 . in the ads space , @xmath142 . figure [ fig : fig05](d ) depicts the variation of @xmath141 with respect to @xmath149 . the curves move below with increase in @xmath139 . the horizontal line with @xmath185 indicate a flat space with @xmath186 . table [ table2 ] shows the dimensionless variables and color of the plot used among with the action obtained from fig . [ fig : fig05 ] . from the numerical data , one can infer that the magnitude of @xmath143 decreases as @xmath139 increases . we carry out the action integral in the range @xmath187 numerically . in the ds space , the solution and the background have their own periods for @xmath149 , which we take the period as the integration limit . for the background ds space , we take @xmath188 . the action for @xmath146 and @xmath147 are positive or zero as long as @xmath189 . the background action is zero for @xmath190 . in this work , we do not check for the special case @xmath191 . simply , the action has a negative value for the ds space . it is related to the fact that the euclidean action for einstein gravity is not bounded from below , and this is known as the conformal factor problem in euclidean quantum gravity @xcite . it was argued in @xcite that the conformal divergence due to the unboundedness of the action might get cancelled with a similar term of opposite sign caused by the measure of the path integral . however , the difference between the action of the solution and that of the background remains positive - valued . the oscillating instanton and the bounce solutions with an @xmath24 symmetry between the ds - ds vacuum states was first studied in ref.@xcite , in which the authors found the solutions in a fixed background geometry and showed how does the maximum allowed number @xmath192 depend on the parameters of the theory , where @xmath8 denotes the crossing number of the potential barrier by the oscillating solutions . the oscillating bounce solutions in the presence of gravity was also studied in ref . @xcite , in which the authors analyzed the negative modes and the fluctuations around the oscillating solutions . the instanton was interpreted as the thermal tunneling @xcite . the oscillating instanton solutions under a symmetric double - well potential in the curved space with an arbitrary vacuum energy was also investigated in detail in @xcite , where a numerical solution is possible as long as the local maximum value of the potential remains positive . the solutions have a thick wall and can be interpreted as a mechanism for the nucleation of the thick wall for topological inflation @xcite . similarly , the process for the tunneling without a barrier in curved space , was studied in ref.@xcite . the existence of numerical solutions was shown in ref . @xcite , in which the case representing the tunneling from flat to ads space shows an oscillating behavior . the solution oscillates around @xmath9 in the inverted potential and the oscillating behavior die away unlike the case under a harmonic potential . in the present paper , the oscillation means that the field in the solutions oscillates around the minimum of the inverted potential and die away asymptotically to the minimum @xmath9 for the case with @xmath111 . thus the resulting geometry of the initial state has wrinkles due to the variation of the volume energy density and the instanton simultaneously . the behavior of the solutions representing the resulting geometry with wrinkles is quite different from those in ref . @xcite . figure [ fig : fig06 ] shows the numerical solutions representing an oscillatory behavior in ( a ) the initial flat background and ( b ) the initial ads background . we take @xmath193 , @xmath190 ( for the flat case ) , and @xmath194 ( for the ads case ) , respectively . figures [ fig : fig06](a ) and ( c ) illustrate the numerical solutions of @xmath138 . the lower right box in those figure shows the magnification of a small region representing the behavior of the solution around @xmath144 . the peak corresponds to the first turning point of the particle similar to a classical mechanics problem in the presence of an inverted potential . for the case with @xmath195 the first turning point reaches furthermost point away from @xmath144 among all the other cases , as one can easily see from the figure . the curves oscillate around @xmath144 and eventually stop at @xmath144 in the flat and ads space . we take the initial point as an arbitrary @xmath143 , which means that the number of oscillations for each solution can be different . however , there is the tendency that the number of oscillations is decreased as the value of @xmath143 is decreased in the given @xmath137 . figures [ fig : fig06](b ) and ( d ) illustrate the numerical solutions for @xmath141 . the upper left box in those figure shows the magnification of an initial small region representing behavior of the curves which move below with the decrease in @xmath143 . ccccccc @xmath94 & @xmath117 ( ads ) & @xmath118 ( ads ) & b ( ads ) & b ( flat ) + @xmath196 & @xmath197 & @xmath198 & @xmath199 & @xmath200 + @xmath201 & @xmath202 & @xmath198 & @xmath203 & @xmath204 + @xmath205 & @xmath206 & @xmath198 & @xmath207 & @xmath208 + @xmath209 & @xmath210 & @xmath198 & @xmath211 & @xmath212 + @xmath213 & @xmath214 & @xmath198 & @xmath215 & @xmath216 + @xmath217 & @xmath218 & @xmath198 & @xmath219 & @xmath220 + @xmath221 & @xmath222 & @xmath198 & @xmath223 & @xmath224 + table [ table3 ] shows the initial values of @xmath138 , the actions for the ads , and flat background which are obtained from fig . [ fig : fig06 ] . in the flat case , the background action is zero as @xmath185 and therefore @xmath146 is equal to @xmath145 . in the present case , we cut all the data at a certain point which is @xmath225 . we now analyze the behavior of the solutions using a phase diagram method . after plugging the value of @xmath226 from eq.([eqcon ] ) into eq . ( [ erho ] ) and using @xmath227 , the equation becomes @xmath228 first , we consider the situation where the kinetic energy is small compared to the potential energy such that @xmath229 , @xmath230 and the term @xmath231 is smaller than other terms . in other words , the last term is the most dominant among other terms in the numerator . then the equation reduces to the form @xmath232 the above relation shows that the first stage of the curve has got such kind of form . second , we consider the situation where @xmath233 , i.e. with vanishing acceleration and then we impose all the above mentioned conditions . it will then describe the special points in the phase diagram . then the equation reduces to the form @xmath234 third , we consider the situation where @xmath235 , i.e. a negative constant . we impose all the above mentioned conditions among with @xmath236 . thus , we obtain the above equation again . this relation implies that the special points with a vanishing acceleration and some of the region with a negative constant acceleration in the phase diagram have got a linear function type behavior in the phase diagram as shown in figs . [ fig : fig07](a ) and ( c ) . figure [ fig : fig07 ] illustrates the behavior of the solutions in the @xmath138-@xmath93 plane . each trajectory represents the behavior of the solution in the phase diagram . the trajectories begin with zero velocity as @xmath237 shown in figs . [ fig : fig07](a ) and ( c ) . the velocity increases rapidly to the maximum and then decreases linearly up to the turning point . figures [ fig : fig07](b ) and ( d ) show the magnification of the small region representing the behavior of the solution around @xmath237 and @xmath144 . basically , the fubini solution has an asymptotic condition to be satisfied . we expect that there exist an oscillating solutions although the ds background has got a finite size in the euclidean signature . however , if we consider an analytic continuation not of the angle parameter @xmath238 but of the euclidean evolution parameter @xmath239 , then the meaning becomes clearer . when there is an ` even ' symmetry for the oscillating instantons , we can see the half - way point @xmath240 as @xmath241 . then , we can paste the lorentzian manifold @xmath242 at the @xmath243 surface . this is possible only for the case @xmath241 , because of the cauchy - riemann theorem of complex analysis ; otherwise , the lorentzian manifold should be complex valued functions ( for exceptional cases , we might be able to consider complex valued instantons , the so - called fuzzy instantons @xcite ) . in this procedure , an event shows a spontaneous creation of the universe from ` nothing ' @xcite , in which nothing means a state without the concept of classical spacetime @xcite . we already know that there is such a solution when the scalar field is exactly on top of the local maximum . however , now we observe a creation from nothing with highly non - trivial field dynamics . this is worthwhile to be highlighted and we postpone further analysis for the future work . we now shift our attention to the new type of solutions in the initial background as the ds space , i.e. @xmath112 . the euclidean ds space has a compact geometry . thus the solutions can have @xmath0 symmetry . we consider the boundary conditions in eq . ( [ ourbc-3 ] ) . to obtain the solutions with @xmath0 symmetry , we need to impose additional conditions . for the background geometry , @xmath244 at @xmath245 . on the other hand , for the scalar field , we impose @xmath9 at @xmath245 for the solutions with odd number of crossings of the potential well and @xmath246 at @xmath245 for the solutions with even number of crossings . the solutions with odd number of crossing have the opposite state of the value @xmath41 at @xmath247 and @xmath77 , i.e. @xmath248 . the solutions with even number of crossing have the same state of the value @xmath41 at @xmath247 and @xmath77 , i.e. @xmath249 . we stress that the boundary conditions in eq . ( [ ourbc-3 ] ) gives rise to completely new type of solutions of fubini instanton . figure [ fig : fig08 ] shows the numerical solutions of the fubini instanton with @xmath0 symmetry . we take @xmath250 and @xmath251 . thus the ds region in the @xmath138-space spans the region @xmath252 . we consider four cases with different initial positions of @xmath138 . figure [ fig : fig08](a ) illustrates the numerical solution of @xmath138 . the trajectories with the blue and red color go back to the same position of @xmath138 in the presence of the inverted potential , i.e. they have even number of crossings . the trajectories with the black and green color go back to the opposite position of @xmath138 , i.e. they have odd number of crossings . figure [ fig : fig08](b ) depicts the numerical solution of @xmath141 . figures [ fig : fig08](c ) and ( d ) illustrate the behavior of the solutions in the @xmath138-@xmath93 plane . each trajectory represents the behavior of the solution in the phase diagrams . the blue and red lines indicate that the interior part of two instantons has got the same state as @xmath138 , whereas the green and black lines indicate that the interior part has got the opposite state of @xmath138 . figure [ fig : fig08](d ) illustrates the magnification of a small region representing the behavior of the solution around @xmath237 and @xmath144 . figure [ fig : fig08](e ) illustrates the volume energy density , where the density has got a form @xmath253 . the box shows the magnification of a small region representing behavior of curves . the densities in each of the case have got positive values near the initial starting point @xmath143 far away from the point @xmath144 , because the densities have the form @xmath253 and @xmath254 . the solutions oscillate in the ds region as found the present work . the density always negative values for the case of the black line . figure [ fig : fig08](f ) illustrates the euclidean energy @xmath255 for each slice of constant @xmath149 values . the negative energy parts in each of the case signifies a rolling state in the ds region . table [ table4 ] shows the initial value of @xmath138 , colors of the plot used , and the actions obtained from fig . [ fig : fig08 ] . ccccc @xmath94 & color of plot & @xmath117 & @xmath118 & b + @xmath256 & black & @xmath257 & @xmath258 & @xmath259 + @xmath260 & red & @xmath261 & @xmath258 & @xmath262 + @xmath263 & green & @xmath264 & @xmath258 & @xmath265 + @xmath266 & blue & @xmath267 & @xmath258 & @xmath268 + in this section , we briefly outline the causal structure of the solutions in the lorentzian signature . due to the pressure difference , the nucleated ads region will expand over the background and hence the boundary of the nucleated ads region will be time - like . figure [ fig : fig09 ] shows the schematic diagrams representing the causal structures of the fubini instantons and the related solutions . the @xmath269 surface can be analytically continued to the surface @xmath242 in the lorentzian signature . the lower vacuum region in the instanton ( green colored region ) will be unstable during the lorentzian time evolution ( orange colored region ) . due to the instability of the fubini type potential , the whole causal structure may depend on the shape of the potential or the vacuum structure i.e. whether the left or the right side of the potential has true vacua or not . therefore , the followings are meaningful only as reasonable estimations for general behavior and these may be different for special examples . figure [ fig : fig09](a ) illustrates the instanton solution in an ads background . it will form time - like @xmath270 and @xmath271 boundaries in the lorentzian signature . however , the ads region may be unstable to form a kind of singularity . figure [ fig : fig09](b ) illustrates the instanton solution in the ds background . the ds region has a cosmological horizon and will this form a future infinity . the ads region ( orange colored region ) will expand over the ds region due to the pressure difference . figure [ fig : fig09](c ) is the pair creation by the oscillating instanton solutions . therefore , in the instanton part , the ds region around the @xmath272 is surrounded by the ads ( green colored region ) part . in the lorentzian signature , we can interpret these two ads parts as being nucleated in a ds background . in fig.[fig : fig09](c ) , we infer that , there still remains a ds region and a future infinity . the pair creation of the instantons in this work is quite different from the ordinary quantum process of pair creation of particles . we take the initial background as the ds space , i.e. @xmath112 . the euclidean ds space has a compact geometry . thus , the geometry has two poles . if one object is created on the north pole and the other on the south pole , we can interpret that process as the pair creation of objects . as an example of the process , the two - crossing solution between the sides of the potential barrier in the double - well potential was considered as a type of a double - bounce solution or an anti - double - bounce solution @xcite , in which the authors interpreted the double - bounce solution as the spontaneous pair - creation of true vacuum bubbles , one at each pole in the ds space . we adopt a similar interpretation for our solutions with @xmath0 symmetry . in this paper we have studied fubini instantons of a self - gravitating scalar field representing the tunneling without a barrier . there are two kinds of euclidean solutions representing the tunneling without any barrier . one of them is the tunneling from the local maximum of the potential to the vacuum state . the other one is the tunneling from the maximum to any arbitrary state . the latter corresponds to the fubini instanton solution . we have shown that there exist several new kinds of fubini instanton solutions of a self - gravitating scalar field found as numerical solutions , which are possible only if gravity is taken into account . we also computed the action difference @xmath273 , in each case , between the euclidean action corresponding to a classical solution @xmath117 and the background action @xmath118 for the rate of decay . in sec . 2 , we reviewed the fubini instanton in the absence of gravity from the viewpoint of a tunneling problem . we have presented a numerical solution including the euclidean energy density for example . we analyzed the structure of the solution in a theory with the potential having only a quartic self - interaction term . these solutions can be considered as a ball consisting of only a thick wall except for the one point at the center of the solution with a lower arbitrary state than the outer vacuum state unlike a vacuum bubble which consists of an inner part with a lower vacuum state and a wall . in sec . 3 , we have studied the instanton solutions in curved space . we performed careful numerical study to solve the coupled equations for the gravity and the scalar field simultaneously . we have shown that there exist numerical solutions without any oscillation in the initial ads space for the potential with only the quartic term . we have also shown that there exist numerical solutions for the potential with both a quartic and a quadratic term irrespective of the value of the cosmological constant . for this particular case , there is no solution with an @xmath24 symmetry when gravity is switched off . in order to estimate the decay rate of the background state , we calculated the action difference between the action of the solution and that of the background obtained using numerical means . we have obtained oscillating fubini instantons as new types of solutions . we have shown that there exist oscillating numerical solutions for the potential with only the quartic term in the flat and ads space , except for the solution without oscillation in the initial ads space with the specific value of a cosmological constant and the parameters . we have analyzed the behavior of the solutions using the phase diagram method . the oscillation dies away asymptotically in both the flat and the ads space . we have obtained numerical solutions representing the fubini instanton with @xmath0 symmetry . we stress that they represent completely new type of solutions of fubini instanton . these solutions can be interpreted as the pair creation with each one having the same state and with each one having the opposite state , respectively . the solutions can lead to more interesting interpretation as follows : any arbitrary state can tunnel into another arbitrary state with an @xmath24-symmetry in the curved spacetime , although no vacuum state exists as the instanton solution . the solutions are possible as long as the maximum of the potential remains positive . the subject on the pair creation of bubbles was first considered in ref . the numerical solution representing the pair of solutions is in fig . 2 in ref . @xcite , which can be interpreted as the pair creation of the bubbles , one at each pole in the ds space . however , there is a different interpretation on the solutions @xcite , in which the authors studied a decay channel of de sitter vacua . the solutions with @xmath274 symmetry can be understood as describing tunneling in a finite horizon volume at finite temperature . the solutions maybe correspond to thermal production of a bubble in their interpretation . in this stage , the comparative analysis between the o(4)-symmetric solution and o(3)-symmetric solution with respect to the pair creation is needed to be studied more . we leave this for future work . in sec . 4 , we have analyzed the schematic diagrams representing the causal structures of the fubini instantons and the related solutions in the lorentzian signature . for the special case representing the solution with @xmath0 symmetry , the ds region around the @xmath272 is surrounded by an ads part . we now mention on the negative mode problem . it was known that the bounce solution has one negative mode in the spectrum of small perturbations about the solution @xcite . the bounce solution with one negative mode corresponds to the tunneling process in the lowest wkb approximation . in ref . @xcite , coleman argued that the euclidean solution with only one negative mode is related to the tunneling process in the flat minkowski spacetime . however , there is no rigorous proof on extension of coleman s argument to the curved space claiming the physical irrelevance of the solutions with additional negative modes . for example , the time - translation invariance , or zero modes , is one of crucial elements to prove the uniqueness of the negative mode in his argument . however , the existence of zero modes is not guaranteed in curved spacetime . another point is that the euclidean time interval is at most of @xmath275 in de sitter space . hence , only a finite number of the bounces can be placed far apart from each other . therefore , the dilute gas approximation may become invalid easily , which leads to the breakdown of the wkb approximation @xcite . there appears diverse situations on the negative modes when the gravity is taken into account @xcite . although , the bounce solution with one negative mode in curved space dominates the tunneling process , the solutions with additional negative modes may also contribute to the tunneling process . there exist some works including the physical interpretation on the oscillating solutions with more than one negative mode . one can naturally interpret the system in de sitter background as a thermal system . the authors in refs . @xcite interpreted that the existence of additional negative modes represents the solutions as unstable intermediate thermal configuration . they seem to observe the clue to support this idea on the other point of view . it is known that the @xmath276 times oscillating solutions have @xmath276 negative modes @xcite . the even numbers of negative modes of the form @xmath277 and @xmath278 do not have imaginary part of the energy , while the odd numbers of negative modes of the form @xmath279 have the imaginary part of the energy with the wrong sign . however , @xmath280 negative modes may have a meaning for a tunneling process even if the solution may not be related to the lowest wkb approximation . recently the analysis on the negative modes of oscillating instantons has been investigated @xcite . the oscillating instantons as homogeneous tunneling channels have been also studied @xcite . in conclusion , we believe many euclidean solutions in curved space with zero and negative modes may have physical significance and deserves further investigation . in summary , we illustrate the following finding in our new contribution regarding this issue : 1 . in the absence of gravity , a @xmath281-type potential has infinitely many instanton solutions whereas a @xmath282-type potential has no instanton solution . however , _ the inclusion of the gravity changes all the situation abruptly _ : for the former case , the solution space get reduced to a finite space and for the latter case , there exists solutions . we also confirm that @xmath281-type potentials have oscillating instanton solutions as well as the solutions with @xmath0 symmetry . therefore , the fubini instanton is one of the few examples that shows the effect of gravity bringing drastic changes to the tunneling process . there can be more applications of the oscillating instantons and we confirm that the fubini - type potentials also contribute largely towards these processes . we postpone any possible application of such oscillating solutions including the phase space of solutions for our future work @xcite . we would like to thank andrei linde for kind historic comments on the inflationary multiverse scenario and fubini instanton . we would like to thank erick j. weinberg , george lavrelashvili , hongsu kim , yunseok seo , and dong - il hwang for helpful discussions and comments , and thank chaitali roychowdhury for a careful english revision of the manuscript . we would like to thank manu b. paranjape and richard mackenzie for their hospitality during our visit to universit de montral . this work was supported by the korea science and engineering foundation ( kosef ) grant funded by the korea government(mest ) through the center for quantum spacetime(cquest ) of sogang university with grant number r11 - 2005 - 021 . wl was upported by basic science research program through the national research foundation of korea(nrf ) funded by the ministry of education , science and technology(2012r1a1a2043908 ) . dy is supported by the jsps grant - in - aid for scientific research ( a ) no . we appreciate apctp for its hospitality during completion of this work . a. d. linde , _ nonsingular regenerating inflationary universe _ , cambridge university preprint , print-82 - 0554 . a. vilenkin , phys . d * 27 * , 2848 ( 1983 ) ; arxiv : gr - qc/0409055 . a. d. linde , phys . b * 175 * , 395 ( 1986 ) . a. h. guth , phys . * 333 - 334 * , 555 ( 2000 ) . s. winitzki , lect . notes phys . * 738 * , 157 ( 2008 ) . r. bousso and j. polchinski , j. high energy phys . 06 ( 2000 ) 006 . l. susskind , arxiv : hep - th/0302219 . j. garriga and a. vilenkin , phys . d * 57 * , 2230 ( 1998 ) . a. borde , a. h. guth , and a. vilenkin , phys . lett . * 90 * , 151301 ( 2003 ) . a. mithani and a. vilenkin , arxiv:1204.4658 ; l. susskind , arxiv:1204.5385 ; arxiv:1205.0589 . s. kachru , r. kallosh , a. linde , and s. p. trivedi , phys . d * 68 * , 046005 ( 2003 ) . s. k. ashok and m. r. douglas , j. high energy phys . 01 ( 2004 ) 060 . c. m. hull , classical quantum gravity * 2 * , 343 ( 1985 ) . r. kallosh , a. linde , s. prokushkin , and m. shmakova , phys . d * 65 * , 105016 ( 2002 ) ; _ ibid . _ d * 66 * , 123503 ( 2002 ) . r. kallosh and a. linde , phys . d * 67 * , 023510 ( 2003 ) . h. kim , j. high energy phys . 01 ( 2003 ) 080 . a. a. belavin , a. m. polyakov , a. s. schwartz , and yu . s. tyupkin , phys . * 59b * , 85 ( 1975 ) . s. coleman , _ aspects of symmetry _ ( cambridge university press , cambridge , england , 1985 ) . i. herbut , _ a modern approach to critical phenomena _ ( cambridge university press , cambridge , 2007 ) . m. b. voloshin , i. yu . kobzarev , and l. b. okun , yad . 20 * , 1229 ( 1974 ) [ sov . j. nucl . phys . * 20 * , 644 ( 1975 ) ] . s. coleman , phys . d * 15 * , 2929 ( 1977 ) ; _ ibid . _ d * 16 * , 1248(e ) ( 1977 ) . s. coleman and f. de luccia , phys . d * 21 * , 3305 ( 1980 ) . s. parke , phys . lett . * 121b * , 313 ( 1983 ) . k. lee and e. j. weinberg , phys . rev . d * 36 * , 1088 ( 1987 ) . lee and w. lee , classical quantum gravity * 26 * , 225002 ( 2009 ) . s. fubini , nuovo cimento a * 34 * , 521 ( 1976 ) . a. d. linde , nucl . * b216 * , 421 ( 1983 ) ; _ ibid . _ * b223 * , 544(e ) ( 1983 ) . k. lee and e. j. weinberg , nucl . * b267 * , 181 ( 1986 ) ; k. lee , nucl . * b282 * , 509 ( 1987 ) . l. g. jensen and p. h. steinhardt , nucl . * b317 * , 693 ( 1989 ) . . lee , c. h. lee , w. lee , and c. oh , phys . d * 85 * , 024022 ( 2012 ) . s. kanno , m. sasaki , and j. soda , classical quantum gravity * 29 * , 075010 ( 2012 ) ; prog . phys . * 128 * , 213 ( 2012 ) . j. garriga , x. montes , m. sasaki , and t. tanaka , nucl . b * 551 * , 317 ( 1999 ) . s. khlebnikov , nucl . b * 631 * , 307 ( 2002 ) . f. loran , mod . a * 22 * , 2217 ( 2007 ) . s. de haro and a. c. petkou , j. high energy phys . 12 ( 2006 ) 076 ; s. de haro , i. papadimitriou , and a. c. petkou , phys . * 98 * , 231601 ( 2007 ) . j. l. f. barbon and e. rabinivici , j. high energy phys . 04 ( 2010 ) 123 ; j. high energy phys . 04 ( 2011 ) 044 . a. n. kuznetsov and p. g. tinyakov , mod . a * 11 * , 479 ( 1996 ) ; phys . d * 56 * , 1156 ( 1997 ) . a. a. yurova and a. v. yurov , phys . a * 372 * , 4222 ( 2008 ) . c. m. bender and t. t. wu , phys . d * 7 * , 1620 ( 1973 ) ; j .- q . liang and h. j. w. mller - kirsten , phys . d * 45 * , 2963 ( 1992 ) ; _ ibid . _ d * 48 * , 964(e ) ( 1993 ) . l. n. lipatov , sov . jetp * 45 * , 216 ( 1977 ) . i. affleck , nucl . * b191 * , 429 ( 1981 ) . j. w. york , jr . lett , * 28 * , 1082 ( 1972 ) ; g. w. gibbons and s. w. hawking , phys . d * 15 * , 2752 ( 1977 ) ; j. w. york , jr . , found . * 16 * , 249 ( 1986 ) . c. w. misner , k. s. thorne , and j. a. wheeler , _ gravitation _ ( freeman , san francisco , 1973 ) . b. k. berger , gen . grav . * 38 * , 625 ( 2006 ) . w. h. press , s. a. teukolsky , w. t. vetterling , and b. p. flannery , _ numerical recipes in fortran _ ( cambridge university press , cambridge , england , 1992 ) . c. g. callan , jr . and s. coleman , phys . d * 16 * , 1762 ( 1977 ) . e. j. weinberg , phys . d * 47 * , 4614 ( 1993 ) ; j. baacke and g. lavrelashvili , phys . d * 69 * , 025009 ( 2004 ) ; g. v. dunne and h. min , phys . d * 72 * , 125004 ( 2005 ) . lee , w. lee , d. ro , and d .- h . yeom , in progress . g. w. gibbons , s. w. hawking , and m. j. perry , nucl . * b138 * , 141 ( 1978 ) . a. dasgupta and r. loll , nucl . b * 606 * , 357 ( 2001 ) . j. c. hackworth and e. j. weinberg , phys . d * 71 * , 044014 ( 2005 ) ; e. j. weinberg , aip conf . proc . , 259 ( 2005 ) . lavrelashvili , phys . d * 73 * , 083513 ( 2006 ) ; g. v. dunne and q .- h . wang , phys . d * 74 * , 024018 ( 2006 ) . a. r. brown and e. j. weinberg , phys . d * 76 * , 064003 ( 2007 ) . a. vilenkin , phys . rev . lett . * 72 * , 3137 ( 1994 ) ; a. linde , phys . b * 327 * , 208 ( 1994 ) . j. b. hartle , s. w. hawking , and t. hertog , phys . rev . * 100 * , 201301 ( 2008 ) ; phys . d * 77 * , 123537 ( 2008 ) . hwang , h. sahlmann , and d .- h . yeom , classical quantum gravity * 29 * , 095005 ( 2012 ) ; d .- hwang , b .- h . lee , h. sahlmann , and d .- h . yeom , classical quantum gravity * 29 * , 175001 ( 2012 ) . a. vilenkin , nucl . * b252 * , 141 ( 1985 ) . r. bousso and a. linde , phys . d * 58 * , 083503 ( 1998 ) . j. garriga and a. megevand , phys . d * 69 * , 083510 ( 2004 ) ; int . j. theor . phys . * 43 * , 883 ( 2004 ) . a. masoumi and e. j. weinberg , phys . d * 86 * , 104029 ( 2012 ) . s. r. coleman , nucl . * b298 * , 178 ( 1988 ) . t. tanaka and m. sasaki , prog . phys . * 88 * , 503 ( 1992 ) ; t. tanaka , nucl . * b556 * , 373 ( 1999 ) . a. r. brown and a. dahlen , phys . d * 84 * , 105004 ( 2011 ) . g. v. lavrelashvili , v. a. rubakov , and p. g. tinyakov , phys . * 161b * , 280 ( 1985 ) . a. khvedelidze , g. lavrelashvili , and t. tanaka , phys . d * 62 * , 083501 ( 2000 ) ; g. lavrelashvili , nucl . b , proc . suppl . * 88 * , 75 ( 2000 ) . l. battarra , g. lavrelashvili and j .- l . lehners , phys . d * 86 * , 124001 ( 2012 ) . lee , w. lee , and d .- h . yeom , arxiv:1206.7040 [ hep - th ] .

we present new uv spectra of the nucleus of the seyfert 1 galaxy ngc 5548 , which we obtained with the space telescope imaging spectrograph at high spectral resolution , in conjunction with simultaneous _ chandra x - ray observatory _ spectra . taking advantage of the low uv continuum and broad emission - line fluxes , we have determined that the deepest uv absorption component covers at least a portion of the inner , high - ionization narrow - line region ( nlr ) . we find nonunity covering factors in the cores of several kinematic components , which increase the column density measurements of n v and c iv by factors of 1.2 to 1.9 over the full - covering case ; however , the revised columns have only a minor effect on the parameters derived from our photoionization models . for the first time , we have simultaneous n v and c iv columns for component 1 ( at km s ) , and find that this component _ can not _ be an x - ray warm absorber , contrary to our previous claim based on nonsimultaneous observations . we find that models of the absorbers based on solar abundances severely overpredict the o vi columns previously obtained with the _ far ultraviolet spectrograph _ , and present arguments that this is not likely due to variability . however , models that include either enhanced nitrogen ( twice solar ) or dust , with strong depletion of carbon in either case , are successful in matching all of the observed ionic columns . these models result in substantially lower ionization parameters and total column densities compared to dust - free solar - abundance models , and produce little o vii or o viii , indicating that none of the uv absorbers are x - ray warm absorbers .

astro-ph0305154

mass outflow in the form of uv and x - ray absorbers has been shown to be a common phenomenon in the active nuclei of seyfert galaxies ( reynolds 1997 ; george et al . 1998 ; crenshaw et al . 1999 ) . with the advent of the space telescope imaging spectrograph ( stis ) on the _ hubble space telescope _ ( _ hst _ ) , the _ far ultraviolet spectroscopic explorer _ ( _ fuse _ ) , and the _ chandra x - ray observatory _ ( _ cxo _ ) , it has become possible to study the absorption systems at unprecedented spectral resolution . early studies with these instruments ( e.g. , crenshaw & kraemer 1999 ; kaastra et al . 2000 ; kaspi et al . 2000 , 2002 ; kriss et al . 2000 ) proved successful in isolating kinematic components of the absorption and constraining the physical conditions in the gas . the absorbers are often variable in their ionic column densities , and it was quickly realized that simultaneous uv , far - uv , and x - ray observations provide the most valuable information , since they cover an enormous range in ionization , as well as address the uncertain relationship between the uv and x - ray absorbers . such observations in at least two of the three above wave bands are available for ngc 4051 ( collinge et al . 2001 ) , ngc 3783 ( kaspi et al . 2002 ; gabel et al . 2003 ) , ngc 3516 ( kraemer et al . 2002 ; netzer et al . 2002 ) , and mrk 509 ( kraemer et al . 2003 ; yaqoob et al . 2003 ) . in crenshaw & kraemer ( 1999 , hereafter ck99 ) , we presented the first stis echelle observations of the seyfert 1 galaxy ngc 5548 . we confirmed the presence of five kinematic components of absorption in ly@xmath2 , n v , and c iv at significant blueshifts with respect to the systemic velocity , which were first detected in goddard high resolution spectrograph ( ghrs ) observations ( crenshaw et al . 1999 ; mathur et al . compared to the ghrs spectra , the column densities of n v and c iv in the stis spectra had decreased substantially in components 1 and 3 ( at respective radial velocities of @xmath01040 and @xmath0530 km s@xmath1 , relative to an adopted systemic redshift of 0.01676 [ ck99 ] ) . we attributed these variations to changes in the total column densities of these components , but could not rule out variable ionization for component 1 . in ck99 , we calculated photoionization models for the uv absorption components . based on the c iv and n v column densities from the stis 1998 data , we found that the ionization parameters and total ( ionized plus neutral ) hydrogen column densities of components 2 5 were too low to produce significant o vii and o viii columns . since we only had an upper limit for the c iv column density in component 1 from the stis 1998 spectrum , we relied on the nonsimultaneous ghrs observations of c iv and n v to produce a model for this component . in this case , the column density ratio of n v / c iv was @xmath318 , which together with the actual columns yielded a high ionization parameter ( u @xmath4 2.4 ) , as well as the observed o vii and o viii column densities from earlier observations by _ asca _ ( reynolds 1997 ; george et al . 1998 ) . on this basis , we suggested that component 1 was likely to be the x - ray warm absorber , with the caveat that none of the uv or x - ray observations were simultaneous . in our conclusions ( ck99 ) , we noted that the ghrs c iv and n v observations were obtained @xmath36 months apart , the _ asca _ observations were @xmath33 years earlier , and that our findings needed to be checked with simultaneous uv and x - ray observations . based on the o vi and h i column densities in _ fuse _ observations of ngc 5548 , brotherton et al . ( 2002 ) found that component 1 could not be an x - ray warm absorber , and suggested that our original claim could be due in fact to nonsimultaneous observations . arav et al . ( 2002 ) examined the ghrs and stis spectra , and suggested that our c iv column densities for component 4 were underestimated by a factor @xmath5 4 , due to covering factor effects . given the importance of resolving these issues , we obtained simultaneous _ stis and _ chanda _ observations of ngc 5548 at significantly longer exposure times than those for previous observations . we present the results on the uv observations in this paper , and the x - ray observations in a companion paper ( steenbrugge et al . 2003 , in preparation , hereafter paper ii ) . we obtained stis echelle spectra of the nucleus of ngc 5548 through a 02 x 02 aperture on two consecutive days ( 2002 january 22 , 23 ) . we list the details of these observations and previous high - resolution observations obtained by _ hst _ in table 1 . the setup of the new observations was identical to that of our first stis observations of ngc 5548 ( ck99 ) , covering the full uv range from 1150 3119 at a velocity resolution ( fwhm ) of 7 10 km s@xmath1 . table 1 shows we were able to devote a much larger exposure time to the new e140 m spectra , which contain all of the previously detected intrinsic absorption lines . we also list the previous ghrs observations of the c iv ( crenshaw et al . 1999 ; mathur et al . 1999 ) and n v ( savage , sembach , & lu 1997 ; crenshaw et al . 1999 ) regions obtained on separate occasions ; we have reanalyzed these spectra for this paper . we reduced the stis spectra using the idl software developed at nasa s goddard space flight center for the stis instrument definition team . the data reduction includes a procedure to remove the background light from each order using a scattered light model devised by lindler ( 1998 ) . the individual orders in each echelle spectrum were spliced together in the regions of overlap . we found no evidence for variability in the e140 m spectra obtained on consecutive days , so we averaged these spectra together for further analysis . to place the new spectra into context , we have retrieved all of the previous low - resolution spectra of ngc 5548 that contain the 1360 continuum region , which are those obtained by the _ international ultraviolet explorer _ ( _ iue _ ) , _ the hopkins ultraviolet telescope _ ( _ hut _ ) , and the faint object spectrograph ( fos ) on _ hst_. we give a brief log of these observations in table 2 . we retrieved the most recently processed versions of these spectra from the multimission archive at the space telescope science institute ( mast ) , except for those obtained during the _ hst _ fos monitoring on 1993 april 19 may 27 , for which we used the reprocessed versions in korista et al . we measured continuum fluxes by averaging the points in a bin centered at 1360 ( observed frame ) with a width of 30 , and determined the one - sigma flux errors from the standard deviations . for the _ iue _ spectra , this method overestimates the errors ( clavel et al . 1991 ) , and we therefore scaled them by a factor of 0.5 to ensure that observations taken on the same day agree to within the errors on average . figure 1 shows the uv continuum light curve of ngc 5548 , which spans a 24-year time period . the variations are dramatic , but undersampled , except for the 1988/1989 _ iue _ campaign around jd 2,447,600 ( clavel et al . 1991 ) and the 1993 _ hst _ fos/ _ iue _ campaign around 2,449,100 ( korista et al . 1995 ) . the last two observations in figure 1 are from stis , and demonstrate that the large amplitude variations in the uv have continued . our previous stis observation was obtained at a high continuum state [ f@xmath6(1360 ) @xmath4 6.61 ( @xmath70.31 ) x 10@xmath8 ergs s@xmath1 @xmath9 @xmath1 ] , whereas the most recent one was obtained at a low state [ f@xmath6(1360 ) @xmath4 1.97 ( @xmath70.20 ) x 10@xmath8 ergs s@xmath1 @xmath9 @xmath1 ] . for comparison , the all - time low occurred on 1992 july 5 ( jd 2,448,809 ) at a flux level of f@xmath6(1360 ) @xmath4 1.07 ( @xmath70.17 ) x 10@xmath8 ergs s@xmath1 @xmath9 @xmath1 ( crenshaw , boggess , & wu 1993 ) , a factor of @xmath32 lower than our stis 2002 observation . the ghrs fluxes in figure 1 are estimates , since the continuum regions surrounding the broad emission lines were not observed , due to the small wavelength coverage . in these cases , we used separate fits to the continua and broad emission lines from the stis spectra , and tried different linear combinations of these fits until we obtained an accurate match to the observed ghrs spectra . the inferred continuum levels for the two ghrs spectra are quite different : f@xmath6(1360 ) @xmath4 2.7 ( @xmath70.4 ) x 10@xmath8 and 5.6 ( @xmath70.4 ) x 10@xmath8 ergs s@xmath1 @xmath9 @xmath1 for the spectra of the n v and c iv regions , respectively . this has significant ramifications , since we assumed in our original modeling of the intrinsic absorption ( ck99 ) that the ghrs spectra were obtained in a similar state . we will discuss the implications of this result in subsequent sections . we have also retrieved the _ far - ultraviolet spectroscopic explorer _ spectrum of ngc 5548 , covering the range 905 1187 , which was obtained on 2000 june 7 ( jd 2,451,703 ) by brotherton et al . ( 2002 ) . the continuum level at this time was even lower than that of the stis 2002 spectrum . in the region of overlap , the stis 2002 spectrum has a continuum flux that is 1.5 times that of the _ fuse _ spectrum . in these low continuum states , the broad emission - line fluxes are also low , and due to the resulting contrast , the contributions of the narrow - line region ( nlr ) to the emission - line profiles is much easier to determine . in particular , the narrow components of o vi @xmath111031.9 , 1037.6 were identified in the _ fuse _ spectrum by brotherton et al . ( 2002 ) , and the narrow components of ly@xmath2 @xmath12 1215.7 , c iv @xmath111548.2 , 1550.8 , and c iii ] @xmath121908.7 were previously identified and measured in the low - state fos spectrum by crenshaw et al . ( 1993 ) , and are also easily detected in the stis 2002 spectra . in figure 2 , we show the c iv profile from the three high - resolution spectra of this region , and label the kinematic components identified in our previous papers ( crenshaw et al . 1999 ; ck99 ) . all of the previously identified components are still present in the new spectrum , but the strengths of some components have changed . as noted in ck99 , absorption components 1 and 3 showed evidence for a significant decrease in their c iv and n v columns between the ghrs 1996 and stis 1998 observations ( although we could only place an upper limit on the c iv column for component 1 in the stis 1998 data ) ; the other components were consistent with no change . this trend has continued @xmath34 years later ; components 1 and 3 are still highly variable , with substantial _ increases _ in the c iv and n v columns since the stis 1998 observation ( for n v , compare figure 3 with plots in our earlier papers ) . thus , the pattern for components 1 and 3 has been an increase in ionic column densities with decreasing continuum flux . however , there are only three observations of each ion , and this does not prove a direct connection between column changes and variations in the ionizing flux ( see 4.5 ) . figure 3 shows a plot of the stis 2002 data in the regions where intrinsic absorption is detected . as in the past , absorption is detected in the lines of ly@xmath2 @xmath12 1215.7 , n v @xmath11 1238.8 , 1242.8 , and c iv @xmath111548.2 , 1550.8 . at this low state , and with a higher signal - to - noise spectrum , we have still not detected any absorption components from lower ionization lines , such as si iv @xmath111393.8 , 1402.8 or mg ii @xmath11 2796.3 , 3803.5 . component 6 , near the systemic velocity , is still only seen in ly@xmath2 , and is likely due to the interstellar medium or halo of the host galaxy ( ck99 ) . components 2 and 5 are relatively narrow and uncomplicated . there is considerable structure in components 1 , 3 , and 4 that has persisted over @xmath36 years . components 1 and 4 each show a strong subcomponent in their red wings , and component 3 appears to consist of at least three roughly equally - spaced subcomponents . figure 2 shows that it is the two subcomponents on either side of the central trough in component 3 that have increased dramatically in strength in the latest stis spectrum . the procedures we used to measure the absorption lines follow those of crenshaw et al . ( 1999 ) . to determine the shape of the underlying emission , we fit cubic splines separately to the continuum and the emission lines . for the continuum , we fit four wavelength regions unaffected by emission or absorption features spanning the full range of the spectrum . for this low - state spectrum , the broad and narrow emission components are easily distinguished by inflections in the profiles ( see also crenshaw et al . we therefore performed a single spline fit to each broad and narrow component of ly@xmath2 and c iv , again using segments unaffected by other emission or absorption lines . to fit the n v emission , which is relatively weak and blended with ly@xmath2 , we reproduced the c iv broad and narrow profiles at their expected n v positions ( assuming the flux ratio of n v @xmath121238.8/@xmath121242.8 @xmath4 2 ) , and scaled the profiles until a suitable fit was obtained . we note that our philosophy is to use the lowest reasonable emission - line profiles needed to fit the regions surrounding the absorption accurately . to normalize the absorption profiles , we divided the observed spectra by our spline fits . as described in crenshaw et al . ( 1999 ) , we determined uncertainties in the absorption parameters from photon statistics and different reasonable placements of the continuum and emission - line profiles . figure 3 shows our fits to the spectrum . we also show the narrow emission - line profiles separately , in order to gauge their fluxes above zero as a function of radial velocity . the narrow - line profile of c iv ( and likely n v ) has a strong blue tail , which is common for seyfert galaxies ( vrtilek & carleton 1985 ) , particularly for high - ionization lines from the inner nlr ( kraemer & crenshaw 2000 ) . the ly@xmath2 profile shows a small bump in its red wing the nature of the bump is uncertain . the centroids of the n v and c iv profiles are shifted to the red , relative to ly@xmath2 . this is due to the contribution from the weaker member of the doublets in emission , which is most noticable in n v. an important aspect of figure 3 is that it shows that component 4 dips below the narrow emission profiles in all three ions . arav et al . ( 2002 ) find that in the earlier stis data , their fits to the narrow - line profiles just skim the bottom of the absorption troughs from component 4 . these authors suggest that the narrow emission lines are therefore not covered by the absorption components , and that component 4 in particular is likely to be heavily saturated . however , with the advantage of having a low - state spectrum , we find that the fluxes of the narrow - line profiles do exceed the troughs of component 4 . thus , _ at least _ a portion of the nlr in this small aperture ( 02 x 02 ) is covered by absorption component 4 . table 3 shows our measurements of the radial velocity centroids , widths ( fwhm ) , and covering factors for each component . the radial velocities , widths , and associated one - sigma uncertainties are based on averages of the unblended components in n v and c iv . a comparison with our results on the previous ghrs and stis observations ( ck99 ) shows that the velocity centroids have not changed ( outside of the uncertainties ) over the previous six years . however , the widths of components 1 and 3 have increased substantially since the first stis observations : from 85 to 222 km s@xmath1 for component 1 and from 78 to 159 km s@xmath1 for component 3 . we attribute these increases in fwhm to increases in the depths of the subcomponents , relative to the central cores . we find that the widths of these components at the continuum level did _ not _ change , which supports our conclusion that the increases in fwhm are due to changes in the ionic columns rather than changes in the coverage of velocity space by the components . in addition to the above measurements , we have taken advantage of the higher signal - to - noise ratio of the stis 2002 spectrum to determine the covering factors ( c@xmath13 ) for the cores of several components . we used the technique of hamann et al . ( 1997 ) for doublets to determine c@xmath13 , which is the fraction of continuum plus emission that is occulted by the absorber in our line of sight . since we have no information to the contrary , our measurements implicitly assume that the continuum , broad emission , and narrow emission are equally covered by each absorption component . figure 2 shows that the only component of absorption in c iv that is not blended with another is component 4 . however , for n v , the doublets of 2 , 3 , and 4 are clean ( n [email protected] component 5 is blended with n [email protected] component 1 , see crenshaw et al . furthermore , the wings of these components are blended with those from other components . thus , we can only reliably determine covering factors for the cores of components 2 , 3 , and 4 . our derived covering factors are given in table 3 . components 2 , 3 , and 4 all show nonunity covering factors in their cores ; previously , we had only been able to determine that component 4 had a nonunity covering factor ( crenshaw et al . the actual values in this table are from the n v doublet , whereas the lower limits are from the residual intensities in l@xmath2 . these values are consistent with our estimates or lower limits from the previous ghrs and stis spectra . gabel et al . ( 2003 ) show that it is possible for the total covering factor ( i.e. , that for continuum plus emission ) to differ among different lines ( e.g. , c iv and n v ) . this effect arises as a result of different covering factors for the continuum and emission - line regions , combined with different contributions of the continuum and emission - line fluxes to the underlying emission . the only direct test that we have of this effect is for component 4 ; in this case we find that the covering factors from the c iv doublet is 0.91 @xmath7 0.03 , which is the same as that from the n v doublet to within the errors . for components 2 and 3 , we must assume that the c iv covering factors are the same as those derived for n v. since we can not measure covering factors for components 1 and 5 , we assume that c@xmath13 @xmath4 1 for these two components . brotherton et al . ( 2002 ) independently derived covering factors for the o vi absorption components , and their values agree with ours to within the uncertainties . in addition , they obtained c@xmath13 @xmath4 0.98 ( @xmath70.12 ) and 0.94 ( @xmath70.14 ) for components 1 and 5 , respectively , which indicates that our assumption of unity covering factors for these two components is a reasonable one . to determine the ionic column densities for each component , we converted each normalized profile to optical depth as a function of radial velocity , and integrated across the profile , as described in crenshaw et al . ( 1999 ) . for components 2 , 3 and 4 , we determined the optical depths using the measured c@xmath13 values and the formalism of hamann et al . for the other components , we assumed c@xmath13 @xmath4 1 . as we stated in the preceding paragraph , this appears to be a reasonable assumption for components 1 and 5 , and a necessary assumption for component 6 if it is indeed due to the interstellar medium or halo of the host galaxy . we list the ionic column densities for each component in table 4 . for the ghrs and stis 1998 spectra , we have remeasured the columns for components 2 , 3 , and 4 assuming the covering factors we determined from the stis 2002 spectra . we give upper limits to the si iv absorption , since they provide useful constraints for the photoionization models . we consider the h i column densities from l@xmath2 to be lower limits . as mentioned in ck99 , the components are more blended in ly@xmath2 than those in the other lines and are therefore more heavily saturated . furthermore , the h i column densities from ly@xmath14 in the fuse spectrum ( brotherton et al . 2002 ) are 5 8 times higher than the values in table 1 . it is unlikely that this could be due to variability of the h i column alone , since we do nt see evidence for strong variations in other lines of components 2 , 4 , and 5 . the most straightforward explanation is that the ly@xmath2 lines are indeed heavily saturated , and our measurements are lower limits . it would be interesting to obtain a longer _ fuse _ exposure to detect higher - order lyman lines , to check for saturation in ly@xmath14 . to investigate the effects of the nonunity covering factors on the measured ionic columns , we compared the values in table 4 with those we determined previously for the ghrs and stis 1998 data ( ck99 ) . this comparison reveals that the remeasured c iv columns are 1.2 to 1.5 times larger than those previously determined assuming c@xmath13 @xmath4 1 . the effects on the n v measurements are the same or slightly larger ; the remeasured n v columns are 1.2 to 1.9 times larger . our earlier statements on absorption variability based on the observed spectra are borne out by the measurements in table 4 . component 1 shows huge ( factor of 6 10 ) changes in the n v and c iv column densities , and component 3 shows significant ( factor of @xmath32 ) variations in both ions . other components show no evidence or only marginal evidence for column density variations , given the uncertainties . for the first time , we have direct measurements of the n v and c iv column densities in component 1 from simultaneous observations ( the ghrs observations were not simultaneous , and the previous stis observation could only place a lower limit on the c iv column ) . thus , we are able to check our previous suggestion that component 1 might be the x - ray warm absorber , based on its high n v / c iv ratio ( @xmath4 18 ) from the ghrs observations . in table 4 , we see that component 1 s n v / c iv ratio in the stis 2002 spectrum is only 2.5 , which is close to the ratios for components 2 , 4 , and 5 . based on our models ( 4 ) , this value does not produce significant o vii or o viii columns , and component 1 in the stis 2002 spectrum _ can not _ be an x - ray warm absorber at that epoch . given this finding , we can no longer assume that the ghrs n v / c iv ratio provides an appropriate constraint on the physical conditions in component 1 . the ionic column densities in table 4 are strongly variable over 2 4 years , and it is certainly possible that they varied over the six - month interval between the ghrs observations . furthermore , in 2.2 we showed that the ghrs c iv and n v observations were obtained at different continuum states , so that one can not call upon a similar ionization state of the gas as a reason for using this ratio . finally , photoionization models of component 1 using simultaneous observations of the o vi and ly@xmath14 absorption in _ fuse _ data ( brotherton et al . 2002 ) indicate that this component has an ionization parameter too low to be a warm absorber at the epoch of observation ( 2000 june 7 ) . thus , there is no longer any significant evidence that component 1 is , or was , a warm absorber . note that this does not rule out the presence of x - ray absorption from a component of more highly ionized gas at this radial velocity , and in fact , kaastra et al . ( 2002 ) present such evidence based on c vi absorption in _ cxo _ spectra at the same approximate radial velocity as component 1 ( see 4.4 ) . furthermore , in our simultaneous hetgs spectrum , the high ionization si xiv ly@xmath2 line is dominated by a velocity component compatible with uv component 1 ( paper ii ) . as noted earlier , arav et al . ( 2002 ) suggest that the narrow emission lines in the stis aperture are not covered by the absorption components . they conclude that component 4 , which has the deepest troughs , is therefore much more saturated than it appears . in particular , they suggest that the c iv column densities for component 4 are at least four times greater than we had previously measured ( crenshaw et al . 1999 ) . however , with the new low - state stis spectrum , we find that the peaks of the narrow - line profiles do extend above the troughs of component 4 in each line ( l@xmath2 , c iv , and n v ) . thus , _ at least _ a portion of the nlr is occulted by the uv absorber associated with component 4 . as discussed earlier , the inclusion of nonunity covering factors does increase the measured column densities , but only by a factor of 1.2 to 1.9 . in particular , the c iv column density for component 4 increased by a factor of only 1.2 with the inclusion of the nonunity covering factor . thus , we find no evidence for the severe underestimation of column densities suggested by arav et al . ( 2002 ) . there are a couple of caveats that we should mention concerning the above discussion . first , we have implicitly assumed that the continuum source , blr , and nlr are all covered by the same fraction , but it could be that only the nlr is partially covered . in the latter case , one could devise emission - line profiles for the uncovered nlr that skim the bottoms of the component 4 troughs , and thereby produce much higher ionic column densities ; however , we have no evidence that this is actually the case . second , we have not been able to accurately determine the variation of covering factor as a function of radial velocity for any component , due to a combination of blending and insufficient signal - to - noise . thus , the column densities could be higher if the covering factor decreases in the wings . however , we expect this effect would be small , since it seems unlikely that saturation effects in the wings would be as large as those in the troughs , which are not severe . an accurate determination of the covering factor as a function of radial velocity should be possible with a substantially longer exposure time . for the _ fuse _ data , brotherton et al . ( 2002 ) generated fits to the o vi emission profile assuming a `` covered nlr '' and an``uncovered nlr '' case , and prefer the uncovered case . however , it looks to us as though the nlr flux from either fit exceeds the trough of component 4 , and the covered nlr is therefore preferable . in addition , the width of their narrow o vi profile is @xmath3400 km s@xmath1 for the uncovered case and @xmath3650 km s@xmath1 for the covered case ; the latter is similar to our values of @xmath3700 km s@xmath1 for c iv and 590 km s@xmath1 for ly@xmath2 . thus , it appears that the covered case is much more likely for o vi , as well as the other lines . however , we can not rule out the possibility that some fraction of the narrow o vi emission is not covered , particularly in the large ( 30@xmath15 x 30@xmath15 ) _ fuse _ aperture . in figure 4 , we show the _ fuse _ spectrum in the region of the o vi lines . the spectrum was normalized by dividing by the covered nlr fit of brotherton et al . ( 2002 ) ; each member of the doublet is plotted as a function of radial velocity . the positions of the stis components align well with the absorption troughs ( at most there is a 10 km s@xmath1 offset ) . brotherton et al . find that there is a velocity offset of @xmath3100 km s@xmath1 between the stis and _ fuse _ components , and suggest this offset is probably due to calibration uncertainties . the source of the discrepancy is actually the different systemic redshifts used ; brotherton et al . used the h i 21-cm value of 0.017175 , whereas we use the emission - line redshift of 0.01676 to be consistent with our earlier papers , which amounts to a velocity offset of 125 km s@xmath1 . for components 1 , 2 , 4 , and 5 , the blue member of the o vi doublet in figure 4 is clearly deeper than the red member , indicating that these lines are not heavily saturated . however , for component 3 ( brotherton et al.s components 2.5 plus 3 ) , the two lines lie on top of each other , which suggests heavy saturation in o vi across the profile and a covering factor of @xmath30.45 ( in agreement with brotherton et al.s value for component 2.5 ) . we suggest that the o vi absorption profile for component 3 is heavily saturated , and should therefore be considered a lower limit . the inferred large o vi column is consistent with the high n v / c iv ratio ( @xmath4 9.2 ) that we find for component 3 , as discussed in 4.2 and 4.3 . photoionization models for this study were generated using the code cloudy90 ( ferland et al . we have modeled the absorbers as matter - bounded slabs of atomic gas , irradiated by the ionizing continuum radiation emitted by the central source . as per convention , the models are parameterized in terms of the ionization parameter , u , the ratio of the density of photons with energies @xmath5 13.6 ev to the number density of hydrogen atoms at the illuminated face of the slab , and the total hydrogen column density , n@xmath16 ( @xmath4 n@xmath17 @xmath18 n@xmath19 ) . for n@xmath16 @xmath20 10@xmath21 @xmath22 , the predicted ionic columns densities are not sensitive to density . for the models , we have assumed only thermal broadening , since 1 ) the absorption line widths could be due to the superposition of unresolved kinematic components and 2 ) comparison models assuming turbulent velocities of @xmath23 300 km s@xmath1 predict nearly identical ionic columns . the models were deemed successful when the predicted n v and c iv column densities matched those observed to within the measurement errors . for the spectral energy distribution ( sed ) of the ionizing radiation , we used the x - ray continuum given in paper ii . after correcting for galactic reddening ( e@xmath24 @xmath4 0.03 ) and intrinsic reddening ( e@xmath24 @xmath4 0.04 , see kraemer et al . 1998 ) , we fit the uv continuum fluxes from 1170 to 3000 with a power law of index @xmath25 @xmath26 @xmath01.0 . we assumed that the uv continuum turns over at the lyman limit , with @xmath25 @xmath4 @xmath01.4 , joining the the euv and x - ray continua at @xmath3100 ( 124 ev ) . figure 5 shows our adopted sed , which is similar to that used in ck99 ; tests show that the two seds yield essentially identical results . we use our simultaneous observations of n v and c iv as the principal constraints on the models . additional simultaneous lower limits on h i ( from ly@xmath2 ) and upper limits on si iv are also helpful . since the ghrs c iv and n v observations were not simultaneous and were , in fact , obtained during different continuum states , we do not calculate models for the ghrs data , but use the observed columns to investigate the absorption variability . for additional constraints , we make use of the simultaneous _ cxo _ observations , as well as the nonsimultaneous observations with _ fuse _ , keeping in mind that the ionic column densities are almost certainly variable . for models of the stis columns , we begin with the most well - constrained data , which are those from the 2002 observation . for the stis 2002 data , we initially assumed that the uv absorbing gas was free of cosmic dust . we also assumed solar elemental abundances ( see grevesse & anders 1989 ) , which are , by number relative to h , as follows : he @xmath4 0.1 , c @xmath4 3.4 x 10@xmath27 , n @xmath4 1.2 x 10@xmath27 , o @xmath4 6.8 x 10@xmath27 , ne @xmath4 1.1 x 10@xmath27 , mg @xmath4 3.3 x 10@xmath28 , si @xmath4 3.1 x 10@xmath28 , s @xmath4 1.5 x 10@xmath28 , and fe @xmath4 4.0 x 10@xmath28 . the resulting parameters from our best fitting models are listed in table 5 ( each designated as component number followed by `` solar '' ) , while the predicted ionic column densities are given in table 6 , along with the observed values for comparison . for each component , we matched the observed column densities n@xmath29 and n@xmath30 to well within the measurement errors . in each case , the model predictions for n@xmath17 exceeded the lower limits determined from ly@xmath2 and the predictions for n@xmath31 are @xmath32 10@xmath33 @xmath9 ( lower than the observational upper limits ) . as mentioned previously , we have been able to determine simultaneous n v and c iv columns for component 1 for the first time . the derived ionization parameter and column density for component 1 are lower than those from ck99 by factors of @xmath313 and @xmath3200 , respectively , and the predicted o vii and o viii column densities are undetectable . this supports our conclusion ( 2.5 ) and that of brotherton et al . ( 2002 ) that the gas responsible for the uv absorption in component 1 can not produce the observed x - ray absorption . again , this does not exclude the presence of more highly ionized gas at the same velocity . table 6 also lists the observed columns of h i and o vi from the _ fuse _ spectrum ( brotherton et al . 2002 ) obtained @xmath319 months earlier . based on similar velocity coverages , we have added their components 0.5 and 1 together for our component 1 , and combined their components 2.5 and 3 for our component 3 . as noted previously , we use the o vi value for component 3 as a lower limit . although we were able to obtain good fits to the n v and c iv columns and a reasonable match to the h i columns ( except for component 3 ) , the dust - free models predict o vi columns far in excess of those measured , by similar factors of 6 8 . this discrepancy can not be due to the ionizing continuum used in the model since , unless there is extreme structure in the ionizing continuum in the vicinity of the o v ( 113.9 ev ) and o vi ( 138.1ev ) threshold energies , the model predictions for n@xmath34 do not depend strongly on sed . we consider it highly unlikely that variability could be the main source of the o vi discrepancies , for the following reasons . 1 ) none of the other ions show significant variability in components 2 , 4 , and 5 . 2 ) the _ fuse _ spectrum was in a lower continuum state than that for the stis 2002 spectrum , which would result in even higher predicted o vi columns for components 1 and 3 , according to our models . 3 ) it seems very unlikely that the observed o vi columns would all be low by roughly the same amount as a result of extreme decreases in the total column density for each component during this epoch . thus , it seems reasonable that these models severely overpredict the o vi columns , even though the observations are not simultaneous . brotherton et al . ( 2002 ) suggested that models based on n@xmath30/n@xmath29 alone may be unreliable due to super - solar nitrogen abundances . we investigate this possibility in the next section and the possibility that the discrepancies can be explained by depletion ( primarily of carbon ) into dust grains in 4.3 . we generated a second set of models with a nitrogen abundance that is two times solar , and assumed that the enhanced nitrogen is at the expense of carbon alone . this is the simplest reasonable adjustment of elemental abundances that can be made , for the following reason . if enrichment of the interstellar medium above solar abundances is due to intermediate - mass stars , the carbon gets converted to nitrogen first through the cno process , whereas the oxygen goes to nitrogen at later times and higher temperatures ( maeder & meynet 1989 ) . thus , for these models , we keep the same abundances as before , except that c @xmath4 2.2 x 10@xmath27 and n @xmath4 2.4 x 10@xmath27 by number relative to h. the model results are listed in tables 5 and 6 under the designation `` 2xn '' . we were able to match the observed n v and c iv columns as before to well within the measurement errors ( and n@xmath31 is @xmath32 10@xmath33 @xmath9 in all cases ) . furthermore , we have obtained a reasonable ( within a factor of two ) match to the n@xmath34 for components 1 , 2 , 4 , and 5 , as well as a value for component 3 that is consistent with the lower limit . in addition , the predicted h i column for component 3 is much closer to the observed value from _ fuse_. as can be seen in table 5 , the principal effect of enhancing nitrogen and depleting carbon is that the ionization parameter and column density of each component must be lowered substantially to match the observed n v / c iv ratio and their columns . the values for u and n@xmath16 are now in approximate agreement with those determined by brotherton et al . ( 2002 ) , except for component 3 . based on the high n@xmath30/n@xmath29 ratio ( @xmath4 9.2 ) for component 3 , we determined a much larger column density and ionization parameter than did brotherton et al . ( 2002 ) , and our prediction for n@xmath34 is @xmath38 times greater than their estimate . however , given evidence that the o vi absorption is heavily saturated and the quoted column density is a lower limit , our prediction is consistent with the observations . thus , for all of the components , the enhanced nitrogen models provide a reasonably good match to all of the ionic columns . we generated a third set of models that starts with solar abundances , but includes dust within the uv absorbers . we assumed the following depletions from solar - abundance gas phase ( see snow & witt 1996 ) : c , 65% ; o , 25% ; mg , 80% ; si , 94% ; and fe , 95% ( n is not depleted ) . the results are listed in tables 5 and 6 under the designation `` dusty '' . again , we were able to match the observed n v and c iv columns , the n@xmath34 for components 1 , 2 , 4 , and 5 , and the lower limit to n@xmath34 in component 3 . the predicted h i columns are also close to the _ fuse _ values . thus , the dusty uv models provide an equally good match to the ionic columns as those provided by the enhanced nitrogen models . the ionization parameters are nearly identical for the two cases , whereas the n@xmath35 columns for the dusty models are 2 3 times those of the enhanced nitrogen models . as noted above , in kraemer et al . ( 1998 ) we estimated an intrinsic reddening of the nlr in ngc 5548 of e@xmath24 @xmath26 0.04 mag . assuming a galactic ratio of reddening to the total hydrogen column ( see shull & van steenberg 1985 ) , this corresponds to a total hydrogen column density of 2.1 x 10@xmath36 @xmath9 . this in remarkable agreement with our total model hydrogen column of 2.3 x 10@xmath37@xmath9 . hence , it is possible that the reddening of the continuum and emission lines occurs in the same gas in which the uv absorption lines arise . kraemer et al . ( 1998 ) place the high - ionization inner nlr at a distance of @xmath31 pc from the central source , which is comfortably outside the dust sublimation radius ( r@xmath38 ) . following barvainis ( 1987 ) , and assuming a dust sublimation temperature of 1500 k ( salpeter 1997 ) , r@xmath38 @xmath4 1.3 x l@xmath39 , where l@xmath40 is the uv x - ray luminosity of the central source in units of 10@xmath41 erg s@xmath1 . based on our continuum fit , l@xmath40 @xmath26 0.01 , and hence r@xmath38 @xmath3 0.13 pc , which is well inside the inner nlr and the majority of the uv absorbing gas . from our contemporaneous _ cxo _ letgs spectrum ( see paper ii ) , we determined an o vi column density of 2 x 10@xmath42 @xmath9 from the absorption lines at 21.79 and 22.01 . we predict a total n@xmath34 @xmath4 1.1 x 10@xmath42 @xmath9 for the nitrogen - enhanced models and 1.8 x 10@xmath43@xmath9 for the dusty models , mostly from component 3 , which are close to the letgs value . this suggests that at least half of the observed o vi column comes from the absorbers detected in the uv . notably , brotherton et al . ( 2002 ) derive a total n@xmath34 @xmath26 4.1 x 10@xmath44 @xmath9 , significantly below the letgs - derived value , which we suggest is further evidence that the o vi lines in component 3 are saturated . regarding the `` classic '' warm absorbers , the total o vii and o viii columns from the contemporaneous _ cxo _ spectrum are 5 x 10@xmath45 @xmath9 and 1.5 x 10@xmath46 @xmath9 , respectively . the uv components do not make significant contributions to the o vii ( @xmath47 15% ) and o viii ( @xmath47 1% ) columns , confirming the presence of more highly ionized gas than found in the uv . in kaastra et al . ( 2002 ) , the c vi ly@xmath2 line detected in the 1999 letgs spectrum was fitted using a sum of narrow components at the velocities of the five uv absorbers . the c vi profile in our contemporaneous letgs spectrum shows less structure ( see paper ii ) , and it was best fit with three velocity components at @xmath01050 km s@xmath1 , @xmath0520 km s@xmath1 , and @xmath0160 km s@xmath1 . the combined c vi column density from these components is 1.4 x 10@xmath45 @xmath9 . based on our dusty uv models , only component 3 possesses a significant c vi column ( 7.4 x 10@xmath44 @xmath9 for the nitrogen - enhanced model , 1.3 x 10@xmath42 @xmath9 for the dusty model ) , and thus the uv components contribute @xmath4710% of the observed c vi . since c vi requires a substantially higher ionization potential for its creation than o vi ( 392.1 ev and 113.9 ev , respectively ) , this is an indication of more highly ionized gas _ at the radial velocities of the uv absorbers _ , as also seen in ngc 3783 ( kaspi et al . 2002 ) and mrk 509 ( yaqoob et al . 2003 ; kraemer et al . 2003 ) . as noted previously , components 1 and 3 show strong variations in their ionic columns , whereas the other components show little evidence for variations . as in ck99 , we can ask if the ionic column variations can be attributed to either variable ionization or variable total column ( e.g. , from bulk motion of gas across the line of sight ) . as shown in figure 1 , different uv continuum flux levels were observed in the four high - resolution spectra of ngc 5548 , with the ghrs civ and stis 1998 spectra taken during relatively high states and the ghrs n v and stis 2002 spectra taken during low states . if the absorbers respond to changes in the ionizing continuum , we would expect both _ n v and c iv columns and _ larger _ n@xmath30/n@xmath29 ratios during high flux states . this is clearly not the case for component 3 ; although the columns are smaller in 1998 compared to 2002 , the n v / c iv ratio is smaller as well ( 6.8 in 1998 compared to 9.2 in 2002 ) . although we only have an upper limit for n@xmath29 in the stis 1998 data for component 1 , n@xmath30 requires an ionizing continuum a factor of @xmath36 higher during the stis 1998 observations . however , the observed uv flux was only 3.4 times as bright . hence , either the amplitude of the variations in the euv ionizing radiation is about twice as large as that in the uv , or the observed changes in the ionic columns for component 1 are not consistent with the changes in the continuum flux . given the lack of response of the absorbers to changes in the continuum flux ( with the possible exception of component 1 ) we can derive upper limits to their electron densities and lower limits to their distances from the central continuum source . the timescale for the recombination of an ion x@xmath48 is given by the following expression ( krolik & kriss 1995 ; bottorff , korista , & shlosman 2001 ) : @xmath49^{-1},\ ] ] where @xmath50 is the recombination coefficient for ion @xmath51 , @xmath52 ) is the fraction of element @xmath53 in ionization state @xmath54 , and n@xmath55 is the electron density . based on our model predictions , components 1 , 2 , 4 , and 5 have similar conditions , with t @xmath26 2.4 x 10@xmath56 k , f(c v)/f(c iv ) @xmath26 6 , and f(n vi)/f(n v ) @xmath26 1.3 . for component 3 , which is at higher ionization , we predict t @xmath26 4.3 x 10@xmath56 k , f(c v)/f(c iv ) @xmath26 70 , and f(n vi)/f(n v ) @xmath26 16 . assuming only radiative recombination ( see shull & van steenberg 1982 ) , and a timescale of @xmath57 4 yrs ( the interval between stis observations ) , we obtain the following upper limits on electron density : for component 3 , 15 @xmath22 and 50 @xmath22 ( from c iv and n v , respectively ) ; for the lower ionization components , we obtain 150 @xmath22 and 650 @xmath22 . based on our estimate of the central source luminosity above the lyman limit ( see kraemer et al . 1998 ) and our ionization parameters , the lower limit to the radial distances of the low ionization components is between 20 and 250 pc , while that for component 3 is between 150 and 250 pc from the central source . these constraints place the absorbers at large enough distances that they may cover much of the entire nlr . however , we note that the continuum and absorption variations are obviously not well characterized , with just a few high - resolution observations widely separated in time . since the ionic columns for components 2 , 4 , and 5 have not varied significantly , despite large variations in the continuum fluxes , the above constraints are likely valid . for the variable components 1 and 3 , however , we can not rule out changes in ionization in response to unobserved continuum variations on time scales smaller than years . thus , we consider the limits on the densities and radial locations of components 1 and 3 to be loose constraints ; more frequent monitoring at high spectral resolution are needed to test them . since there is no clear signature of a response to the changes in the continuum flux , we can investigate whether or not the variations in ionic column densities are consistent with changes in the total column densities of the absorbers . pursuing this , we generated sets of nitrogen - enhanced and dusty models for components 2 5 in the stis 1998 data , with the same ionization parameters used for the stis 2002 models , but adjusting the total column densities ( see table 7 ) . in each case , the predictions for n@xmath29 and n@xmath30 match the measured values to within the measurement errors . components 2 , 4 , and 5 show only small changes in the total columns , consistent with our earlier claims of no or marginal variability of these components . component 3 shows a much smaller column in the 1998 data , and the agreement between observed and predicted columns indicates that the variations are likely dominated by changes in the total column ( by a factor of @xmath32.6 ) . we did not attempt to model component 1 , since we only had an upper limit for n@xmath29 from the stis 1998 spectra of course , we can reproduce the observed n v column in 1998 by reducing the total column of gas by a factor of @xmath36 . obviously , the larger c iv and n v columns in components 1 and 3 for the two ghrs observations ( compared to the stis 1998 values ) could also be attributed to changes in the total columns , but we have no constraints on possible ionization changes . new _ hst _ stis echelle spectra , combined with simultaneous _ cxo _ observations , place tighter constraints on the uv absorbers of ngc 5548 than were possible in the past . simultaneous detection and measurement of the c iv and n v absorption lines in component 1 show that this component is _ not _ an x - ray warm absorber , in agreement with the conclusion of brotherton et al . ( 2002 ) based on _ fuse _ observations of o vi and ly@xmath14 , and despite our earlier claim based on nonsimultaneous ghrs observations ( ck99 ) . the new stis observations were obtained when the uv continuum and broad emission lines fluxes were low , allowing us to easily distinguish the narrow components of the emission lines . we present direct evidence that uv absorption component 4 covers at least a portion of the nlr . our derivation of the covering factors indicates that our previous c iv and n v column densities were underestimated by factors of 1.2 1.9 , and not the @xmath584 factors suggested by arav et al . ( 2002 ) . our dust - free , solar - abundance models of the uv absorbers in the stis 2002 data match the c iv and n v columns well , but they severely overpredict the observed o vi columns in the nonsimultaneous _ fuse _ data . models that include enhanced nitrogen or elemental depletion into dust , in both cases reducing the carbon in gas phase , provide reasonable ( within a factor of two ) matches to the o vi columns . absorption lines ( or lack thereof ) from heavier refractory elements that tend to be depleted into dust ( e.g. , fe , mg , si ) , would be extremely useful for discriminating between these two models , but unfortunately there are no suitable high - ionization lines from these elements in the uv . in particular , si iv , with a creation ionization potential of 33.5 ev , has an insignificant column even in our dust - free models . if the dusty models are correct , then the total hydrogen column from the absorbers is exactly the amount needed to redden the inner nlr by the observed amount , e(b@xmath0v ) @xmath4 0.04 ( kraemer et al . 1998 ) , assuming a galactic dust - to - gas ratio . this is a piece of evidence in favor of the dusty models , although it is possible that the nlr clouds provide their own reddening . there is strong evidence for dust in uv absorbers at distances of @xmath57100 pc in ngc 3227 ( crenshaw et al . 2001 ) and akn 564 ( crenshaw et al . 2002 ) . however , in these cases , the absorbers appear at radial velocities close to systemic and are probably due to the host galaxy , as opposed to the outflowing absorbers seen in ngc 5548 . our absorber models indicate that none of the uv components contribute significantly to the o vii and o viii columns seen in the x - rays , in agreement with the conclusions of brotherton et al . ( 2002 ) . hence , none of the uv absorbers is a classic x - ray warm absorber . however , the uv components , particularly the higher ionization component 3 , do contribute @xmath5850% of the o vi column and 5 10% of the c vi column measured in the _ cxo _ spectra ( paper ii ) . regarding the source of the ionic column variations in components 1 and 3 , we come to the same conclusions as in ck99 . the variations in component 3 are dominated by changes in the total column density , whereas those in component 1 are probably dominated by column changes , although we can not rule out variable ionization of the gas for this component . as discussed in ck99 , changes in total column due to bulk motion of the gas across our line of sight in ngc 5548 require transverse velocities of @xmath3 1500 km s@xmath1 . there are a couple of indications that the uv components 2 , 4 , and 5 are at significant distances from the active nucleus of ngc 5548 . we have presented direct evidence that component 4 covers at least a portion of the high - ionization , inner nlr , at a distance of @xmath31 pc from the central continuum source ( kraemer et al 1998 ) . also , these absorbers apparently do not respond to continuum variations over time scales of years , which suggests that they are at distances between 20 and 250 pc from the nucleus . desirable observations in the future include coordinated uv and fuv observations at high spectral resolution , to check our conclusions on the nonsolar abundances in the gas phase of the absorbers . longer exposure times on the above would be desirable , to derive covering factors as a function of radial velocity and investigate saturation effects in the wings . monitoring observations at high spectral resolution would place tighter constraints on the densities and distances of the absorbers . we thank nahum arav for helpful discussions . sbk and dmc acknowledge support from nasa grant nag5 - 4103 . support for proposal 9279 was provided by nasa through a grant from the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , under nasa contract nas 5 - 26555 . some of the data presented in this paper were obtained from the multimission archive at the space telescope science institute ( mast ) . support for mast for non - hst data is provided by the nasa office of space science via grant nag5 - 7584 and by other grants and contracts . the space research organisation of the netherlands ( sron ) is supported financially by nwo , the netherlands organisation for scientific research . ccccrl stis & e140 m & 1150 1730 & 46,000 & 7639 & 2002 january 22 + stis & e230 m & 1607 2366 & 30,000 & 2700 & 2002 january 22 + stis & e140 m & 1150 1730 & 46,000 & 7639 & 2002 january 23 + stis & e230 m & 2274 3119 & 30,000 & 2700 & 2002 january 23 + & & & & & + stis & e140 m & 1150 1730 & 46,000 & 4750 & 1998 march 11 + stis & e230 m & 1607 2366 & 30,000 & 2295 & 1998 march 11 + stis & e230 m & 2274 3119 & 30,000 & 1905 & 1998 march 11 + ghrs & g160 m & 1554@xmath59 1590 & 20,000 & 13,600 & 1996 august 24 + ghrs & g160 m & 1232@xmath60 1269 & 20,000 & 4607 & 1996 february 17 + lcrl iue swp & 1150 1978 & @xmath3250 & 1978 june 27 1995 may 15@xmath61 + fos g130h & 1150 1605 & @xmath31200 & 1992 july 5 + fos g130h & 1150 1605 & @xmath31200 & 1993 february 3 + fos g130h & 1150 1605 & @xmath31200 & 1993 april 19 may 27@xmath62 + hut & 820 1840 & @xmath3450 & 1995 march 14 + lccc 1 & @xmath01041 ( @xmath711 ) & 222 ( @xmath718 ) & @xmath580.65 + 2 & @xmath0667 ( @xmath74 ) & 43 ( @xmath74 ) & 0.77 ( @xmath70.03 ) + 3 & @xmath0530 ( @xmath719 ) & 159 ( @xmath727 ) & 0.70 ( @xmath70.02 ) + 4 & @xmath0336 ( @xmath74 ) & 145 ( @xmath74 ) & 0.92 ( @xmath70.02 ) + 5 & @xmath0166 ( @xmath76 ) & 61 ( @xmath713 ) & @xmath580.69 + 6 & @xmath1878 ( @xmath76 ) & 68 ( @xmath710 ) & @xmath580.20 + clccc 1 & h i & & @xmath581.34 & @xmath582.31 + & n v & 1.98 ( 0.29 ) & 0.44 ( 0.18 ) & 2.76 ( 0.10 ) + & c iv & 0.11 ( 0.04 ) & @xmath470.17 & 1.12 ( 0.12 ) + & si iv & & @xmath470.10 & @xmath470.06 + & & & & + 2 & h i & @xmath580.76 & @xmath580.61 & @xmath581.18 + & n v & 0.90 ( 0.11 ) & 1.11 ( 0.16 ) & 1.45 ( 0.13 ) + & c iv & 0.39 ( 0.04 ) & 0.51 ( 0.07 ) & 0.56 ( 0.03 ) + & si iv & & @xmath470.10 & @xmath470.06 + & & & & + 3 & h i & @xmath581.79 & @xmath582.01 & @xmath581.65 + & n v & 7.40 ( 0.51 ) & 3.04 ( 0.16 ) & 8.64 ( 0.56 ) + & c iv & 0.79 ( 0.23 ) & 0.45 ( 0.11 ) & 0.94 ( 0.26 ) + & si iv & & @xmath470.10 & @xmath470.06 + & & & & + 4 & h i & @xmath583.30 & @xmath582.97 & @xmath583.64 + & n v & 7.41 ( 0.87 ) & 10.23 ( 0.79 ) & 7.26 ( 0.78 ) + & c iv & 3.43 ( 0.35 ) & 3.44 ( 0.47 ) & 2.70 ( 0.23 ) + & si iv & & @xmath470.10 & @xmath470.06 + & & & & + 5 & h i & @xmath580.77 & @xmath580.66 & @xmath580.92 + & n v & 1.05 ( 0.23 ) & 1.18 ( 0.24 ) & 1.45 ( 0.10 ) + & c iv & 0.41 ( 0.18 ) & 0.31 ( 0.13 ) & 0.58 ( 0.05 ) + & si iv & & @xmath470.10 & @xmath470.06 + & & & & + 6 & h i & 0.13 ( 0.04 ) & 0.13 ( 0.04 ) & 0.13 ( 0.03 ) + lcc 1 `` solar '' & 3.3 x 10@xmath63 & 0.18 + 1 `` 2xn '' & 3.3 x 10@xmath46 & 0.03 + 1 `` dusty '' & 7.0 x 10@xmath46 & 0.03 + & & + 2 `` solar '' & 2.1 x 10@xmath63 & 0.21 + 2 `` 2xn '' & 1.7 x 10@xmath46 & 0.03 + 2 `` dusty '' & 3.8 x 10@xmath46 & 0.03 + & & + 3 `` solar '' & 1.1 x 10@xmath64 & 1.24 + 3 `` 2xn '' & 6.6 x 10@xmath63 & 0.22 + 3 `` dusty '' & 2.0 x 10@xmath36 & 0.24 + & & + 4 `` solar '' & 1.1 x 10@xmath36 & 0.21 + 4 `` 2xn '' & 8.7 x 10@xmath46 & 0.03 + 4 `` dusty '' & 1.9 x 10@xmath63 & 0.03 + & & + 5 `` solar '' & 1.9 x 10@xmath63 & 0.19 + 5 `` 2xn '' & 1.9 x 10@xmath46 & 0.03 + 5 `` dusty '' & 3.7 x 10@xmath46 & 0.03 + lccccccc 1 `` solar '' & 6.27 & 2.83 & 1.12 & 47.9 & 5.43 & 140 & 19.3 + 1 `` 2xn '' & 5.25 & 2.72 & 1.09 & 5.34 & 0.64 & 2.01 & 0.03 + 1 `` dusty '' & 9.66 & 2.79 & 1.09 & 9.05 & 0.89 & 3.96 & 0.07 + & ( 9.41 ) & ( 2.76 ) & ( 1.12 ) & ( 7.55 ) & & & + 2 `` solar '' & 3.36 & 1.45 & 0.53 & 26.7 & 35.9 & 92.6 & 15.7 + 2 `` 2xn '' & 2.74 & 1.42 & 0.57 & 2.79 & 0.37 & 1.05 & 0.01 + 2 `` dusty '' & 5.06 & 1.50 & 0.57 & 5.03 & 0.50 & 2.28 & 0.04 + & ( 3.25 ) & ( 1.45 ) & ( 0.56 ) & ( 3.15 ) & & & + 3 `` solar '' & 100 & 8.68 & 0.98 & 486.0 & 1.0e04 & 2.1e04 & 3.6e04 + 3 `` 2xn '' & 10.1 & 8.61 & 0.99 & 80.8 & 74.0 & 293.4 & 51.9 + 3 `` dusty '' & 23.7 & 8.63 & 0.94 & 140 & 130 & 680 & 160 + & ( 16.02 ) & ( 8.64 ) & ( 0.94 ) & ( @xmath5810.0 ) & & & + 4 `` solar '' & 16.9 & 7.39 & 2.71 & 140 & 460 & 460 & 76.7 + 4 `` 2xn '' & 13.1 & 7.21 & 2.71 & 15.2 & 1.89 & 6.13 & 0.11 + 4 `` dusty '' & 23.9 & 7.25 & 2.70 & 25.0 & 2.53 & 11.7 & 0.24 + & ( 20.39 ) & ( 7.26 ) & ( 2.70 ) & ( 16.92 ) & & & + 5 `` solar '' & 3.39 & 1.50 & 0.57 & 26.4 & 31.4 & 130 & 12.5 + 5 `` 2xn '' & 2.83 & 1.47 & 0.58 & 2.87 & 0.35 & 1.09 & 0.02 + 5 `` dusty '' & 5.11 & 1.47 & 0.58 & 4.78 & 0.47 & 2.10 & 0.04 + & ( 5.60 ) & ( 1.45 ) & ( 0.58 ) & ( 3.69 ) & & & + lccccc 2 `` 2xn '' & 0.03 & 1.4 x 10@xmath46 & 2.26 & 1.17 & 0.47 + 2 `` dusty '' & 0.03 & 2.9 x 10@xmath46 & 3.87 & 1.15 & 0.44 + & & & ( @xmath580.61 ) & ( 1.11 ) & ( 0.51 ) + 3 `` 2xn '' & 0.22 & 2.5 x 10@xmath63 & 3.82 & 3.25 & 0.37 + 3 `` dusty '' & 0.24 & 7.8 x 10@xmath63 & 8.61 & 3.30 & 0.35 + & & & ( @xmath582.01 ) & ( 3.04 ) & ( 0.38 ) + 4 `` 2xn '' & 0.03 & 1.2 x 10@xmath63 & 18.32 & 10.2 & 3.83 + 4 `` dusty '' & 0.03 & 2.6 x 10@xmath63 & 33.7 & 10.2 & 3.82 + & & & ( @xmath582.97 ) & ( 10.2 ) & ( 3.44 ) + 5 `` 2xn '' & 0.03 & 1.3 x 10@xmath46 & 2.02 & 1.05 & 0.41 + 5 `` dusty '' & 0.03 & 2.8 x 10@xmath46 & 3.87 & 1.12 & 0.44 + & & & ( @xmath580.66 ) & ( 1.18 ) & ( 0.31 ) +

recently kaniel & itin proposed a gravitational model with the wave type equation\vartheta^\alpha=0 $ ] as vacuum field equation , where denotes the coframe of spacetime . they found that the viable yilmaz - rosen metric is an exact solution of the tracefree part of their field equation . this model belongs to the _ teleparallelism _ class of gravitational gauge theories . of decisive importance for the evaluation of the kaniel - itin model is the question whether the variation of the coframe commutes with the hodge star . we find a master formula for this commutator and rectify some corresponding mistakes in the literature . then we turn to a detailed discussion of the kaniel - itin model . _ file kaniel21.tex , 1998 - 01 - 12 _ keywords : gravity , gauge theory , teleparallelism , variation , hodge duals

gr-qc9801036

we were very much surprised when we learned during the 8@xmath2 marcel grossmann meeting in jerusalem @xcite that kaniel & itin @xcite were able to propose a gravitational model which looks viable at a first sight even if it had neither an einstein - hilbert type of lagrangian nor the schwarzschild metric as an exact solution . their gravitational potential is represented by a quartet of 1-forms @xmath3 or , for short , by @xmath1 , which constitutes the coframe field of spacetime . their vacuum field equation is simply the wave equation with an additional ` massive ' contribution depending on some scalar field @xmath4 : @xmath5\vartheta^\alpha=0\,.\ ] ] they show that the yilmaz - rosen metric @xcite solves the _ tracefree _ part of exactly . let us be a bit more specific : the yilmaz - rosen metric , in isotropic coordinates , is given by @xmath6 where @xmath7 . if we introduce an orthonormal coframe , @xmath8 then the following coframe , up to arbitrary local lorentz transformations , represents the yilmaz - rosen metric : @xmath9 the tracefree part of will be determined in sec . [ sec:44 ] and turns out to be @xmath10\vartheta^\alpha=0\,.\ ] ] the coframe solves the tracefree field equation exactly . we have verified this by means of our computer algebra program kaniti.exi displayed in the appendix in sec . [ sec : ca ] . kaniel & itin tried to derive the field equation from a suitable lagrangian . for that purpose they had to assume specifically that the variation @xmath11 of the coframe @xmath1 commutes with the hodge star : @xmath12 . however , such a commutativity is only valid for internal yang - mills fields . it is violated for the coframe and the metric . therefore the kaniel - itin model is based on somewhat shaky foundations . in the light of the results mentioned so far , the following questions come to mind : ( i ) what is the source on the right hand side of the field equation ? ( ii ) can the yilmaz - rosen metric also be adjusted to the trace part of and , more generally , to a possible source term on the right hand side of ? ( iii ) is there a consistent variational principle available which would allow to derive , including a source term , from a suitable lagrangian ? ( iv ) what is the ( geometrical ? ) meaning of the constrained variations of kaniel & itin ? the purpose of this article is to try to answer these questions . moreover , along our way , we will discuss some unclear points on the commutativity of variation and hodge star which led to some ( so far uncorrected ) mistakes in the literature. in sec . [ sec : model - ki ] we provide some background material on how to derive wave equations of the general type from lagrangians in maxwell s theory and in theories of other internal fields . here and in the following _ internal _ fields are those which do not depend on the spacetime geometry ( in contrast to @xmath1 and @xmath13 ) . in this way we are able to understand how the lagrangian of kaniel - itin comes up in the first place . but since they identify the _ gravitational potential _ with the _ coframe _ , we run into trouble from the point of view of finding a suitable lagrangian . any reasonable gauge approach to gravity contains in some way the gauging of the translation group . the simplest gauge theories of gravity are teleparallel theories with only the translation group as gauge group . they already require the knowledge of how to vary the hodge dual of forms . in a teleparallel theory , spacetime can be described by an _ coframe @xmath1 as the only gravitational field variable , which is interpreted as translational gauge potential , see @xcite . and this gauge potential was used by kaniel - itin in their model . accordingly , their lagrangian is a special teleparallelism lagrangian with the additional postulate of constrained variations . the commutativity of @xmath14 and @xmath15 is , in general , _ not _ fulfilled for gauge theories of _ external _ ( or spacetime ) groups , i.e. , for gravitational gauge theories . in this case it is important to know the commutator @xmath16 of the variation @xmath14 and the hodge star @xmath17 . therefore , in sec . [ sec : hodge ] we will derive the master formula for @xmath16 . we will include general variations of the components @xmath13 of the riemannian metric @xmath18 besides those of a ( not necessarily orthonormal ) coframe @xmath1 . if we _ insist _ , in accordance with the kaniel - itin postulate , on commutativity of @xmath14 and @xmath17 , then the variations @xmath19 of the components of the metric are no longer independent and can be expressed in terms of the variation @xmath11 of the coframe , see . in sec . [ sec : tele ] we give a short overview of teleparallelism theories and the relevant quadratic lagrangians . we will discuss the viable set of lagrangians and display the results in table 1 . we will show that the ki - lagrangian , for arbitrary variations , is not viable . some errors in the literature ( see schweitzer et al . @xcite ) are rectified . in sec . [ sec : ki ] , we evaluate the model of kanin & itin @xcite . the field equation of the constrained variational principle is the _ antisymmetric part _ of a wave equation for @xmath1 , in contrast to the full wave equation as claimed by kaniel and itin . the yilmaz - rosen metric , found by yilmaz ( * ? ? ? * eqs.(18 ) and ( 20 ) ) in 1958 as a solution in the context of a scalar field theory of gravitation , also turned out to be a solution of the bi - metric theory of gravitation of rosen @xcite ; cf . also @xcite . and , in the kaniel - itin model , it solves the tracefree wave equation . in sec . [ sec : yr - sol ] we compare the yilmaz - rosen with the schwarzschild metric and give , in sec . [ yr - mot ] , a motivation for the emergence of the yilmaz - rosen metric . finally , we investigate the implications that would arise if the yilmaz - rosen metric is considered to be a solution of the field equation of kaniel & itin including its trace . in sec . 6 we collect our arguments . the kinetic part of the lagrangian of a yang - mills theory is conventionally built from the first derivative of the gauge potential @xmath20 and the corresponding hodge dual . for an internal gauge group , such as for the @xmath21 or the @xmath22 , the gauge potential @xmath20 is _ independent _ of the metric @xmath18 or the coframe @xmath1 of the underlying spacetime manifold . then the variation @xmath14 of @xmath20 commutes with the hodge star operator @xmath15 . let us illustrate this for maxwell s theory , i.e. , for @xmath21-gauge theory in minkowski spacetime . the maxwell lagrangian is given by @xmath23 the variation of the 1-form @xmath20 is independent of the variations @xmath11 or @xmath24 ; furthermore , it commutes with the exterior derivative , since the variation is defined in this way . therefore , with the _ _ co__derivative @xmath25 , we find @xmath26 thus the vacuum field equation reads : @xmath27 additionally , we take the lorentz condition @xmath28 as a gauge condition . then , introducing the dalembertian @xmath29 the vacuum field equation can be rewritten as @xmath30 one could try to derive ( [ eq : wave - max ] ) directly by supplementing ( [ u1lagrangian ] ) with a suitable lagrangian . the choice @xmath31 looks suggestive . it leads to the field equation @xmath32which should be compared with . consequently the sum of the lagrangians and , enriched by a matter lagrangian , @xmath33 would yield directly the wave equation:@xmath34 however , the lagrangian ( [ eq : lmax - dagger ] ) is not gauge - invariant : for the regauging by means of the arbitrary function @xmath35 , @xmath36 one finds - forms , see @xcite . then , instead of , one has @xmath37 as gauge transformation , since @xmath38 . and the new `` lorentz condition '' is @xmath39 . ] @xmath40 accordingly , the lagrangian has to be rejected . we can obtain the wave equation only in the _ special _ gauge , _ after _ the derivation of the field equation by means of the variational principle with the lagrangian . if we used massive fields , then we would have no difficulties with lack of gauge invariance , because the mass term is not gauge invariant anyway . since we want to study gravity la kaniel - itin , we start with a quartet of 1-form fields @xmath41 , where @xmath42 is an internal index with @xmath43 . we again derive a wave type equation as in the last subsection , but we now add a massive term for each of the four fields : @xmath44 we vary with respect to @xmath41 and find as the euler - lagrange equation : @xmath45 one could also think of an additional higgs - type ( or ` cosmological ' ) term . then we would have @xmath46 and , as field equation , @xmath47 this is as near as we can approach the field equation . since currents are 3-forms , we take the hodge dual of and remember @xmath48 . furthermore we put a source term on its right hand side . in gravitational theory , this can be only the matter current @xmath49 , representing the energy - momentum flux of matter . then the completed kaniel - itin field equation reads : @xmath50{}^\star \vartheta_\alpha=\frac{\delta l_{\text{mat } } } { \delta\vartheta^\alpha } = : \sigma_\alpha\,.\ ] ] since @xmath4 is a _ function _ , it can not be identified with some constant mass @xmath51 . also an interpretation of @xmath4 as a cosmological constant is obviously meaningless . therefore the equations and have to be carefully distinguished . in future , we will refer to as the ( completed ) kaniel - itin field equation . . represents the heart of their theory . kaniel - itin proposed in ( * ? ? ? * eq.(8 ) ) the following lagrangian for the derivation of the vacuum version of : @xmath52\,.\ ] ] we used our notation here . we want to correct this lagrangian in two respects : ( i ) if we compare with , it is clear that we should _ add _ the first two terms , instead of subtracting them . ( ii ) the lagrange multiplier term is a hoax since the expression multiplying @xmath4 _ vanishes _ identically : we have quite generally @xmath53 , with @xmath54 , and the trace of this equation proves our contention . taking care of both objections , we will call @xmath55 the ( corrected ) kaniel - itin lagrangian . if we recall that we were able to derive from only because the @xmath41 was an internal field , the variation of which commutes with the star , then it becomes clear that kaniel - itin field equation _ ( [ ki - fieldeq ] ) _ , for @xmath56 , _ is not the euler - lagrange equation of _ _ ( [ ki - lagrangian])_. in the kaniel - itin model , the hodge star no longer commutes with the variation @xmath11 since their gravitational potential @xmath1 is inseparably connected to the spacetime manifold . a postulate of kaniel - itin to the opposite , see ( * ? ? ? * statement between eqs.(7 ) and ( 8) ) , is without foundation , at least from a geometrical point of view . consequently , we will give up this postulate . the kaniel - itin lagrangian belongs to the so - called teleparallelism models of gravity . we will come back to this in sec . [ sec : tele ] . the addition of the adjoint piece in does not break the translational invariance since @xmath57 , and thus @xmath58 , can be rewritten in terms of @xmath59 ( shown here for an orthonormal coframe ) : @xmath60 an analogous procedure is _ not _ available in the maxwellian case for @xmath61 , that is , @xmath61 can not be expressed in terms of @xmath62 . therefore @xmath63 is not gauge invariant , see , and has to be rejected as a decent lagrangian . in order to install the _ variation _ @xmath14 as a derivation , we demand that it fulfills an even leibniz rule,[sec : rules ] @xmath64 where @xmath65 and @xmath66 are arbitrary exterior differential forms . the leibniz rule is even , because the variation does not change the degree of the form . in contrast to this , the interior product @xmath67 ( here @xmath68 is a vector ) and the exterior derivative @xmath69 decrease or increase , respectively , the degree of the form by one and fulfill an _ odd _ leibniz rule . furthermore , we need a relation between the variation and the exterior derivative . according to the definition of the variation , they simply commute : @xmath70 = 0\;. \label{eq : var - comm}\ ] ] let us now turn to the hodge star operator , see @xcite . it maps a @xmath71form @xmath72 into an @xmath73form @xmath74 ; here @xmath75 is the dimension of the manifold , i.e. , in our case @xmath76 . in terms of components we have @xmath77 where @xmath78 is the levi - civita symbol . besides the @xmath79-basis @xmath80 , having the hodge star at our disposal , we may define the so - called @xmath81-basis : @xmath82 now we can derive the desired expression for @xmath83 , for an arbitrary @xmath71-form @xmath84 . we can saturate the @xmath73-form @xmath85 with coframes @xmath86 such as to arrive at the @xmath75-form @xmath87 we vary . then the even leibniz rule for the variation leads to @xmath88 thus , apparently , we know how to vary the hodge star , provided we know how to vary the @xmath81-basis . the variation of the @xmath73-form @xmath89 is computed in the appendix in sec . [ sec:2.2 ] . it turns out to be @xmath90 incidentally , for the special choice of an orthonormal tetrad , we have @xmath91 , and the last two terms vanish ; in particular , we then have @xmath92 . but we will not introduce this specialization at the present stage . if we resolve ( [ varphi ] ) with respect to @xmath93 , then , after some intermediary algebra , see sec . [ deduc33 ] of the appendix , we find for arbitrary @xmath71-forms @xmath84 the master formula : @xmath94 \label{eq : var - hodge } \\ & + \delta g_{\alpha\beta}\big[\vartheta^{(\alpha}\ ! \wedge\ ! ( e^{\beta ) } \rfloor \ensuremath{{}^\star } \phi ) - \frac{1}{2}\,g^{\alpha\beta}\ ; \ensuremath{{}^\star } \phi\big ] \;. \end{split}}\ ] ] again , for orthonormal ( co)frames , the terms in the second line vanish since @xmath91 . * proposition * : _ the condition @xmath95 for an arbitrary @xmath71-form @xmath84 is equivalent to the following relation between the variation of the metric and the coframe : @xmath96 therefore , for an orthonormal coframe , the allowed variations are of the lorentz type , i.e. , @xmath97 _ to prove this equivalence we first assume that is valid . we apply this constrained variation to the volume @xmath75-form @xmath98 , defined in , @xmath99 since the constant 1 is not varied . in turn , for the identity @xmath100 we find : @xmath101 the commutation rule applied to @xmath102 yields @xmath103 thus , by exterior multiplication with @xmath1 we arrive at @xmath104 on substitution into , we find @xmath105or , since @xmath106 , @xmath107 the 1-form @xmath11 can be expanded with respect to the coframe : @xmath108 we insert into . then we find @xmath109 to investigate the reverse part of the proposition , we apply the general rule for the variations of hodge dual forms and use and : @xmath110\big ] \label{eq : var - hodge - sub } \\ & \quad -2 \omega^{(\beta\alpha)}\,\vartheta_{\alpha}\ ! ! ( e_{\beta } \rfloor \ensuremath{{}^\star } \phi ) + \omega_\gamma{}^\gamma\ ; \ensuremath{{}^\star } \phi\\&= \;\omega^{\beta\alpha}\big[-\vartheta_\alpha \wedge \left(e_\beta \rfloor \ensuremath{{}^\star } \phi \right)- \ensuremath{{}^\star } \big[\vartheta_\beta \wedge \left ( e_\alpha \rfloor \phi \right)\big]+g_{\alpha\beta}\,\ensuremath{{}^\star } \phi\big ] \\&= \;\omega^{\beta\alpha}\big[e_\beta\rfloor(\vartheta_\alpha \wedge \ensuremath{{}^\star } \phi ) - \ensuremath{{}^\star } \big[\vartheta_\beta \wedge \left ( e_\alpha \rfloor \phi \right)\big]\big]\\ & \stackrel{\text{\eqref{eq : hodge - coframe1 } , \eqref{eq : hodge - coframe3}}}{= } \omega^{\beta\alpha}\,e_\beta\rfloor{}^\star(e_\alpha\rfloor\phi)\big [ \frac{1}{(-1)^{p-1}}-(-1)^{p-1}\big]= 0\;. \end{split}\ ] ] thus the proposition is proved . a minkowski space is invariant under rigid translations . in order to make a manifold locally translation invariant , one can introduce a gauge potential by means of a dynamical coframe @xmath1 , see @xcite . such a spacetime carries a torsion , but no curvature : it is a so - called _ weitzenbck spacetime_. then , by picking a suitable frame , the connection @xmath111 can always globally be transformed to zero . therefore , in a teleparallel theory , the coframe @xmath1 is the basic gravitational field variable . furthermore , let a metric @xmath18 be given of minkowskian signature , i. e. of index 3 ( number of negative eigenvalues of the metric ) . for the rest of this paper we choose the coframe to be _ orthonormal _ , @xmath112 , with @xmath113 , and raise and lower the frame indices by means of @xmath114 . according to rumpf @xcite , a general quadratic lagrangian for the coframe @xmath1 can be expanded in terms of the _ gauge - invariant _ translational lagrangians ( @xmath115 planck length , @xmath116 cosmological constant @xmath117 ) : @xmath118}v\ ; , \label{eq : decomp - rumpf}\ ] ] with [ eq : rumpf ] @xmath119}v & = \frac{1}{4 } \ , \vartheta^\alpha \wedge \ensuremath{{}^\star } \vartheta_\alpha = \eta \;,\\ \,{}^{[1]}v & = d\vartheta^\alpha \wedge \ensuremath{{}^\star } d\vartheta_\alpha \qquad \text{(pure yang - mills type)}\ ; , \\\ , { } ^{[2]}v & = \big(d\vartheta_\alpha \wedge \vartheta^\alpha \big ) \wedge \ensuremath{{}^\star } \left ( d\vartheta_\beta \wedge \vartheta^\beta \right ) \ ; , \\\ , { } ^{[3]}v & = \left(d\vartheta^\alpha \wedge \vartheta^\beta \right ) \wedge \ensuremath{{}^\star } \left ( d\vartheta_\alpha \wedge \vartheta_\beta \right ) = d\vartheta^\alpha \wedge \vartheta^\beta \wedge \left ( e_\beta \rfloor \ensuremath{{}^\star } d\vartheta_\alpha \right ) = 2 \ , { } ^{[1]}v \ ; , \\\ , { } ^{[4]}v & = \left(d\vartheta_\alpha \wedge \vartheta^\beta \right ) \wedge \ensuremath{{}^\star } \big ( d\vartheta_\beta \wedge \vartheta^\alpha \big ) \;.\end{aligned}\ ] ] since @xmath120}v = 2 \,{}^{[1]}v$ ] , we can always put @xmath121 . alternatively , the field strength @xmath59 can be decomposed into three pieces which transform irreducibly under the lorentz group : @xmath122 here we defined ( in parentheses we are mentioning the corresponding names of our computer algebra programs ) : [ eq : irred ] @xmath123 in terms of the numbers of components involved , we have the decomposition @xmath124 . then we can write @xmath125\;. \label{eq : decomp - irred}\ ] ] we substitute and into . then a comparison between and yields @xmath126 and , additionally @xmath127 . these relations were checked by means of a computer algebra program , see sec . [ sec : ca ] of the appendix . the field equation of a general translation invariant lagrangian reads @xcite @xmath128 with @xmath129 in , the partial derivatives are implicitly defined by means of the variation of the lagrangian : @xmath130 if we use the abbreviations , we do nt need our master formula for the computation of the field equation . alternatively , one can take the lagrangian together with and vary the resulting expression by using with @xmath91 . this yields the explicit form of the field equation , cf . kopczyski @xcite . we have the following translation rules : @xmath131}v$ ] , @xmath132}v$ ] , @xmath133}v$ ] . in , the second derivatives of the coframe are exactly the same ( for @xmath134 ) as those in the corresponding three equations of kopczyski ( * ? ? ? * top of p. 503 ) . ] : @xmath135 \\ & + \rho_2\;\big[2 d\vartheta_\alpha \wedge { } ^\star \left ( d\vartheta^\beta \wedge \vartheta_\beta \right ) \\ & \quad + e_\alpha \rfloor \left ( d\vartheta^\gamma \wedge \vartheta_\gamma \wedge { } ^\star \left ( d\vartheta^\beta \wedge \vartheta_\beta\right ) \right ) - 2\left ( e_\alpha \rfloor d\vartheta^\beta\right ) \wedge \vartheta_\beta \wedge { } ^\star \left ( d\vartheta^\gamma \wedge \vartheta_\gamma\right)\big ] \\ & + \rho_4\big[2 d\vartheta_\beta \wedge { } ^\star \left ( \vartheta_\alpha \wedge d \vartheta^\beta\right)\\&\quad + e_\alpha \rfloor \left ( \vartheta_\gamma \wedge d\vartheta^\beta \wedge { } ^\star \left ( d\vartheta^\gamma \wedge \vartheta_\beta \right)\right ) - 2 \left ( e_\alpha \rfloor d\vartheta^\beta\right ) \wedge \vartheta_\gamma \wedge { } ^\star \left ( d\vartheta^\gamma \wedge \vartheta_\beta \right)\big ] \ , . \end{split}\ ] ] in the first line of this equation we displayed the leading terms containing second derivatives of the coframe . in the remaining terms there enter only first derivatives . in order to better recognize the structure of the ki - lagrangian , one can decompose it in its irreducible pieces as well as into the rumpf lagrangians.[sec : ki - decomp ] the first term of the ki - lagrangian is exactly the yang - mills type lagrangian @xmath136}v$ ] , cf . with . the term @xmath137 is left for scrutiny . using formula and other rules , see , e.g. , @xcite and sec . [ formulae ] for details , we find @xmath138 \\ & = \vartheta^\beta \wedge \ensuremath{{}^\star } d\vartheta_\beta \wedge \vartheta^\alpha \wedge \left [ e_\alpha \rfloor \left ( e_\gamma \rfloor d\vartheta^\gamma \right)\right ] \\ & = \ensuremath{{}^\star } d\vartheta_\beta \wedge \vartheta^\beta \wedge \left ( e_\gamma \rfloor d\vartheta^\gamma\right ) = \ensuremath{{}^\star } \left [ \vartheta^\beta \wedge \left ( e_\gamma \rfloor d\vartheta^\gamma\right ) \right ] \wedge d\vartheta_\beta\\ & = - d\vartheta_\beta \wedge \ensuremath{{}^\star } \left [ e_\gamma \rfloor \left ( \vartheta^\beta \wedge d\vartheta^\gamma\right)\right ] + d\vartheta_\gamma \wedge \ensuremath{{}^\star } d\vartheta^\gamma \\ & = - d\vartheta_\alpha \wedge \vartheta_\beta \wedge \ensuremath{{}^\star } \left ( \vartheta^\alpha \wedge d\vartheta^\beta \right ) + d\vartheta^\alpha \wedge \ensuremath{{}^\star } d\vartheta_\alpha = { } ^{[1]}v - { } ^{[4]}v\;. \end{split}\ ] ] accordingly , the ki - lagrangian can be rewritten as @xmath139\;,\ ] ] and we can read off the @xmath140 coefficients as follows : @xmath141 by _ subtracting _ the adjoint term we would have @xmath142 and @xmath143 , i.e. , we would get the von der heyde lagrangian @xcite . the coefficients of the decomposition into irreducible pieces , by using , turn out to be @xmath144 ( in the von der heyde case , we have @xmath145 , @xmath146 , @xmath147 ) . accordingly , the ki - lagrangian can be rewritten in the form @xmath148 with the translational `` excitation '' @xmath149 and the coefficients . the lagrangian @xmath150 is _ not _ locally lorentz invariant . rather , a locally lorentz invariant theory results from the following choice of the parameters : @xmath151 this represents the teleparallel _ equivalent _ of einstein s general relativity . the form of a general quadratic lagrangian was displayed in eqs . ( [ eq : decomp - irred ] ) and ( [ eq : decomp - rumpf ] ) . various choices of parameters @xmath152 or @xmath140 correspond to various teleparallel theories of gravity . we call a specific lagrangian _ viable _ if it leads to a theory which fulfills the following conditions : ( i ) it has the correct newtonian approximation . ( ii ) it agrees with the first post - newtonian approximation of general relativity . ( iii ) it has the schwarzschild metric as exact solution in the case of spherical symmetry . the question which parameters yield a viable lagrangian has already been discussed in the literature , see for example @xcite . the result is that we have viable lagrangians for @xmath153 or @xmath154 . the arbitrary @xmath155 or @xmath156 pieces , respectively , represents the axial square contribution @xmath157 of the torsion , with @xmath158 and @xmath159 . deviations between viable theories due to different axial pieces only show up in fifth order of the post - newtonian approximation @xcite . therefore , on a phenomenological level , all viable teleparallel theories are indistinguishable . in table 1 we have listed some quadratic torsion lagrangians . it follows from the end of the last subsection , see also , that only the teleparallel equivalent of einstein s general relativity is both viable and locally lorentz invariant . it has not yet been clearly answered if in this context local lorentz invariance is obligatory or merely an aesthetic feature . ._this table lists the @xmath152 and the @xmath140 coefficients for different teleparallel lagrangians . @xmath160 , spelled out in the first column , represents a viable gravitational model , the same is true for the von der heyde case . obviously , the kaniel - itin lagrangian , in the framework of the conventional variational procedure , is not viable . we have @xmath161 and @xmath162 . _ [ cols="^,>,>,>,>,>,>",options="header " , ] schweizer and straumann @xcite and schweizer , straumann , and wipf @xcite investigated the von der heyde teleparallelism lagrangian @xcite , see also @xcite . in particular they showed that , to first post - newtonian order , the von der heyde theory predicts the same gravitational radiation loss as general relativity . however , they assumed in some places the incorrect commutation rule @xmath163 . therefore some equations in these articles must be corrected . we stress that the corrections do not influence their overall results , though . we present the corrected formulas of the article by schweizer , straumann , and wipf @xcite in the numbering used there , but in our notation . the correcting terms are printed in bold . the wrong variation first shows up in the explicit expressions of the canonical energy - momentum tensors derived from the teleparallel version of the hilbert - einstein lagrangian , @xmath164 , and the difference between the von der heyde lagrangian and the teleparallel version of the hilbert - einstein lagrangian , @xmath165 : @xmath166 - d\vartheta^\beta \wedge \ensuremath{{}^\star } \left ( d\vartheta_\alpha \wedge \vartheta_\beta \right ) \\ & \quad \notag + \frac{1}{2 } d \left\ { \vartheta_\alpha \wedge \ensuremath{{}^\star } \left ( d\vartheta^\beta\wedge \vartheta_\beta \right ) \right\ } + \frac{1}{2 } d\vartheta_\alpha \wedge \ensuremath{{}^\star } \left ( d\vartheta^\beta\wedge \vartheta_\beta \right ) \\ & \notag \quad \qquad \qquad \boldsymbol{+ \frac{1}{2 } e_\alpha \rfloor \left ( \vartheta_\gamma \wedge d\vartheta^\beta \right ) \wedge \ensuremath{{}^\star } \left ( d\vartheta^\gamma \wedge \vartheta_\beta \right ) } \\ & \notag \quad \qquad \qquad \boldsymbol{+ \frac{1}{2 } d\vartheta^\gamma \wedge \vartheta_\beta \wedge e_\alpha \rfloor \ensuremath{{}^\star } \left ( \vartheta_\gamma \wedge d\vartheta^\beta \right ) } \\ & \notag \quad \qquad \qquad \boldsymbol{- \frac{1}{4 } e_\alpha \rfloor \left ( d\vartheta^\beta \wedge \vartheta_\beta \right ) \wedge \ensuremath{{}^\star } \left ( d\vartheta^\gamma \wedge \vartheta_\gamma \right ) } \\ & \notag \quad \qquad \qquad \boldsymbol { - \frac{1}{4 } d\vartheta^\beta \wedge \vartheta_\beta \wedge e_\alpha \rfloor \ensuremath{{}^\star } \left ( d\vartheta^\gamma \wedge \vartheta_\gamma \right)}\ ; , \\ \tag*{(2.13),\cite{straumann2 } } \delta \epsilon_\alpha & = d \vartheta_\alpha \wedge \ensuremath{{}^\star } \left ( d\vartheta^\beta\wedge \vartheta_\beta \right ) - \frac{1}{2 } \vartheta_\alpha \wedge d\ , \ensuremath{{}^\star } \left ( d\vartheta^\beta\wedge \vartheta_\beta \right ) \\ & \notag \quad \qquad \qquad \boldsymbol{- \frac{1}{4 } e_\alpha \rfloor \left ( d\vartheta^\beta \wedge \vartheta_\beta \right ) \wedge \ensuremath{{}^\star } \left ( d\vartheta^\gamma \wedge \vartheta_\gamma \right ) } \\ & \notag \quad \qquad \qquad \boldsymbol{-\frac{1}{4 } d\vartheta^\beta \wedge \vartheta_\beta \wedge e_\alpha \rfloor \ensuremath{{}^\star } \left ( d\vartheta^\gamma \wedge \vartheta_\gamma \right)}\;.\end{aligned}\ ] ] since the additional terms do not influence the antisymmetric part of @xmath165 , we only need to correct the symmetric part : @xmath167 \,\eta_\beta\;. \\\intertext{hence we find the corrected field equation as } \tag*{(3.3),\cite{straumann2 } } \epsilon_\alpha^e & - \frac{(\lambda -1)}{2 } \ ; \ensuremath{{}^\star } \ , \bigg [ \left ( d\vartheta_\alpha \wedge \vartheta^\beta + d\vartheta^\beta \wedge \vartheta_\alpha \right ) \wedge \ensuremath{{}^\star } \left ( d\vartheta^\gamma \wedge \vartheta_\gamma\right ) \\ & \notag \qquad \qquad \qquad \boldsymbol { -\frac{1}{2 } \vartheta^\beta \wedge d\vartheta^\gamma \wedge \vartheta_\gamma \wedge e_\alpha \rfloor \ensuremath{{}^\star } \left ( d\vartheta^\delta \wedge \vartheta_\delta\right ) } \\ & \notag \qquad \qquad \qquad \qquad \qquad \left . \boldsymbol { + \frac{1}{2 } d\vartheta^\gamma \wedge \vartheta_\gamma \wedge \ensuremath{{}^\star } \left ( d\vartheta^\delta \wedge \vartheta_\delta\right ) \ , \delta^\beta_\alpha } \right ] \,\eta_\beta = - t^\alpha \;.\end{aligned}\ ] ] in our units , we have for the @xmath168-parameter of schweitzer et al . @xmath169 . like the terms discussed in @xcite , the additional terms are at least quadratic in @xmath170 ( which is the symmetric part of @xmath171 in the expansion @xmath172 ) , such that the arguments of @xcite remain unchanged . the formulas ( * ? ? ? * ( 4.1 ) ) and ( * ? ? ? * ( 4.6 ) ) have to be corrected similarly as the last two equations , replacing @xmath173 by @xmath174 . since these explicit expressions are not used in the remainder of @xcite , the conclusions remain valid therein . finally , the quantities @xmath175 , @xmath176 , and @xmath177 ( formulas ( * ? ? * ( 5.4b)(5.4d ) ) ) in the expansion of the quadratic and higher - order terms @xmath178 would need corrections . since these quantities are not used explicitly in @xcite , we did not display the exact expressions here . we come back to the kaniel - itin lagrangian : @xmath179 its euler - lagrange equation can be read off from by substituting the coefficients : @xmath180 + 2 \left ( e_\alpha \rfloor d\vartheta^\beta\right ) \wedge \vartheta_\gamma \wedge { } ^\star \left ( d\vartheta^\gamma \wedge \vartheta_\beta \right ) \;. \end{split}\ ] ] in the first line , we displayed the second derivatives of the gravitational potential . as we already saw in table 1 , this field equation is not viable . if we use the _ the constrained variations _ in , then we commute @xmath14 and the star @xmath17 and find by simple algebra : @xmath181 since the variations are constrained , we have to turn to our proposition and to use : @xmath182 , with @xmath183 . on substitution in , the field equation turns out to be proportional to the _ antisymmetric _ part of the wave equation , @xmath184 } = \ell^2\,\vartheta_{[\alpha } \wedge \sigma_{\beta ] } \label{eq : ki - correct}\;,\ ] ] rather than to the wave equation itself . accordingly , the constrained variations also lead to a dead end and we have to turn back our attention again to the ki - field equation . let us split the full wave equation into its different pieces . for that purpose , we most conveniently start from the decomposition of the energy - momentum current[sec:44 ] as a covector - valued 3-form with 16 independent components : @xmath185 here @xmath186 is its _ symmetric traceless _ part , the second term on the right hand side its _ antisymmetric _ and the last term its _ trace part _ , see ( * ? ? ? * eq.(5.1.15 ) ) . we come into better known territory if we decompose @xmath49 with respect to the @xmath81-basis : @xmath187 the @xmath188 s are the components of the energy - momentum tensor . we can ` saturate ' the 3-form @xmath189 by means of the 1-form @xmath1 : @xmath190 or @xmath191 the analog of is , of course , the following splitting of the energy - momentum tensor : @xmath192}+\frac{1}{4}\,t^{\gamma}{}_\gamma \;g_{\alpha\beta}\;.\ ] ] coming back to , after some algebra , we find the following decomposition : [ eq : wave ] @xmath193 } \ ; , \label{eq : ki - antisymm}\\ \vartheta^\gamma \wedge \square \eta_\gamma + 4\lambda(x)\,\eta & = \ell^2\,\vartheta^\gamma \wedge \sigma_\gamma \;. \label{eq : ki - trace}\end{aligned}\ ] ] we can combine the antisymmetric and the symmetric - tracefree part of the wave equation in order to get its tracefree part . the fine splitting simplifies to the tracefree and the trace part of the field equation ( [ eq : ki - correct ] ) : @xmath194 sometimes it may be useful to rewrite by using @xmath195 thereby finding @xmath196 \eta_\alpha = \ell^2\,\sigma\kern-1.1em\nearrow_\alpha\ ; , \label{eq : decomp - second}\]]or , alternatively : @xmath197 \eta_\alpha = \ell^2\,\sigma\kern-1.1em\nearrow_\alpha\;. \label{eq : decomp - second'}\ ] ] the yilmaz - rosen metric and the corresponding orthonormal coframe were displayed in to . in order to compare the yilmaz - rosen metric with the schwarzschild metric , we transform the former one from the isotropic coordinates used in into schwarzschild coordinates and taylor expand it : @xmath198 for the schwarzschild metric in schwarzschild coordinates we find : @xmath199 their @xmath200 components are equal up to second order . the radial components @xmath201 begin to differ slightly in the second order . therefore the yilmaz - rosen solution is consistent with the classical tests of general relativity and can , in particular , describe the post - newtonian perihelion advance correctly , see synge ( * ? ? ? * , footnote 1 ) . it requires further investigations to decide whether the two solutions can be observationally distinguished by strong gravity effects in close binary pulsar systems . a viable theory of gravitation should be consistent with the _ local _ equivalence principle . let us consider an electromagnetic wave of frequency @xmath202 in the gravitational field of a point mass @xmath203 . the frequency shift due to the propagation from a point with radial coordinate @xmath204 to one with @xmath205 reads ( @xmath206 ) : @xmath207 according to mashhoon @xcite , yilmaz effectively proposed to extend the local equivalence principle to a _ global _ one . with this idea in mind , we can tentatively integrate : @xmath208 we solve the integrals and find @xmath209\;.\ ] ] using the newtonian potential explicitly , @xmath210 , and taking the limit @xmath211 , yields @xmath212 since @xmath213 is not influenced by gravity , one can directly read off @xmath214 following the pattern of the components of the schwarzschild metric , we now define @xmath201 as the inverse of @xmath200 : @xmath215 however , the taylor approximation of this metric ( which is _ not _ the yilmaz - rosen metric of ) reads : @xmath216 the @xmath217 component differs from @xmath218 already to second order . and the deviation of @xmath219 from @xmath220 is doubled in comparison to that of the yilmaz - rosen @xmath221 . therefore , in order to approximate the experimentally well verified schwarzschild metric in an optimal way , we choose the forefactors of as metric components in _ isotropic _ coordinates , which eventually leads to the yilmaz - rosen metric : @xmath222 the yilmaz - rosen metric , keeping in mind and , fulfills the tracefree field equation [ or , alternatively , or ] with vanishing source . for a proof compare the corresponding computer algebra program in sec . [ sec : ca ] . as we saw , the yilmaz - rosen solution is consistent with the classical tests of general relativity . however , its integration constant @xmath203 can not be directly identified with the source of a spherical body . for that reason , we take recourse to the trace equation . in the vacuum case we find @xmath223 the right hand side of this equation can be easily calculated with our computer algebra program , see also @xcite . therefore the yilmaz - rosen metric solves the vacuum field equation provided the ` cosmological ' function is prescribed as follows : @xmath224 since such an ad hoc structure looks too implausible to us , we can change horses at this moment : we reject the ` cosmological ' function of kaniel - itin and put it to zero . then we can mimic this function @xmath4 by means of the energy - momentum trace @xmath225 , cf . and : @xmath226 this energy - momentum trace is plotted in fig . [ fig : matter ] . therefore , the yilmaz - rosen solution fulfills the field equations @xmath227 taking an ideal fluid for the description of matter , then , for vanishing pressure , @xmath228 , we have @xmath229 . therefore we can understand the above computation as a matter distribution which can be viewed as ( probably unphysical ) star model . the matter of such a star reaches to infinity , but it decreases exponentially . we find the maximum of the distribution at @xmath230 , compare with fig . [ fig : matter ] . most of the star mass is concentrated inside the schwarzschild radius @xmath231 . the volume integral over @xmath232 yields the total mass @xmath203 of the star : @xmath233 where @xmath234 is the mass in conventional units . as a result , we can interpret the constant @xmath203 of the yilmaz - rosen metric as the mass of a star , but this mass is distributed in a probably unphysical way . we posed four questions about the kaniel - itin model in sec . 1 . we will try to answer them in turn : * we can put the energy - momentum 3-form of matter on the right hand side of the ki - field equation , see : @xmath235\eta_\alpha=\ell^2\ , \frac{\delta l_{\text{mat}}}{\delta\vartheta^\alpha}=:\ell^2\,\sigma_\alpha\,.\ ] ] * the yilmaz - rosen metric can be accommodated to in the following sense : we decompose in vacuum into + @xmath236 + these equations are fulfilled by the yilmaz - rosen metric , provided we prescribe the ` cosmological ' function @xmath4 in the following way : @xmath237 if one used a wave equation as field equation , i.e. , if in one put @xmath238 , then one could find the yilmaz - rosen solution for the matter distribution @xmath239 . * there does nt exist a consistent variational principle for arriving at . maybe one is able to find one for the tracefree vacuum equation . * the constraints of kaniel - itin on the variations amount to getting rid of the independence of the variations of the metric , provided the variations of the coframe are prescribed . is the model of kaniel & itin viable ? well , it is presently in intensive care and it is a beautiful model anyways . [ [ acknowledgments ] ] acknowledgments : + + + + + + + + + + + + + + + + we are grateful to shmuel kaniel and yakov itin for interesting discussions and most helpful remarks . we thank bahram mashhoon , eckehard mielke , and yuri obukhov for useful comments and hints . we now collect some rules for calculations with the hodge star , where @xmath240 and @xmath84 are forms of the _ same _ degree @xmath71 ( see @xcite ) : @xmath241 the _ index _ @xmath242 of a metric is the number of minus signs if it is in diagonal form . furthermore one has the useful rules [ eq : hodge - coframe ] @xmath243 sometimes we also need the formula : @xmath244 with these rules one can determine the @xmath81-basis , cf . : [ eq : eta - def - all ] @xmath245 two helpful rules , which connect the different elements of the @xmath81-basis , read @xmath246 in case of independent variations of the metric components @xmath13 , we need the rules @xmath247 & = \det\left(g_{\mu\nu}\right ) g^{\alpha\beta } \ , \delta g_{\alpha\beta}\;. \end{aligned}\ ] ] with these definition and rules , we can compute[sec:2.2 ] a general variation of the @xmath81-basis , involving the fields @xmath1 and @xmath13 : @xmath248{\ensuremath{\eqref{eq : eta - def}}}}}{= } \quad \frac{1}{(n - p)!}\ , \eta^{\beta_1\dots\beta_p}{}_{\beta_{p+1}\dots\beta_n } \delta \left ( \vartheta^{\beta_{p+1}}\wedge \dots \wedge \vartheta^{\beta_n } \right ) + \frac{1}{(n - p)!}\\ & \;\ : \times \delta \left ( \sqrt{|\det g_{\mu\nu}|}\ , g^{\alpha_1\beta_1}\cdots g^{\alpha_p\beta_p}\ , \epsilon_{\alpha_1\cdots\alpha_p\beta_{p+1}\cdots \beta_n}\right)\ , \vartheta^{\beta_{p+1 } } \wedge \dots \wedge \vartheta^{\beta_n } \\ & \stackrel{\text{\makebox[0pt]{\ensuremath{\substack{\eqref{eq : var - leibniz } , \eqref{eq : var - metrikdet},\\ \eqref{eq : var - metrikinv } , \eqref{eq : eta - def}}}}}}{= } \qquad \frac{n - p}{(n - p)!}\ , \eta^{\beta_1\dots\beta_p}{}_{\beta_{p+1}}{}_{\beta_{p+2}\dots\beta_n } \left(\delta\vartheta^{\beta_{p+1}}\right ) \wedge \left ( \vartheta^{\beta_{p+2}}\wedge \dots \wedge \vartheta^{\beta_n } \right ) \\ & \qquad \qquad \qquad + \left ( \frac{1}{2 } g^{\kappa\lambda } \eta^{\beta_1\dots \beta_p } - \sum_{i=1}^p g^{\kappa\beta_i } \eta^{\beta_1\dots \beta_{i-1}\lambda \beta_{i+1 } \dots\beta_p } \right)\ , \delta g_{\kappa\lambda } \\ & \stackrel{\text{\makebox[0pt]{\ensuremath{\eqref{eq : eta - def}}}}}{= } \quad \delta\vartheta^{\beta_{p+1 } } \wedge \eta^{\beta_1\dots\beta_p}{}_{\beta_{p+1 } } + \left ( \frac{1}{2 } \ , g^{\kappa\lambda } \eta^{\beta_1\dots \beta_p } \right . \\ & \qquad \qquad \qquad \qquad \left.- \sum_{i=1}^p ( -1)^{p - i } g^{\kappa\beta_i } g^{\lambda\rho } e_\rho \rfloor \eta^{\beta_1\dots \beta_{i-1}\beta_{i+1 } \dots\beta_p } \right)\ , \delta g_{\kappa\lambda } \\ & \stackrel{\text{\makebox[0pt]{\ensuremath{\eqref{eq : rettung}}}}}{= } \quad \delta\vartheta^{\beta_{p+1 } } \wedge \eta^{\beta_1\dots\beta_p}{}_{\beta_{p+1 } } \\ & \qquad \qquad + \left ( \frac{1}{2 } g^{\kappa\lambda } \eta^{\beta_1\dots \beta_p } - g^{\lambda\rho } e_\rho \rfloor \left(\vartheta^\kappa \wedge \eta^{\beta_1 \dots \beta_p } \right ) \right)\ , \delta g_{\kappa\lambda}\;. \end{split}\ ] ] hence @xmath249 we start with . we abbreviate @xmath250 and get , using the variation of @xmath251 : @xmath252{\ensuremath{\eqref{eq : eta - def } , \eqref{eq : hodge - wedge}}}}}{= } \qquad \psi_p\wedge \ensuremath{{}^\star } \left(\delta \phi\right ) - \delta \psi_p \wedge \ensuremath{{}^\star } \phi + \phi\wedge \delta\vartheta^\mu \wedge \left ( e_\mu \rfloor \eta^{\beta_1\dots\beta_p}\right ) \\ & \qquad \qquad + \left(\phi \wedge \vartheta^\kappa \wedge \left ( g^{\lambda\rho } e_\rho \rfloor \eta^{\beta_1\dots\beta_p}\right ) - \frac{1}{2 } \,\psi_p \wedge g^{\kappa\lambda}\ ; \ensuremath{{}^\star } \phi \right ) \ , \delta g_{\kappa\lambda } \\ & = \psi_p\wedge \left ( \ensuremath{{}^\star } \left(\delta \phi\right ) - \frac{1}{2}\ , \ensuremath{{}^\star } \phi \ ; g^{\kappa\lambda}\ , \delta g_{\kappa\lambda } \right ) - \delta\vartheta^\mu \wedge \left ( e_\mu \rfloor \psi_p\right ) \wedge \ensuremath{{}^\star } \phi \\ & \quad + ( -1)^p \left ( \ , \delta\vartheta^\mu \wedge \phi \wedge \left(e_\mu \rfloor \ensuremath{{}^\star } \psi_p \right ) + \ , g^{\lambda\rho } e_\rho \rfloor \left ( \phi \wedge \vartheta^\kappa \right ) \wedge \ensuremath{{}^\star } \psi_p \ ; \delta g_{\kappa\lambda}\right ) \\ & = \psi_p\wedge \left ( \ensuremath{{}^\star } \left(\delta \phi\right ) - \frac{1}{2}\ , \ensuremath{{}^\star } \phi \ ; g^{\kappa\lambda}\ , \delta g_{\kappa\lambda } \right ) + \delta\vartheta^\mu \wedge \big ( e_\mu \rfloor \left ( \phi \wedge \ensuremath{{}^\star } \psi_p \right ) \\ & \qquad \quad - \left ( e_\mu \rfloor \phi \right ) \wedge \ensuremath{{}^\star } \psi_p - e_\mu \rfloor \left ( \psi_p \wedge \ensuremath{{}^\star } \phi \right ) + ( -1)^p \psi_p \wedge \left ( e_\mu \rfloor \ensuremath{{}^\star}\phi\right)\big ) \\ & \qquad \qquad \qquad \qquad \qquad \qquad + \psi_p \wedge \vartheta^\lambda \wedge \ensuremath{{}^\star } \left ( \phi \wedge \vartheta^\kappa \right ) \ , \delta g_{\kappa\lambda } \\ & \stackrel{\text{\makebox[0pt]{\ensuremath{\eqref{eq : hodge - wedge}}}}}{= } \quad \psi_p\wedge \ensuremath{{}^\star } \left(\delta \phi\right ) - \psi_p \wedge \ensuremath{{}^\star } \left ( \delta \vartheta^\mu \wedge \left ( e_\mu \rfloor \phi \right ) \right ) + \psi_p \wedge \delta \vartheta^\mu \wedge \left ( e_\mu \rfloor \ensuremath{{}^\star } \phi\right)\\ & \qquad \qquad - \psi_p \wedge \frac{1}{2}\ , \ensuremath{{}^\star } \phi \ ; g^{\kappa\lambda}\ , \delta g_{\kappa\lambda } + \psi_p\wedge \vartheta^\lambda \wedge g^{\kappa\rho } e_\rho \rfloor \left ( \ensuremath{{}^\star } \phi \right ) \ , \delta g_{\kappa\lambda } \;. \end{split}\ ] ] since @xmath250 is constructed with @xmath71 arbitrary @xmath1 s , we conclude for an arbitrary @xmath71-form @xmath84 : @xmath253 \label{eq : var - hodge ' } \\ & + \delta g_{\alpha\beta}\big[\vartheta^{(\alpha}\ ! \wedge\ ! ( e^{\beta ) } \rfloor \ensuremath{{}^\star } \phi ) - \frac{1}{2}\,g^{\alpha\beta}\ ; \ensuremath{{}^\star } \phi\big ] \;. \end{split}\ ] ] for the special choice of an _ orthonormal _ coframe , we have @xmath254 . in this case the two last summands vanish : @xmath255\;. \label{eq : var - hodge - ortho}\ ] ] the following ` reduce ` program was written with the help of the ` excalc ` package , see @xcite . it verifies ( i ) the decompositon ( [ eq : base - rumpf ] ) of the ki - lagrangian , ( ii ) that the yilmaz - rosen metric fulfills the tracefree ki - vacuum field equation ( 74a ) and ( 77 ) , and ( iii ) the validity of eq . ( 89 ) . v0rumpf4 : = o(a ) ^ # o(-a)/4 $ v1rumpf4 : = d o(a ) ^ # d o(-a)$ v2rumpf4 : = ( d o(-a ) ^ o(a ) ) ^ # ( d o(-b ) ^ o(b ) ) $ v3rumpf4 : = ( d o(a ) ^ o(b ) ) ^ # ( d o(-a ) ^ o(-b))$ v4rumpf4 : = ( d o(-a ) ^ o(b ) ) ^ # ( d o(-b ) ^ o(a ) ) $ f. gronwald and f.w . hehl : `` on the gauge aspects of gravity '' . in : _ international school of cosmology and gravitation : _ 14@xmath258 course : quantum gravity , held may 1995 in erice , italy . proceedings . bergmann et al.(eds . ) ( world scientific , singapore 1996 ) pp . 148198 . los alamos eprint archive gr - qc/9602013 . hehl , j.d . mccrea , e.w . mielke , and y. neeman : `` metric - affine gauge theory of gravity : field equations , noether identities , world spinors , and breaking of dilation invariance '' , _ physics reports _ * 258 * ( 1995 ) 1171 . e. schrfer , f.w . hehl , and j.d . mccrea : `` exterior calculus on the computer : the reduce package excalc applied to general relativity and the poincar gauge theory '' , _ gen . j. _ * 19 * ( 1987 ) 197218 .

current adaptive mesh refinement ( amr ) simulations require algorithms that are highly parallelized and manage memory efficiently . as compute engines grow larger , amr simulations will require algorithms that achieve new levels of efficient parallelization and memory management . we have attempted to employ new techniques to achieve both of these goals . patch or grid based amr often employs ghost cells to decouple the hyperbolic advances of each grid on a given refinement level . this decoupling allows each grid to be advanced independently . in astrobear we utilize this independence by threading the grid advances on each level with preference going to the finer level grids . this allows for global load balancing instead of level by level load balancing and allows for greater parallelization across both physical space and amr level . threading of level advances can also improve performance by interleaving communication with computation , especially in deep simulations with many levels of refinement . while we see improvements of up to on deep simulations run on a few cores , the speedup is typically more modest ( ) for larger scale simulations . to improve memory management we have employed a distributed tree algorithm that requires processors to only store and communicate local sections of the amr tree structure with neighboring processors . using this distributed approach we are able to get reasonable scaling efficiency ( ) out to 12288 cores and up to 8 levels of amr - independent of the use of threading .

1112.1710

the development of amr @xcite was meant to provide high resolution simulations for much lower computational cost than fixed grid methods would allow . the use of highly parallel systems and the algorithms that go with them were also meant to allow higher resolution simulations to be run faster ( relative to wall clock time ) . the parallelization of amr algorithms , which should combine the cost / time savings of both methods is not straight forward however and there have been many different approaches @xcite . while parallelization of a uniform mesh demands little communication between processors , amr methods can demand considerable communication to maintain data consistency across the unstructured mesh as well as shuffling new grids from one processor to another to balance workload . in this paper we report the development and implementation of new algorithms for the efficient parallelization of amr designed to scale to very large simulations . the new algorithms are part of the astrobear package for simulation of astrophysical fluid multi - physics problems @xcite . the new algorithmic structure described in this paper constitutes the development of version 2.0 of the astrobear code . amr methods come in many varieties . meshes can either be unstructured or semi - structured . semi - structured methods can be further divided into those which allow grids to be of arbitrary size ( patch based ) and those which require grids to be of a fixed size ( block based or cell - based if the block size is 1 ) . with block ( or cell ) based amr , the additional constraints imposed on the structure of the mesh allow for a simpler type of connectivity within a tree . for example in block based amr , any given block will have exactly 8 children or none ( if it is a leaf ) and will have at most 6 face sharing neighbors . with patch based amr , there is no limit to the number of children or neighbors . in addition , the operation of regridding in block based amr is much simpler . as the grid changes a given block will either persist if the physical region continues to require refinement or be destroyed . in patch based amr , a given region may subsequently be better covered by patches of a different shape requiring transfer of data between physically overlapping previous patches and new patches . this adds an additional dimension to the tree structure and increases the complexity of maintaining a distributed tree . for both block ( or cell ) and patch based amr , the actual grid data ( fluid variables etc . ) are always distributed across the various processors . usually some overlap in grid data ( guard / ghost cells ) is desired to allow for frequent access to neighboring values without the need for additional communication . but the metadata that describes the shape and distribution of the grid data is usually stored on every processor . for 100 s or 1000 s of cores , this global tree typically requires less memory than that required for the local grid data and it allows for easy access to data from any part of the domain . for instance after regridding , the entire tree can be updated and stored locally and then finding new neighbors for local cells / blocks / patches can be done without the need for any further communication . for patch - based amr it also allows for better load balancing as each processor can determine from the entire tree which section of data it should be responsible for and can use various knap - sack type algorithms to optimize the degree of interprocessor communication . cell based amr engines typically use an octree data structure to handle the tree metadata and implementations have been developed that support a memory - distributed tree @xcite , or various ways of compressing the global tree @xcite that rely on the simple structure of the tree . for patch based amr , algorithms for implementing a distributed tree have not yet been published though implemented in the package samrai @xcite as well as astrobear . in addition the chombo library @xcite recently has developed a method of compressing the metadata to avoid having to distribute the tree . here we document the distributed tree algorithm used in astrobear 2.0 in which no processor has access to the entire tree but rather each processor is only aware of the amr structure it needs to manage in order to carry out its computations and perform the necessary communications . while currently , this additional memory is small compared to the resources typically available to a cpu , future clusters will likely have much less memory per processor similar to what is already seen in gpu s . additionally each processor only sends and receives the portions of the tree necessary to carry out its communication as opposed to a pruning approach in which every processor receives every new patch , but only keeps those necessary to maintain the local tree . astrobear 2.0 also uses extended ghost cells to decouple advances on various levels of refinement . as we show below this allows for each level s advance to be computed independently on separate threads . such inter - level threading allows for total load balancing across all refinement levels instead of balancing each level independently . independent load balancing becomes especially important for deep simulations ( simulations with low filling fractions but many levels of amr ) as opposed to shallow simulations ( high filling fractions and only a few levels of amr ) . processors with coarse grids can advance their grids simultaneously while processors with finer grids advance theirs . without such a capability , each level would need to have enough cells to be able to be distributed across all of the processors . variations in the filling fractions from level to level can make the number of cells on each level very different . if there are enough cells on the level with the fewest to be adequately distributed , there will likely be far too many cells on the highest level to allow the computation to be completed in a reasonable wall clock time . this often restricts the number of levels of amr that can be practically used . with inter - level threading this restriction is lifted . inter - level threading also allows processors to remain busy while waiting for messages from other processors . in what follows we provide descriptions of the new code and its structure as well as providing tests which demonstrate its effective scaling . in section [ amr_alg ] we review patch based amr . in section [ distributedtree ] we describe in detailt the distributed tree algorithm for patch - based amr , in section [ threading ] we will discuss the inter - level threading of the advance , in section [ loadbalancing ] we will discuss the load balancing algorithm , and in section [ results ] we will present our scaling results and we will conclude in section [ conclusion ] . here we give a brief overview of patch based amr introducing our terminology along the way . the fundamental unit of the amr algorithm is a patch or grid . each grid contains a regular array of cells in which the fluid variables are stored . grids with a common resolution or cell width @xmath3 belong to the same level @xmath4 and on all but the coarsest level are always nested within a coarser parent " grid of level @xmath5 and resolution @xmath6 where @xmath7 is the refinement ratio . the collection of grids comprises the amr mesh , an example of which is shown in figure [ meshtree ] . in addition to the computations required to advance the fluid variables , each grid needs to exchange data with its parent grid ( on level @xmath5 ) as well as any child grids ( on level @xmath8 ) . grids also need to exchange data with physically adjacent neighboring grids ( on level @xmath4 ) . in order to exchange data , the physical connections between parents , children , and neighboring grids are stored in the amr tree as connections between nodes . each grid has a corresponding node in the amr tree . thus there is a one to one correspondence between nodes and grids . the grids hold the actual fluid dynamical data while the nodes hold the information about each grid s position and its connections to parents , children and neighbors . figure [ meshtree ] shows one example of an amr mesh made of grids and the corresponding amr tree made of nodes . note that what matters in terms of connections between nodes is the physical proximity of their respective grids . while siblings share a common parent , they will not necessarily be neighbors , and neighbors are not always siblings but may be 1st cousins , 2nd cousins , etc ... [ meshtree ] additionally since the mesh is adaptive there will be successive iterations of grids on each level as the simulation progresses . thus the fluid variables need to be transferred from the previous iteration of grids to the current iteration as shown in figure [ overlaps ] . thus nodes can have `` neighbors '' within a 4 dimensional space - time . nodes that are temporally adjacent ( belonging to either the previous or next iteration ) and spatially coincident are classified as preceding or succeeding overlaps respectively instead of temporal neighbors , reserving the term neighbor to refer to nodes of the same iteration that are spatially adjacent and temporally coincident . nodes on level @xmath4 therefore have a parent connection to a node on level @xmath5 , child connections to nodes on level @xmath8 , neighbor connections to nodes on level @xmath4 of the same iteration , and overlap connections to nodes on level @xmath4 of the previous or successive iteration in time . [ overlaps ] we use the term iteration " to refer to successive generations of the grid distribution for each level . thus at some point in the simulation a distribution of level @xmath4 grids will cover some fraction of the computational space . when , as the simulation proceeds , the grid generation subroutine is called again the next iteration of level @xmath4 grids will be laid down over the computational domain . while in principle level @xmath4 could be regridded each level @xmath4 time step , it is convenient for restricting data to parent grids , to wait until after @xmath7 level @xmath4 steps , or equivalently after @xmath9 parent level @xmath10 step so that child and parent grids have advanced to the same time . therefore for each level @xmath11 time step , there is @xmath9 iteration of level @xmath9 grids , @xmath7 iterations of level @xmath12 grids , @xmath13 iterations of level @xmath14 grids , and so on . since new child grids are created after each parent step , each parent will have multiple iterations of children . this additional branching of the amr tree in time allows for temporally adjacent ( overlap ) grids to be classified as temporal siblings or 1st , 2nd , 3rd temporal cousins etc ... additionally parents will have connections to multiple iterations of children ( although only the two most current need to be kept ) . note that there are different ways of indexing the iteration of grids on a given level . one can chose a global indexing that begins with the initiation of the simulation , i.e the iterations can be indexed by counting the number of successive iterations of grids on that level from the beginning of the simulation . another method is to index iterations relative only to the level above ( ie which iteration of children from a given parent does a grid correspond to . ) we will use the latter indexing scheme , since it is relevant to the way overlap connections between grids of different iterations are formed . figure [ overlapfigure ] shows this indexing system ( where we have assumed @xmath15 for simplicity ) . here the circles represent iterations of the entire level ( i.e. the entire collection of nodes for a given level at the specified iteration ) . since figure [ overlapfigure ] shows the evolution of the amr tree _ in time _ , it s useful to make a connection with the the way the tree appears at any given moment _ in space _ in terms of spatial connections between nodes ( figure [ meshtree ] ) . to understand the relation between figure [ meshtree ] and figure [ overlapfigure ] imagine taking the tree in figure [ meshtree ] and rotating it into the plane . neighbor relationships would now disappear and all of the nodes of a given level would visually merge . each visible circle in figure [ overlapfigure ] represents a collection of nodes of the same level and iteration as shown in figure [ meshtree ] . each level @xmath16 has multiple iterations that stretch both forward and backward in time . note that the level 0 grids are static and there are no successive iterations . level 1 iterations go from @xmath9 to @xmath17 since the level 0 grids continue to create successive iterations of children . level 2 iterations and higher are indexed from @xmath9 to @xmath15 . it is important to stress here , that each level s iteration in figure [ overlapfigure ] represents a collection of nodes on that level . additionally each level s iteration represents multiple time steps on that level . the nodes of the level 0 iteration actually take many time steps , while all other level iterations take @xmath15 time steps . preceding overlap connections ( those going backward in time ) are only needed before the 1st of @xmath7 time steps , and the succeeding overlap connections ( those going forward in time ) are only needed after the last of @xmath7 time steps . after the 1st time step and before the last , ghost overlap data is shared between neighboring grids of the same iteration instead of preceding / succeeding overlaps . while there will be many iterations of grids on each level , only the two most current iterations are stored in the tree . after a grid finishes its advance it becomes old " . when this occurs the previous generation of old grid iterations are discarded . the current and old grids are shown schematically in figure [ overlapfigure ] . the current level @xmath4 iteration will always contain children of the current level @xmath5 iteration . for the old level @xmath4 iteration there are two cases . first the old level @xmath4 iteration may be the old iteration of children of the current @xmath5 iteration as is the case for the old level 3 grids in figure [ overlapfigure ] . secondly , the old level @xmath4 iteration may be the last iteration of children of the old level @xmath5 iteration as is the case for the old level 2 grids in figure [ overlapfigure ] . that is , they are either old children of the current parent level grids , or last children of the old parent level grids . [ overlapfigure ] the execution of the various stages in the amr algorithm is shown in the upper panel of figure [ threadingfig ] ( for @xmath15 ) . there are 5 steps in the algorithm : overlaps ( o ) ; prolongation ( p ) ; advances ( a ) ; synchronization ( s ) and restriction ( r ) . a pseudo - code description of their order of operations with @xmath18 being the highest level of refinement is written below . note each root step is initialized by calling amr(@xmath11 ) . recursive subroutine amr(@xmath4 ) * * o**verlap data from preceding iterations of grids on level @xmath4 do @xmath19=1 , @xmath7 * * a**dvance grids on level @xmath4 if ( @xmath20 ) then * * p**rolongate initial ( pre - advanced ) data from level @xmath4 to level @xmath8 call amr(@xmath8 ) * * r**estrict data from level @xmath8 end if * * s**ynchronize fluxes between neighboring level @xmath4 grids and update ghost cells end do end recursive subroutine the subroutine above describes a process in which grids are created , populated with data , advanced and brought in line with higher and lower refinement representations of the data . to be more specific , after each iteration of level @xmath4 grids are created their cells ( including ghost regions ) are initialized with a combination of prolongated data from the parent grid as well as data from the preceding set of level @xmath4 grids that physically overlap with the computational space that the new grids now describe . ghost regions are needed to update the fluid variables within the grid . the new level @xmath4 grids then determine which cells to refine and lay down a first set of level @xmath8 child grids to cover those cells . the parent level @xmath4 grids take one step of @xmath21 . meanwhile their child grids advance @xmath7 steps of @xmath22 . next the level @xmath4 grids merge restricted data from their children with their own updated data since both levels have now reached the same time @xmath23 . these level @xmath4 grids then synchronize fluxes with any adjacent neighboring grids ( also of level @xmath4 ) and update any ghost cells . the level @xmath4 grids are now ready to repeat the process ( creating children ; advance a time step ; restricted child data ; synchronize fluxes with neighbors ) . when the level @xmath4 grids have completed @xmath7 steps relative to the level above ( @xmath5 ) , their own data and fluxes are restricted and applied to their parent ( @xmath5 ) grids . after completing its advances , the data from each level @xmath4 grid is stored until it can be copied onto the next iteration of level @xmath4 grids . after the data is copied onto _ succeeding _ overlaps , the old level @xmath4 grids are destroyed . throughout a level @xmath4 grid s lifetime it must therefore share data with its parent grid @xmath10 , its preceding overlaps ( @xmath4 ) , multiple iterations of it own child grids ( @xmath24 , its neighbor grids @xmath25 , and succeeding overlaps @xmath25 . these connections for a given grid are held by its node and the web of connections between grids / nodes form the amr tree . while a grid / node may have many successive iterations of children , only connections that belong to the two most current iterations of the child level are kept . refined grids need to take multiple ( @xmath7 ) time steps . for the first step , prolongated ghost cells from the coarser grid can be used to advance the internal cells . however further time steps can not be taken without updating the ghost cells in time as well . if a grid is surrounded by neighboring grids , it can update its ghost cells in time by copying the advanced data from its neighbors internal cells . however , if a grid is isolated ( ie . has no neighbors ) or even if it is partially isolated , it must advance those ghost cells in another way . there are two solutions to this problem shown in figure [ extendedghostcells ] . one is to use the time derivative for the fluid variables calculated on the coarse grid to update the ghost cells where needed . this solution requires parent grids to advance before their children and can lead to large errors if shocks are not completely resolved since spatial discontinuities lead to delta functions for time derivatives which , in turn , produce large errors upon discretization . the alternative method is to use extended ghost cells that allow grids to successively update smaller and smaller regions so that there is always enough ghost cells to complete the final step . in general if @xmath26 is the number of ghost cells needed to take a time step and @xmath7 is the refinement ratio , then the extended ghost region needs to initially be @xmath27 cells wide . while only isolated grids need to update extended ghost cells , it can be difficult to implement efficient advance schemes for partially isolated grids , and each grid is often assumed to be isolated . while this can result in a large overhead for small grids , ( especially when the refinement ratio @xmath7 , or @xmath26 is large ) the ability to allow coarse grids to be advanced independently of fine grids can be exploited to increase the speed of the code . we have taken this approach in astrobear to allow for more efficient distribution of grids as well as the creation of multiple advance threads described in section [ threading ] . [ extendedghostcells ] many current amr codes store the entire amr tree on each processor . this , however , can become a problem for simulations run on many processors . to demonstrate this let us first assume that each amr grid requires @xmath28 bytes per node to store its meta data ( ie 6 bytes to store its physical bounds for a 3d simulation and 1 byte to store the processor containing the grid ) . we also assume that each grid requires on average @xmath29 bytes for the actual data . if there are , on average , @xmath30 grids on each of @xmath31 processors , then the memory per processor would be @xmath32 . the second term @xmath33 represents the meta - data for the _ entire _ amr tree . the memory requirement for just the nodes in the amr tree without storing any connections becomes comparable to the local actual data when @xmath34 . if we assume a 3d isothermal hydro run where each cell contains @xmath35 with a typical average grid size of 8x8x8 then @xmath36 . while this additional memory requirement is negligible for problems run using typical cpu s on 100 s of processors , it can become considerable when the number of processors @xmath37 . since it is expected that both efficient memory use and management will be required for ever larger hpc ( high performance computing ) clusters down the road , astrobear 2.0 is designed to use a _ distributed _ tree algorithm . in this scheme each processor is only aware of the section of the tree containing nodes that connect to its own grids nodes . additionally , new nodes are communicated to other processors on a proscriptive `` might need to know basis '' . since maintaining these local trees as the mesh adapts is not trivial , we describe the process below . because of the nested nature of grids , neighbor and overlap relationships between nodes can always be inherited from parent relationships . first consider the neighbors of the @xmath38 iteration of a node s children . the nested nature of the grids restricts each of the child s neighbors to either be a sibling of that child ( having the same parent node and be of the same iteration ) , or to be a member of a neighbor s @xmath38 iteration of children . thus the neighbors of a level @xmath4 node s children ( on level @xmath8 ) will always be a child of that level @xmath4 node s neighbors . for overlaps ( temporal neighbors ) the situation is a bit more complicated . if we consider figure [ overlapfigure ] we can identify two different types of overlap connections . first there are overlap connections between iterations with a common parent which we will refer to as temporal siblings. for example the current and old iterations shown on level 3 are different iterations of children of the same level 2 iteration . there are also overlap connections between children that come from different iterations of the parent grids . these can be thought of as temporal cousins. for example the current and old child iterations shown on level 2 come from different iterations of the parent nodes one level above them . in principle temporal cousins can be 1st cousins , 2nd cousins , and so on , but the distinction does not matter for the way connectivity is treated in the algorithm . overlaps between temporal siblings on level @xmath4 will be between successive child iterations of the current level @xmath5 iteration . note that grids in the current @xmath5 iteration do not physically overlap . since grids are always nested , the overlaps between the grid s current ( @xmath38 ) iteration of children must belong to the grid s old ( @xmath39 ) iteration of children ( for completeness these are what are referred to as preceding overlaps ) . however , while grids do not physically overlap , their ghost cells may . thus overlapping data for a level @xmath4 grid s children may also come from a neighbor level @xmath4 grid s old children . thus the ( preceding ) overlaps for a node s children may be an old child of the node s neighbors . likewise , every node s old child s succeeding overlap may be a node s neighbor s current child . for temporal cousins , the current iteration of level @xmath5 grids / nodes will be the first iteration of children of the current level @xmath4 grids / nodes . and the old iteration of level @xmath5 grids will be the last iteration of children of the old level @xmath4 grids . because of the nested nature of the grids , we can say that for the first iteration of children , every node s child s preceding overlap must be a node s preceding overlap s child . the same follows for the succeeding overlaps of a node s last iteration of children . that is , every node s child s succeeding overlap must be a node s succeeding overlap s child . these relationships are summarized for @xmath15 in table [ simpleinheritance ] and a generalized table is given in table [ geninheritance ] . neighbors & node s child s preceding overlaps & node s child s succeeding overlaps + 1@xmath40 & node s neighbors 1@xmath40 children & node s preceding overlaps 2nd children & node s neighbor s 2@xmath41 children + 2@xmath41 & node s neighbors 2@xmath41 children & node s neighbors 1@xmath40 children & node s succeeding overlaps 1@xmath40 children + for parallel applications , the grids are distributed across the processors . in addition to data for the local grids , each processor needs to know where to send and receive data for the parents , neighbors , overlaps , and children of those local grids . this is information contained within the nodes . in order for a processor to know where to send data , each one must maintain a local sub - tree containing its own local " nodes ( corresponding to local grids ) as well as all remote nodes ( living on other processors ) directly connected to the local nodes . it is also possible , though not desirable , that an individual processor have data from disjoint regions of the simulation . in that case each processor would have multiple disjoint sections of the amr tree , but these disjoint sections would collectively be considered the processor s sub - tree . each time new grids on level @xmath8 are created ( by local parent grids on level @xmath4 ) , each processor determines how the new child grids should be distributed ( i.e. which processor should get the new grids ) . this distribution is carried out in the manner described in section [ loadbalancing ] below . connections between the new level @xmath8 nodes and the rest of the tree must then be formed . because of the inheritability of the neighbor / overlap connections , even if a child grid is distributed to another processor , the connections between that child node and its neighbors / overlaps / parent are first made on the processor that created the grid ( ie the processor containing the grid s parent ) . if a processor s local grid has a remote parent , then that processor will always receive information about that local grid s neighbors / overlaps from the processor containing the remote parent . this is true of neighbor and preceding overlap connections of new grids as well as succeeding overlap connections of old grids . before processors containing parents of level @xmath8 grids ( both new and old ) can send connection information to remote children ( if they exist ) , these processors must first share information about the creation of children with each other . neighbor connections between new nodes on level @xmath8 require each processor to cycle through its local level @xmath4 nodes and identify those with remote neighbors living on other processors . once remote neighbors have been identified the information about new children from the local nodes is sent to the processor(s ) containing the remote neighbors . not all children need to be sent to every remote neighbor . only those that are close enough to potentially be adjacent to the remote neighbor s children are necessary . the information must flow in both directions meaning a individual processor also needs to _ receive _ information about potential new children from all other processors containing remote neighbors . overlaps are again a bit more complicated because of the different situations for temporal siblings and temporal cousins as described above . for temporal cousins ( ie if the new level @xmath8 child grids are the 1st iteration from level @xmath4 parents ) , each processor must cycle through local current level @xmath4 nodes with preceding overlaps that live on remote processors . the processor must then send new children of local nodes to the processor(s ) containing the remote preceding overlaps . similarly , each processor must cycle through the local * old * level @xmath4 nodes with remote succeeding overlaps sending children of those old local nodes to the processor(s ) containing remote succeeding overlaps . again , not all children need to be sent to the processors containing remote overlaps . only information about those nodes close enough to potentially overlap the remote node s children must be sent . the information must flow in both directions and processors also need to receive information about potential children from those same processors with remote overlaps . for overlaps between temporal siblings , no communication is required . this is because previous level @xmath8 nodes that could overlap a new level @xmath8 child node would either be old children of the same parent or be old children of the parent s neighbors . the parent would already be aware of its old children , and the old children of the parent s neighbors would have been previously communicated when establishing the neighbor relationships of the parent s old children . this process is summarized in table [ disttreealg ] . many if not all current amr codes tend to perform grid updates across all levels in a prescribed order that traverses the levels of the amr hierarchy in a sequential manner . thus the code begins at the base grid ( level 0 ) , moves down to the highest refinement level and then cycles up and down across levels based on time step and synchronization requirements(for a simulation with 3 levels the sequence would be : 0 , 1 , 2 , 2 , 1 , 2 , 2 , 0 ... ) in the top panel of figure [ threadingfig ] the basic operations of ( p)rolongating , ( o)verlapping , ( a)dvancing , ( s)ynchronizing , and ( r)estricting are shown for each level along with the single ( serial ) control thread . good parallel performance requires each level update to be independently balanced across all processors ( or at least levels with a significant fraction of the workload ) . load balancing each level , however , requires the levels to contain enough grids to be effectively distributed among the processors . such a requirement demands each level to be fairly large " in the sense of having many grids or allowing each level s spatial coverage be artificially fragmented into small pieces . the former situation leads to broad simulations ( large base grid leaving resources for only a few levels of amr ) , while the later situation leads to inefficient simulations due to the fair amount of overhead required for ghost cell calculations . the problem becomes worse when @xmath42 is large or when using extended ghost cells with refinement ratios @xmath43 . consider an isolated level @xmath4 grid of size @xmath44 inside of a parent grid on level @xmath4 . let s assume that the coarsening ratio @xmath15 and that the grid must therefore take two steps each of which requires @xmath45 ghost cells . on the first step it must update a region that is @xmath46 and then on the second step a region that is @xmath44 for a total of @xmath47 cell updates . if that @xmath44 grid were split into 4 @xmath48 pieces to be distributed on 4 processors , then instead of 160 cell updates there would now be @xmath49 per processor or 416 total cell updates and the parallelization efficiency for that grid would be @xmath50 at best . even if one uses prolongated time derivatives instead of extended ghost cells , there is still a fair amount of overhead for ghost cell computations for small grids and the efficiencies would still be @xmath51 depending on the particular method used . in the bottom panel of figure [ threadingfig ] we show a schematic of the astrobear 2.0 amr algorithm . in this figure basic operations of ( p)rolongating , ( o)verlapping , ( a)dvancing , ( s)ynchronizing , and ( r)estricting are shown again however this time the level advances are independent and exist on separate threads of computation . there is an overarching control thread which handles all of the communications and computations required for prolongation , overlapping , synchronizing , and restricting as well as the finest level advances . each coarser level advance has its own thread and can be carried forward independently with preference being given to the threads that must finish first ( which is always the finer level threads ) . in addition to relaxing the requirement of balancing every level , the existence of multiple threads allows processors to remain busy when the control thread becomes held up because it needs information from another processor . for example , while waiting for ghost cell data for level 3 it can work on advancing levels 2 , 1 , or 0 . the simplest implementation of threading would allow for the operating system kernel to manage the various threads with priority given to the control thread followed by the highest level advance and so on all on the same processor . this would be considered preemptive prioritized threading within a process . unfortunately the pthreads library implementation under linux does not allow for non - privileged users to give priorities to threads nor to limit a set of threads to use only a single processor . gnu portable threads do require the threads to remain on a single processor , but do not allow for preemptive thread scheduling . threads must schedule themselves cooperatively by yielding control to the scheduler or to each other . while gnu portable threads do allow for priorities to be assigned to threads , it is not terribly useful as threads can cooperatively yield control to the appropriate thread without the need for a scheduler . a more complicated implementation , but one that does not require external libraries , is to manually switch ( at the application level ) between grid advances and the control thread . this is somewhat difficult to implement as the grid advances must be interruptable and any intermediate variables within the advance stack have to be manually cached by the application . this caching effectively limits the frequency with which the advance thread can be interrupted . in astrobear 2.0 we have implemented gnu portable threads , manual thread switching ( scheduling ) , as well as the serial version of the amr algorithm . performance results are given in section [ results ] . [ threadingfig ] as was discussed above , threading level advances removes the need for balancing each level independently and instead allows for global load balancing . it also and perhaps more importantly allows for consideration of the progress of coarser advance threads when successively distributing the workload of finer grids . this `` dynamic load balancing '' allows adjustments to be made to finer level distributions to compensate for variations in progress made on ongoing coarser advances . @xmath52 & current workload assigned to processor @xmath31 on level @xmath4 + @xmath53 & amount of current workload @xmath52 completed + @xmath54 & newly created child workload on level @xmath4 by parent grids on level @xmath5 on processor @xmath31 + @xmath55 & average of @xmath52 across processors + @xmath56 & average of @xmath53 across processors + @xmath57 & sum of @xmath52 across processors . also equals sum of @xmath54 across processors . + @xmath58 & desired workload assignments for processor @xmath31 and level @xmath4 + @xmath59 & imbalance in @xmath52 ( @xmath60 ) + @xmath61 & imbalance in @xmath53 ( @xmath62 ) + @xmath63 & predicted remaining workload for entire amr step for processor @xmath31 on level @xmath4 + @xmath64 & current number of remaining steps on level @xmath4 within entire amr step + @xmath65 & workload assigned to processor @xmath66 of children created by grids on processor @xmath31 + here we give a brief example introducing terminology along the way . also see table [ variablelist ] for a complete list of variables used below . consider a simulation run on three processors with two levels of refinement with a refinement ratio @xmath15 and let us assume ( arbitrarily ) that there are @xmath67 cells on level 0 distributed among the three processors as @xmath68 $ ] . each processor may have more then one grid , but here we just count the total number of cells in all of the processor s grids . after spawning the level 0 advance thread , each processor continues along the control thread until it eventually creates new child grids on level 1 . at this point these new child grids must be distributed . let s assume that at this point no work has yet been accomplished on the level 0 advance thread ( @xmath69 $ ] ) . we also assume that the number of new child cells created by grids on each processor is @xmath70 $ ] for a total of ( @xmath71 ) cells created on level 1 to be distributed across all three processors . if we were independently balancing the workload on level 1 the desired distribution @xmath72 would be constant for each processor @xmath73 where @xmath74 is the average new child workload . for our example this would give a distribution of @xmath75 $ ] . there is however a workload imbalance on level 0 of @xmath76 $ ] that we would like to compensate for . in what follows we show how we can successively rewrite @xmath72 in new forms which allow us to better anticipate the best distribution to account for computation within and across levels . if we completely compensate for the level 0 imbalance @xmath77 then the desired workload distribution for level 1 would be @xmath78 $ ] . if the level 1 workload is distributed that way , then while processor 1 waits for processors 2 & 3 to complete their 20 level 1 updates , it should have time to complete 12 of its level 0 updates so @xmath79 $ ] . at this point the remaining workload on level 0 would be balanced at @xmath80 $ ] . however we need to take an additional step on level @xmath9 . thus while the remaining level 0 workload is balanced , we see an opportunity to redefine @xmath72 in a way that accounts for this additional level 1 step . we would like to redistribute the grids on level 1 such that they are balanced at @xmath81 $ ] before taking the second step to avoid processor idling . this redistribution has an additional cost that can easily be avoided if the imbalance on level 0 is first weighted by the ratio of the number of level 0 steps @xmath82 to level 1 steps @xmath83 remaining . during the first distribution of level 1 grids there are 2 remaining level 1 steps @xmath84 and one remaining level 0 step @xmath85 . thus we redefine @xmath86 by modifying equation [ verysimpledistributioneq ] to @xmath87 this would give us @xmath81-[+4,-2,-2]=[12,18,18]$ ] . now while processor 1 waits for processors 2 & 3 to complete their 18 level 1 updates , it should have time to complete 6 of its level 0 updates effectively reducing the imbalance by a factor of 2 . after completing the first level 1 advance , the second distribution of level 1 grids must somehow take into account the coarser workload accomplished . to do so we include a completed workload imbalance @xmath88 $ ] into equation [ distributioneq2 ] . @xmath89 note that now with @xmath90 we get the same desired distribution of @xmath91 $ ] for the second level 1 step as for the first . we could have just used the remaining workload @xmath92 to calculate a remaining workload imbalance @xmath93 and then used equation [ distributioneq2 ] but this would be the same as modifying equation [ simpledistributioneq ] to be @xmath94 . since in our example @xmath95 only when @xmath96 , it makes no difference , but when there are more then two levels of refinement this approach can lead to unnecessary shuffling of grids between processors even with static meshes . for a more detailed analysis see [ staticdist ] . when there are multiple levels of refinement , the workloads on many coarser levels must be taken into account . the generalization of equation [ simpledistributioneq ] is given by @xmath97 a more convenient form of equation [ distributioneq ] can be obtained if we replace @xmath98 and @xmath99 with their expansions and then group terms by whether they involve averages across processors or local quantities on a particular processor . we then arrive at the final general form of the distribution equation @xmath100 where @xmath101 . note that @xmath102 is the predicted remaining work on all coarser levels and that if the actual distribution matches the desired distribution ( ie @xmath103 ) then the total predicted remaining workload including the current level is @xmath104 that is , each distribution attempts to equalize the predicted remaining work over the entire amr advance . it is not always possible to distribute a discrete set of grids perfectly so that @xmath103 , however small differences on coarser levels can be corrected for on finer level distributions . on the finest level the grids are artificially split to balance out all of the coarser level imbalances as described below . thus at each distribution of level @xmath4 grids , processors must collect the new child workloads @xmath54 to calculate @xmath105 ) , and @xmath63 to calculate the desired workload distributions @xmath58 . then the new child workloads @xmath54 are partitioned over @xmath106 as shown in figure [ partition ] to give the desired workload @xmath65 assigned to processor @xmath66 of children created by grids on processor @xmath31 . note that @xmath107 and @xmath108 . each processor @xmath31 can then determine from which processors @xmath66 it should expect to receive new child grids from ( @xmath109 ) as well as which `` child '' processors @xmath110 it should distribute new grids to ( @xmath111 ) here the term child processor does not refer to the addition of new processors ( unlike the addition / creation of new child grids ) but only those existing processors which will receive child grids from other processors ( referred to here as parents " ) . when distributing child grids among child processors , each processor @xmath31 will try to assign to each child processor @xmath66 a collection of level @xmath4 grids with a combined workload @xmath65 . however , arbitrarily partitioning a few discrete grids into exact sizes is often not possible , and the actual workload assigned @xmath112 to each child processor may be different from the desired workload @xmath65 . as a result , the combined child workload from all of a processor s parent processors @xmath113 may be different from the desired workload @xmath114 . this results in some variation in @xmath115 between processors and a temporary predicted load imbalance . however , if this variation is small , it can be corrected on finer level distributions or on the next round of level @xmath4 distributions without adversely effecting performance , provided that there remains enough computational work on coarser advance threads to buffer the imbalance . of new level @xmath4 child workloads @xmath116 onto desired workload shares @xmath117 . shades of gray correspond to processor rank ( 0 - 3).,title="fig : " ] [ partition ] before discussing the details of the distribution of new refined grids , it is instructive to first consider the simpler problem of distributing newly refined cells as is the case for cell - based amr . after newly refined cells are created , a space filling curve is generally used to sort the new cells into a prescribed order based on their position along the curve @xcite . in addition the processors are ordered based on the topology of the network . then the ordered list of cells can be partitioned over the ordered list of processors in a manner similar to that used in constructing the @xmath65 in figure [ partition ] except that now instead of globally distributing @xmath54 over @xmath58 to determine @xmath65 we are locally distributing the individual workload of each child cell @xmath118 over the child processor allocations @xmath65 to determine the processor allocations for each individual child cell @xmath119 . most cells will have only one nonzero allocation @xmath120 and will be assigned to that processor @xmath66 and if a cell has more than one nonzero allocation then the processor with the largest allocation could be used . this `` rounding '' will only lead to load imbalances that are of order @xmath121 . this ordering attempts to keep communication between neighboring cells to a minimum . for the partitioning to work , however , each processor needs to know the global sort index of each of its newly refined cells . sorting the new cells could be done using traditional sorting algorithms provided each processor was aware of all of the cells . a better alternative , is to maintain a strict grouping of cells by sort order on each processor as shown in figure [ hilbertfig ] . that is , a cell with a lower sort index will never be found on a higher numbered processor and vice versa . because of the fractal nature of many space filling curves like the hilbert curves shown in figure [ hilbertfig ] , the ordering of any two child cells of different parent cells will be the same as that of its parents . therefore if the parent cells obey a strict grouping by sort order , the child cells will as well . processors therefore only need to determine the local sort order of their child cells . they can then determine the global sort order if they know the new child cell counts of every other processor . an example of these cell counts can be seen in figure [ hilbertfig ] as well as local and global index of each level 0 and level 1 cells . the global index of each cell can be constructed by adding the cell counts of lower processors to the local index . the fact that child cells will obey the same ordering of their parents can be used to locally sort child cells by first sorting the parent cells and then locally sorting the child cells among their siblings . in cell based amr each processor can usually partition its newly refined cells among child processors " with very little load imbalance since each processor typically has 1000 s of cells to distribute over a few processors . not being able to split a cell might result in imbalances of few tenths of a percent . this is not the case for patch based amr . processors will still have 1000 s of cells , but they will be grouped into a few grids of various sizes . distributing a few grids of various sizes among a few child processors with various workload assignments will in general lead to significant imbalances . often what is done in grid - based amr is a more global distribution of all new child grids among all processors using a knapsack algorithm . this type of distribution does not rely on a space - filling curves or any global ordering of new grids although it may take the communication costs into account . this however , requires global knowledge of the amr tree which can result in memory issues for simulations run on many processors . in astrobear , the amr tree is distributed so a global knapsack type algorithm is not feasible . instead astrobear uses an approach quite similar to that used for cell - based amr . astrobear avoids the issues with load imbalances not by artificially fragmenting grids into small pieces , or by locally shuffling the grids in a local knapsack algorithm , but instead through the use of level threading described above in which load imbalances can be corrected for by subsequent distributions . this approach permits a strict grouping of grids on processors by their sort order . this is true to at least to the extent that grids obey the same type of child order inheritance true for individual cells . that is , while the order of two child cells will always be the same as that of their parent cells , it may not hold for grids . cells have a unique distance along the space filling curve , however grids contain many cells each with different distances . a distance for the grid along the curve can be approximated by averaging that of all of its cells , or some subset of cells that appropriately samples the grid ( ie the four corners or the center most subset ) . however this fuzziness in grid distances can result in child grids occasionally being ordered differently then their respective parents . however , this does not appear to be a major problem and would only result in slight increases in neighbor communication . we note that instead of using a space - filling curve to order the grids , we could implement a local knapsack algorithm in which each processor distributes its own newly created children among its own child processors in an optimal fashion , however we have not investigated whether this improves performance compared to the ordering provided by a space - filling curve . occasionally splitting of a child grid into various pieces to accommodate its processor allocations @xmath119 is desired and there are two such instances in astrobear . first , since most of the workload resides on the finest level grids , imbalances can sometimes be too large to be compensated for by the coarser threads . for this reason astrobear always splits the finest level grids when necessary to achieve global load balancing . in general the number of finest level grids split will be of the order of the number of processors . since the splitting occurs on the finest level , such fragmentation will not result in subsequent artificial fragmentation of higher level grids . second , since the base grid in amr simulations can be quite large it can often contain more then a single processor s share of the entire workload . for this reason the base grid is also split among the processors . the algorithm for splitting begins by taking a single grid and its non - zero processor allocations @xmath120 that were determined by locally partitioning the set of @xmath65 over the individual child workloads @xmath122 . this then gives the share of each individual child grid that should be assigned to each child processor . it then attempts to split the grid into pieces each with a workload equal to the processor allocations @xmath119 . additionally it attempts to construct the pieces so that they are properly ordered along the hilbert curve . this is done through a recursive bisection algorithm described below : 1 . divide the list of weights into two pieces as equal as possible . for example if the list of weights were [ .1 .2 .1 .3 .3 ] they would be split into [ .1 .2 .1 ] and [ .3 .3 ] 2 . determine the two possible split points along each dimension that break the grid into two proportionate pieces ( ie either [ .4 .6 ] or [ .6 .4 ] ) . 3 . for each of the possibilities evaluate the combined advance costs of the resulting grids . this will in general favor splits along the longest grid dimension . 4 . if several possible splits have comparable combined advance costs then select the one that results in the largest difference in space - filling values between the two pieces with the correct sign . 5 . if there were only two weights then we are done . otherwise , continue to recursively bisect the left and right pieces . for a large fixed grid simulation , the above algorithm applied to a square base grid in 2d ( or a cube in 3d ) on @xmath123 processors ( @xmath124 in 3d ) , will yield a distribution that traces out a level @xmath30 hilbert curve . [ hilbertfig ] it is challenging to characterize the performance of an amr code in general terms . this is because the performance on any given problem will depend on many factors that are specific to that problem such as the number of levels of refinement , the number of cells on each level ( filling fractions ) , their spatial distribution , and the number of processors used . here we give a few examples to demonstrate the complexity of the problem . first consider a fixed grid 16x16 problem run on 5 processors . ideally the 16x16 grid could be split into 5 even rectangular pieces to distribute among 5 processors . however , the best arrangement still leaves one processor with 55 cells which is 7% more then the ideal average of 51.2 effectively limiting the performance to 93% . we could split the grid into more than 5 pieces so that the maximum number of cells on any processor is closer to the mean , but the additional overhead with having smaller grids would likely defeat the purpose . the same problem however can be easily broken into 16 even pieces ( each 4x4 ) to be perfectly distributed among 16 processors . however , running the problem on 16 processors instead of 1 will not generally allow the problem to complete in 1/16 of the time for two primary reasons . first , before each 4x4 grid can update , it most share ghost data with the surrounding grids and there is some network latency in transmitting ghost data between grids . additionally , while updating the 4x4 grid , calculations will be done using the ghost cells that will be repeated on the neighboring processors . this redundant calculation is not required if there is only one 16x16 grid . thus updating 16 4x4 grids requires more calculations then updating 1 16x16 grid , so the cpu time will in general increase as more and more processors are used for the same problem even if the geometry of the problem allows for even distribution . now consider the same 16x16 grid on 16 processors and lets assume that there are 16 refined regions ( each 2x2 ) that just happen to lie in the center of each 4x4 grid . these child grids will not be adjacent to other child grids and will only need to communicate with their parent grid . if each child grid is distributed on the same processor containing the parent , then the communication pattern is the same as that of the fixed grid run and the performance should actually be better because each processor now has more work that can be done independently . if however , instead of having 16 2x2 regions marked for refinement , the central 8x8 region was marked for refinement , then the communication pattern of the level 1 grids would be very similar to that of the level 0 although there would now be some additional communication between parent and child grids . thus the simplest and fairest test of an amr code is to keep the number of cells per processor fixed as well as keeping the refined regions spatially connected . furthermore , if we keep the filling fractions at 1/4 for 2d ( 1/8 for 3d ) then each level will have the same number of cells and the number of cells per level per processor will be a constant . one still has the freedom in choosing the number of cells per leve per processor and this can have dramatic effects on the scaling . larger grids will in general allow for better scaling . a fair test will choose a grid size that will allow for a simulation to complete in a reasonable wall time . to that end we chose a resolution low enough so that a simulation on 10,000 cores would only take 8 hours per crossing time . this works out to be @xmath125 cells per processor for the fixed grid run . if we kept the base grid fixed and added 1 additional level of amr , the simulation would take 3 times as long . so we have to reduce the base grid size as we increase the maximum level of amr to keep the overall walltime per crossing time fixed at 8 hours . for 1 additional level of amr this works out to be @xmath126 cells per processor per level , and @xmath127 , @xmath128 , & @xmath129 for 2 , 4 , & 8 levels of amr respectively . as the grids get smaller we expect the additional overhead related to ghost zones to reduce the overall speed - and the scaling will likely also suffer as well . figure [ weakscaling ] shows the results of weak scaling done on kraken . for each level of amr we ran simulations on 12 , 48 , 192 , 768 , 3072 , & 12288 cores using 5 different options for the load balancing / threading scheme . for each simulation we calculated the total number of cell updates divided by the cpu time to get the efficiency of the code in actually solving the problem . other metrics such as the degree to which a processor remains idle , while interesting , are less relevant to the performance of actual simulations . the solid black line shows the traditional serial amr approach which in which each level is independently balanced out of necessity . the solid light and dark grey line shows the results using a scheduling approach to threading in which processors schedule rendevouz times before switching to coarser advance threads . the light line is for a simulation in which each level is balanced independently and the dark line is for a simulation that attempts to balance across levels or globally . the dashed light and dark grey lines are for the simulations that actually use a threading library . this allows for much finer grained switching between the control thread and the coarser advance threads . again the light line is for balancing within a level , and the dark line is for balancing across levels or globally . first there is a secular drop in performance for any number of cores in terms of cell updates per second per cpu as we increase the number of levels of amr . this is due to the decrease in the average grid size required as the number of amr levels increases to keep the wall time constant . the fixed grid run has typical grid sizes of @xmath125 while the run with 8 levels of amr has typical grid sizes of only @xmath129 . advancing cells within a @xmath125 grid can be done 4 times as efficiently as grids with only @xmath129 cells due to the overhead with regards to ghost zones etc . next if we look at the serial amr performance it turns out to be quite good with a worse case performance drop of only @xmath130 from 12 to 12288 cores . we attribute this excellent performance to a fully parallelized tree and the presence of only a few scalar quantities that must be gathered across processors . we also note in the upper right plot in figure [ weakscaling ] that for fixed grid there is no difference between the various approaches as expected since there is only 1 thread and no difference between local and global balancing . there is also a noticeable trend with regards to global vs. local balancing for both the scheduled and threaded approach . the lower right plot of figure [ weakscaling ] shows most clearly this effect . for 12 cores , which on kraken are on a single node , the globally balanced approach out performs the locally balanced approach . the globally balanced approach - while allowing for larger average grid sizes and therefore more efficient updating - reduces the degree to which various prolongation / restriction / synchonization operations are parallelized while also requiring more intercore communication . when the cores reside on the same node , this communication can happen quickly and does not degrade the performance as much as the larger grid sizes improves the performance . however by 48 cores the additional communication has negated any advantage of having larger grid sizes , and trying to keep grids as large as possible to avoid this sort of overhead does not seem to be effective on large numbers of cores . finally if we compare the three level balanced approaches , we see that both the scheduled and the truly threaded approaches out perform the serial amr approach by @xmath131 . there is a tendency for the fully threaded level balanced approach to drop in performance at 12288 cores , but this is likely due to the crude implementation of a non - blocking barrier . the idea behind the threaded approach is that every communication is non - blocking and that while waiting for a message to complete coarser advance threads can be running while periodically checking to see if the communication had completed . however global reductions can not be done using non - blocking calls . we dealt with this by writing a crude non - blocking barrier that could be executed before any global reductions and that would ensure every processor was ready to procede with the global mpi call . the implementation involved every processor sending a ready signal to processor 0 , followed by every processor waiting for a continue signal from processor 0 . this all - to - one followed by a one - to - all would likely cause significant performance degradation at 12288 cores . there are more elegant solutions that could be implemented , such as scheduling these barriers or writing a more efficient branching communication pattern . for @xmath132 levels of amr respectively . each plot shows the performance of the serial amr approach , scheduled advances with level by level balancing as well as global balancing , and the threaded approach with level by level and global balancing . the y - axis shows the number of cells updated per processor per second.,title="fig:",scaledwidth=100.0% ] [ weakscaling ] as discussed in the introduction adaptive mesh refinement methods were designed to allow high resolution simulations to be carried out at low computational cost . highly parallel systems and their algorithms carried the same promise . the parallelization of amr algorithms is , however , not straight forward however and there have been a number of different approaches @xcite , to solve the problem . while parallelization of a uniform mesh demands little communication between processors , amr methods can demand considerable communication to maintain data consistency across the unstructured mesh as well as shuffling new grids from one processor to another to balance workload . in this paper we have described our attempt to design and implement a new strategy for amr parallelization with an eye towards running codes on large machines with @xmath133 processors . we have found that a threaded approach to the amr algorithm significantly improves performance by allowing processors to remain busy while waiting for messages as well as to dynamically adjust distributions based on the progress of ongoing coarser grid updates . we have also shown that a distributed tree algorithm significantly reduces the amount of memory required ( and corresponding communication ) for simulations run on many processors with modest grid sizes . we believe this threaded , distributed tree approach described in this paper holds considerable promise as a methodology for implementing amr on ever - larger numbers of processors . future work will entail elaborating refinements to the methods as we experiment with its implementation on different classes of problems and machines . 1 balsara , d. s. 2001 , journal of computational physics , 174 , 614 brian van straalen , phil colella , daniel t. graves , and noel keen . petascale block - structured amr applications without distributed meta - data m. l. simone , r. m. loy , m. s. shephard and j. e. flaherty tiankai tu , d. r. ohallaron , and o. ghattas b. hariharan , a srinivas , 31:311 331 , 2005 a.m and khokhlov . fully threaded tree algorithms for adaptive refinement fluid dynamics simulations . , 143(2):519 543 , 1998 . marsha j berger and joseph oliger . adaptive mesh refinement for hyperbolic partial differential equations . , 53(3):484 512 , 1984 . berger and p. colella . local adaptive mesh refinement for shock hydrodynamics . , 82(1):64 84 , 1989 . a. j. cunningham , a. frank , p. varnire , s. mitran , and t. w. jones . . , 182:519542 , june 2009 . p. macneice , k. m. olson , c. mobarry , r. de fainchtein , and c. packer . . , 126:330354 , april 2000 . j. mellor - crummey , d. whalley , and k. kennedy . improving memory hierarchy performance for irregular applications . , 29(3 ) , june 2001 . b. w. oshea , g. bryan , j. bordner , m. l. norman , t. abel , r. harkness , and a. kritsuk . . , march 2004 . u. ziegler . . , 179:227244 , august 2008 . in table [ geninheritance ] we outline the procedure for inheriting neighbors , overlaps , from parent relationships for arbitrary number of refinement ratios . and in table [ disttreealg ] we outline the communication steps for maintaining the distributed tree . & node s child s preceding overlaps & node s child s succeeding overlaps + 1 & node s neighbors 1@xmath40 children & node s preceding overlaps r@xmath134 children & node s neighbor s 2@xmath41 children + 2 & node s neighbors 2@xmath41 children & node s neighbors 1@xmath40 children & node s neighbors 3@xmath135 children + 3 & node s neighbors 3@xmath135 children & node s neighbors 2@xmath41 children & node s neighbors 4@xmath134 children + ... & ... & ... & .... + r-1 & node s neighbors r@xmath134 - 1 children & node s neighbors r@xmath134 - 2 children & node s neighbors r@xmath134 children + r & node s neighbors r@xmath134 children & node s neighbors r@xmath134 - 1 children & node s succeeding overlaps 1@xmath40 children + + 1 & _ receive _ new grids & nodes along with their parents , neighbors , and * preceding * overlaps from parent processors & _ receive _ * succeeding * overlaps from parent processors + 2 & + 3 & determine which remote * preceding * nodes might have children that would overlap with its own children and _ send _ the relevant children info & determine which remote * succeeding * nodes might have created children that would overlap with its own children and _ send _ the relevant children info + 4 & + 5 & + 6 & determine which local * preceding * nodes have children that overlap with its own & determine which local * succeeding * nodes have children that overlap with its own + 7 & + 8 & _ receive _ children from remote * preceding * nodes and determine which of the nodes children overlaps with its own & _ receive _ children from remote * succeeding * nodes and determine which of the nodes children overlaps with its own . + 9 & for each remote child , _ send _ the child grid s data as well as information about its parents , neighbors , & * preceding * overlaps . & for each remote child , _ send _ the child s * succeeding * overlaps . + + 10 & + 11 & + 12 & + 13 & + 14 & + 15 & + consider what happens if the mesh is static so that the average workload per level @xmath55 remains constant . initially there will be a desired set of distributions @xmath58 that will result in an actual set of distributions @xmath52 . on the finest level they will agree , but on coarser levels there will be some difference between @xmath58 and @xmath52 . it is desirable that each successive desired distribution should match the original so that each successive actual distributions will also remain constant and grids will not be unnecessarily shuffled around . for example , consider the desired distribution of the highest level @xmath18 after the first level @xmath18 advance . assuming the highest level workload follows the desired workload , @xmath136 . the amount of time each processor will have waiting for every other processor to complete the level @xmath18 advance is just @xmath137 . if this time is spent advancing coarser grids , then the new sum of work done on coarser grids @xmath138 $ ] . this gives a new predicted remaining work @xmath139 $ ] and a new desired distribution @xmath140}{s_l-1 } \nonumber \\ & = & \overline{g_l}-\frac{\eta_l^p-\overline{\eta_l } + g_l^p-\overline{g_l}}{s_l-1}=\overline{g_l}-\frac{\eta_l^{p}-\overline{\eta_l^{p}}}{s_l } = d_l^p \nonumber \\\end{aligned}\ ] ] more generally we can consider the distribution of any given level @xmath4 after completing @xmath141 steps leaving @xmath142 remaining steps . @xmath143 if @xmath144 , then the desired distribution will remain fixed . @xmath145 the workload completed on the coarser grids will be the difference between the maximum work completed on all of the previous grids by any processor and the local work completed on all of the previous grids . this then gives @xmath146 } - \sum_{l'=0}^l(2^{l'}-s_{l'})g_{l'}^p \nonumber \\\end{aligned}\ ] ] and we have @xmath147 now @xmath148 it the initial entire predicted workload and is balanced by the highest level distribution so the value on any processor equals the average and the first term cancels . the second term can be re - written by recognizing that at the moment we are distributing level @xmath4 , for @xmath149 , @xmath150 . this gives @xmath151 therefore if the mesh is static , the distribution will remain static as well .

when using reinforcement learning ( rl ) algorithms to evaluate a policy it is common , given a large state space , to introduce some form of approximation architecture for the value function ( vf ) . the exact form of this architecture can have a significant effect on the accuracy of the vf estimate , however , and determining a suitable approximation architecture can often be a highly complex task . consequently there is a large amount of interest in the potential for allowing rl algorithms to adaptively generate ( i.e. to learn ) approximation architectures . we investigate a method of adapting approximation architectures which uses feedback regarding the frequency with which an agent has visited certain states to guide which areas of the state space to approximate with greater detail . we introduce an algorithm based upon this idea which adapts a state aggregation approximation architecture on - line . assuming states , we demonstrate theoretically that provided the following relatively non - restrictive assumptions are satisfied : ( a ) the number of cells in the state aggregation architecture is of order or greater , ( b ) the policy and transition function are close to deterministic , and ( c ) the prior for the transition function is uniformly distributed our algorithm can guarantee , assuming we use an appropriate scoring function to measure vf error , error which is arbitrarily close to zero as becomes large . it is able to do this despite having only space complexity ( and negligible time complexity ) . we conclude by generating a set of empirical results which support the theoretical results . example.eps gsave newpath 20 20 moveto 20 220 lineto 220 220 lineto 220 20 lineto closepath 2 setlinewidth gsave .4 setgray fill grestore stroke grestore

1703.07940

most commonly used reinforcement learning ( rl ) algorithms store an estimate of what s known as the value function ( vf ) . the vf corresponds to a particular policy , and is a mapping from each state - action pair to a real value which reflects the amount of reward the agent will obtain starting from that state - action pair and following the policy in question ( sutton and barto 1998 ) . in order for an rl algorithm to perform well ( i.e. to achieve a high reward ) , it is important that the vf estimate is as accurate as possible , since it is this estimate which governs how the algorithm will update its policy . traditional rl algorithms such as td(@xmath4 ) or @xmath5-learning can generate exact vf estimates when dealing with small state and action spaces . however , when environments are complex ( with large state or action spaces ) , applying such algorithms directly becomes too computationally demanding . as a result it is common to introduce some form of architecture with which to approximate the vf , for example a parametrised set of functions ( sutton and barto 1998 ; bertsekas and tsitsiklis 1996 ) . one issue when introducing vf approximation , however , is that the accuracy of the algorithm s vf estimate is highly dependent upon the exact form of the architecture chosen ( it may be , for example , that no element of the chosen set of parametrised functions closely fits the vf ) . accordingly , a number of authors have explored the possibility of allowing the approximation architecture to be _ learned _ by the agent , rather than pre - set manually by the designer see buoniu et al ( 2009 ) for an overview . what we might hope to achieve by employing such an approach is to create an rl algorithm which still has relatively low computational demands , but at the same time has increased flexibility , allowing us to apply the algorithm to a wider set of problems without needing to invest time to suitably adapt it in each case . if we assume that the approximation architecture being adapted is linear ( so that the vf is represented as a weighted sum of basis functions ) such methods are known as _ basis function adaptation_. a simple and perhaps , as yet , under - explored method of basis function adaptation involves using an estimate of the frequency with which an agent has visited certain states to determine which states to more accurately represent . such methods are `` unsupervised '' in the sense that no direct reference to the reward or to any estimate of the vf is made . the concept of using visit frequencies in an unsupervised manner is not completely new ( menache et al 2005 ; bernstein and shimkin 2010 ) however it remains relatively unexplored compared to methods which seek to measure the error in the vf estimate explicitly and to then use this error as feedback ( munos and moore 2002 ; bertsekas and yu 2009 ; di castro and mannor 2010 ; mahadevan et al 2013 ) . as we will demonstrate , however , unsupervised methods have some distinct advantages : ( a ) estimates of visit frequencies are cheap to calculate and to store , ( b ) accurate estimates of visit frequencies can be generated with fewer samples than accurate estimates of , for example , temporal differences ( which are a common form of feedback used to adapt basis functions ) , and ( c ) under suitable conditions , and applying an appropriate scoring function , such methods can in fact generate very accurate vf estimates , guaranteeing in certain cases scores arbitrarily close to zero . our overarching objective in this article is to more closely examine unsupervised methods and seek to quantify , where possible , these advantages . it is point ( c ) which is perhaps the most surprising , and it forms the substance of the article s main result . for any fixed policy you will have a stationary distribution describing the probability of being in each state . suppose ( i ) the policy and transition function for a given problem are close to deterministic , and ( ii ) the prior for the transition function is uniformly distributed ( this will be more precisely described below ) . we will show that , under these conditions , an agent which follows an arbitrary policy will , on average , tend to spend almost all of its time in a small subset of the state space . indeed , if the state space is of size @xmath0 , there is a theoretical upper bound on the average size of this subset : @xmath6 . provided we have enough basis functions to individually represent the states in this subset , unsupervised basis function adaptation methods can ensure that the vf is arbitrarily well estimated over this subset . if the scoring function we apply is weighted by the probability of visiting each state , we can then guarantee an arbitrarily low score . the implication of this is that , under these circumstances , unsupervised methods will perform _ at least as well _ as other more complex methods , but , compared to these other methods , will do so _ more cheaply _ ( in the sense of sampling required and also computational demands ) . whilst conditions ( i ) and ( ii ) encompass many important and general problems , we will also explore the potential to generalise condition ( ii ) to encompass a larger set of possible priors . we also explore these ideas experimentally . our experimental results suggest that our techniques provide a powerful advantage in many real world settings . over a set of realistic parameter settings the techniques ( when compared to fixed state aggregation ) can reduce vf error by an amount in the range of @xmath7-@xmath8% . in some cases the experimental results also suggest that the assumptions required by the theory can be relaxed . as noted above , unsupervised techniques at present are relatively unexplored . menache et al ( 2005 ) provided a brief evaluation of an unsupervised algorithm ( to provide a comparison with two more complex adaptation algorithms ) in the setting of policy evaluation . berstein and shimkin ( 2009 ) examined an algorithm where a set of kernels are progressively split ( `` once - and - for - all '' ) based on the visit frequency for each kernel . their algorithm includes policy updates which incorporate knowledge of uncertainty in the vf estimate . the algorithm we propose below works in conjunction with a state aggregation approximation architecture , employing a form of `` cell - splitting '' to give state space regions more or less resolution . this bears similarities to a number of approaches examined in the literature ( the main difference in our approach is the rules under which cells are split or joined ) . moore and atkeson ( 1995 ) provide an early algorithm based on updating a state aggregation architecture , whilst whiteson et al ( 2007 ) examined a basis function construction method involving cell splitting . our analysis is new both in terms of the details of the unsupervised algorithm we outline ( it is designed so as to minimise memory requirements , in particular to ensure that space complexity is roughly of the order of the number of basis functions , whilst still permitting the approximation architecture to be continuously adapted on - line ) and in terms of the theoretical concepts we derive . in the remainder of this section we outline the formal framework we will be using . in section [ unsupervised ] we set out the details of our algorithm ( pasa , short for `` probabilistic adaptive state aggregation '' ) which performs unsupervised basis function adaptation based on state aggregation . in section [ theoretic ] we outline our main theoretical results . finally , in section [ simulation ] we set out some empirical results designed to both support and extend the results in section [ theoretic ] . we assume that we have an agent which interacts with an environment over a sequence of iterations @xmath9 . for each @xmath10 the agent will be in a particular state to hold in the case of a continuous state space , additional assumptions would need to be introduced . ] @xmath11 ( @xmath12 ) and will take a particular action @xmath13 ( @xmath14 ) . each action is taken according to a _ policy _ @xmath15 whereby the probability the agent takes action @xmath13 in state @xmath11 is denoted as @xmath16 . we are considering the problem of policy evaluation , so we will always assume that the agent s policy @xmath15 is fixed ( i.e. does not change as a function of @xmath10 ) . the _ reward function _ @xmath17 is a mapping from each state - action pair @xmath18 to a real number , such that if the agent is in state @xmath11 and takes action @xmath13 , then it will receive a _ reward _ @xmath19 . the reward function is assumed to be deterministic and bounded , i.e. @xmath20 for all @xmath21 . the _ transition function _ @xmath22 defines how the agent s state evolves over time . if the agent is in state @xmath11 and takes an action @xmath13 in iteration @xmath10 , then the probability it will transition to the state @xmath23 in iteration @xmath24 is given by @xmath25 . both @xmath22 and @xmath17 are taken as unknown , however we assume we are given a prior distribution for both . the _ value function _ @xmath26 , which maps each of the @xmath27 state - action pairs to a real value , is defined as follows : @xmath28 where the expectation is taken over the distributions of @xmath22 and @xmath15 ( i.e. for a particular instance of @xmath22 , not over its prior distribution ) and where @xmath29 is known as a _ discount factor_. we have used superscript brackets to indicate dependency on the iteration @xmath10 . initially the vf is unknown , our objective , given some @xmath15 , is to learn the vf for a particular instance of @xmath22 . assuming we are able to store an explicit real value for each state - action pair , traditional rl algorithms provide a means of estimating @xmath26 . these estimates will converge to the correct value as @xmath10 becomes large ( sutton and barto 1998 ) . in cases where @xmath0 or @xmath30 are large , though , it may be impossible to store @xmath27 real values . hence , when dealing with such cases , it is common to employ vf approximation . once we approximate the vf , however , even if the underlying rl algorithm still converges to generate an estimate of the vf ( which is not always guaranteed ) , we can no longer rely on the estimate being arbitrarily close to the true value @xmath26 . one form of vf approximation , _ parametrised value function approximation _ , involves generating an approximation of the vf using a parametrised set of functions . the goal of the rl algorithm then becomes to find a value for the parameters so that the vf estimate is as near to the true value as possible . the approximate vf is denoted as @xmath31 , and , assuming we are approximating over the state space only and not the action space , this value is parametrised by a matrix @xmath32 of dimension @xmath33 ( where @xmath34 ) . such an approximation architecture is _ linear _ if @xmath31 can be expressed in the form @xmath35 , where @xmath36 is the @xmath37th column of @xmath32 and @xmath38 is a fixed vector of dimension @xmath1 for each pair @xmath18 . the @xmath39 distinct vectors of dimension @xmath0 given by @xmath40 are called _ basis functions_. it is common to assume that @xmath41 for all @xmath37 , in which case we have only @xmath1 distinct basis functions , and @xmath42 . a _ state aggregation _ approximation architecture see , for example , singh et al ( 1995 ) and whiteson et al ( 2007 ) is a simple linear approximation architecture which we can define as being a mapping @xmath43 from each state @xmath11 to a _ cell _ @xmath44 ( @xmath45 ) . defining the architecture as a `` mapping '' implies that every state corresponds to exactly one cell . we will denote as @xmath46 the set of states in the @xmath47th cell . given a state aggregation approximation architecture , the underlying rl algorithm can not distinguish states in the same cell , and hence @xmath48 will be the same for all states which are in the same cell ( @xmath32 in this context can be interpreted as a set of weights , one given to each cell - action pair ) . if we want to design an algorithm to adapt an approximation architecture , we need a means to assess how well it is doing this . a _ scoring function _ is used to assess the accuracy of a vf estimate ( and can also therefore be used to evaluate an algorithm designed to generate a vf estimate ) . many basis function adaptation algorithms use a scoring function as a form of feedback to help guide how the basis functions should be updated . in such cases it is important that the score is something which can be measured computationally . one commonly used score is the squared error in the vf estimate for each state - action , weighted by the probability of each state - action occurring ( menache et al 2005 ; bertsekas and yu 2009 ; di castro and mannor 2010 ) . we will refer to this as the _ mean squared error _ ( mse ) : @xmath49 this is where @xmath50 is a vector of the probability of each state given the stationary distribution associated with @xmath15 ( given some fixed policy @xmath15 , the transition matrix obtained from @xmath15 and @xmath22 has a corresponding stationary distribution , provided the transition matrix is irreducible and aperiodic ) . note that the true vf @xmath26 appears in ( [ mseexact ] ) . this value , however , is unknown . therefore , another commonly used scoring function ( which , unlike mse , can be estimated empirically ) uses @xmath51 , the _ bellman operator _ , to obtain an approximation of the mse . this scoring function we denote as @xmath52 ( this is a weighted sum of what is sometimes known as the _ bellman error _ at each state - action ) : @xmath53 where : @xmath54 our results in section [ theoretic ] will be stated in relation to mse and @xmath52 . it will be crucial to all of our results that the scoring function is weighted by @xmath50 . two important comments should be made in relation to this . the first is that , whilst applying such a weighting appears natural , a scoring function does not necessarily need to be weighted by @xmath50 ( or an approximation of @xmath50 ) . there may be circumstances under which a more appropriate measure of the accuracy of a vf estimate would , for example , weight every state equally ( the most appropriate measure to use in each situation would depend on a number of complex factors ) . we acknowledge this as a limitation of the analysis in section [ theoretic ] . the second is that , in investigating unsupervised basis function adaptation methods , we are implicitly making a comparison with methods of basis function adaptation which use explicit scores as a source of feedback ( these could be called `` supervised '' methods ) . if an algorithm uses a scoring function as feedback , then ( irrespective of which scoring function is most appropriate ) it is best evaluated in terms of how well it minimises _ that particular scoring function_. the scoring function @xmath52 is an attractive choice to provide feedback for supervised methods since , by weighting the error by the probability of visiting a state , it is possible to generate an estimate of the score without knowing the probability of visiting each state . in fact , any feedback based on a scoring function which is _ not _ weighted by @xmath50 would implicitly require some way of normalising the score for each state . this in turn would require an estimate for @xmath50 , which implies that @xmath55 distinct values need to be recorded . hence , if unsupervised methods can perform comparatively well in terms of minimising probability weighted scoring functions , this is of great importance when comparing such methods to supervised alternatives . in the definitions above both mse and @xmath52 are weighted by the probability of each action occurring . we could redefine mse and @xmath52 to weight each action equally when sampling for @xmath52 this would not be an issue computationally since @xmath15 is known to the algorithm , i.e. we can simply divide each sample by @xmath16 ( acknowledging that samples from rarely chosen actions would contribute more significantly to the variance of the estimate ) . our results in section [ theoretic ] will extend to @xmath52 under such an alternative definition . as we noted in the introduction , circumstances exist under which an agent ( adopting a random , or in some cases an arbitrary , policy ) will , on average , tend to spend almost all of its time in a relatively small subset of the state space . we examine these circumstances more closely in section [ theoretic ] . the underlying idea of pasa is to make the vf representation as detailed as possible over this relatively small subset ( whilst allowing the representation to be coarser over the remainder of the state space ) . pasa is designed to function in conjunction with a state aggregation approximation architecture , and accordingly it updates a mapping @xmath43 over time . provided this mapping converges to some fixed mapping @xmath56 ( we prove below that it will ) , then if an rl algorithm such as sarsa ( which we have used for the experiments in section [ simulation ] ) is used to update @xmath31 for the state aggregation architecture associated with @xmath56 , the estimate @xmath31 will also converge , due to the fact that sarsa , as well as many other rl algorithms , will , with a linear approximation architecture , converge for fixed policies ( bertsekas and tsitsiklis 1996 ) . we can then assess @xmath31 using our scoring function . suppose , momentarily , we have sufficient computational space to be able to generate an estimate @xmath57 of each @xmath58 ( which we could do , for example , by having @xmath0 weights and using a stochastic approximation algorithm ) . we could then design our algorithm roughly as follows : ( i ) start with an initial course set of @xmath59 cells , ( ii ) calculate @xmath60 for @xmath61 , ( iii ) split the cell with the largest value of @xmath62 , ( iv ) recalculate @xmath62 for the new @xmath63 cells , and ( v ) continue in the same fashion until we have @xmath1 cells . we could rerun this splitting process every iteration , or , if we prefer , at discrete intervals . provided @xmath58 is large for only a small set of states ( i.e. the variance of the elements of @xmath50 is high ) the resulting set of cells ( basis functions ) will tend to have a more detailed representation of the vf in areas of the state space with high stationary probability . this is the essence of how pasa works , except that , to avoid storing @xmath55 weights ( which we ve implicitly assumed is impossible ) , we instead measure the probability of visiting the @xmath64 `` base cells '' and of visiting @xmath65 additional cells ( which may change over time ) which are progressively split from these base cells . we can then estimate the probability of visiting each of the @xmath1 resulting cells by subtracting estimates from one another . the consequence , as will become apparent , is that pasa converges to the same point as the algorithm described in the paragraph above , whilst requiring only @xmath66 space complexity . the trade off is that it may take a longer amount of time to converge , although in practical terms this difference would appear to be marginal . before setting out how pasa works in detail , it is worthwhile highlighting one aspect of our algorithm . note that many methods examined in the literature involve what has been termed _ basis function construction _ ( buoniu et al 2009 ) , where a set of basis functions are determined by an initial process `` once - and - for - all '' , which occurs prior to the agent beginning to function `` as normal '' examples include munos and moore ( 2002 ) and whiteson et al ( 2007 ) . such methods work , for example , by progressively adding basis functions until some criteria is satisfied . the pasa algorithm falls into an alternative class of methods which have been termed _ basis function optimisation _ ( buoniu et al 2009 ) . the assumption with such methods is that there is a fixed number of basis functions which are progressively updated throughout the whole period the agent functions . this approach has the advantage of being more flexible the basis functions can adapt to policy changes , or indeed to changes in the environment . pasa will store a vector @xmath67 of integers of dimension @xmath65 , where @xmath59 . suppose we start with a fixed partition of the state space into @xmath64 cells , indexed from @xmath68 to @xmath64 , each of which is approximately the same size . using @xmath67 we can now define a new partition by splitting ( as evenly as possible ) the @xmath69th cell in the original partition . we leave one half of the @xmath69th cell with the index @xmath69 and give the other half the index @xmath63 ( all other indices stay the same ) . taking this new partition ( consisting of @xmath63 cells ) we can create a further partition by splitting the @xmath70th cell . continuing in this fashion we will end up with a partition containing @xmath1 cells ( which gives us the mapping @xmath43 ) . we need some additional mechanisms to allow us to update @xmath67 . denote by @xmath71 the set of states in the @xmath72th cell of the @xmath37th partition ( so @xmath73 and @xmath74 ) . the algorithm will store a vector @xmath75 of real values of dimension @xmath1 ( initialised as a vector of zeroes ) . this will record the approximate frequency with which certain cells have been visited by the agent . we define a new vector @xmath76 of dimension @xmath1 accordingly : @xmath77 where @xmath78 is the _ indicator function _ for a logical statement such that @xmath79 if @xmath30 is true . we can interpret @xmath76 as follows : @xmath80 will be equal to @xmath68 if and only if the current state @xmath81 falls into any cell which has been split from the original cell @xmath82 or @xmath83 ( depending on @xmath72 ) . the resulting mapping from each state to a vector @xmath76 we denote as @xmath84 ( there is a simple mapping from each vector @xmath76 to one of the @xmath1 final cells : simply take the mapped - to cell as the highest index @xmath47 such that @xmath85 ) . we then update @xmath75 in each iteration as follows ( i.e. using a simple stochastic approximation algorithm ) : @xmath86 this is where @xmath87 $ ] is a constant step - size parameter . to update @xmath67 , at certain intervals @xmath88 the pasa algorithm performs a sequence of @xmath65 operations . a temporary copy of @xmath75 is made , which we call @xmath89 . we also store an @xmath1 dimensional boolean vector @xmath90 and set each entry to zero at the start of the sequence . this keeps track of whether a particular cell has only one state , as we do nt want the algorithm to try to split singleton cells . at each stage @xmath91 of the sequence we update @xmath67 as follows ( for @xmath67 , if multiple indices satisfy the @xmath92 function , we take the lowest index ) : @xmath93 where @xmath94 and where @xmath95 is a constant designed to ensure that a ( typically small ) threshold must be exceeded before @xmath67 is adjusted . we also , after @xmath67 , update : @xmath96 the idea behind each step in the sequence is that the non - singleton cell @xmath97 with the highest value @xmath98 ( an estimate of visit frequency which is recalculated at each step ) will be split . details of these steps , as well as the overall pasa process , are outlined in algorithm [ pasa ] . note that the algorithm calls a procedure to cells . this procedure simply updates @xmath84 and @xmath90 given the latest value of @xmath67 . it also calls a procedure , which converts the mapping @xmath84 to a mapping @xmath43 . both of these procedures are computationally very straightforward . a diagram illustrating the main steps is at figure [ cellsplit ] . @xmath99 @xmath100 @xmath101 @xmath102 @xmath103 @xmath104 @xmath105 @xmath106 @xmath107 @xmath108 pasa requires only a modest increase in computational resources compared to fixed state aggregation . in relation to time complexity , @xmath75 can be updated in parallel with the rl algorithm s update of @xmath32 ( and the update of @xmath75 would not be expected to have any greater time complexity than the update to @xmath32 if using a standard rl algorithm such as sarsa or @xmath5-learning ) . the vector @xmath67 can be updated at intervals @xmath109 ( and this update can also be run in parallel ) . in practice @xmath109 can be large because this allows time for @xmath75 to converge . the mapping from state to cell has a very low order of time complexity : @xmath110 for an rl algorithm using pasa compared to a minimum of @xmath111 for @xmath1 equally sized cells . hence , pasa involves effectively no increase in time complexity . pasa does involve additional space complexity with respect to storing the vector @xmath75 : we must store @xmath1 real values . if we also store @xmath43 and @xmath84 ( as well as @xmath89 temporarily ) the overall space complexity becomes @xmath3 , although @xmath43 must also be stored for fixed state aggregation . the rl component has space complexity @xmath112 ( reflecting the @xmath33 cell - action pairs ) , so that the introduction of pasa as a pre - processing algorithm will not materially impact the overall space complexity . ( note also that the space complexity of pasa is independent of @xmath30 . ) regarding sampling efficiency , two points can be made . the first is that , since visit frequencies do not depend on each individual action , reward or subsequent trajectory , they can be estimated quickly , much more quickly than , for example , temporal differences . the second point arises when we compare pasa to methods based on explicitly estimating the vf error ( supervised methods ) . once pasa has converged ( we argue below that it will ) , then the estimate @xmath31 only needs to converge _ once_. in contrast , if the estimate @xmath31 is used to update the basis functions , then a new value of @xmath31 must , in principle , be generated each time there is an update to the basis functions . this could have serious consequences for the time required for the process to converge , as @xmath31 may take a very long time to generate an accurate estimate for a particular set of basis functions . this is particularly so if @xmath113 is near @xmath68 or when @xmath0 or @xmath30 are large . is large or @xmath15 heavily favours certain actions then an accurate estimate of the bellman error ( obtained , for example , using temporal differences ) will require an even larger number of samples , since rarely taken actions will have a high amount of variance . ] in relation to the convergence properties of pasa , if we are considering a situation where @xmath15 is updated , then provided @xmath15 continues to change pasa will not converge and instead will continuously update @xmath43 . however , for fixed @xmath15 , pasa will converge in a particular sense which we now describe . our outline of pasa assumed a single fixed step - size parameter @xmath114 . for our proof below it will be easier to suppose that we have a distinct fixed step - size parameter @xmath115 for each element @xmath116 of @xmath75 , each of which we can set to a different value ( fixed as a function of @xmath10 ) . for the remainder of this section @xmath114 should be understood as referring to this vector of step - size parameters . we will say that some function of @xmath10 , @xmath117 , _ becomes @xmath118-fixed over @xmath119 after @xmath120 _ provided @xmath120 is such that , for all @xmath121 , the value @xmath122 will remain the same for all @xmath123 satisfying @xmath124 with probability at least @xmath125 ( where the probability is taken over the prior distributions for @xmath22 and @xmath17 ) . for every @xmath126 and @xmath127 there exists @xmath114 and @xmath120 such that the mapping @xmath128 generated by pasa will become @xmath118-fixed over @xmath119 after @xmath120 . suppose that @xmath129 is a sequence of real numbers such that @xmath130 for all @xmath72 and that @xmath131 is a sequence of closed intervals on the real line such that @xmath132 for all @xmath72 . provided we have suitably specified @xmath133 , we can denote : @xmath134 for any @xmath135 , @xmath136 and @xmath137 we can set the elements @xmath138 small enough , find @xmath139 and find some @xmath140 large enough so that @xmath141 remains within the interval @xmath139 for @xmath135 iterations with probability at least @xmath142 provided @xmath140 iterations have elapsed . in this way @xmath143 becomes @xmath144-fixed over @xmath135 after @xmath140 . this follows by relying on results regarding stochastic approximation algorithms with fixed step - sizes . by replacing each @xmath145 with a time scaled equivalent , so that @xmath146 , each @xmath116 will , as @xmath147 , have an associated deterministic ordinary differential equation ( ode ) with a unique solution . furthermore , the difference between each @xmath116 and this ode converges ( again as @xmath147 ) weakly in distribution to an ornstein - ulenbeck process , which has a normal stationary distribution with a scaling factor of @xmath148 ( bucklew et al 1993 ) . hence , due to the fact that each @xmath116 is bounded by the unit interval , we can select a scalar @xmath149 so that our requirement is satisfied provided @xmath150 for all @xmath61 . suppose we have chosen @xmath151 . then @xmath69 will also become @xmath144-fixed over @xmath135 after @xmath140 . this is for the following reason . suppose each @xmath98 for @xmath152 remains within an interval of size @xmath153 for @xmath154 . define @xmath155 ( taking the lowest index if this is satisfied by more than one index ) . now , for all @xmath156 such that @xmath157 , and for all @xmath158 , we have : @xmath159 which implies that @xmath69 will not change for @xmath154 ( recalling that @xmath160 is the threshold which must be exceeded before the pasa algorithm will update @xmath67 ) . we now proceed via an induction argument . we claim that , provided for all @xmath161 , @xmath162 and @xmath163 there exists @xmath164 , @xmath133 and @xmath165 such that @xmath166 and @xmath167 are @xmath163-fixed over @xmath161 after @xmath165 , then for all @xmath168 , @xmath169 and @xmath170 there exists @xmath171 , @xmath172 and @xmath173 such that @xmath174 and @xmath175 are @xmath170-fixed over @xmath168 after @xmath173 . the case for @xmath176 we have already shown . to see our claim holds , suppose we are given values for @xmath168 , @xmath169 and @xmath170 . first note that for any @xmath177 , @xmath178 we can find @xmath179 and @xmath180 such that , assuming @xmath166 remains fixed for all @xmath10 , @xmath181 will remain in an interval @xmath182 of size @xmath183 over @xmath168 iterations with probability at least @xmath184 after @xmath180 iterations have elapsed ( using the same argument as above regarding stochastic approximation algorithms with fixed step - sizes ) . by assumption , for any @xmath161 , @xmath162 and @xmath163 , we can find @xmath164 and @xmath165 so that @xmath166 will remain the same , and @xmath185 will remain in an interval of size @xmath162 , for @xmath161 iterations with probability at least @xmath186 in both cases , provided @xmath165 iterations have elapsed . accordingly we choose @xmath162 and @xmath183 such that @xmath187 , and we choose @xmath161 such that @xmath188 . hence @xmath189 will remain the same , and @xmath190 will remain in an interval of size @xmath169 , over @xmath168 iterations both with probability at least @xmath191 after @xmath192 iterations have elapsed . for any @xmath170 we can choose @xmath163 and @xmath193 so that @xmath194 , and so our claim holds . hence we can choose @xmath114 and @xmath120 such that the vector @xmath67 becomes @xmath118-fixed over @xmath119 after @xmath120 , and so the same holds for @xmath128 . we have taken care to allow the vector @xmath114 to remain fixed as a function of @xmath10 . in practical applications , fixed step - sizes will allow an agent to continue to adapt in response to , for example , changes in the environment , and we use fixed step - sizes in our experiments below . whilst in our experiments in section [ simulation ] we use only a single step - size parameter ( as opposed to a vector ) , the details of the proof point to why there may be merit in using a vector of step - size parameters as part of a more sophisticated implementation of the ideas underlying pasa ( i.e. allowing @xmath115 to take on larger values for larger values of the index @xmath47 , for @xmath195 , may allow the algorithm to converge more rapidly ) . we now set out our main theoretical results . the key idea is that , in many important circumstances , when following a fixed policy an agent will have a tendency to spend nearly all of its time in only a small subset of the state space . we can use this property to our advantage . it means that by focussing on this small area ( which is what pasa does ) we can eliminate most of the terms which significantly contribute to the expected squared vf error . the trick will be to quantify this tendency . the fact that we have adopted a cell splitting approach will be of critical importance , because it easily permits us to create cells which contain only a single state ( allowing us to estimate the vf restricted to single state - action pairs with complete accuracy ) . we must make the following assumptions : ( 1 ) @xmath22 is `` close to '' deterministic , i.e. @xmath22 , interpreted as an operator with three arguments , can expressed as follows : @xmath196 , where @xmath197 is a deterministic transition function and @xmath198 is an arbitrary transition function , and where @xmath199 is small ( what constitutes `` small '' will be made clearer below ) , ( 2 ) @xmath22 has a uniform _ prior _ distribution , in the sense that , according to our prior distribution for @xmath22 , the random vector @xmath200 is independently distributed for all @xmath21 and each random variable @xmath25 is identically distributed for all @xmath201 , and ( 3 ) @xmath15 is also `` close to '' deterministic ( i.e. the probability of taking an _ off - policy _ action is no greater than @xmath199 , where an _ on - policy _ action is the action with the highest probability for a given state , and all other actions are off - policy ) . momentarily putting aside these assumptions , we can make the following observation . if @xmath15 and @xmath22 are deterministic , and we pick a starting state @xmath202 , then the agent will create a path through the state space and will eventually revisit a previously visited state , and will then enter a cycle . call the set of states in this path ( including the cycle ) @xmath203 and call the set of states in the cycle @xmath204 . denote as @xmath205 and @xmath206 the number of states in the path ( including the cycle ) and the cycle respectively . of course @xmath207 . if we now place the agent in a state @xmath208 ( arbitrarily chosen ) it will either create a new cycle or it will terminate on the path or cycle created from @xmath202 . call @xmath209 and @xmath210 the states in the second path and cycle ( and @xmath211 and @xmath212 the respective numbers of states , noting that @xmath213 is possible if the new path terminates on @xmath203 , and in fact that @xmath214 is also possible , if @xmath215 ) . if we continue in this manner we will have @xmath0 sets @xmath216 . call @xmath217 the union of these sets , denote as @xmath218 the number of states in @xmath217 , and call @xmath219 the number of sets @xmath220 which are not empty . we denote as @xmath165 the event that the @xmath72th path created in such a manner terminates on itself , and note that , if this does not occur , then @xmath221 . in the discussion which follows , until stated otherwise , we assume that ( 2 ) holds . [ moments ] @xmath222 and @xmath223 . choose any state @xmath202 . we must have : @xmath224 this means that , for large @xmath0 , the expected value of @xmath206 ( over the prior distribution for @xmath22 ) can be approximately expressed , making use of stirling s approximation @xmath225 , as : @xmath226 we have also used the fact that a poisson distribution with parameter @xmath0 will , as @xmath0 becomes sufficiently large , be well approximated by a normal distribution with mean @xmath0 and standard deviation @xmath227 . hence we can replace the first and second sum in the third equality by the second and first raw moment respectively of a half normal distribution with variance @xmath0 . ( the error associated with the stirling approximation is less than order @xmath68 . ) similarly for the variance , we first calculate the expectation of @xmath228 : @xmath229 as a result : @xmath230 note that the expectation can also be derived from the solution to the `` birthday problem '' : the solution to the birthday problem gives the expectation of @xmath205 , and since each cycle length ( less than or equal to @xmath205 ) has equal probability when conditioned on this total path length , we can divide the average by @xmath231 . [ genmoments ] @xmath232 and @xmath233 . we will have : @xmath234 and for the variance : @xmath235 where we have used the fact that the covariance term must be negative for any pair of lengths @xmath236 and @xmath237 , since if @xmath236 is greater than its mean the expected length of @xmath237 must decrease , and vice versa . supposing that @xmath15 and @xmath22 are no longer deterministic then we can still define the sets @xmath238 and @xmath220 for @xmath15 and @xmath22 by considering the _ most probable _ action and the _ most probable _ transition . indeed , if ( 1 ) and ( 3 ) hold , we can set @xmath199 sufficiently low so that the agent will spend an arbitrarily large proportion of its time in @xmath217 . note that if @xmath22 is deterministic , the transition matrix generated by @xmath15 and @xmath22 is not guaranteed to be irreducible or aperiodic , in which case @xmath50 may not exist . this is just a technicality , as a state distribution which is periodic but still restricted to a small number of states does not violate any of our conclusions . the pasa algorithm , provided @xmath1 is large enough and provided a subset of the state space has sufficiently high probability and is sufficiently small , will be such that the majority of states with high probability will be represented individually in the final set of basis functions . we can now use this fact , in conjunction with the results we ve generated above regarding the distribution of @xmath50 , to demonstrate that pasa will generate a set of basis functions which will be such that the resulting estimate @xmath239 will have an arbitrarily low error @xmath52 . [ error ] for all @xmath137 and @xmath240 , there is sufficiently large @xmath0 and sufficiently small @xmath199 such that pasa in conjunction with a suitable rl algorithm will provided @xmath241 for some @xmath242 generate , with probability no less than @xmath142 , a vf estimate with @xmath243 . using chebyshev s inequality , and lemmas [ moments ] and [ genmoments ] , for any @xmath242 we can choose @xmath0 sufficiently high so that @xmath244 with probability no greater than @xmath144 . to see this , take @xmath245 , @xmath246 and @xmath247 . we have : @xmath248 where in the first and second inequalities we assume @xmath0 is sufficiently large so that all but the highest order terms in lemma [ genmoments ] can be ignored . since @xmath17 is bounded and @xmath249 then for any @xmath127 and @xmath0 we can also choose @xmath199 so that @xmath52 summed only over states not in @xmath217 is no greater than @xmath250 . we choose @xmath199 so that this is satisfied , but also so that @xmath251 for all elements of @xmath252 . this is possible since each such @xmath58 will be bounded from below for all @xmath253 . now provided that @xmath254 then each state in @xmath217 will eventually be in its own cell . the rl algorithm will have no error for each such state so @xmath52 will be no greater than @xmath250 . we can prove in almost identical fashion that , under the same assumptions , both ( i ) mse _ and _ ( ii ) @xmath52 redefined so that all actions are weighted equally , i.e. equation ( [ mseapprox ] ) with the factor @xmath15 removed , will similarly be arbitrarily close to zero with arbitrarily high probability . the bound on @xmath1 provided represents a significant reduction in complexity when @xmath0 starts to take on a size comparable to many real world problems , and could make the difference between a problem being tractable and intractable . it also seems likely that the bound on @xmath1 in theorem [ error ] can be improved upon ( see the discussion in section [ simulation ] ) . conditions ( 1 ) and ( 3 ) are commonly encountered in practice , in particular ( 3 ) which can be taken to reflect a `` greedy '' policy . condition ( 2 ) can be interpreted as the transition function being `` completely unknown '' ( and can be generalised somewhat , see below ) . we might ask what happens when @xmath15 is no longer generated randomly , but rather , for example , according to a process of policy iteration . it is intuitively plausible that this should typically have the effect of reducing @xmath255 . we will not offer a general proof . however note that , if for example @xmath256 states have reward of one , for @xmath257 , and all other states have reward of zero , the agent must determine a shorter path on average than a random path between each pair of such states ( otherwise its performance will be no better than a random policy ) . denote as @xmath258 the average path length between reward states determined by an agent which is updating its policy so as to increase reward . by the reasoning we ve just stated @xmath258 will not increase above a bounded value as @xmath0 increases . we can then apply the same reasoning as the random policy case to conclude that at most @xmath259 such states will be visited under such a policy on average , and that @xmath260 . on a final related note , _ fixed _ state aggregation will _ also _ tend to have zero error if @xmath261 and all our other assumptions continue to hold . this is because the probability of more than one state falling into a cell tends to zero if the number of cells grows at a faster rate than the size of @xmath217 . of course the probability of all states falling into a unique cell increases much more slowly than is true for pasa as @xmath0 is increased . alterations can be made to the pasa algorithm ( with no impact on computational complexity ) that can remove the @xmath262 factor in theorem [ error ] . however such an alternative algorithm is more complex to describe and unlikely to perform noticeably better in a practical setting . there is some scope to extend our results beyond uniform transition function priors . some knowledge about the value of @xmath22 ( i.e. a non - uniform prior ) can have important implications for our results . if we know , for example , that @xmath263 for all @xmath264 and all @xmath37 , and @xmath265 for all @xmath37 , then @xmath266 , and lemma [ moments ] clearly does nt hold . there are , however , more general sets of priors for which the results will hold . our interest is in how the prior influences the values of the moments which we calculated ( for uniform priors ) in lemmas [ moments ] and [ genmoments ] . we again assume a deterministic transition function , and a fixed deterministic policy @xmath15 . in our discussion around uniform priors , we were also able to assume that the sequence of states @xmath267 used to generate the sets @xmath268 were selected arbitrarily . in a more general setting we may need to assume that this sequence is generated according to some specific probability distribution . since , in the discussion below , we will generally assume that the prior distribution of the transition probabilities is identical for different actions , we will be able to continue to assume that @xmath15 is arbitrary . at the expense of imposing heavier constraints on @xmath15 . ] appealing to techniques which make use of the notion of schur convexity ( marshall et al 2011 ) , it s possible to show that , if the random vector @xmath200 is independently distributed for all @xmath21 , and @xmath269 for all @xmath270 , then , given an arbitrary policy @xmath15 , and assuming the sequence of starting states @xmath267 are distributed uniformly at random , @xmath271 and @xmath272 are minimised where the prior for @xmath22 is uniform . using this fact , theorem [ error ] can be extended to such priors ( we omit the details , though the proof uses arguments substantially equivalent to those used in theorem [ error ] ) . if we continue to assume an arbitrary policy and that the starting states are selected uniformly at random , we can consider a yet more general class of priors , using a result from karlin and rinott ( 1984 ) . their result can be used to demonstrate that , of the set of priors which satisfy the following three conditions ( i ) that @xmath273 for all @xmath274 , ( ii ) that : @xmath275 for all @xmath276 , and ( iii ) the random vector @xmath200 is independently distributed for all @xmath21 the uniform prior will again maximise the values @xmath271 and @xmath272 . since @xmath277 , an equivalent result to theorem [ error ] ( though not necessarily with the same constant @xmath278 ) can similarly be obtained for this even larger set of priors ( we again omit the details and note that the arguments are substantially equivalent to those in theorem [ error ] ) . both these results assume a degree of similarity in the transition prior probabilities for each state . a perhaps more interesting potential generalisation is as follows . we can define a _ balanced _ prior for a deterministic transition function @xmath22 as any prior such that the random vector @xmath200 is independently distributed for all @xmath21 , and we have @xmath273 , @xmath279 and @xmath280 for all @xmath274 . in essence , the prior probability of transitioning from state @xmath11 to @xmath23 is the same as transitioning in the reverse direction from @xmath23 to @xmath11 . this sort of prior would be reflective of many real world problems which incorporate some notion of a geometric space with distances , such as navigating around a grid and as such represents an important generalisation . the difference to the uniform prior is that we now have an expectation that some states will be `` close '' to one - another , and other states will be further apart . however similar to the uniform prior there is no inherent `` flow '' creating cycles which have larger expected value than @xmath218 in the uniform case . it is not hard to conceive of examples where @xmath281 may be significantly reduced for a particular balanced prior compared to the uniform case . it would furthermore appear plausible that , amongst the set of all balanced priors , @xmath271 would be maximised for the uniform prior . indeed investigation using numerical optimisation techniques demonstrates this is the case for @xmath282 , even when a fixed arbitrary starting state is selected relative to the balanced set of transition probabilities . the techniques used for the generalisations stated above can not be used to prove a similar result for balanced priors . is maximised for all @xmath47 by a uniform prior , from which our conclusions regarding the moments follow . in the case of balanced priors such a strong result does not hold , which can be seen by , for large @xmath0 , and taking @xmath283 , comparing a uniform prior to a prior where all transitions outside of a single fixed cycle covering all @xmath0 states have probability zero , and where the prior probability of a transition in either direction along this cycle is @xmath284 . ] we conjecture , based on our numerical analysis , that the uniform prior does maximise @xmath271 for all @xmath0 , which would carry the implication , since @xmath285 , that lemma [ moments ] can be used to argue @xmath286 and @xmath287 . even if this conjecture holds , we can not extend theorem [ error ] to balanced priors , which we can see with a simple example . set @xmath288 for all @xmath21 where @xmath118 is small . provided that the transition matrix associated with @xmath15 and @xmath22 is irreducible , then @xmath289 for all @xmath0 . notwithstanding that a formal result equivalent to theorem [ error ] is unavailable for balanced priors , by selecting suitable parameter values we should still be able to exploit the apparent tendency of the agent to spend a majority of the time in a small subset of the state space . the main difference is that this small subset may change slowly over time . the situation may become further altered in our favour once policy iteration is introduced , and policies are no longer random but rather target states with reward . our experimental results focus on transition functions drawn from a uniform distribution , however further research could help indicate the extent to which pasa can generate a vf estimate with low error , subject to the slightly altered dynamics introduced by balanced priors . our main objectives in this section are to : ( a ) test empirically the tendency for an agent to spend the majority of its time in a small subset of the state space , and ( b ) conduct an experimental comparison of pasa to fixed state aggregation , to demonstrate that , on average , pasa will help generate vf estimates with lower mse ( and therefore that the theoretical advantages to unsupervised basis function adaptation can be realised in a practical setting ) . whilst the bound we placed on @xmath1 in theorem [ error ] may be of help when @xmath0 is very large , note that if we set @xmath290 ( i.e. such that we have a fifty per cent chance of being guaranteed to be able to represent every high probability state ) , then we will have @xmath291 up until the point where @xmath292 . it appears likely that the bound in theorem [ error ] is a loose bound . we conduct some simple experiments to examine how @xmath218 behaves , in particular for lower values of @xmath0 . in figure [ empiricalc ] we report on the outcome of a sequence of @xmath293 independent trials where @xmath218 was calculated explicitly , in order to test our theoretical bounds . the results demonstrate that the estimates for @xmath281 and @xmath294 are accurate , but that the bound in lemma [ genmoments ] is ( perhaps not surprisingly ) generous . in figure [ empiricalc ] we have also run equivalent tests on a square grid world environment for all @xmath72 and @xmath37 ( the agent is forced to move in one of four directions : up , down , left or right ) . this violates the exact balanced definition , however would only serve to increase @xmath295 . ] ( we have used @xmath296 , @xmath297 and @xmath298 to distinguish these values from those obtained for the uniform prior ) . this helps to reinforce the discussion around balanced transition function priors ( we can see that @xmath299 is very small compared to the uniform case , and may in fact be bounded as a function of @xmath0 , however that @xmath300 , and therefore the value @xmath298 , appears to increase linearly as a function of @xmath0 ) . [ cols="^,^ " , ] we can also examine the empirical distribution of @xmath50 given a sample trajectory , to see the extent to which our section [ theoretic ] assumptions do indeed result in only a few states having high probability . this is worth exploring since the value of @xmath199 strictly required by theorem [ error ] may be too low to be realistic in practice ( i.e. , given a too - large value for @xmath199 , we might expect to see the stationary distribution `` leak '' to include states outside @xmath217 ) . assume that the agent follows an @xmath301-greedy policy . we define : @xmath302 hence @xmath303 is the minimum number of states required to amass a proportion @xmath304 $ ] of the total probability of the stationary state distribution ( as @xmath305 approaches @xmath68 and @xmath199 approaches @xmath306 , the value can be considered as a rough experimental equivalent of @xmath218 ) . figure [ empiricalm ] demonstrates that , when @xmath301 is held constant as @xmath0 is increased , we start to see what appears to be a linear increase in @xmath303 as a function of @xmath0 . it seems likely that this is because of leakage into states outside @xmath218 . as @xmath307 becomes smaller and @xmath301 remains the same , this leakage will become more pronounced . whilst this does nt invalidate any of the underlying principles , in a practical setting it is something to remain aware of . even where @xmath301 is reasonably high , e.g. @xmath308 , and where we require a high value of @xmath305 , e.g. @xmath309 , only a reasonably small number of states is required to obtain a total probability of @xmath305 , e.g. around @xmath310% of states for @xmath311 , implying that in general we can set @xmath1 comparatively low . the implication of these results is that we expect we should be able to use pasa to effectively reduce error in the vf estimate in a practical setting ( including where @xmath0 is low , for example less than @xmath293 ) . this is what we next investigate . as a final comment , whilst we need @xmath312 to theoretically guarantee capturing every high - probability state individually , the factor of @xmath313 is not likely to be essential in a practical setting ( as it covers the worst case scenario in terms of how the high - probability states are arranged ) . we have run a series of experiments to test the performance of pasa compared with simple fixed state aggregation . before stating the results it will be helpful to make some comments first . if @xmath113 approaches @xmath68 and @xmath199 is too large as a function of @xmath113 , then we will find that all values @xmath314 will approach the same value , for any reward function @xmath17 . this is because the discounted reward is heavily weighted into the future , and due to randomness in the agent s trajectory , future reward will not be heavily impacted by each individual decision . hence , under such conditions , both mse and @xmath52 for fixed state aggregation will _ also _ tend towards arbitrarily low values . furthermore , if , given a state aggregation architecture , the number of high probability states in each cell is large , then the average value for each state - action over each cell will also approach the same value ( this is due to the law of large numbers ) . the second point is important when calculating @xmath52 , since if @xmath43 is such that all cells are large ( and particularly if @xmath113 is close to @xmath68 ) then each estimate @xmath315 will tend to a single mean value , and @xmath52 will take on a low value ( even if mse is high , and the vf estimate is poor ) . for these reasons , we have taken care not to set @xmath113 too close to @xmath68 ( keeping in mind the value of @xmath199 ) , and we have used mse to compare pasa to fixed state aggregation . since mse is a more accurate measure , this will only serve to strengthen the results . the only downside is that , since an exact vf estimate is impossible to calculate for large @xmath0 , we can only perform our comparison for relatively low values of @xmath0 . we then use our theoretical insights to infer that similar performance differences will exist for large @xmath0 . in all cases we have reported the square root of the mse ( as this provides a less distorted comparison ) . the two algorithms we will be comparing are pasa combined with a sarsa which we then weighted by the inverse of the probability of the action selected . such normalisation is not strictly necessary , however it allows the vf estimate for infrequently selected actions to converge more quickly . the random policy @xmath15 was chosen such that the same action @xmath13 is the `` preferred '' action ( the action taken with probability @xmath316 ) for all states in each particular cell . ] rl algorithm , and a sarsa rl algorithm with fixed state aggregation , where the cells are sized as equally as possible . it can be shown that sarsa will generate a vf estimate with minimal mse given a fixed policy and a fixed linear approximation architecture ( bertsekas and tsitsiklis 1996 ) . we have tested both algorithms with sequences of randomly generated environments ( where @xmath22 is deterministic and with a uniform prior , the prior for @xmath17 is such that the reward for each state - action pair has an independent standard normal distribution , and @xmath15 is @xmath301-greedy ) , and then estimated ( by sampling ) the mse of the estimate vf generated by the rl algorithm . for each algorithm we ran eight separate experiments , for @xmath0 set at @xmath317 , @xmath318 , @xmath319 and @xmath293 , and for @xmath320 and @xmath321 , for @xmath322 randomly generated trials each . the running time was slightly less than seven hours for a single trial on an intel(r ) xeon(r ) cpu e5 - 4650 0 @ 2.70ghz for both algorithm variants , although this includes the time to calculate beforehand an exact solution ( to allow for mse to be calculated ) . ] in all cases both algorithms were tested against an identical set of environments . for each value of @xmath0 listed above we set @xmath1 equal to @xmath8 , @xmath322 , @xmath323 and @xmath324 respectively ( and @xmath64 equal to @xmath325 , @xmath326 , @xmath327 and @xmath328 ) , guided in part by figure [ empiricalm ] and the other comments above ( but also in part heuristically ) . to approximately half @xmath1 appears to be a reliable way of ensuring that a large number of high probability states fall into a singleton cell . smaller values of @xmath64 tend to exhaust many cells in seeking to isolate ( into smaller and smaller cells ) one or two individual states , whereas greater values of @xmath64 may mean that pasa has too little control over the final set of basis functions . ] the results of these experiments are shown in table [ resultstable ] ( and figure [ timeseries ] for @xmath320 ) . the values of other parameters used were as follows : @xmath329 , @xmath330 , @xmath331 , @xmath332 and @xmath333 . with respect to sampling the mse , we compared the vf estimate @xmath239 for each state visited to the exact @xmath26 for the policy @xmath15 , which for each experiment was calculated in advance exactly by solving the system of equations involving @xmath32 using least squares . quite a large number of iterations were used in each trial ( @xmath318 million ) . this was in part to accommodate the most complex environment ( @xmath334 ) , and also to provide clear evidence that the estimates have stabilised . such long trials are not necessary to see a significant difference between the two algorithms ( see figure [ timeseries ] ) . minimising the number of iterations required to see a strong difference between pasa and fixed state aggregation would depend on optimising parameters like @xmath160 , @xmath301 and @xmath114 subject to the choices for @xmath113 and @xmath0 . p1.3cmp0.9cmp0.4cmlllllll experiment & @xmath0 & @xmath1 & & & + & & & @xmath335 & @xmath336 & @xmath335 & @xmath336 & @xmath335 & @xmath336 + @xmath68 & * 250 * & @xmath8 & @xmath337 & @xmath338 & @xmath339 & @xmath340 & * 67.7% * & * 56.9*% + @xmath231 & * 500 * & @xmath322 & @xmath341 & @xmath342 & @xmath343 & @xmath344 & * 63.2% * & * 55.6*% + @xmath345 & * 750 * & @xmath323 & @xmath346 & @xmath347 & @xmath343 & @xmath348 & * 59.4% * & * 45.8*% + @xmath349 & * 1,000 * & @xmath324 & @xmath350 & @xmath351 & @xmath352 & @xmath353 & * 55.9% * & * 42.4*% + we take the average mse over the last fifth of each trial . since the theoretic results assume , for example , that @xmath114 and @xmath301 can be made arbitrarily low , in a practical setting we wo nt see mse falling to zero . however table [ resultstable ] demonstrates that , at a low computational cost , we are able to significantly reduce mse from the fixed state aggregation case . the table shows us that for @xmath320 error can be reduced by around @xmath7 to @xmath354 per cent , and for the less complex case of @xmath321 by as much as approximately @xmath326 to @xmath8 per cent . as noted above , the effect of pasa becomes slightly less pronounced as @xmath113 approaches @xmath68 ( in particular if @xmath301 is comparatively large ) . figure [ timeseries ] shows that , as @xmath0 increases , pasa can make the algorithm take slightly longer to generate a vf with low mse ( which might be expected since the pasa component needs time to converge ) , however in the longer term mse is significantly reduced . optimising parameters such as @xmath114 , @xmath301 or even the value of @xmath1 ( which we ve set well below the bound in section [ theoretic ] ) might be expected to further increase the disparity between pasa and fixed state aggregation ( the latter can not be improved since it has no parameters to adjust outside the underlying rl algorithm ) . as we noted in the introduction , whilst basis function adaptation methods have been the subject of a large amount of research recently , unsupervised methods have not received much attention ( compared to methods which seek to estimate vf error explicitly ) . this may be an oversight . as we have seen there are many circumstances reflecting real - world problems where unsupervised methods can be very effective in creating an approximation architecture , and thereby help to generate more accurate vf estimates ( both in theory and in practice ) . what distinguishes unsupervised methods from more complex alternatives , however , is their simplicity . as we have seen , an algorithm such as pasa carries only minimal additional costs ( both in terms of sampling and computational demands ) compared to the rl algorithm which it supports . it is their effectiveness combined with their low cost which , in our view , make such techniques a promising candidate for further research . in the setting of policy improvement the advantages offered by unsupervised methods have the potential to be particularly important . each policy update requires a vf estimate ( where the vf estimate is based on the current policy ) however each policy update will change the vf ( perhaps significantly ) , requiring a new estimate . so accurately and _ efficiently _ estimating @xmath26 for each new policy @xmath15 is critical . some initial experimentation suggests that the pasa algorithm can have a significant impact on rl algorithm performance where policy updates are introduced . this should perhaps not come as a surprise given the reduced error in @xmath239 shown in section [ simulation ] . the pasa algorithm represents only a relatively naive or simplistic application of the idea of unsupervised basis function adaptation ( i.e. the idea of using visit frequency to guide which parts of the vf to represent in more detail ) . it appears likely that a number of improvements could be made to the algorithm in order to further optimise its performance . is binomially distributed , and this number will have low variance if @xmath64 is substantially lower than @xmath255 , it seems likely that , in choosing a cell to split , we could initially only consider a subset of the set of all available cells . this would potentially allow pasa to converge more quickly , and be less likely to change as a result of random fluctuation in the value of @xmath75 . ] from a theoretical perspective the assumption around scoring functions being weighted by @xmath50 is crucial . the nature of the vf estimate generated by pasa and its associated rl algorithm is that the vf will be well estimated for states which are frequently visited under the existing policy . however this results in poorer estimates of the value of deviating from the current policy . thus , even though the expected vf error may be low , it does not immediately follow that an algorithm can use this to optimise its policy via standard policy iteration ( since the consequences of deviating from the current policy are less clearly represented ) . ultimately , however , the theoretical implications of the improved vf estimate in the context of policy iteration are complex , and would need to be the subject of further research .

in the context of mod - gaussian convergence , as defined previously in our work with j. jacod , we obtain asymptotic formulas and lower bounds for local probabilities for a sequence of random vectors which are approximately gaussian in this sense , with increasing covariance matrix . this is motivated by the conjecture concerning the density of the set of values of the riemann zeta function on the critical line . we obtain evidence for this fact , and derive unconditional results for random matrices in compact classical groups , as well as for certain families of-functions over finite fields .

0912.3237

it is well - known ( see , e.g. , ( * ? ? ? * th . 11.9 ) ) that , for @xmath1 , the set of values @xmath2 , @xmath3 , is dense in the complex plane . in fact , much more is true : it was proved by bohr and jessen that there exists a borel probability measure @xmath4 on @xmath5 , such that the support of @xmath4 is the whole complex plane , and such that the convergence in law @xmath6 holds for @xmath7 continuous and bounded . the corresponding density question for @xmath8 is , however , still open ( it was apparently first raised by ramachandra during the 1979 durham conference , but seems to appear in print only in heath - brown s note in @xcite ) : the difficulty is that the values @xmath9 , @xmath10 , do not have a limiting distribution , as evidenced already by the hardy - littlewood asymptotic @xmath11 or by selberg s result that @xmath12 , @xmath10 , is asymptotically normal with variance _ growing to infinity _ ( see also the work of ghosh @xcite for the imaginary part , and @xcite for a recent proof ) . in other words , `` most '' values of @xmath9 are rather large , though the zeta function is zero increasingly often as the imaginary part grows . in this paper , we show ( corollary [ cor - zeta ] ) how the density of values of zeta on the critical line would follow rather directly from a suitable version of the keating - snaith moment conjectures , which we viewed in our previous work with j. jacod @xcite as a refined version of the gaussian model . in fact , under suitable assumptions , we could prove a quantitative result , bounding from above the smallest @xmath13 for which @xmath9 lies in a given open disc in @xmath5 . this argument is based on a very general probabilistic estimate proved in section [ sec - proba ] , which throws some light on the nature of the mod - gaussian convergence that we defined in @xcite . we hope that this result will be of further use . in another paper ( jointly with f. delbaen , see @xcite ) , it will be seen that one can weaken considerably the assumption needed in order to prove the density of values of @xmath9 ( but without quantitative information ) . as applications of the general result , we will also prove the following theorems in section [ sec - examples ] ( the precise versions are given there ) . [ th - rmt ] let @xmath14 be arbitrary , @xmath15 such that @xmath16 . there exists @xmath17 , which can be bounded explicitly , such that @xmath18 provided @xmath19 , where @xmath20 denotes probability haar measure on the unitary group @xmath21 , and the implied constant is absolute . [ th - eulerproduct ] define @xmath22 for @xmath23 and @xmath3 . let @xmath14 be arbitrary , @xmath15 such that @xmath16 . there exists @xmath17 , explicitly bounded , such that @xmath24 for all @xmath19 , where @xmath25 is the lebesgue measure and the implied constant is absolute . in a different direction , we obtain some evidence for the density of @xmath9 by looking at special values of families of @xmath0-functions over finite fields . in doing so , we also consider the analogue of theorem [ th - rmt ] for symplectic and orthogonal matrices . we refer to section [ sec - ff ] for precise statements and definitions , and only state here one appealing ( qualitative ) corollary : [ th - ff ] the set of central values of the @xmath0-functions attached to non - trivial primitive dirichlet characters of @xmath26 $ ] , where @xmath27 ranges over primes , is dense in @xmath5 . * notation . * as usual , @xmath28 denotes the cardinality of a set . by @xmath29 for @xmath30 , or @xmath31 for @xmath30 , where @xmath32 is an arbitrary set on which @xmath33 is defined , we mean synonymously that there exists a constant @xmath34 such that @xmath35 for all @xmath36 . the `` implied constant '' refers to any value of @xmath37 for which this holds . it may depend on the set @xmath32 , which is usually specified explicitly , or clearly determined by the context . similarly , @xmath38 for @xmath30 means @xmath29 and @xmath39 , both for @xmath30 . we write @xmath40 the pochhammer symbol . * acknowledgments . * the first version of this paper was written while the first author was on sabbatical leave at the institute for advanced study ( princeton , nj ) ; many thanks are due to this institution for its support . this material is based upon work supported by the national science foundation under agreement no . dms-0635607 . the second author was partially supported by snf schweizerischer nationalfonds projekte nr . 200021 119970/1 . thanks to k. soundararajan for pointing out that the density conjecture for @xmath9 might be a good problem to study using complex moments of the zeta function , and to r. heath - brown for explaining us the history of the question . many thanks also to n. katz for discussions and explanations surrounding issues of monodromy computations . many thanks also to f. delbaen for useful discussions concerning the underlying probabilistic framework . the graphs were produced using sage 4.2 @xcite , relying on the barnes function routines in the mpmath package . in this section , which is purely probabilistic , we present two versions of `` local '' bounds for probabilities in the case of sufficiently uniform mod - gaussian convergence of sequences of random vectors . this may be compared with the local central limit theorem ( see , e.g. , @xcite for the one - dimensional case ) . in fact , as f. delbaen pointed out , one can obtain qualitative statements that generalize both the standard local central limit theorem and recover our results below under weaker assumptions . however , our emphasis is on explicit quantitative lower bounds for local probabilities . we first introduce the definition , generalizing @xcite to random vectors . fix some integer @xmath41 , and let @xmath42 be a sequence of @xmath43-valued random variables defined on a probability space @xmath44 ( as is the case for convergence in law , we could work without change with random variables defined on different probability spaces ) . let @xmath45 be a sequence of non - negative quadratic forms on @xmath43 . the sequence @xmath42 is then said to be convergent in the mod - gaussian sense with covariance @xmath46 and limiting function @xmath47 if @xmath48 locally uniformly for @xmath49 ; @xmath47 is then a function continuous at @xmath50 and @xmath51 . here , @xmath52 denotes the standard inner product on @xmath43 . the intuitive meaning is that , in some sense , @xmath53 is `` close '' to a ( centered ) gaussian vector @xmath54 with covariance matrix @xmath46 . as in @xcite , this notion is of most interest if the covariance `` goes to infinity '' . however , in contrast with the case of @xmath55 , this can mean different things because there is more than a single variance parameter involved . to discuss this , we diagonalize @xmath46 in an orthonormal basis . ] in the form @xmath56 where @xmath57 is the necessary ( orthogonal ) change of variable . then `` @xmath46 goes to infinity , '' in the weakest sense , means that the largest eigenvalue @xmath58 goes to @xmath59 as @xmath60 . we are interested in the distribution of values of @xmath53 as @xmath61 grows ; clearly , if ( say ) the @xmath46 are already diagonalized in the canonical basis and @xmath62 is constant , these values will have first coordinate much less spread out than the last ones . to simplify our discussion , and because this is the situation in our applications , we will assume this behavior does not occur and that in fact the smallest eigenvalue goes to infinity . for simplicity , we will assume in fact that for some fixed @xmath63 , we have @xmath64 so that also @xmath65 ( we say that the convergence is _ balanced _ ) . in our main applications , this will be the case with @xmath66 . note in particular that it follows that the discriminant @xmath67 goes to infinity as @xmath60 , and moreover @xmath68 our question is now the following : given @xmath42 , @xmath69 , as above , with this type of mod - gaussian convergence , can we bound from below the probability @xmath70 where @xmath71 is a fixed open set in @xmath43 ? denoting by @xmath72 the dual quadratic form , the gaussian model suggests that , if @xmath71 is relatively compact ( e.g. , some non - empty open ball ) , we could expect @xmath73 as @xmath60 , since @xmath65 implies that @xmath74 for all @xmath75 . we will confirm that this holds in certain conditions at least . we strive especially for lower bounds on @xmath76 , which we wish to be quantitative , so that we can determine some @xmath77 ( depending explicitly on @xmath71 ) for which @xmath78 note that such a quantitative result must depend on the location of the open set @xmath71 , whereas the limit itself only depends on the volume , as seen above . the specific hypothesis we use may seem somewhat arbitrary , but they turn out to be satisfied ( with room to spare ) in the later applications . [ th-2 ] let @xmath41 be fixed and let @xmath42 be a sequence of @xmath43-valued random variables defined on @xmath44 , such that @xmath42 converges in the mod - gaussian sense with covariance @xmath69 , and that the convergence is @xmath79-balanced with @xmath63 , with @xmath80 for all @xmath61 . let @xmath81 be gaussian random variables with covariance matrices given by @xmath69 , so that @xmath82 assume moreover the following three conditions : _ ( 1 ) _ there exist constants @xmath83 , @xmath84 and @xmath85 such that , for any @xmath23 and @xmath49 such that @xmath86 , we have @xmath87 _ ( 2 ) _ the function @xmath47 is of class @xmath88 on @xmath89 . _ ( 3 ) _ for some @xmath90 and @xmath91 , we have @xmath92 . let @xmath93 be any number such that @xmath94 then , for any fixed non - empty open box @xmath95 with width @xmath96 such that @xmath97 , we have @xmath98 for @xmath23 , where the implied constant depends only on @xmath99 and the implied constant in _ ( [ eq - strong - convergence])_. in particular , for any fixed non - empty open set @xmath100 , we have @xmath101 provided @xmath19 , where @xmath77 and the implied constant depend on @xmath71 and the same data as above . note the following elementary lower bound , valid if @xmath102 : @xmath103 where the implied constant depends only on @xmath104 ; this is where the location of @xmath71 enters , since the error term will only be smaller than this , roughly , when @xmath105 . the growth condition ( [ eq - order-2 ] ) is in fact a consequence of the uniform mod - gaussian convergence , at least provided the sequence @xmath106 does not grow too fast . for instance , if @xmath107 for all @xmath23 , for some constants @xmath108 , @xmath109 , we can obtain ( [ eq - order-2 ] ) with @xmath110 . indeed , we can write @xmath111 if @xmath61 is large enough and @xmath86 . note then that @xmath112 we now fix @xmath23 minimal such that @xmath113 and if this value of @xmath61 is large enough , we get @xmath114 , and hence @xmath115 as desired . on the other hand , if this chosen @xmath61 is too small , @xmath116 is bounded , and the desired estimate is trivial . let @xmath117 be the smallest eigenvalue of @xmath46 , so that @xmath118 for all @xmath49 and @xmath23 . for simplicity , we denote also @xmath119 as in ( [ eq - strong - convergence ] ) . we now first fix @xmath120 such that @xmath121 , and then fix a smooth , compactly supported function @xmath122 on @xmath123 such that @xmath124 where @xmath125 and the implied constant depends only on @xmath126 ( we will define @xmath120 to be a function of @xmath61 at the end , and hence must be careful to have estimates uniform in terms of @xmath120 ; this is provided by using only the above properties of @xmath122 ; the maximal value of @xmath126 used will also be bounded only in terms of @xmath104 , @xmath47 and the data in ( [ eq - strong - convergence ] ) ) . it is classical that such a function exists ( examples are constructed in @xcite ) . then define @xmath127 for @xmath75 . it follows that @xmath128 next , we define @xmath129 and we start our argument with the obvious inequality @xmath130 where @xmath131 is the law of @xmath53 . applying the plancherel formula , we get @xmath132 where @xmath133 is the characteristic function of @xmath53 and @xmath134 denotes the fourier transform of @xmath33 ( the smoothness of @xmath33 guarantees that @xmath135 is in @xmath136 , so the plancherel formula is valid by a simple fubini argument ) . we have @xmath137 since @xmath138 for @xmath139 and @xmath140 ( by repeated integration by parts ) , we find that @xmath141 the implied constant depending on @xmath126 . using the formula for @xmath142 , selecting for given @xmath143 an index @xmath126 so that @xmath144 and applying the second upper bound above for this index only , if @xmath145 , we derive @xmath146 for any fixed @xmath147 , where the implied constant depends only on @xmath104 and @xmath148 . in particular , for @xmath149 , we will use simply the upper bound @xmath150 . we now proceed to approximate . first of all , for any radius @xmath151 , we can estimate the contribution of those @xmath143 with @xmath152 using the estimate above with a value of @xmath147 which will be determined later . after integrating over @xmath153 , we obtain @xmath154 for any @xmath151 . after selecting @xmath155 , we obtain @xmath156 for @xmath23 , the implied constant depending only on @xmath157 . on the other hand , provided @xmath158 , we use ( [ eq - strong - convergence ] ) and ( [ eq - order-2 ] ) to get @xmath159 for @xmath23 , using again the definition of @xmath33 , the implied constant depending on @xmath47 . by ( [ eq - uniformity ] ) , the last term can be bounded by @xmath160 for @xmath23 if @xmath161 , the implied constant depending on @xmath162 . we then split the first term further in two parts , namely where @xmath163 , and where @xmath164 . the parameter @xmath165 will be chosen later , in such a way that the region @xmath166 ( which is inside @xmath167 ) contains the region @xmath163 ( which is a neighborhood of @xmath50 that contracts to @xmath50 as @xmath60 , if @xmath165 does not grow too fast , since it is an ellipsoid with longest axis @xmath168 ) . the second part of the integral is bounded by @xmath169 and in the first part , we use the approximation @xmath170 for @xmath166 , coming from the @xmath88 assumption on @xmath47 , to get @xmath171 the implied constant depending on @xmath47 , where the last integral was estimated using latexmath:[\[\|t\|^2{\leqslant}\frac{1}{\delta_n}q_n(t){\leqslant}\frac{\kappa_n^2}{\delta_n},\quad\quad we can rewind the computation for the first term , with the gaussian @xmath54 instead of @xmath53 : for @xmath23 , we have @xmath173 where the implied constant is absolute , and then we write @xmath174 for @xmath23 , where the implied constant depends only on @xmath104 ( the last step is obtained using the density of the gaussian @xmath54 ) . summarizing , we have found @xmath175 ( where we recall that @xmath176 ) . now if @xmath177 , and @xmath178 we have first ( see ( [ eq - mmu ] ) ) the condition @xmath179 and moreover the third error term is absorbed in the second one , while the last becomes @xmath180 thus if we select any @xmath181 , the previous conditions on @xmath148 hold , and we find that @xmath182 now , we attempt to select @xmath120 to equalize the error terms involving it , i.e. , so that @xmath183 which translates to @xmath184 and two cases arise : \(i ) if @xmath185 , we have @xmath186 ( as is necessary to define @xmath120 in this way ) , and we obtain from the above that @xmath187 where @xmath188 ( note that , since @xmath189 is assumed to be @xmath190 , we have @xmath191 . ) \(ii ) if we have @xmath192 , we take simply @xmath193 and obtain @xmath194 where the implied constant depends on @xmath104 and @xmath148 . the combination of the two cases leads to the lower - bound in ( [ eq - quantitative ] ) . the upper bound is proved similarly , using instead of @xmath122 a function which is @xmath195 for @xmath196 and @xmath197 for @xmath198 for some suitable @xmath199 ; we leave the details to the reader . [ rm - lower - bound ] if we are interested in obtaining a lower bound only ( which is the most interesting aspect in a number of applications ) , it is simpler and more efficient to fix , e.g. , @xmath193 , from the beginning of the proof . for @xmath200 this leads for instance to @xmath201 for any @xmath202 ( by taking @xmath148 large enough depending on @xmath203 ) , where @xmath204 and the implied constant depends on @xmath47 , @xmath205 and @xmath203 . [ rm - tv ] from the probabilistic point of view , this proposition gives one answer , quantitatively , to the following type of question : given a sequence @xmath42 of ( real - valued ) random variables and parameters @xmath206 such that @xmath207 converges in law to a standard centered gaussian variable ( with variance @xmath208 ) , to what extent is @xmath53 itself distributed like a gaussian with variance @xmath209 ? here , we perform the comparison by looking at @xmath76 , @xmath71 a fixed open set . and this shows clear limits to the gaussian model : for example , any integer - valued random variable @xmath53 will have @xmath210 for any open set not intersecting @xmath211 , and yet may satisfy a central limit theorem ( e.g. , the @xmath61-th step of a symmetric random walk on @xmath211 ) . even more precisely , denoting @xmath212 the kolmogorov distance , there exist integer - valued random variables @xmath53 with @xmath213 where @xmath54 is a centered gaussian with variance @xmath61 , which indicates a close proximity and yet , again , @xmath210 if @xmath214 . for instance , take @xmath53 with distribution function @xmath215 given by @xmath216 in particular , it is also impossible to prove results like theorem [ th-2 ] using only this type of assumption on the kolmogorov distance . of course , if we assume that @xmath217 with @xmath218 ( arbitrarily slowly ) , we get @xmath219 for @xmath220 $ ] , @xmath221 . but such an assumption is unrealistic in practice . for instance , if one assumes that @xmath42 converges in the mod - gaussian sense with covariance @xmath222 , and if the limiting function @xmath47 is @xmath88 and the convergence holds in @xmath88 topology , one can straightforwardly estimate the kolmogorov distance by @xmath223 which is comparable to ( [ eq - dist - k ] ) , but one can also check that this can not be improved in general to something like ( [ eq - impossible ] ) . in example [ ex - eigenvalue - count ] in section [ sec - examples ] , we will also describe a much deeper and more illuminating situation concerning the limits of what can be hoped , even with something like mod - gaussian convergence . [ rm - variant ] other variants could easily be obtained . in particular , it is clear from the proof that if @xmath47 has sub - gaussian growth , i.e. , we can take @xmath224 in ( [ eq - order-2 ] ) , the results can be substantially improved . however , in our main applications , this condition fails . also , one could use test functions @xmath33 which decay at infinity faster than polynomials to weaken the uniformity requirement in the convergence condition ( [ eq - strong - convergence ] ) ( for instance , for @xmath55 , it is possible to find @xmath33 which is smooth , non - negative and compactly supported in any fixed open interval and satisfies @xmath225 for any @xmath15 , as constructed , e.g. , in ( * ? ? ? 1.3.5 ) or @xcite ) . again , for our main unconditional applications , our conditions hold with room to spare , so we avoided this additional complexity . we can also introduce a linear term ( corresponding roughly to the expectation of @xmath53 ) in addition to the covariance terms in the definition of mod - gaussian convergence ( as in @xcite for @xmath55 ) , but this amounts to saying that @xmath42 converges in the mod - gaussian sense with covariance @xmath69 and mean @xmath226 , @xmath227 , if the sequence @xmath228 converges in our original sense above . but note that the interpretation of a lower bound for @xmath229 , as given by theorem [ th-2 ] for a fixed @xmath100 , is quite different , if @xmath230 is itself `` large '' . maybe one should see the statements in that case as giving natural examples of sets @xmath231 for which one knows that @xmath232 has the specific decay behavior @xmath233 as @xmath60 . in particular , lower bounds for @xmath229 do not give control of @xmath76 , and indeed this may be zero for all @xmath61 large enough ( see , for instance the example in section [ ssec - symplectic ] below of values at @xmath208 of characteristic polynomials of unitary symplectic matrices , which is always @xmath234 ) . we present now some applications of theorem [ th-2 ] ( in particular , proving theorems [ th - rmt ] and [ th - eulerproduct ] ) . we also give an example that illustrates the limitations of such results , suggesting strongly that one can not replace mod - gaussian convergence with the existence of the limits ( [ eq - converge ] ) only for @xmath143 in a neighborhood of the origin . [ ex - keating - snaith ] one of the canonical motivating examples of mod - gaussian convergence is due to keating and snaith @xcite . let @xmath235 where @xmath236 is a haar - distributed random unitary matrix in the compact group @xmath237 ; we view these random variables as @xmath238-valued ( via the real and imaginary parts ) , so if @xmath239 , we have @xmath240 we first clarify the choice of the branch of logarithm : @xmath53 is defined almost everywhere ( when @xmath208 is not an eigenvalue of @xmath236 ) , and for @xmath241 with @xmath242 , such that @xmath243 where the last logarithms are given by the taylor expansion around @xmath50 . this is the same convention as in ( * ? ? ? after ( 7 ) ) . keating and snaith show that @xmath42 satisfies @xmath244 for @xmath245 ( note the asymmetry between @xmath246 and @xmath247 ) , see ( * ? ? ? * eq . ( 71 ) ) , taking into account a slightly different normalization : their @xmath143 is our @xmath246 and their @xmath248 is our @xmath247 . it is also useful to observe that , in the connection with tplitz determinants ( see @xcite ) , this characteristic function is the @xmath61-th tplitz determinant corresponding to the symbol which is a pure fisher - hartwig singularity of type @xmath249 ( this is denoted @xmath250 in @xcite , where the formula ( [ eq - ks ] ) is stated as eq . ( 41 ) ; see @xcite for two elementary computations of the corresponding tplitz determinants . ) we rewrite this in terms of the barnes function @xmath251 , as is customary . we recall that @xmath252 is an entire function of order @xmath253 , such that @xmath254 and @xmath255 for all @xmath256 , and that its zeros are located at the negative integers . in particular , it satisfies @xmath257 for all @xmath23 and @xmath258 . thus , we have @xmath259 ( see , e.g. , ( * ? ? ? * eq . ( 41 ) ) ) . we now get from ( * ? ? ? 3.2 ) that @xmath260 where @xmath261 hence we have complex mod - gaussian convergence with limiting function @xmath262 here we can take @xmath66 in ( [ eq - grow - not - too - much ] ) . note that this is _ not _ the product of the two individual limiting functions for mod - gaussian convergence of the real and imaginary parts of @xmath53 separately ( which are @xmath263 and @xmath264 ) , although after normalizing , one obtains convergence in law of @xmath265 to _ independent _ standard gaussian variables , as noted by keating and snaith . we now check that theorem [ th-2 ] can be applied to the sequence of random variables @xmath42 . the fact that @xmath266 is of class @xmath88 on @xmath238 is clear in view of the analytic properties of the barnes function . condition ( 3 ) is also obvious . a uniformity estimate like ( [ eq - strong - convergence ] ) is not found in @xcite or @xcite , though it is proved for @xmath143 in a fixed compact region of @xmath238 in ( * ? ? ? 3.2 ) . in proposition [ pr - app ] in the appendix , we prove @xmath267 for @xmath268 . in view of @xmath269 , this is compatible with ( [ eq - strong - convergence ] ) and ( [ eq - uniformity ] ) , with @xmath270 arbitrarily large , @xmath271 and @xmath189 ( defined by ( [ eq - order-2 ] ) ) can be taken to be any @xmath272 . thus the constant @xmath273 in ( [ eq - quantitative ] ) can be any number @xmath274 moreover , since @xmath275 we see by using remark [ rm - lower - bound ] ( namely , ( [ eq - quant - lower ] ) ) and ( [ eq - lower - easy ] ) that we have the following corollary : for @xmath276 we have @xmath277 for all @xmath61 with @xmath278 where both implied constants are absolute . ( the first condition on @xmath61 ensures the main term is @xmath279 , while the second ensures that the error term is smaller ; we have taken @xmath280 for definiteness ; any number @xmath281 can replace @xmath282 ) . now we appeal to the following elementary lemma : [ lm - log ] let @xmath283 and @xmath15 , and denote @xmath284-\pi,\pi].\ ] ] then , provided @xmath16 , we have @xmath285 for all @xmath286 such that @xmath287 for given @xmath288 , non - zero , we get from this and ( [ eq - dense - quant - log ] ) , applied to @xmath289 and to @xmath290 instead of @xmath96 , the following explicit form of theorem [ th - rmt ] : [ th - rmt - quant ] let @xmath14 be arbitrary , @xmath15 such that @xmath16 . we have @xmath291 for @xmath292 , where @xmath293 where @xmath37 and the implied constants are absolute . it follows from asymptotic formulas for the barnes function , e.g. @xcite , that we have @xmath294 which is illustrated in figure 1 . this super - gaussian behavior is the main cause of difficulty in the proof of theorem [ th-2 ] . , @xmath295,width=384 ] note that if one only wants to say that @xmath296 , for @xmath61 growing , has dense image in @xmath5 , much simpler topological arguments suffice . another conspicuous example of mod - gaussian convergence is the arithmetic euler factor in the moment conjecture for @xmath9 . in @xcite , we considered this factor for the real part @xmath297 only , and we first generalize this as in the previous section . consider a sequence @xmath298 of independent random variables uniformly distributed on the unit circle and indexed by prime numbers , and let @xmath299 where the logarithm is given here by the taylor expansion around @xmath50 . for each individual term @xmath300 , we have @xmath301 with @xmath302 . expanding by the binomial theorem and picking up the constant term in the expansion in fourier series , we obtain @xmath303 in terms of the gauss hypergeometric function . arguing as in @xcite , we see now that @xmath304 and hence , denoting @xmath305 we get @xmath306 as @xmath60 , with limiting function @xmath307 we have here also @xmath66 in ( [ eq - grow - not - too - much ] ) . now , to check the uniformity required in ( [ eq - strong - convergence ] ) , we write @xmath308 with @xmath309 if we expand the @xmath27-factor using the binomial theorem , we obtain @xmath310 and if we assume that @xmath311 with @xmath90 , crude bounds show that this @xmath27-factor is @xmath312 where the implied constant is absolute , so that if @xmath313 , for instance , we get @xmath314 although this is crude , it already gives much more than ( [ eq - strong - convergence ] ) , both in terms of range of uniformity and sharpness of approximation . since condition ( 3 ) is also obviously valid here , theorem [ th-2 ] ( or rather ( [ eq - quant - lower ] ) ) applies with @xmath189 any real number @xmath315 , @xmath270 and @xmath37 arbitrarily large , and shows that @xmath316 for any @xmath317 and @xmath318 , provided @xmath319 for some large constant @xmath320 . from this we deduce easily the more arithmetic - looking statement of theorem [ th - eulerproduct ] . indeed , let @xmath321 be given by ( [ eq - pnt ] ) . for fixed @xmath61 , it is well - known that the random variables @xmath322 on the probability spaces @xmath323,t^{-1}\lambda)$ ] converge in law , as @xmath324 , to @xmath325 where @xmath326 are as above ( independent and uniformly distributed on the unit circle ; the independence is due to the fundamental theorem of arithmetic ) . for any open set @xmath327 , is in the boundary of @xmath327 is zero or not , we do not claim or need to claim an equality ; see , e.g. , ( * ? ? ? 2.1 , ( iv ) ) . ] it follows that @xmath328 applying lemma [ lm - log ] as in the proof of theorem [ th - rmt - quant ] , we obtain theorem [ th - eulerproduct ] with @xmath329 for some absolute constant @xmath37 . again , the density of values of @xmath321 for @xmath23 and @xmath3 ( or of @xmath330 , which amounts to the same thing ) is an easier matter that can be dealt with using topological tools . [ ex - zeta ] the two previous examples are of course motivated by their conjectural relation with the behavior of the riemann zeta function on the critical line ( this is the arithmetic essence of @xcite ) . indeed , keating and snaith conjecture that : [ cj - ks ] define @xmath331 , when @xmath332 is not the ordinate of a non - trivial zero of @xmath333 , by continuation along the horizontal line @xmath334 , with limit @xmath50 when @xmath335 . for any @xmath245 , we have @xmath336 as @xmath324 , where @xmath52 is the inner product on @xmath238 as in _ ( [ eq - inner - product])_. hence , we see in particular that the following holds : [ cor - zeta ] assume there exist @xmath84 , @xmath337 and @xmath338 such that conjecture _ [ cj - ks ] _ holds with the error term @xmath339 replaced by @xmath340 uniformly for @xmath341 . then the set of values @xmath9 is dense in the complex plane . in fact , there exists @xmath85 , @xmath342 , such that , for any @xmath14 and @xmath16 , there exists @xmath143 with @xmath343 such that @xmath344 of course , such a strong conjecture concerning the imaginary moments of @xmath9 looks quite hopeless at the current time : there is no known non - trivial result available , even assuming the riemann hypothesis . but example [ ex - eigenvalue - count ] below suggests that ( with this approach ) it is indeed necessary to require that the characteristic function converge uniformly for @xmath143 in a region growing with @xmath345 . in @xcite , jointly with f. delbaen , we will explain how the weaker qualitative statement @xmath346\,\mid\ , \zeta({{\textstyle{\frac{1}{2}}}}+iu)\in v\})\gg \frac{1}{\log\log t}\ ] ] for a fixed open set @xmath327 and for @xmath345 large enough ( which of course suffices to give a positive answer to ramachandra s question ) can be proved under much weaker assumptions than a uniform version of conjecture [ cj - ks ] . another remark concerning conjecture [ cj - ks ] has to do with the factored form of the limiting function @xmath347 , which seems to imply some asymptotic independence property . recall that the real and imaginary parts of @xmath348 and @xmath266 are themselves asymptotically independent _ after renormalization _ , but are not products of the limiting functions for the two parts separately . so conjecture [ cj - ks ] , if correct , is evidence of quite particular probabilistic behavior . and a few other results do provide evidence of a factored limiting function , with `` random matrix '' term split from the euler factor . the mod - poisson analogy is also consistent with this , in the case of the number of prime factors of an integer , as discussed in detail in @xcite . ] [ ex - eigenvalue - count ] our assumptions in theorem [ th-2 ] are probably not optimal . we now describe an illuminating ( counter)-example in the direction of understanding when a result like this could be true . we again look at random matrices @xmath236 in the compact group @xmath237 ( as in example [ ex - keating - snaith ] ) , but this time we consider the random variables counting the number of eigenvalues in certain fixed arcs of the unit circle : fix @xmath3490,1/2[$ ] , and let @xmath350 then let @xmath53 be the number of eigenvalues @xmath351 of @xmath236 such that @xmath352 . note that @xmath53 is an integer - valued random variable . it was proved by costin and lebowitz that @xmath353 converges in law to a standard normal random variable . wieand @xcite gave a proof based on asymptotics of tplitz determinants with discontinuous symbols ; as noted by basor , this gives the asymptotic @xmath354 as @xmath60 , for all @xmath143 _ with _ @xmath355 ( see , e.g. , ( * ? ? ? * th . 5.47 ) , applied with @xmath356 , @xmath357 , @xmath358 , @xmath359 , and the condition on @xmath143 is equation ( 5.79 ) in loc . cit . , or @xcite ) . this asymptotic is of course of the form ( [ eq - converge ] ) for these values of @xmath143 , but the restriction @xmath355 is necessary , since the characteristic function of @xmath53 is @xmath360-periodic for all @xmath61 . the convergence is sufficiently uniform for @xmath143 close to @xmath50 to allow the deduction of the renormalized normal behavior ( as wieand did , using the laplace transform instead of the characteristic function ) , but when @xmath203 is rational , the set of possible values of @xmath361 for @xmath23 is a discrete set in @xmath123 . is true for irrational @xmath203 , by dirichlet s approximation theorem , and by the existence of matrices in @xmath237 where the number of eigenvalues in @xmath362 takes any value between @xmath50 and @xmath61 we now consider examples related to @xmath0-functions over finite fields . our main input will be deep results of deligne and katz , and we are of course motivated by the philosophy of katz and sarnak @xcite . the goal is to make statements about the distribution of values at the central point @xmath363 of @xmath0-functions over finite fields . the appealing aspect is that these form discrete sets , hence proving that they are dense in @xmath5 ( as in theorem [ th - ff ] ) , for instance , is obviously interesting and meaningful . we consider examples of our results for the three basic symmetry types in turn : unitary , symplectic , and orthogonal . for the last two , this means first obtaining a suitable analogue of example [ ex - keating - snaith ] . the corresponding limiting functions have already been studied in some respect by keating - snaith @xcite and conrey - farmer @xcite , though our expressions seem somewhat more natural . let @xmath364 be a finite field with @xmath365 elements . unitary symmetry arises ( among other cases ) for certain types of one - variable exponential sums over finite fields , which are associated to dirichlet characters of @xmath366 $ ] , which we now describe ; these will lead to a proof of theorem [ th - ff ] . let @xmath364 be a finite field with @xmath365 elements of characteristic @xmath367 . a _ dirichlet character _ modulo @xmath368 $ ] is a map @xmath369{\rightarrow}{{{\mathbf c}}},\ ] ] defined by @xmath370 where @xmath371 is a group homomorphism @xmath372/g{\mathbf{f}}_q[x])^{\times}{\longrightarrow}{{{\mathbf c}}}^{\times}.\ ] ] this character is non - trivial if @xmath373 , and primitive if it can not be defined ( in the obvious way ) modulo a proper divisor of @xmath374 . the associated @xmath0-function is defined by the euler product @xmath375 for @xmath376 , where the product ranges over monic irreducible polynomials in @xmath366 $ ] and @xmath377 . one shows quite easily that this is in fact a polynomial ( which we denote @xmath378 ) in the variable @xmath379 of degree @xmath380 , if @xmath381 is primitive modulo @xmath374 and non - trivial . the examples used in proving theorem [ th - ff ] arise from the following well - known construction . for any integer @xmath382 , with @xmath383 , any non - trivial multiplicative character @xmath384 such that @xmath385 , and any squarefree polynomial @xmath386 $ ] of degree @xmath387 , we let @xmath388 where @xmath389 is defined to be @xmath50 . these are multiplicative exponential sums , and have been studied intensively , due in part to their many applications to analytic number theory ( for their generalizations to multiple variables , see the paper @xcite of katz ) . it is also well - known that one can construct a non - trivial dirichlet character @xmath390 , primitive modulo @xmath374 , such that @xmath391 where @xmath392 denotes the `` companion '' sums over extensions of @xmath364 , namely @xmath393 where @xmath394 is the norm map . we will denote @xmath395 the corresponding @xmath0-function . moreover , we have the riemann hypothesis for these @xmath0-functions ( due to a. weil ) , which gives the link with random unitary matrices : there exists a unique conjugacy class @xmath396 in the unitary group @xmath397 such that @xmath398 ( so that , in particular , we recover the weil bound @xmath399 by looking at the trace of @xmath396 ) . for all this , one can see , for instance , @xcite , which contains a self - contained account . we will first prove the following theorem , which is clearly a stronger form of theorem [ th - ff ] in view of the preceding remarks : [ th - ff - strong ] for @xmath382 and @xmath400 , let @xmath401 $ ] . for @xmath27 prime , let @xmath402 denote the set of pairs @xmath403 where @xmath404 is non - trivial character of @xmath405 and @xmath406 . let @xmath283 and @xmath15 with @xmath16 be given . for all integers @xmath407 , we have @xmath408 where @xmath409 where @xmath34 and the implied constants are absolute . this result depends on the mod - gaussian convergence for characteristic polynomials on @xmath237 ( i.e. , on example [ ex - keating - snaith ] ) . indeed , denoting by @xmath410 the space of conjugacy classes in @xmath237 , we have the following : [ th - katz ] for any integer @xmath411 , any odd prime @xmath27 with @xmath412 , the conjugacy classes @xmath413 become equidistributed in @xmath414 as @xmath415 , with respect to haar measure . this is an easy consequence of results of katz ( see ( * ? ? 5.13 ) ) , the only `` twist '' being the extra averaging over all non - trivial dirichlet characters to obtain unitary instead of special unitary ( or similar ) equidistribution . first of all , it is easy to check that if @xmath412 and @xmath406 is such that @xmath416 , the polynomial @xmath417 $ ] is a strong deligne polynomial in one variable ( in the language of @xcite , these are called `` weakly - supermorse '' polynomials ) . hence the conjugacy classes in the statement are well - defined . for simplicity , denote @xmath418 the open subset of the affine @xmath143-line where @xmath416 . now , according to the weyl equidistribution criterion , we must show that @xmath419 for any ( fixed ) non - trivial irreducible unitary representation @xmath420 of the compact group @xmath397 . we isolate in the sum those characters @xmath404 where @xmath421 : there are at most @xmath422 of them . for any other character @xmath423 , the inner sum over @xmath424 is of the type handled by the deligne equidistribution theorem . let @xmath425 be the order of the dirichlet character @xmath426 , where @xmath427 is the real character modulo @xmath27 . by ( * ? ? * th . 5.13 , ( 2 ) ) ( the restriction @xmath428 ensures the assumptions hold ) , provided @xmath412 , e.g. , @xmath429 , the relevant geometric monodromy group is equal to @xmath430 with maximal compact subgroup @xmath431 for simplicity , we write @xmath432 , @xmath433 . the restriction of @xmath420 to @xmath434 is a finite sum of irreducible representations of this group ( possibly including trivial components ) . applying ( * ? ? ? * th . 9.2.6 , ( 5 ) ) to each of the non - trivial ones ( and the obvious identity for the trivial components ) , we find that @xmath435 where the implied constant is _ absolute _ ( this is because we have a one - parameter family , so we can apply @xcite and the fact proved in @xcite that the relevant sheaf is everywhere tame , so the swan - conductor contribution is zero ; the parameter curve @xmath418 has @xmath387 points at infinity , which give the factor @xmath387 above ) . using frobenius reciprocity or direct integration ( using , e.g , ( * * lemma ad.7.1 ) ) , we find that the multiplicity of the trivial representation in @xmath420 ( restricted to @xmath434 ) satisfies @xmath436 for a given @xmath420 , this is equal to @xmath208 only if @xmath437 , so @xmath438 for some @xmath439 , and if @xmath404 is such that @xmath440 . the number of such characters @xmath404 is therefore @xmath441 , the implied constant depending on @xmath420 . hence we find , after adding back the characters with @xmath421 , that @xmath442 where the implied constant depends on @xmath420 . this confirms the claimed equidistribution . this an easy consequence of theorem [ th - katz ] : for any open set @xmath443 , we have first that @xmath444 as @xmath27 goes to infinity , and then we can write @xmath445 and then we apply theorem [ th - rmt - quant ] . in theorem [ th - katz ] , we performed the average over @xmath404 , because for `` standard '' families of exponential sums ( those parametrized by points of algebraic varieties ) , the ( connected component of the ) geometric monodromy group is always semisimple , so its center is finite and its maximal compact subgroup can never be @xmath237 . however , one may expect that @xmath143 could be fixed in the example above , and that ( for instance ) the conjugacy classes @xmath446 corresponding to the exponential sums @xmath447 ( where @xmath448 , @xmath412 ) would be already equidistributed in @xmath414 as @xmath415 . very recently , n. katz @xcite has indeed shown that for such families there is always an a - priori `` sato - tate law '' , i.e. , that the conjugacy classes become equidistributed in @xmath449 for some compact group @xmath450 . in fact , in this work , katz proves the required equidistribution for ( most ) fixed @xmath143 _ in the vertical direction _ where one looks at characters of @xmath451 for fixed @xmath365 and @xmath452 ( see ( * ? ? ? 17.6 , remark 17.7 ) ) . it seems quite possible that the `` horizontal '' direction we are interested in will also follow from these new techniques . similarly , it is likely that families of hyper - kloosterman sums of certain types exhibit full unitary monodromy , e.g. , for any integer @xmath453 , any additive character @xmath454 , any multiplicative character @xmath455 , one can consider the sums @xmath456 for which the basic theory ( due to deligne @xcite ) shows that the associated @xmath0-function ( unitarily normalized so the central point is @xmath457 ) is @xmath458 for some unique conjugacy class @xmath459 . one would then expect ( this was suggested by katz , and is again potentially in the realm of his recent work @xcite ) that if we fix a character @xmath460 for @xmath405 ( e.g. , @xmath461 ) and define @xmath462 for all @xmath41 , the sets of conjugacy classes @xmath463 become equidistributed in @xmath464 as @xmath452 . a typical example of symplectic symmetry involves families of @xmath0-functions of algebraic curves over finite fields . for simplicity , we will consider one of the simplest ones , but we first start by proving distribution results for characteristic polynomials of symplectic matrices , which are of independent interest . we first remark that for @xmath465 , the characteristic polynomial can be expressed in the form @xmath466 for some eigenangles @xmath467 , @xmath468 , and it follows that @xmath469 this positivity is reflected in a shift in expectation in the mod - gaussian convergence ( it also means that the argument is not an interesting quantity here ) . we obtain : [ pr - symplectic ] for @xmath470 , let @xmath471 where @xmath472 is a haar - distributed random matrix in the unitary symplectic group @xmath473 . then @xmath474 converges in mod - gaussian sense with @xmath475 and limiting function @xmath476 indeed , we have @xmath477 for @xmath478 , where the implied constant is absolute . figure 2 is a graph illustrating the logarithmic growth of @xmath479 . , @xmath480,width=384 ] ( compare ( * ? ? ? 4.9 ) ) keating - snaith ( * ? ? ? * ( 10 ) ) compute that @xmath481 which , together with the formula ( [ eq - ratio - gamma ] ) , gives @xmath482 by applying proposition [ pr - app ] , ( 3 ) in the appendix , we get @xmath483 as claimed . in particular , the central limit theorem for @xmath484 takes the form of the convergence in law @xmath485 so , for any @xmath486 , we have @xmath487 on the other hand , by applying theorem [ th-2 ] , as we can according to the previous proposition , we can control the probability of the values of @xmath484 in much smaller ( dyadic or similar ) intervals : let @xmath488a , b[$ ] with @xmath486 real numbers . we have @xmath489 for @xmath490 , where the implied constant is absolute . in particular @xmath491 if @xmath492 , we apply theorem [ th-2 ] , with the constants @xmath66 , @xmath189 arbitrarily close to @xmath253 , @xmath270 arbitrarily large and @xmath493 , so that @xmath273 can be any number with @xmath494 and in particular @xmath495 is valid . if @xmath496 , we split the interval @xmath497a , b[$ ] into @xmath498 intervals of length @xmath499 and apply the previous case to the interior of those intervals . since the joint distribution of eigenvalues of @xmath472 is absolutely continuous with respect to lebesgue measure , the probability of falling on one of the missing endpoints is zero , and summing over these intervals gives the result . we now deduce an arithmetic corollary , using families of hyperelliptic curves over finite fields . for any odd @xmath365 , any integer @xmath470 and any squarefree monic polynomial @xmath500 $ ] of degree @xmath501 , let @xmath502 be the smooth projective model of the affine hyperelliptic curve @xmath503 the number of @xmath451-rational points on @xmath502 satisfies @xmath504 where @xmath427 is the quadratic character of @xmath505 and the notation is as in section [ ssec - unitary ] . the associated @xmath0-function ( the numerator of the zeta function ) is defined by @xmath506 or , in other words , we have @xmath507 weil proved that @xmath508 is a polynomial in @xmath509 $ ] , of degree @xmath510 , all roots of which have modulus @xmath511 , and which is _ symplectic _ : there is a unique conjugacy class @xmath512 in @xmath473 such that @xmath513 [ th - hyperelliptic ] let @xmath514 be the set of squarefree , monic , polynomials of degree @xmath501 in @xmath366 $ ] . fix a non - empty open interval @xmath497\alpha,\beta[\subset ] 0,+\infty[$ ] . for all @xmath374 large enough , we have @xmath515\alpha,\beta [ \bigr\}\bigr| } \gg \frac{1}{\sqrt{\log g}}.\ ] ] ( note that this is in fact a very weak version of what we can prove ) . let first @xmath516 be the set of @xmath517 for which @xmath518 . in ( * ? 4.9 ) , we showed , using the relevant equidistribution computation in @xcite that the ( real - valued ) random variables @xmath519 on @xmath516 ( with counting measure ) converges in law to @xmath471 where @xmath472 is a random matrix in the unitary symplectic group @xmath473 , distributed according to haar measure . the previous proposition shows that theorem [ th-2 ] is applicable to @xmath474 with covariance @xmath475 and limiting function @xmath520 . letting @xmath521 as in the previous section , we get @xmath522\alpha,\beta[\bigr\}\bigr| \gg \frac{1}{\sqrt{\log g}}\ ] ] for @xmath374 large enough . since @xmath523 for fixed @xmath374 and @xmath521 ( by an easy application of the equidistribution , see ( * ? ? ? 4.9 ) ) , we get the result stated by exponentiating . the lower bound ( [ eq - symplectic ] ) is good enough to combine with various other statements proving arithmetic properties of @xmath0-functions which hold for `` most '' hyperelliptic curves . for instance , from ( * ? ? ? * prop . 1.1 ) ( adapted straightforwardly to all hyperelliptic curves instead of special one - parameter families ) , it follows that if we denote by @xmath524 the set of @xmath517 such that the eigenvalues of @xmath525 satisfy no non - trivial multiplicative relation ,- linear independence of the ordinates of zeros of the riemann zeta function ; non - trivial refers to a relation that can not be deduced from the fact that , if @xmath526 is an eigenvalue , so is its inverse @xmath527 . ] then we have @xmath528 for some @xmath529 , and hence we get @xmath530\alpha,\beta [ \bigr\}\bigr| } \gg \frac{1}{\sqrt{\log g}}\ ] ] for @xmath374 large enough . orthogonal symmetry , in number theory , features prominently in families of elliptic curves . in contrast with symplectic groups , there are a number of `` flavors '' involved , due to the `` functional equation '' @xmath531 for an orthogonal matrix @xmath532 ( the standard maximal compact subgroup of the orthogonal group @xmath533 ) , which implies that @xmath534 is zero for `` trivial '' reasons if @xmath61 is even and @xmath535 or @xmath61 is odd and @xmath536 . when this happens , it is of great interest to investigate the distribution of the first derivative at @xmath537 of the reversed characteristic polynomial . for simplicity , however , we restrict our attention here to @xmath61 even and matrices with determinant @xmath208 , i.e. , to the subgroup @xmath538 of @xmath539 , where @xmath23 . in that case , it is also true that eigenangles come in pairs of inverses , and therefore we have @xmath540 . as in the previous section , we start with random matrix computations . [ pr - orthogonal ] for @xmath23 , let @xmath541 where @xmath236 is a haar - distributed random matrix in the special orthogonal group @xmath538 . then @xmath53 converges in mod - gaussian sense with @xmath542 and limiting function @xmath543 indeed , we have @xmath544 for @xmath545 , where the implied constant is absolute . using ( * ? ? ? * ( 56 ) ) and ( [ eq - ratio - gamma ] ) , we get @xmath546 and by applying proposition [ pr - app ] , ( 4 ) in the appendix , we get the desired formula @xmath547 if we compare with the symplectic case , we observe the ( already well - established ) phenomenon that the value @xmath534 , for @xmath548 tend to be small , whereas they tend to be large for symplectic matrices in @xmath473 . our arithmetic corollary is based on families of quadratic twists of elliptic curves over function fields , and we select a specific example for concreteness ( see @xcite ) ; the basic theory , which we illustrate here , is again due to katz @xcite . for any odd prime power @xmath549 , any integer @xmath23 , we consider the elliptic curves over the functional field @xmath550 given by the weierstrass equations @xmath551 where @xmath552 is a parameter such that @xmath256 is not a critical value of @xmath553 . katz proved that the associated @xmath0-function ( which is now defined by the `` standard '' euler product over primes in @xmath554 $ ] , with suitable ramified factors ) is of the form @xmath555 where @xmath556 is a unique conjugacy class in @xmath539 . [ th - orthogonal ] fix a non - empty open interval @xmath497\alpha,\beta[\subset ] 0,+\infty[$ ] . for all @xmath61 large enough , we have @xmath557\alpha,\beta [ \bigr\}\bigr| } \gg \frac{1}{\sqrt{\log n}}.\ ] ] as recalled in ( * ? ? ? 4.4 and before ) , for all @xmath558 and primes @xmath27 with @xmath559 and @xmath560 , the conjugacy classes @xmath561 , for @xmath562 not a critical value , become equidistributed in @xmath563 for the image of haar measure ( precisely , this is stated for the `` vertical direction '' where @xmath27 is fixed and finite fields of characteristic @xmath27 and increasing degree are used ; however , because the parameter variety is a curve with @xmath564 points at infinity and the relevant sheaf is tame , we can recover the horizontal statement as in the proof of theorem [ th - katz ] ) . in particular , there is a subset @xmath565 with @xmath566 where @xmath567 and those restricted conjugacy classes become equidistributed in @xmath568 . hence , for @xmath61 large enough , we get @xmath569\alpha,\beta [ \bigr\}\bigr| \\{\geqslant}\mu_{so(2n,{{{\mathbf r}}})}(\{a\,\mid\ , \log \det(1-a)\in]\log \alpha,\log\beta[\ } ) \gg \frac{1}{\sqrt{\log n}},\end{gathered}\ ] ] as desired . obviously , this result ( or its generalizations to other families of quadratic twists over function fields ) has interesting consequences concerning the problem of the distribution of the order of tate - shafarevich groups of the associated elliptic curves , through the birch and swinnerton - dyer conjecture ( which is known to be valid in its strong form for many elliptic curves over function fields over a finite field with analytic rank @xmath50 or @xmath208 ) . we hope to come back to this question , and its conjectural analogue over number fields , in another work . we present in this appendix some uniform analytic estimate for the barnes function , which are needed to verify the strong convergence assumption ( [ eq - strong - convergence ] ) for sequences of random matrices in compact classical groups . note that we did not try to optimize the results . if we expand the logarithm , defined using the taylor expansion of @xmath578 at the origin , we have @xmath579 for @xmath580 , with an absolute implied constant . hence we obtain @xmath581 with an absolute implied constant , for @xmath453 and @xmath582 , hence for @xmath571 we get @xmath583 since @xmath584 for @xmath453 , with absolute implied constants . hence , we have @xmath585 for @xmath571 , for some absolute implied constant , and we get the stated formula @xmath586 \(2 ) note first that the conditions @xmath23 and @xmath268 ensure that the values of the barnes function in the denominator are non - zero . next , let @xmath587 , @xmath588 ; we can express the ratio of barnes function as @xmath589 and @xmath268 gives @xmath590 , @xmath591 , allowing us to apply ( [ eq - power - ratio ] ) three times . the exponential terms cancel out , leading to @xmath592 which gives the first part of the proposition . \(3 ) we use a similar computation , applying ( [ eq - power - ratio ] ) six times with the parameters @xmath593 @xmath594 leading , after an easy calculation , to a main term @xmath595 for the ratio of barnes functions , and some further computation leads to the stated result ( each parameter @xmath256 has @xmath596 , so the error term is also as given ) . e. kowalski : _ the large sieve , monodromy , and zeta functions of algebraic curves , 2 : independence of the zeros _ , international mathematics research notices ( 2008 ) , article rnn091 , 57 pages , doi:10.1093/imrn / rnn091

let be a knot in a closed orientable irreducible 3manifold and let be a heegaard splitting of the knot complement of genus at least two . suppose is a bridge surface for and let denote a regular neighborhood of . then either * , or * can be isotoped to be disjoint from so that after the isotopy is a heegaard surface for that is isotopic to a possibly stabilized copy of .

math0703474

a heegaard splitting of a compact 3manifold @xmath1 is a decomposition of the manifold into two compression bodies , @xmath7 and @xmath8 . if the manifold is closed , @xmath7 and @xmath8 are handlebodies . the common boundary of @xmath7 and @xmath8 is called a _ heegaard surface _ which we denote by @xmath2 . we will write @xmath9 . the distance between any two essential simple closed curves @xmath10 and @xmath11 in a heegaard surface @xmath2 is the smallest integer @xmath12 so there is a sequence of essential simple closed curves @xmath13 in @xmath2 such that for each @xmath14 , @xmath15 and @xmath16 can be isotoped to be disjoint in @xmath2 . the _ distance _ of a heegaard splitting @xmath17 , @xmath18 , defined by hempel @xcite , is the smallest integer @xmath19 so that there is an essential curve @xmath10 in @xmath2 which bounds a disk in @xmath7 and an essential curve @xmath11 which bounds a disk in @xmath8 and @xmath20 . if @xmath21 then we say that the heegaard splitting @xmath22 is _ reducible _ , if @xmath23 then we say that @xmath22 is _ weakly reducible _ and if @xmath24 then we say that @xmath17 is _ strongly irreducible_. obtaining bounds on the distance of a heegaard splitting @xmath9 is an interesting problem and much progress has been made recently . the first such bound is due to hartshorn @xcite who uses a closed essential surface in the manifold to give a bound on @xmath18 . this result was extended by scharlemann in @xcite to allow for possible boundary components of the essential surface . [ thm : essentialsurface ] ( * ? ? ? * theorem 3.1 ) suppose @xmath2 is a heegaard surface for a compact orientable manifold @xmath1 and @xmath25 is a connected essential surface . then @xmath26 scharlemann and tomova , @xcite , gave a bound on @xmath18 using a second heegaard surface . [ thm : alternate ] ( * ? ? ? * corollary 3.5 ) suppose @xmath2 and @xmath3 are distinct heegaard splittings for the compact orientable @xmath27-manifold @xmath1 . then either @xmath28 or @xmath3 is isotopic to a stabilization of @xmath2 . suppose a manifold @xmath1 contains a knot @xmath0 and let @xmath29 be the knot complement . there has been interest in obtaining a bound on the distance of heegaard splittings of @xmath29 using a bridge surface for the knot . we say that a surface @xmath3 is a bridge surface for a knot @xmath0 in a manifold @xmath1 if @xmath3 is a heegaard surface for @xmath30 and @xmath0 intersects each of the compression bodies @xmath31 and @xmath32 in arcs that are simultaneously parallel to @xmath3 . recently johnson and thompson ( * ? ? ? * theorem 1 ) used theorem [ thm : alternate ] to show that if a knot @xmath33 with bridge surface @xmath3 is such that @xmath0 can not be isotoped to lie in @xmath3 , then the distance of any heegaard splitting of the knot complement is bounded by @xmath34 where @xmath4 is a regular neighborhood of @xmath0 . they used this theorem to obtain tunnel number one knots that have arbitrarily high bridge number with respect to any bridge surfaces of genus 1 . kobayashi and rieck ( * ? ? ? * proposition 2.6 ) extended this result to obtain the same bound on the distance of a heegaard splitting in the case when the genus of the bridge surface @xmath3 is strictly less than the minimum genus of any heegaard splitting of the knot complement . their result allows for the knot to be isotopic to an embedded curve in @xmath3 . they used this distance bound to prove the existence of counterexamples to morimoto s conjecture ( * ? ? ? * conjecture 1.5 ) . in this paper we remove the restriction on the genus of @xmath3 as well as restrictions on whether the knot is isotopic into @xmath3 . more precisely : [ thm : main ] let @xmath0 be a knot in a closed oriented irreducible 3manifold @xmath1 and let @xmath2 be a heegaard splitting of the knot complement of genus at least two . suppose @xmath3 is a bridge surface for @xmath0 and let @xmath4 denote a regular neighborhood of @xmath0 . then either * @xmath35 , or * @xmath0 can be isotoped to be disjoint from @xmath3 so that after the isotopy @xmath3 is a heegaard surface for the knot complement that is isotopic to a possibly stabilized copy of @xmath2 . in some sense this theorem completes the solution to the problem as it is clear that a stabilized copy of @xmath2 will not carry any information about @xmath18 and these are the only bridge surfaces which we exclude . this result was used by minsky , moriah and schleimer in @xcite to show that for any pair of integers @xmath36 and @xmath37 there is a knot with tunnel number @xmath36 such that every bridge surface of genus @xmath36 for the knot intersects the knot in at least @xmath38 points . the proof of theorem [ thm : main ] uses different methods than those used in @xcite and @xcite and is independent of them . in this paper , unless otherwise specified , we will consider a closed irreducible orientable 3manifold @xmath1 containing a knot @xmath0 . if @xmath31 is any subset of @xmath1 we will denote by @xmath39 the set @xmath40 where @xmath4 is an open tubular neighborhood of @xmath0 . in particular the knot exterior @xmath6 will be denoted by @xmath29 . note that @xmath29 is a compact orientable manifold with a single torus boundary component . throughout this paper @xmath2 will be a surface splitting @xmath1 into handlebodies @xmath7 and @xmath8 so that @xmath0 is entirely contained in @xmath7 . we will further assume that @xmath41 is a compression body , that is @xmath2 is also a heegaard surface for @xmath29 and @xmath42 . we will further assume that @xmath2 has genus at least 2 as otherwise distance is always infinite with our definition . recall that a simple closed curve in a compact surface @xmath43 is _ essential _ if it does not bound a disk in the surface and it is not parallel to a boundary component of the surface . suppose @xmath1 is a closed manifold containing a knot @xmath0 and @xmath43 is a surface in @xmath1 transverse to @xmath0 . a disk @xmath44 is a compressing disk for @xmath45 if @xmath46 and @xmath47 is an essential simple closed curve in @xmath45 . a disk @xmath48 in @xmath1 is a _ cut disk _ for @xmath45 if @xmath49 , @xmath50 is essential in @xmath45 and @xmath48 intersects @xmath0 in a single point . c - disk _ is a cut or a compressing disk . a properly embedded surface @xmath51 in a 3manifold is _ essential _ if it is incompressible and has at least one component that is not parallel to @xmath52 . a surface @xmath51 in @xmath1 is called a _ splitting surface _ if @xmath1 can be written as the union of two 3manifolds @xmath53 and @xmath54 along @xmath51 . if @xmath51 is a splitting surface it is _ bicompressible _ if it is compressible in both @xmath53 and @xmath54 and @xmath51 is _ c - bicompressible _ if it has c - disks in both @xmath53 and @xmath54 . if @xmath51 is a splitting surface for @xmath1 , we will call @xmath51 _ c - weakly incompressible _ if it is c - bicompressible and any pair of c - disks for @xmath51 on opposite sides of the surface intersect along their boundaries . if a c - bicomressible surface @xmath51 is not c - weakly incompressible , it is _ c - strongly compressible_. suppose @xmath0 is a knot in a closed orientable 3manifold @xmath1 . we say that a surface @xmath3 is a bridge surface for @xmath0 if @xmath3 is a heegaard surface for @xmath30 and @xmath0 intersects each of the handlebodies @xmath31 and @xmath32 in arcs that are simultaneously parallel to @xmath3 . more generally , if the manifold @xmath1 has boundary , we require that @xmath0 intersect each of the compression bodies @xmath31 and @xmath32 in arcs that are simultaneously parallel into @xmath3 . the arcs are called _ bridges _ and the disks of parallelism are called _ bridge disks_. a handlebody @xmath55 intersecting a knot @xmath0 in a collection of bridges will be called a @xmath0-handlebody and will be denoted by @xmath56 . a _ spine _ @xmath57 of a handlebody @xmath55 is any graph that @xmath55 retracts to . removing a neighborhood of a spine from a handlebody results in a manifold that is homeomorphic to @xmath58 . more generally the spine @xmath59 of a compression body @xmath60 is the union of @xmath61 and a 1complex @xmath62 so that @xmath60 collapses to @xmath59 . suppose @xmath56 is a @xmath0-handlebody and let @xmath63 , @xmath64 be the bridges . the spine @xmath65 of @xmath56 is the union of a spine of the handlebody @xmath55 , @xmath57 , together with a collection of straight arcs @xmath66 , where one endpoint of each @xmath66 lies in @xmath63 and the other endpoint lies in @xmath57 . then @xmath67 . as in the handlebody case , spines of @xmath0-handlebodies are not unique . a bridge surface @xmath3 for @xmath0 is called _ stabilized _ if there is a pair of compressing disks for @xmath3 , one in @xmath31 and one in @xmath32 , that intersect in exactly one point . the surface @xmath3 is called _ meridionally stabilized _ if there is a compressing disk for @xmath3 in @xmath31 and a cut - disk for @xmath3 in @xmath32 ( or vice versa ) that intersect in exactly one point . finally @xmath3 is called _ perturbed _ if there is a pair of bridge disks , @xmath68 and @xmath69 , such that @xmath70 and @xmath71 . by @xcite if @xmath3 is stabilized , meridionally stabilized or perturbed , then there is a bridge surfaces @xmath72 for @xmath0 such that @xmath73 . furthermore it is easy to see that if @xmath3 is stabilized , meridionally stabilized or perturbed then it is c - strongly compressible , see @xcite . suppose @xmath3 is a bridge surface for a knot @xmath33 . we say that @xmath0 is _ removable _ with respect to @xmath3 if it can be isotoped to lie in @xmath3 and there is a meridian disk for one of the two handlebodies @xmath31 and @xmath32 that intersects @xmath74 in a single point . in this case , by lemma 3.3 in @xcite , there is a heegaard splitting @xmath75 for @xmath29 such that @xmath76 . we will also need the following result , shown in @xcite . [ thm : essentialexists ] suppose @xmath1 is a closed orientable irreducible 3manifold containing a knot @xmath0 . if @xmath3 is a c - strongly compressible bridge surface for @xmath0 then one of the following is satisfied : * @xmath3 is stabilized * @xmath3 is meridionally stabilized * @xmath3 is perturbed * @xmath0 is removable * @xmath1 contains an essential meridional surface @xmath51 such that @xmath77 . recall that we are considering a heegaard splitting @xmath78 of a knot complement @xmath29 where @xmath1 is a closed manifold . how surfaces in a manifold restrict distance of a heegaard splitting in various settings has been studied in several papers , for example see @xcite , @xcite and @xcite . we will take advantage of some of these results . * proposition 2.5 ) [ prop : proximity2.5 ] suppose @xmath1 is an irreducible 3manifold , @xmath79 is a compressible boundary component of @xmath1 and @xmath80 is a properly embedded essential surface containing no disk components and with at least one essential component incident to @xmath79 . let @xmath81 be the set of essential curves in @xmath79 that bound disks in @xmath1 and let @xmath82 be any component of @xmath83 . then either @xmath84 or @xmath82 lies in the boundary of @xmath85-parallel annulus component of @xmath51 . we will use the above proposition in a very specific situation . the manifold @xmath1 will either be the handlebody @xmath8 or the compression body @xmath41 and @xmath79 will be @xmath86 . then the above proposition says roughly that the distance between the boundary curves of an essential surface in a compression body and the boundaries of the compressing disks for the compression body is bounded above by the euler characteristic of the essential surface . the situation becomes considerable more complicated if we allow the surface @xmath51 to have compressing disks . however , if we restrict our attention to weakly incompressible surfaces , useful information about @xmath87 can still be obtained , @xcite , @xcite . first we recall the definition of a tube - spanned recessed collar , @xcite . let @xmath88 be two connected compact subsurfaces in the same component @xmath79 of @xmath52 , with each component @xmath89 , @xmath90 essential in @xmath52 and @xmath91 . let @xmath92 , @xmath90 be the properly embedded surface in @xmath1 obtained by pushing @xmath93 , _ rel _ @xmath89 into the interior of @xmath1 , so the region @xmath94 lying between @xmath95 and @xmath96 is naturally homeomorphic to @xmath97 . the properly embedded surface obtained by tubing @xmath95 and @xmath96 along an @xmath98-fiber of @xmath97 that is incident to @xmath95 is called a _ tube - spanned recessed collar _ in @xmath1 . the properties of these surfaces are described in detail in @xcite . it turns out that tube spanned recessed collars are the only weakly incompressible surfaces that do nt carry information about distance . more precisely : ( * ? ? ? * theorem 5.4)[thm : proximity5.4 ] suppose @xmath1 is an irreducible 3manifold , @xmath79 is a compressible boundary component of @xmath1 and @xmath80 is a bicompressible , weakly incompressible splitting surface with a bicompressible component incident to @xmath79 . let @xmath81 be the set of essential curves in @xmath79 that bound disks in @xmath1 and let @xmath82 be any component of @xmath99 . then either * @xmath84 in the curve complex of @xmath79 or * @xmath82 lies in the boundary of @xmath85-parallel annulus component of @xmath51 or * one component of @xmath51 is a tube spanned recessed collar ; all other components of @xmath51 incident to @xmath79 are incompressible and @xmath85-parallel . again , we will only be considering the situation when @xmath1 is the handlebody @xmath8 or the compression body @xmath41 . note that in these cases , if the surface @xmath51 consists of a single component which is a tube - spanned recessed collar with boundary in @xmath100 or in @xmath101 , then there is a spine for @xmath41 or @xmath8 that is entirely disjoint from @xmath51 and @xmath51 has a compressing disk ( a meridional disk for the tube ) that lies on the same side of @xmath51 as the spine and is disjoint from the spine . the above two results tell us that if there is a surface with certain properties in a compression body , then the boundary curves are not far " from the boundaries of the compressing disks for the compression body . thus if @xmath102 is a heegaard splitting for a manifold , @xmath103 and @xmath104 are the collections of simple closed curves in @xmath94 that bound disks on sides @xmath53 and @xmath54 respectively and @xmath43 is a surface that intersects @xmath94 in a particular way , then we can hope to obtain a bound on the distance of @xmath94 by summing @xmath105 and @xmath106 . the next lemma makes this idea precise . * lemma 2.6 ) [ lem : essentialdist ] suppose @xmath94 is a heegaard splitting for a compact manifold @xmath1 , dividing @xmath1 into compression bodies @xmath53 and @xmath54 . let @xmath103 and @xmath104 be the collections of simple closed curves in @xmath94 that bound disks on sides @xmath53 and @xmath54 respectively . let @xmath107 be a properly embedded connected surface transverse to @xmath94 and let @xmath108 , @xmath109 . suppose @xmath43 satisfies the following conditions : * all curves of @xmath110 are essential in @xmath43 and @xmath94 . * there is at least one curve @xmath111 such that @xmath112 and any curve in @xmath113 for which the inequality does not hold is the boundary of an annulus component of @xmath114 that is parallel into @xmath94 . * there is at least one curve @xmath115 such that @xmath116 and any curve in @xmath113 for which the inequality does not hold is the boundary of an annulus component of @xmath117 that is parallel into @xmath94 . then @xmath118 . the following lemma is similar to lemma 3.6 in @xcite where the result was proven in the context of a @xmath0-handlebody . we will need the result for a compression body so some modifications in the proof are needed . [ lem : disjointdisk ] let @xmath1 be a closed orientable irreducible 3manifold containing a knot @xmath0 . suppose @xmath2 and @xmath3 are heegaard surface for @xmath119 . in addition , suppose that @xmath2 is also a heegaard surface for @xmath42 and @xmath3 is a bridge surface for @xmath0 . if there is a spine @xmath120 for @xmath8 lying in @xmath32 and @xmath121 has a c - disk @xmath122 in @xmath123 that is disjoint from @xmath120 then either * @xmath3 is c - strongly compressible , or * @xmath0 is removable with respect to @xmath3 , or * @xmath124 and @xmath0 is the unknot . as already noted , @xmath125 has a natural product structure . use this structure to push @xmath121 and @xmath122 to lie entirely in @xmath41 . as @xmath126 , @xmath121 always compresses in @xmath127 and thus we may assume that @xmath128 is a c - weakly incompressible surface . * case 1 : * suppose first that @xmath122 is a disk . in this case @xmath121 is a weakly incompressible surface lying in the compression body @xmath41 . maximally compress @xmath121 in @xmath129 and let @xmath130 be the resulting surface . note that by the construction , @xmath130 separates @xmath2 and @xmath121 . it is a classical result that maximally compressing a weakly incompressible surface results in an incompressible surface , see for example ( * ? ? ? * lemma 5.5 ) . as @xmath130 is an incompressible surface in a compression body and @xmath131 , each component of @xmath130 must be parallel to @xmath132 , that is , each component is an annulus or a torus . as @xmath130 separates @xmath2 and @xmath121 some component @xmath133 of @xmath130 also separates @xmath121 and @xmath2 . * subcase 1a : * suppose first that @xmath133 is an annulus . then the corresponding closed component @xmath134 of @xmath72 bounds a ball in @xmath7 containing @xmath3 . as @xmath130 is parallel to @xmath4 , @xmath0 intersects this ball in a trivial arc . now consider @xmath133 as a surface in the @xmath0-handlebody @xmath32 . as handlebodies are irreducible , @xmath133 must then also bound a ball in @xmath32 and by lemma 3.2 of @xcite , the knot intersects this ball in a trivial arc . thus @xmath1 is the three sphere and @xmath0 is a one bridge knot with respect to the bridge sphere @xmath134 , thus @xmath0 is the unknot . * subcase 1b : * suppose then that @xmath134 is a torus bounding a solid torus @xmath54 in @xmath7 which is a regular neighborhood of @xmath0 . as @xmath135 is obtained from @xmath121 by compressing along a collection of disks , the original surface @xmath121 can be recovered from @xmath135 by tubing along a graph @xmath62 with edges dual to these disks . if @xmath136 , @xmath134 is the only torus component of @xmath130 and all edges of @xmath62 have endpoints on two different components of @xmath130 . pick a meridional disk @xmath51 for @xmath134 that intersect @xmath0 exactly once . isotope @xmath0 to lie in @xmath134 using the parallelism of @xmath134 and @xmath137 , this isotopy is not proper . after the isotopy of @xmath0 each edge of @xmath62 has at least one endpoint on a sphere that bounds a ball in @xmath54 . by shrinking @xmath62 we can guarantee that the graph is disjoint from @xmath51 . thus @xmath51 is also a compressing disk for @xmath3 that intersects @xmath0 in a single point along its boundary , i.e. , @xmath0 is removable with respect to @xmath3 . if @xmath138 , consider the two components of @xmath139 . one component is @xmath39 and so it is a handlebody intersecting the knot in trivial arcs . the other component can be obtained by attaching 1handles corresponding to the edges of @xmath62 to a color of @xmath135 . the result is a compression body @xmath140 which intersects @xmath0 in arcs that are parallel into @xmath141 . thus @xmath142 is a bridge surface for @xmath0 lying in the manifold @xmath54 . as @xmath143 , by @xcite this implies that @xmath144 is stabilized or perturbed . thus @xmath121 is c - strongly compressible in @xmath1 as desired . * case 2 : * suppose now that @xmath122 is a cut disk . as @xmath41 is a compression body , there is a vertical annulus @xmath145 in @xmath7 with one boundary component in @xmath0 and the other in @xmath2 . as @xmath146 , it follows that @xmath147 and in particular @xmath3 intersects @xmath145 in at least one arc . as @xmath148 any such arc must have both of its endpoints in @xmath0 . choose @xmath145 so that @xmath149 is minimized , in particular any circle of intersection that is inessential in @xmath145 is essential of @xmath3 . now choose @xmath122 so that @xmath150 is minimal . by a standard innermost circle argument we may assume that @xmath151 consists only of circles parallel to the core of @xmath145 and arcs , see figure [ fig : annulusa ] . * claim 1 : * we may assume that every component of @xmath152 is either a circle parallel to the core of @xmath145 or is adjacent to an arc of @xmath151 . suppose @xmath153 contains either an inessential circle in @xmath145 or an arc with both endpoints of @xmath0 . in either case assume that this curve is disjoint from @xmath122 . pick @xmath154 to be either an innermost such circle or an outermost such arc and let @xmath155 be the disk in @xmath145 that @xmath154 bounds . by minimality of @xmath153 if @xmath154 is a circle , @xmath155 is a compressing disk for @xmath3 . if @xmath154 is an arc , then @xmath155 is a bridge disk and so there is a compressing disk for @xmath3 contained in the boundary of a regular neighborhood of @xmath155 . in either case we have found a compressing disk for @xmath3 contained in @xmath41 and disjoint from @xmath2 . if this disk is contained in @xmath31 , then @xmath121 is c - strongly compressible as the disk is disjoint from @xmath122 . if the disk is contained in @xmath32 , the result follows by case 1 . * claim 2 : * there is no arc of @xmath151 that is parallel in @xmath145 to a subarc of @xmath153 . suppose there is such an arc and let @xmath10 be an outermost one , i.e. , @xmath10 has both endpoints in a component @xmath156 of @xmath3 such that @xmath157 bounds a disk @xmath158 in @xmath145 whose interior is disjoint from @xmath122 , see figure [ fig : annulusa ] . by claim 1 , the interior of @xmath158 is also disjoint from @xmath3 . the boundary of a regular neighborhood of @xmath159 contains at least one c - disk @xmath51 for @xmath3 , see figure [ fig : annulusb ] . this c - disk has at least one fewer intersections with @xmath145 than @xmath122 does . if @xmath51 is a cut - disk , this will contradict the choice of @xmath122 , if @xmath51 is a compressing disk , then case 1 can be applied . this concludes the proof of claim 2 . as @xmath122 is a cut disk , it intersects the knot exactly once . let @xmath160 be the arc of @xmath151 that has one endpoint in @xmath0 and the other in some component @xmath161 of @xmath153 . * subcase a * @xmath161 is circle parallel to the core of @xmath145 , see figure [ fig : annulusa ] . as @xmath162 , there must be some arc @xmath154 of @xmath163 with both endpoints in @xmath0 . this arc is contained in the disk bounded by @xmath0 , @xmath161 and two copies of @xmath160 . by claim 2 , @xmath154 is disjoint from @xmath122 and so claim 1 completes the proof of this case . * subcase b * @xmath161 is a circle that is inessential in @xmath145 . let @xmath158 be the disk in @xmath145 that @xmath161 bounds . by claim 2 , the interior of @xmath158 is disjoint from @xmath122 and so by claim 1 we may assume that the interior of @xmath158 is also disjoint from @xmath3 . then @xmath158 is a compressing disk for @xmath121 lying in @xmath31 that intersects @xmath122 in exactly one point . by theorem 2.1 of @xcite @xmath121 is c - strongly compressible . * subcase c * @xmath161 is an arc , see figure [ fig : arc]a . let @xmath155 be the disk that @xmath161 cuts from @xmath145 . as before the boundary of a regular neighborhood of @xmath155 contains a compressing disk for @xmath3 in @xmath41 . if this disk lies in @xmath32 , case 1 can be applied so we may assume @xmath155 lies in @xmath31 . thus the arc @xmath160 lies outside of @xmath155 . let @xmath158 be the disk cobounded by @xmath160 , a subarc @xmath164 of @xmath161 and a subarc of @xmath0 . then the boundary of a regular neighborhood of @xmath159 contains a cut disk parallel to @xmath122 and a disk @xmath165 , see figure [ fig : arc]b . if @xmath83 bounds a disk in @xmath3 , then @xmath47 bounds a once punctured disk contradicting the fact that @xmath122 is a cut - disk . if @xmath83 bounds a punctured disk @xmath166 in @xmath3 , the @xmath167 is a sphere that intersects the knot once , a contradiction . thus @xmath83 is essential in @xmath121 and so @xmath168 is a compressing disk for @xmath121 in @xmath41 therefore we can apply case 1 . finally , the following easy proposition will be used repeatedly in the proofs to follow . [ prop : cutimpliescompressing ] suppose @xmath2 is a heegaard surface for @xmath42 with genus at least 2 and @xmath48 is a cut disk for @xmath2 in @xmath41 . then there is a compressing disk @xmath122 for @xmath2 in @xmath41 such that @xmath169 . consider cut - compressing @xmath41 along @xmath48 , i.e. remove a small open neighborhood of @xmath48 from @xmath41 . the resulting 3manifold is a @xmath0-handlebody containing a single bridge - the result of cutting the closed loop @xmath0 . as @xmath170 , the @xmath0-handlebody has genus at least 1 and thus there is a compressing disk @xmath122 which can be taken to be disjoint from the bridge . this disk is the desired compressing disk for @xmath2 in @xmath41 . let @xmath0 be a nontrivial knot in @xmath1 and consider how @xmath2 ( a heegaard splitting of @xmath42 ) and @xmath3 ( a bridge surface for @xmath0 contained in @xmath30 ) intersect in @xmath1 . let @xmath171 , @xmath172 , @xmath173 and @xmath174 and let @xmath175 and @xmath176 be the collections of curves in @xmath2 that bound compressing disks in @xmath7 and @xmath8 respectively . [ def : removable ] suppose @xmath43 and @xmath177 are two properly embedded surfaces in a 3manifold @xmath1 containing a knot @xmath0 and assume @xmath43 and @xmath177 intersect the knot transversely . let @xmath178 be a simple closed curve bounding possibly punctured disks @xmath179 and @xmath180 . if @xmath181 intersects @xmath182 only in curves that are inessential in @xmath182 and @xmath183 intersects @xmath45 only in curves that are inessential in @xmath45 we say that @xmath184 is _ removable_. [ rmk : remove ] 1 . all removable curves can be removed via an isotopy of the surfaces that does not affect any essential curves of intersection . if all curves of intersection between @xmath2 and @xmath121 are either essential in both surfaces or inessential in both surfaces then all inessential curves are removable . isotope @xmath2 and @xmath3 so as to remove all removable curves of intersection . we will associate to a position of @xmath2 and @xmath3 one or more of the following labels . * label @xmath7 ( resp @xmath8 ) if some component of @xmath185 is the boundary of a compressing disk for @xmath2 lying in @xmath41 ( resp @xmath8 ) . * label @xmath186 if some component of @xmath185 is the boundary of a cut disk for @xmath2 lying in @xmath41 . ( as @xmath187 , no label @xmath188 can occur ) . * label @xmath31 ( resp @xmath32 ) if there is a compressing disk for @xmath121 lying in @xmath39 ( resp @xmath189 ) that is disjoint from @xmath2 and the configuration does not already have labels @xmath7 , @xmath186 or @xmath8 . * @xmath190 ( resp @xmath191 ) if there is a cut disk for @xmath121 lying in @xmath39 ( resp @xmath189 ) that is disjoint from @xmath2 and the configuration does not already have labels @xmath7 , @xmath186 or @xmath8 . * @xmath192 ( resp @xmath193 ) if some spine @xmath194 or @xmath195 lies entirely in @xmath189 ( resp @xmath39 ) and the configuration does not already have labels @xmath7 , @xmath186 or @xmath8 . we will use the superscript @xmath196 to denote the possible presence of superscript @xmath197 , for example we will use @xmath198 if there is a label @xmath199 or both . [ lem : nounlabelled ] if a configuration of @xmath2 and @xmath3 has no labels , then @xmath200 . suppose a configuration has no labels . first note that this implies @xmath201 as if @xmath2 is entirely contained in @xmath31 say , then the region would have a label @xmath193 . if there is a curve of @xmath202 that is inessential in @xmath121 but essential in @xmath2 an innermost such curve in @xmath121 would give rise to label a @xmath198 or @xmath8 . if there is a curve of @xmath202 that is inessential in @xmath2 but essential in @xmath121 an innermost such curve in @xmath2 would give rise to label a @xmath203 or @xmath204 . thus all curves of @xmath202 are either essential in both surfaces or inessential in both . by remark [ rmk : remove ] we may assume all curves of @xmath202 are essential in both surfaces . furthermore , as no labels @xmath203 or @xmath204 are associated to this configuration , @xmath205 and @xmath206 are both c - incompressible and as there are no labels @xmath192 or @xmath193 , they each have at least one component that is not parallel to @xmath2 . by proposition [ prop : proximity2.5 ] both @xmath205 and @xmath206 satisfy the hypothesis of lemma [ lem : essentialdist ] from which we deduce that @xmath207 . [ lem : noaandb ] if a configuration of @xmath2 and @xmath3 has labels @xmath198 and @xmath8 , then @xmath208 . if a configuration has labels @xmath7 and @xmath8 , there are curves of @xmath209 , both essential in @xmath2 , that bound compressing disks for @xmath2 in @xmath41 and @xmath8 . as @xmath2 and @xmath121 are embedded surfaces , these two curves are disjoint and thus @xmath210 . if a configuration has labels @xmath186 and @xmath8 , using the triangle inequality and proposition [ prop : cutimpliescompressing ] , we deduce that @xmath211 . [ lem : noxandy ] if a configuration has labels @xmath192 and @xmath204 , then at least one of the following holds : * @xmath121 is c - strongly compressible , or * @xmath0 is removable with respect to @xmath3 , or * @xmath207 . assume @xmath121 is c - weakly incompressible and @xmath0 is not removable . we will show that @xmath207 . from the label @xmath192 we may assume , with no loss of generality , that there exists a spine for @xmath41 or @xmath8 contained in @xmath189 . note that the spine of @xmath41 contains the frontier of an open neighborhood of @xmath0 and @xmath212 thus we conclude that @xmath213 . from the label @xmath204 we know that @xmath121 has a c - disk in @xmath214 , call this disk @xmath158 . by lemma [ lem : disjointdisk ] , @xmath215 so in particular @xmath216 . note that this also implies @xmath217 . as no label @xmath198 or @xmath8 is present , any curve of intersection that is essential in @xmath2 must also be essential in @xmath121 . suppose there is a curve that is essential in @xmath121 and inessential in @xmath2 . let @xmath122 be the disk in @xmath2 an innermost such curve bounds . by lemma [ lem : disjointdisk ] the disk @xmath122 can not be in @xmath189 . if @xmath122 is in @xmath39 , then it is disjoint from @xmath158 giving a c - strong compression for @xmath121 contrary to our assumption . thus we may assume that all curve of intersection are essential in both @xmath2 and @xmath121 . consider first @xmath218 . it is incompressible in @xmath41 because a compression into @xmath189 would violate lemma [ lem : disjointdisk ] and a compression into @xmath39 would provide a c - weak compression of @xmath121 . if @xmath218 is not essential in @xmath41 then every component of @xmath218 is parallel into @xmath2 so in particular @xmath219 . as it is always the case that @xmath220 this implies that @xmath221 . as @xmath3 is a bridge surface for @xmath0 that is not possible . we conclude that @xmath218 is essential in @xmath41 so by proposition [ prop : proximity2.5 ] for each component @xmath156 of @xmath222 that is not the boundary of a @xmath2-parallel annulus in @xmath41 , the inequality @xmath223 holds , i.e. , @xmath218 satisfies the hypotheses of lemma [ lem : essentialdist ] . by lemma [ lem : disjointdisk ] @xmath206 does not have compressing disks in @xmath224 so it either has no compressing disks in @xmath225 at all or has a compressing disk lying in @xmath39 . if @xmath226 is incompressible in @xmath227 , then either it satisfies the hypothesis of proposition [ prop : proximity2.5 ] or each component of @xmath226 is parallel to a subset of @xmath2 . in the latter case the compressing disk @xmath158 of @xmath206 in @xmath228 can be extended via this parallelism to give a compressing disk for @xmath2 that is disjoint from all @xmath229 . hence @xmath230 as long as @xmath206 is not a collection of @xmath2-parallel annuli . if that is the case , then @xmath231=0 for at least one @xmath232 so @xmath233 as desired . thus in this case @xmath206 satisfies the hypothesis of lemma [ lem : essentialdist ] and we obtained the desired distance bound . if @xmath206 is bicompressible in @xmath8 , then @xmath206 is weakly incompressible as every compressing disk for @xmath206 is also a compressing disk for @xmath3 . because every compressing disk for @xmath206 in @xmath189 intersects @xmath120 , @xmath206 can not be a tube - spanned recessed collar . by theorem [ thm : proximity5.4 ] @xmath206 again satisfies the hypothesis of lemma [ lem : essentialdist ] and we can deduce the desired distance bound . [ lem : noxandy ] if a configuration has labels @xmath203 and @xmath204 , then at least one of the following holds : * @xmath121 is c - strongly compressible , or * @xmath0 is removable with respect to @xmath3 , or * @xmath207 . by the labelling we can conclude that @xmath121 has c - disks in both @xmath31 and @xmath32 that are disjoint from @xmath2 . again we will assume that @xmath121 is c - weakly incompressible and @xmath0 is not removable and show that the distance bound must hold . as before we may assume that all curves of @xmath209 are essential in both surfaces . a curve that is essential in @xmath2 but inessential in @xmath121 would lead to a label @xmath198 or @xmath8 ; a curve that is inessential in @xmath2 but essential in @xmath121 would lead to c - strong compression of @xmath121 as @xmath121 has c - disks in both @xmath31 and @xmath32 with boundaries disjoint from @xmath2 . thus all curves are either essential or inessential in both surfaces and all curves of intersection inessential in both surfaces have already been removed . the c - disks for @xmath121 giving rise to the labels @xmath203 and @xmath204 must both be contained in @xmath41 or @xmath8 by c - weak incompressibility of @xmath121 . suppose @xmath218 is bicompressible and @xmath226 is incompressible , the other case is similar . we may assume that @xmath226 is essential as otherwise a label @xmath192 or @xmath193 would be present and lemma [ lem : noxandy ] would apply . thus @xmath206 satisfies the hypothesis of lemma [ lem : essentialdist ] . on the other hand @xmath218 is a bicompressible c - weakly incompressible surface in a 3manifold . by theorem [ thm : proximity5.4 ] , @xmath218 satisfies the hypothesis of lemma [ lem : essentialdist ] unless @xmath218 has one component that is a tube spanned recessed collar and all other components are boundary parallel . it this case @xmath218 would be disjoint from a spine of @xmath41 and thus the region would have a small label . the result then follows by lemma [ lem : noxandy ] [ lem : noxandy ] if a configuration has labels @xmath192 and @xmath193 , then at least one of the following holds : * @xmath121 is c - strongly compressible , or * @xmath0 is removable with respect to @xmath3 , or * @xmath207 . again assume that @xmath121 is c - weakly incompressible and @xmath0 is not removable . we can also assume that the region does not have any capital labels as otherwise we can apply lemma [ lem : noxandy ] . we may assume that all curves of intersection are essential in both @xmath2 and @xmath121 for a curve that is essential in @xmath2 and inessential in @xmath121 would give rise to a label @xmath198 or @xmath8 and a curve that is essential in @xmath121 and inessential in @xmath2 would give rise to a label @xmath203 or @xmath204 . moreover , as labels @xmath203 or @xmath204 do nt occur , @xmath218 and @xmath206 are c - incompressible in @xmath234 . every spine of @xmath41 intersects @xmath3 so both labels @xmath192 and @xmath193 must be due to a spine of @xmath8 . supposed there is a spine @xmath235 and a spine @xmath236 . let @xmath237 be a component of @xmath206 that lies between the two spines . this implies that @xmath237 is parallel into @xmath2 on both its sides , i.e. , that @xmath238 . as @xmath170 , @xmath239 is not an annulus . let @xmath240 be an essential arc in @xmath237 with endpoints in @xmath2 . then @xmath241 is a meridian disk @xmath122 for @xmath8 that intersects @xmath237 precisely in @xmath240 . consider the frontier of a regular neighborhood of @xmath242 . as @xmath170 the frontier contains at least one curve @xmath161 that is essential in @xmath2 . we can conclude that for every curve @xmath243 , @xmath244 . thus @xmath206 always satisfied the hypothesis of lemma [ lem : essentialdist ] . as we already saw @xmath218 is essential so by proposition [ prop : proximity2.5 ] @xmath245 also satisfies the hypothesis of lemma [ lem : essentialdist ] and we are done by that lemma . let @xmath246 represent any subset of the labels @xmath199 , @xmath247 the label @xmath8 , @xmath248 any subset of the labels @xmath249 and @xmath250 any subset of the labels @xmath251 . then the lemmas in this section can be summarized as follows . [ thm : exactlyone ] suppose @xmath121 is a c - weakly incompressible bridge surface for a knot @xmath0 such that @xmath0 is not removable with respect to @xmath3 and @xmath2 is a heegaard surface for the knot exterior . let labels @xmath252 and @xmath250 be defined as above . then either every configuration of @xmath2 and @xmath121 has exactly one of the labels @xmath253 and @xmath250 associated to it or @xmath254 . let @xmath255 and @xmath256 continue to denote the spines for the @xmath0-handlebody @xmath31 and @xmath32 . then there is a map @xmath257 that is a homeomorphism except over @xmath258 and , near @xmath259 , the map @xmath55 gives a mapping cylinder structure to a neighborhood of @xmath260 . little is lost and some brevity gained if we restrict @xmath55 to @xmath261 . the map @xmath55 is then called a _ sweep - out _ associated to @xmath3 . similarly there is a sweep - out associated to the heegaard surface @xmath2 between the spines @xmath262 and @xmath120 . consider simultaneous sweep - outs of @xmath2 , between @xmath194 and @xmath263 and of @xmath121 between @xmath264 and @xmath265 . this two - parameter sweep - out can be described by a square where each point in the interior of the square represents a position of @xmath2 and @xmath121 . inside the square is a graphic @xmath62 which represents all points where the intersection of the two surfaces is not generic . at each edge of @xmath62 there is a single tangency between @xmath2 and @xmath121 and at each valence 4 vertex there are two tangencies . each component of the complement of @xmath62 will be called a region and to each region we can associate labels as defined in section [ sec : config ] . two regions are adjacent if they share an edge . for the moment we will restrict our attention to the case when @xmath121 is a c - weakly incompressible surface and @xmath0 is not removable with respect to @xmath3 , thus by theorem [ thm : exactlyone ] we may assume that either @xmath200 or every region has exactly one of the labels @xmath252 and @xmath250 . in the former case we have achieved the desired distance bound . as the two regions are adjacent , we can transform the configuration caring the labels @xmath246 into the configuration caring the label @xmath247 by isotoping @xmath2 through a single tangency of @xmath121 . notice that the curves of intersection of @xmath2 and @xmath121 before the isotopy are disjoint from the curves of intersection after the isotopy . thus there is a c - disk for @xmath41 , @xmath266 and a compressing disk for @xmath8 , @xmath267 such that @xmath268 . if @xmath266 is actually a compressing disk , that would imply that @xmath210 . if @xmath266 is a cut - disk , then we apply proposition [ prop : cutimpliescompressing ] and the triangle inequality to conclude that @xmath211 . [ lem : noadjacentxandy ] suppose a configuration is changed by passing through a saddle point , and the bigon @xmath60 defining the saddle tangency lies in @xmath269 ( the case when the bigon lies in the handlebody @xmath8 is similar ) . then * no label @xmath192 or @xmath203 is removed . * no label @xmath193 or @xmath204 is created . * if there is no label @xmath192 or @xmath203 before the move , but one is created after and if there is a label @xmath193 or @xmath204 before the move and none after , then @xmath270 . we first show that no label @xmath192 or @xmath203 is removed . if there is a c - disk for @xmath39 that lies in @xmath41 , a standard innermost disk , outermost arc argument on its intersection with @xmath60 shows that there is a c - disk for @xmath39 in @xmath41 that is disjoint from @xmath60 . the saddle move has no effect on such a disk ( nor , clearly , on a c - disk for @xmath39 that lies in @xmath8 ) . if there is a spine of @xmath41 or of @xmath8 lying entirely in @xmath189 then that spine , too , is unaffected by the saddle move . to prove the third item position @xmath121 so that it is exactly tangent to @xmath2 at the saddle . a bicollar of @xmath121 then has ends that correspond to the position of @xmath121 just before the move and just after . let @xmath272 denote @xmath273 after the move and @xmath274 denote @xmath275 before the move . the bicollar description shows that @xmath272 and @xmath274 have disjoint boundaries in @xmath2 . moreover the complement of @xmath276 in @xmath121 is a regular neighborhood of the singular component of @xmath202 , with euler characteristic @xmath277 . it follows that @xmath278 . first note that any curve of intersection that is inessential in @xmath121 must be inessential in @xmath2 as no labels @xmath198 or @xmath8 are present . to prove the claim , suppose a non - singular curve is inessential in @xmath2 and consider an innermost one . assume first that the disk @xmath122 that it bounds in @xmath2 does not contain the singular curve @xmath279 . if @xmath47 is essential in @xmath121 , then it would give rise to a label @xmath31 or a label @xmath32 that persists from before the move until after the move , contradicting the hypothesis . thus any inessential curve in @xmath2 that does nt bound a disk containing the singular curve @xmath279 is also inessential in @xmath121 and can be removed without affecting the label of the region . suppose then that the disk @xmath280 contains the singular component @xmath279 . by the above , @xmath279 is the only component of @xmath222 in the interior of @xmath122 . when the saddle is pushed through , the number of components in @xmath279 switches from one , @xmath281 , to two , @xmath282 , or vice versa . all three curves are inessential in @xmath2 since they lie in @xmath122 . the curve @xmath283 and at least one of @xmath282 bound disks in @xmath2 whose interiors are disjoint from @xmath121 . if one of these curves was essential in @xmath121 that would give rise to a label @xmath31 or @xmath32 that persists through the isotopy . as no such label exists , both of these curves are inessential in @xmath121 , ie bound a possibly one - punctured disks in @xmath121 . as no sphere in @xmath1 can intersect @xmath0 exactly once , @xmath283 and one of @xmath282 bound disks in @xmath121 . because the three curve cobound a pair of pants in @xmath121 , all three curves @xmath283 and @xmath282 are inessential in @xmath121 . this means that all three curves are removable so passing through this saddle can not have an effect on the labelling . the case in which all three curves are inessential in @xmath2 is covered in the proof of claim 1 . if two are inessential in @xmath2 so is the third . thus the only remaining case is that exactly one of the curves @xmath284 is inessential in @xmath2 and by claim 1 , the disk it bounds in @xmath2 is disjoint from @xmath3 . as before the curve can not be essential in @xmath121 as otherwise it will give rise to a label @xmath31 or @xmath32 that persists through the isotopy . thus the curve is inessential in @xmath121 also ( in fact it must bound a disk there ) so it is removable . if this curve is @xmath282 , passing through the saddle can have no effect on the labelling . if the removable curve is @xmath283 , then the curves @xmath282 are parallel in both surfaces . passing through the saddle has the same effect on the labelling as passing an annulus component of @xmath285 across a parallel annulus component @xmath286 of @xmath245 . this move can have no effect on labels @xmath192 or @xmath193 . as there is a label @xmath204 before the move , there is a meridian , possibly punctured disk @xmath183 for @xmath189 that is disjoint from @xmath2 . this disk would persist after the move , unless @xmath287 is in fact the core curve of the annulus @xmath286 . but then the union of @xmath183 and half of @xmath286 would be a possibly punctured meridian disk of @xmath41 bounded by a component of @xmath288 . in other words , there would have to have been a label @xmath198 before the move , a final contradiction establishing claim 2 . the above two claims allow us to assume that all curves of intersection before and after the move are essential in both surfaces . note that @xmath289 and @xmath290 are c - incompressible ( as there are no labels @xmath203 or @xmath204 persisting through the move ) and have at least one component that is not parallel to @xmath2 ( as there are no labels @xmath192 or @xmath193 persisting through the move ) . now apply proposition [ prop : proximity2.5 ] to both sides : let @xmath291 ( resp @xmath292 ) be a boundary component of an essential component of @xmath289 ( resp @xmath290 ) . then @xmath293 @xmath294 as required . suppose there is such a vertex of @xmath296 and assume that @xmath121 is c - weakly incompressible and @xmath0 is not removable with respect to @xmath3 . as we have already established the desired distance bound if any region has more than one label or if two adjacent regions are labelled @xmath246 and @xmath247 or @xmath248 and @xmath250 we may assume that going around the vertex the regions are labelled in the order @xmath297 and @xmath250 . note then that only two saddle moves are needed to move from a configuration labelled @xmath198 to one labelled @xmath8 . the former configuration includes a c - disk for @xmath2 in @xmath41 and the latter a compressing disk for @xmath2 in @xmath8 . recall that all curves of intersection before a saddle move are disjoint from all curves of intersection after the saddle move . using proposition [ prop : cutimpliescompressing ] and the triangle inequality , it follows that @xmath298 , as long as at least one of the regions labelled @xmath248 and @xmath250 contains at least one essential curve . suppose all curves of @xmath209 in the regions with labels @xmath248 and @xmath250 are inessential . consider the region labelled @xmath248 . crossing the edge in the graphic from this region to the region labelled @xmath246 corresponds to attaching a band @xmath299 with both endpoints in an inessential curve @xmath300 . note that attaching this band must produce an essential curve that gives rise to the label @xmath246 , call this curve @xmath301 . similarly crossing the edge from the region @xmath248 into the region @xmath8 corresponds to attaching a band @xmath302 to give a curve @xmath303 . the two bands must be attached to the same curve @xmath184 otherwise @xmath301 and @xmath303 would be disjoint curves giving rise to labels @xmath246 and @xmath247 . as we assumed that in the region labelled @xmath250 all curves of intersection are inessential , attaching both bands simultaneously results in an inessential curve @xmath304 . but that can only occur of @xmath2 is a torus which we know not to be the case . if a label @xmath247 appears in the regions adjacent to the @xmath262 side of @xmath305 , or a label @xmath246 appears along the @xmath306 side , or a label @xmath250 appears along the @xmath255 side , or a label @xmath248 appears along the @xmath256 side , then at least one of the following holds : suppose @xmath121 is a c - weakly incompressible surface and @xmath0 is not removable with respect to @xmath3 . consider first a region that is adjacent to the @xmath262 side of @xmath305 . in such a region @xmath2 is the boundary of a small neighborhood of @xmath262 and @xmath3 either intersects it in meridional circles or does nt intersect it at all . in the former case any curve of intersection @xmath209 would lead to a label @xmath246 so if a label @xmath247 also appears , then @xmath208 by lemma [ lem : noaandb ] . in the latter case no label @xmath247 , or in fact label @xmath246 , is possible . similarly if @xmath2 is near @xmath263 and the configuration carries a label @xmath246 then @xmath208 . suppose now that @xmath121 is near @xmath264 . as @xmath2 intersects @xmath264 transversely , all but a finite number of points of @xmath264 will be disjoint from @xmath2 . thus a label @xmath248 necessarily occurs . if a label @xmath250 also occurs , then @xmath307 by lemma [ lem : noxandy ] . symmetrically the distance bound holds is some region adjacent to the @xmath256 boundary of the square is labelled @xmath248 [ thm : sperner][sperner s lemma ] suppose a square @xmath308 contains a graph @xmath62 such that all vertices of @xmath62 in the interior of the square are valence 4 and all vertices contained in the boundary of the square are valence 1 . suppose each component of @xmath309 is labelled with exactly one of the labels @xmath310 or @xmath311 in such a way that 1 . no region on the east side of @xmath308 is labelled @xmath311 , no region on the west side is labelled @xmath158 , no region on the south side is labelled @xmath79 and no region on the north side is labelled @xmath43 . no two adjacent regions are labelled @xmath158 and @xmath311 nor @xmath79 and @xmath43 . [ thm : c - weaklyincomp ] let @xmath0 be a knot in a closed irreducible 3manifold @xmath1 and let @xmath2 be a heegaard splitting of the knot complement such that @xmath312 . suppose @xmath3 is a bridge surface for @xmath0 . then at least one of the following holds : suppose that @xmath121 is c - weakly incompressible and @xmath0 is not removable with respect to @xmath3 . consider a 2-parameter sweep - out of @xmath2 and @xmath121 and the associated graphic @xmath313 . label the regions of @xmath309 with the labels @xmath253 and @xmath250 as described before . by theorem [ thm : exactlyone ] we may assume that each region has a unique label . by lemmas [ lem : noadjacentaandb ] and [ lem : noadjacentxandy ] we may assume that no two adjacent regions are labelled @xmath246 and @xmath247 or @xmath248 and @xmath250 . finally lemma [ lem : edges ] shows that the labels of regions adjacent to the boundaries of @xmath308 satisfy the conditions that no label @xmath247 appears in the regions adjacent to the @xmath262 side of @xmath305 , no label @xmath246 appears along the @xmath306 side , no label @xmath250 appears along the @xmath255 side , and no label @xmath248 appears along the @xmath256 side of the square . by theorem [ thm : sperner ] there is a valence 4 vertex of @xmath62 that is surrounded by regions carrying all four labels . the distance bound then follows from lemma [ lem : vertexlabels ] . let @xmath0 be a knot in a closed oriented irreducible 3manifold @xmath1 and let @xmath2 be a heegaard splitting of the knot complement of genus at least two . suppose @xmath3 is a bridge surface for @xmath0 . then either if @xmath121 is stabilized , meridionally stabilized or perturbed , as described in section [ sec : defin ] , there is a bridge surface for @xmath0 , @xmath72 , such that @xmath315 . by possibly replacing @xmath121 by @xmath130 we may assume @xmath121 is not stabilized , meridionally stabilized or perturbed . if @xmath121 is removable , then again by the results in section [ sec : defin ] there is a heegaard surface @xmath316 for @xmath29 that is isotopic to @xmath3 in @xmath1 . by hypothesis @xmath75 is not isotopic to a possibly stabilized copy of @xmath2 so the result follows by theorem [ thm : alternate ] . if @xmath121 is c - weakly incompressible , the result follows by theorem [ thm : c - weaklyincomp ] . the only remaining case is that @xmath121 is a c - strongly compressible bridge surface for @xmath0 that is not stabilized , meridionally stabilized , perturbed or removable . by theorem [ thm : essentialexists ] in this case there is an essential surface @xmath51 with @xmath317 . using theorem [ thm : essentialsurface ] we deduce that @xmath318 concluding the proof . in the case when the manifold is @xmath319 we can also eliminate the restriction on the genus of @xmath2 by assuming that @xmath0 is nontrivial as that implies that the genus of any heegaard splitting of @xmath320 is at least 2 . * @xmath314 , or * @xmath0 can be isotoped to be disjoint from @xmath3 so that after the isotopy @xmath3 is a heegaard surface for the knot exterior that is isotopic to a possibly stabilized copy of @xmath2 . notice that if @xmath0 can be isotoped to be disjoint from @xmath3 so that @xmath3 is a heegaard surface for @xmath29 that is isotopic to a possibly stabilized copy of @xmath2 , then @xmath321 . the following corollary then follows immediately from theorem [ thm : main ] if @xmath322 and from corollary [ cor : sphere ] if @xmath323 as the only manifold with a genus 0 heegaard splitting is @xmath319 . [ cor : main ] let @xmath0 be a knot in a closed irreducible 3manifold @xmath1 and let @xmath2 be a heegaard splitting of the knot complement . suppose @xmath3 is a bridge surface for @xmath0 such that @xmath324 . then @xmath314 .

in spiral galaxies , the hi surface density declines with increasing radius to a point where it is seen to truncate dramatically in the best observed cases . it was anticipated that if the ambient radiation field is sufficiently strong , there exists a maximum radius beyond which the cold gas is unable to support itself against ionization . we have now succeeded in detecting ionized gas beyond the observed hi disk in spirals . here , we report on our findings for the sculptor galaxy ngc 253 . the hi disks in sculptor galaxies extend to only about 1.2 although we have detected ionized gas to the limits of our survey out to 1.4 . this has important ramifications for spiral galaxies in that it now becomes possible to trace the gravitational potential beyond where the hi disk ends . the detections confirm that the rotation curve continues to rise in ngc 253 , as it appears to do for other sculptor galaxies from the hi measurements , but there is a hint that the rotation curve may fall abruptly not far beyond the edge of the hi disk . if this is correct , then it suggests that the dark halo of ngc 253 may be truncated near the hi edge , and provides further support for the link between dark matter and hi . the line ratios are anomalous with to h ratios close to unity . while metallicities at these large radii are uncertain , such enhanced ratios compared to solar - abundance hii regions ( / ) are likely to require selective heating of the electron population without further ionization of n . we discuss the most likely sources of ionization and heating , and the possible role of refractory element depletion ( e.g. ca , si , fe ) onto dust grains .

astro-ph9706210

the case for dark matter in galaxies rests primarily on those spiral galaxies where the hi rotation curve has been measured at radii which are several times larger than the optical disk . at these large radii , the observed rotation of the gas is fully a factor of two larger than that expected for circular orbits in the potential field of the luminous galaxy ( van albada 1985 ) . many of these extensive rotation curves are flat or even rising at the outermost points which means the edge of the mass distribution has not yet been found ( puche & carignan 1991 ) . in an attempt to find the edge , several observers have pushed their hi observations to higher sensitivity ( van gorkom 1993 ; corbelli 1989 ) . these observations have shown that hi disks are abruptly truncated at column densities near 10@xmath3 atoms @xmath4 where the rotation curves are still flat . one possible explanation for the truncation is that , at large radii , the thin hi disks become fully ionized by the metagalactic uv background ( bochkarev & sunyaev 1977 ) . if this is the case , the outer parts of gaseous disks should be emitting h@xmath0 photons . maloney ( 1993 ) has shown that the expected h@xmath0 emission measure from the ionized disk beyond the truncation point falls in the range 0.0250.25 @xmath5 pc . after demonstrating that these flux levels are quite feasible with the taurus-2 fabry - perot interferometer ( bland - hawthorn 1994 , hereafter btvs ) , we now use this technique to search for the signature of ionized gas at and beyond the hi edge in sculptor spirals . the implications of a positive detection are profound . if the outer disk could be detected , then we would firstly know how extensive galactic disks ( and dark haloes ) really are and , secondly , we could continue tracking the rotation curve in order to find the edge of the halo and hence find the total mass of galaxies . a positive detection also has important implications for explaining low redshift ly@xmath0 absorption - line systems towards quasars , and for constraining the poorly known metagalactic ionizing background . in @xmath62 , we describe the observations carried out at the aat . in @xmath63 , we derive the expected level of emission at an hi edge , and then describe the experimental procedure to achieve such faint detections . the reduction and analysis steps are briefly outlined in @xmath64 , the results of which are discussed in @xmath65 . in @xmath66 , we attempt a physical interpretation of both the gas kinematics and the gas excitation before drawing our conclusions in @xmath67 . the observations were carried out over three long dark nights ( 1994 aug 9@xmath712 ) at the f/7.91 cassegrain focus of the aat 3.9 m . follow - up observations were carried out at f/14.9 on 1995 sep 27 . the taurus-2 interferometer was used in conjunction with the refurbished hifi 40@xmath8 m gap queensgate etalon and a 4@xmath7cavity blocking filter ( 90% peak transmission ) , centered at @xmath96555 , with a 45 bandpass well matched to the etalon free spectral range ( 54.3 at h@xmath0 ) . the 50 mm filter was placed out of focus close to the focal plane and baffled to give a 5.0 field . the etalon was tilted by 3.4 to shift the optical axis to the field edge . an in - focus , focal plane , colander mask was used to ensure that low - order ghosts fall outside the field of view . the taurus-2 f/8 pupil diameter is 59.9 mm which is oversized for the the 50 mm diameter etalon . due to uncertainties of the precise location of the optical cavity within the etalon , we placed a 45.0 mm aperture stop immediately in front of the etalon . the pupil stop introduced a major loss of light ( 50% ) compounded by a cassegrain hole which is 17% of the total pupil area . the full pupil was used in the f/15 observations . the observational set - up was the same except for a 75 mm blocking filter centered at 6585 which was not baffled thereby producing a field of view similar to the f/8 observations . observations were made at four positions in ngc 253 ( plates 1 and 2 ) with several discrete tunings of the etalon at each position ( table 1 ) . the etalon was used at fixed gap spacings and the resulting ring pattern was imaged onto a tek 1024@xmath10 ccd with pixel scales of 0.594 pix@xmath11 ( f/8 ) and 0.315 pix@xmath11 ( f/15 ) with read noise @xmath12 2.3e@xmath13 . the wavelength range @xmath14 is dispersed quadratically over the 5 field at 1 resolution with blue wavelengths to the north . the instrument and detector were rotated to a position angle of 230 to align the detector with the galaxy major axis . at both foci , we observed two stellar flux standards ( table 1 ) ; at f/8 , we also observed four planetary nebulae . observations were also made of blank fields and in the direction of the smith high velocity cloud ( smith 1963 ; wakker 1991 ) , and twilight flats were taken on all nights with and without the etalon . we now derive the expected flux levels from spiral edges ionized by a metagalactic radiation field . the surface density of hi gas falls exponentially in spiral galaxies . a point is reached where the hi column is no longer able to support itself against ionization by the cosmic radiation field . a nave calculation can be used to predict the expected column density at which this is likely to occur , in addition to the local electron density , , and the expected emission measure , . more sophisticated treatments are given in maloney ( 1993 ) and dove & shull ( 1994 ) . we shall assume that , at large galactocentric radius , the ionization rate is in rough equilibrium with the rate of recombination . therefore , along an axis @xmath15 perpendicular to the galactic disk , we find @xmath16 where the recombination coefficient depends on the neutral hydrogen column and the local electron temperature ( case b : @xmath17 @xmath18 s@xmath11 at t@xmath19k ) . the local electron density ( and therefore the proton density ) , is related to the neutral hydrogen density , , through the ionizing fraction @xmath20 such that @xmath21 . equation [ ion_balance ] assumes that the neutral gas sheet is ionized from both sides . we approximate the cosmic ionizing flux level as @xmath22 where is the metagalactic flux at the lyman limit ( @xmath23 ) in units of . if the cosmic ionizing field ( e.g. haardt & madau 1996 ) is due largely to quasars at high redshift , @xmath24 is approximately unity . for an exponential disk with density profile @xmath25 , the electron density in the plane of the galaxy at the ionized edge is @xmath26 where the vertical scale height of the disk @xmath27 is in parsecs . we assume a value of @xmath28 pc throughout . the particle column density perpendicular to the disk is @xmath29 for a @xmath30 upper limit of @xmath31 ( vogel 1995 ; bland - hawthorn 1997 ) , the gas is fully ionized when @xmath32 @xmath33 and @xmath34 @xmath4 . the latter is in good agreement with observations ( e.g. corbelli 1989 ) . a crude upper limit on the emission measure is @xmath35 this surface brightness , equivalent to 90 millirayleighs ( mr ) and @xmath36 erg @xmath4 s@xmath11 arcsec@xmath37 , is reached at the 3@xmath38 level by the fabry - perot ` staring ' technique in less than one hour of observation ( btvs ) . s@xmath11 sr@xmath11 or 2.41@xmath3910@xmath40 erg @xmath4 s@xmath11 sr@xmath11 at . ] we exploit the ` staring ' technique to obtain a single , extremely deep spectrum of a diffuse source which fills the field of view . for a fixed gap spacing , @xmath41 where @xmath42 is the angle of an incoming ray of wavelength @xmath9 to the optical axis . the spectrum in a narrow band ( @xmath4340 ) is dispersed radially from the optical axis across the field ( see fig . 3 ) . like long - slit spectrometers , the instrumental profile is projected onto the detector but the line fwhm in pixels varies across the field inversely with @xmath42 . complete fabry - perot rings have constant surface brightness and equal spectroscopic resolution . after flatfielding and point source removal , the data are binned azimuthally and resampled to a linear axis to obtain a single deep spectrum ( e.g. figs . 4 and 5 ) . maximum sensitivity is achieved when ( a ) the detector covers the unvignetted field , ( b ) the line - emitting source fills the field of view , and ( c ) the velocity range and internal kinematic dispersion , taken together , are comparable to the instrumental profile width . these criteria determine which galaxies are best matched to a given instrument . the isovelocity contours of a disk undergoing flat rotation show two distinct regimes . the central region of solid body rotation gives rise to contours which are parallel to the kinematic minor axis . at the turnover radius , the contours become radial and therefore subtend a fixed angle as seen on the sky about the kinematic major axis . the spread in velocity within a fixed field of view declines inversely with radius until we reach the intrinsic spread of an isothermal hi disk ( @xmath12 10 ; kamphuis 1993 ) . an important concept is the angular extent of the `` innermost monochromatic field '' ( imf ) for a given instrument , telescope and object . we define this as the field at the smallest radius along the disk major axis within which the observed spread in velocities does not exceed the instrumental resolution . the necessary condition for a galaxy with maximum deprojected velocity inclined at an angle @xmath44 is @xmath45^{-1 } + \sigma^2_{\rm hi}\ \ \leq\ \ \sigma^2_{\rm fp}\ ] ] where is the instrumental width ( in ) and @xmath46 is the radius to the field center ( in kpc ) . truncated hi edges typically occur at radii larger than the semi - major axis distance ( in kpc ) of the @xmath47 mag arcsec@xmath37 contour . for a fixed field ( in kpc ) , this condition is most easily satisfied in large , nearby galaxies where the instrument field of view is a small fraction of the object size . the constant @xmath8 allows for two different approaches to sky subtraction . if this is achieved through a separate exposure , @xmath48 ; otherwise , in a properly matched experiment , we assume that half the field is given over to the sky background for which @xmath49 . the optimal disk inclination is unclear : face - on galaxies have a smaller velocity spread and subtend larger solid angles , whereas an inclined disk has a higher projected column density . a potential problem in edge - on disks is flaring due to the declining stellar surface density ( olling 1995 ) which tends to reduce the projected emission measure . to minimize the velocity spread , we rotate the instrument so that the incomplete ring pattern is convex in the same sense as the projected velocity field , and is centered on the kinematic major axis ( see figs . 1 and 2 ) . a detailed discussion of the many subtleties of data analysis are to be presented elsewhere ; a brief overview is given here . an important step is establishing the optical axis of the incomplete calibration rings to better than 0.1 pix . this was achieved with orthogonal distance regression ( e.g. boggs , byrd & schnabel 1987 ) . next , we azimuthally bin the calibration rings for all nights to ensure that ( a ) there were no opto - mechanical shifts , ( b ) the instrumental response was constant , and ( c ) the etalon gap zero - point was constant from night to night . in ( b ) , the etalon was found to behave reliably except that there was a slow drift in the optical gap during the first night . some calibration spectra showed baseline variations after binning over different parts of the field . this arises from illumination ( vignette ) effects , stray light , chip structure and ccd fringeing which can be divided out reliably with flatfields . the spectral bandpass seen by individual pixels is roughly 1 where the bandpass centroid declines by 40 from the optical axis to the field edge . twilight flats were found to be the most reliable except that the fraunhofer spectrum leaves its imprint in the data . we divide out the solar spectrum from the flatfield by establishing the mean spectrum and then generating a polar image with this spectrum . occasionally , the flatfields leave residual fringe structure in the data . it is possible to remove this with azimuthal smoothing but potentially informative , intensity variations in the data will be washed out . ccd fringeing constitutes the main systematic error in diffuse detection and provides a major challenge to achieving deeper detection limits than that quoted by btvs . the response of the blocking filter is removed in the twilight division . to obtain discrete spectra from the summed ccd images , the data were divided into annular rings two pixels wide . within each annulus , we can calculate the mean , median or mode of the histogram . cosmic ray events appear as outliers and are therefore easily removed . we tried more sophisticated methods ( e.g. the bi - weight statistic ) but the improvement was found to be marginal . the weak underlying galaxy continuum observed in ngc 253 had only a negligible effect on the final spectra . this was examined by subtracting a matching @xmath46 band exposure modulated by the polar response of the blocking filter . this same image was used to identify unresolved sources within the field . sky subtraction is particularly hazardous since atmospheric humidity produces variable water vapour features . if the object fills the field of view , as for ngc 253 , it is necessary to obtain off - field exposures at the same airmass and at comparable humidity . the oh lines are time variable but , most importantly , they vary with the zenith angle of the observation ( kondratyev 1969 ) . this is quite distinct from the geocoronal lines which reach a minimum at local midnight rising sharply towards dawn . the night sky lines allow for wavelength calibration to better than 0.04 ( osterbrock & martel 1992 ) as shown by the difference in wavelength produced by a discharge lamp and solar ly@xmath24 resonance excitation of the exosphere ( yelle & roesler 1985 ) . the velocity centroid of the galactic emission ( ` reynolds layer ' ) , using new wavelengths from spyromilio ( 1995 ) , provides an independent measurement of the earth - sun motion with respect to the local standard of rest . for all four nights , the conditions were consistently photometric as judged by the reynolds layer emission . the photometric calibration was achieved using both line and continuum flux standards ( table 1 ) . the four taurus-2 field positions are shown in figs . 1 and 2 overlaid on a @xmath50 image of ngc 253 and the deepest hi image to date for this galaxy . a magnified section of one interferogram is shown in fig . 3 and summed spectra are shown for the two outermost fields in figs . 4 and 5 . the ` staring ' method allows for an extremely deep spectroscopic detection over pixels that fall within the projected arc defined by a discrete frequency . there are two frequencies of interest so that kinematic measurements are possible at two positions within each field . the etalon was tuned to two different spacings for the two innermost fields so that we have kinematic measurements at no more than 12 independent positions along the major axis . these data are presented in figs . 6 and 7 . optical line emission is detected in all field positions . in fields sw1 and sw2 , bright hii regions were clearly seen in both and . these short exposures were taken to establish independently the systemic velocity of the gas . the kinematic measurements are seen in fig . 6@xmath51 at @xmath52 . the large systematic errors in both axes are due to azimuthally averaging over a large spread in velocities and radial distances within the projected plane at small galaxian radii . field sw3 is the imf defined by the criterion in equation [ imf ] . weak stellar continuum emission is seen over the central third of the field extending from roughly @xmath53 23 mag arcsec@xmath37 ( ne ) to 25 mag arcsec@xmath37 ( sw ) . the and line emission comprise clumpy structure superimposed on a faint diffuse component . only two of the sources are unresolved : these show up on our matched @xmath54 images and are labelled in fig . 3 . the two bright knots at the position of the line are probably faint hii regions . the summed spectrum for field sw3 is shown fig . the difference of the on - object and off - object spectra clearly shows the declining stellar disk along the major axis , in addition to the and lines at 6557.7 and 6572.8 . while these lines fall at different spatial locations within the galaxy , the strength of the blue line with respect to is striking , a result which holds after removing the bright knots . the weakness of the galaxy continuum rules out a significant correction to the line flux due to stellar absorption ( bica & alloin 1987 ) . we achieved the same basic result in the 1995 september 27 observations although the signal - to - noise ratio was roughly half that of the earlier observations . both sets of observations show that the diffuse emission peaks towards the major axis . the summed spectrum for field sw4 is presented in fig . this field lies beyond the hi edge of ngc 253 . unfortunately , we did not manage to obtain a matching sky observation . the data clearly show weak emission and possibly broad diffuse emission . we have attempted to subtract something approximating a background spectrum by dividing up the field into three vertical panels which have equal area when the outer panels are taken together . the central section constitutes our on - object spectrum , the outer quarter panels our off - object spectrum . the weak outer envelope of the galaxy is seen in the difference spectrum . the emission peaks towards the major axis and is not detected in the outermost panels . broad diffuse emission is seen in both regions at comparable intensity and therefore subtracts cleanly in the residual spectrum . that this feature is real can be seen by comparing the lower spectrum in fig . 5 with the off - object spectrum in fig . 4 where such a feature is not seen . the radial velocities for the emission lines in figs . 4 and 5 were corrected for the earth s motion before subtracting the systemic velocity of ngc 253 . these measurements are presented in figs . 6 and 7 . the innermost taurus-2 measurements were used to establish the systemic velocity of the ionized gas . schommer ( 1993 ) show the systematic uncertainties which can arise between hi and ionized gas kinematics . our optical determination ( @xmath55 ) is consistent with the weighted - average hi measurements ( 249@xmath568 ) from the nasa / ipac extragalactic database . we adopt the hi value 245@xmath565 ( puche 1991 , hereafter pcvg ) as we wish to relate the optical data to their data . for the measurements between @xmath57 and @xmath58 , the kinematic error is dominated by the uncertainty in systemic velocity . the smaller error bars compared to the inner taurus-2 measurements reflect the smaller intrinsic radial and kinematic dispersion . that the optical values appear systematically high by 10 is discussed in the next section . for the outermost point , the diffuse broad feature has a large kinematic measurement uncertainty . unlike the other lines , this feature has fairly uniform intensity over the full field . its lack of continuity with the other measurements calls into question its association with the galaxy . the major axis measurements in ngc 253 pose an interesting problem . in fig . 6 , the `` rotation curve '' appears to flatten off before rising at a radius @xmath59 . this effect is seen for both the approaching and receding sides of the disk , with ( fig . 6@xmath60 ) or without ( fig . 6@xmath61 ) the correction for orbit inclination . in fig . 6@xmath51 , the taurus-2 measurements confirm and extend this trend out to @xmath62 . the outer taurus-2 values are somewhat higher than the azimuthally averaged hi rotation curve , but this is a distinctive feature of the approaching side of the disk ( fig . 6@xmath61 ) in that the outer hi measurements here are higher than average . if we assume that the distribution of mass has spherical symmetry , then from poisson s equation , @xmath63 where @xmath64 and @xmath65 are the local mass density and the circular velocity . for the flat region of @xmath65 , @xmath64 declines as @xmath66 ; the sudden rise requires the density profile to become much flatter or even constant with increasing radius . a slightly flattened potential gives the same result although the inferred total mass will be overestimated in this case . the inflexion in the rotation curve can be understood naturally to arise from the potential of an exponential stellar disk with a much larger dark halo ( e.g. carignan & freeman 1985 ) . to illustrate this , we show a variety of possible 3-component fits to the data . in the first model ( fig . 7@xmath61 ) , the hi and stellar disks have the same form . the dark halo is represented by an isothermal sphere , which is fully specified by any two of the halo central density , @xmath67 , the central velocity dispersion @xmath68 , or the core radius @xmath69 . the stellar disk is specified by a central surface density @xmath70 and a scale length @xmath71 . the hi surface density profile is given in pcvg ( fig . 5 ) ; the central hole within @xmath72 causes the gravitational force due to the gas to be directed outward ( e.g. staveley - smith 1990 ) . we have taken @xmath68 @xmath53 253 , @xmath73 kpc , @xmath74 kpc@xmath37 and @xmath75 kpc . these values compare well with pcvg except that the disk and halo scale lengths are 30% smaller than their values . in order to match the observed data , the model rotation curves have been projected to a plane inclined at 72 . we note that such a large core radius is normal for the dark halo of large disk galaxies ( freeman 1993 ) . in the second model ( fig . 7@xmath60 ) , the radial forms of the disk and halo are given by dehnen & binney ( 1997 ) . for the halo , we adopt @xmath76 with @xmath77 the halo parameters are the halo density normalization , @xmath78 , the halo scale radius , @xmath79 , the outer cut - off radius , @xmath80 , and the inner and outer power slopes @xmath81 and @xmath60 . apart from the halo axis ratio ( @xmath82 ) , all of these parameters were obtained from a fit to the rotation curve . both the stellar and gaseous disks are assumed to be exponential in form , i.e. @xmath83 for each of the disks , the adopted parameters are the surface density normalization , @xmath84 , the disk scale height , @xmath85 , and the disk scale radius and inner cut - off radius , @xmath86 and @xmath87 . for the hi disk , these were fixed at @xmath88 kpc@xmath37 , @xmath89 kpc , @xmath90 kpc , and @xmath91 kpc . for the stellar disk , @xmath92 kpc , @xmath93 kpc , and @xmath94 kpc as suggested by the photometry . the surface density normalization for the stellar disk was obtained from the least - squares fit . the fit , kindly undertaken by w. dehnen , ignored the last taurus-2 measurement to yield the following results : @xmath95 kpc@xmath37 , @xmath96 kpc@xmath97 , @xmath98 , @xmath99 , @xmath100 kpc and @xmath101 kpc . of these , @xmath60 was constrained to be greater than or equal to unity . 7@xmath60 shows that the rising halo contribution to the rotation curve gives a better fit to the data overall . but there are strong correlations between the various parameters and the fitted results depend critically on which of the taurus-2 points are included . the outermost taurus-2 measurement in figs . 6@xmath51 and 7 suggests that the rotation curve may be falling beyond a radius of about 10 kpc , but there are certainly other possible explanations for its low observed velocity . for example , _ ( i ) _ it may result from tidal distortion of the outermost regions of ngc 253 , or _ ( ii ) _ from accretion of a faint gas - rich object at large radius . in support of the accretion idea , a stacked photographic image by d. malin shows extended stellar light at faint levels ( @xmath4326 mag arcsec@xmath37 ) to the sw of the disk ; similar emission is not seen to the ne . _ ( iii ) _ while the emission may arise from an outer ionized envelope , it is conceivable that this emission is associated with the group itself . an extragalactic cloud has recently been discovered in the fornax cluster ( bland - hawthorn 1995 ) . mathewson , cleary & murray ( 1975 ) and arp ( 1985 ) found tentative evidence for hi clouds within the sculptor group ( cf . haynes & roberts 1979 ) . it is difficult , however , to reconcile the radial velocity ( @xmath12440 ) with the magellanic stream which passes in front of the sculptor group ( mathewson & ford 1984 ) . finally , we consider the possibility that the rotation curve of ngc 253 is indeed falling beyond 10 kpc . such a rapid decrease suggests that the dark halo may be truncated . we can illustrate the effect of such truncation with a simple model . we take the halo density distribution to be @xmath102 with the rotation curve @xmath103 given by @xmath104\ ] ] for @xmath105 and core radius @xmath106 kpc , the rotation curve for this model is very similar to that of the isothermal halo in fig . 7@xmath61 : the optical galaxy lies well within the core radius of the halo , and the halo rotation curve is close to solid - body . we now truncate the halo at @xmath107 kpc ; its rotation curve is then keplerian for @xmath108 kpc . 7@xmath51 shows how the resulting total rotation curve now provides an acceptable fit to all of the hi and taurus-2 data . the model is artificial ( the halo is sharply truncated , and we did not truncate the stellar and hi disks ) , but the point is obvious enough : truncation of the dark halo near the observed outer edge of the hi disk can produce the apparent falling rotation curve . if this is all correct so far , there are some interesting consequences . the properties of dark halos are best studied in disk galaxies for which the hi distribution extends well beyond the optical distribution . in the outer regions of these galaxies , the ratio of ( dark matter surface density)/(hi surface density ) is roughly constant ( bosma , 1978 ; carignan 1991 ) . ngc 253 is not such a galaxy . its hi extent is similar to its optical extent ( pcvg ) . our data provides the first hint that , in galaxies like ngc 253 where the hi and the light are co - extensive , the dark matter also may not extend much beyond the optical distribution . it is tentative evidence of the apparent link between the dark matter and the hi . we emphasize that the dark matter is still essential to generate the observed rotation curve for ngc 253 : the inferred mass ratio of dark matter to luminous matter is about 5 . our point is simply that the dark matter distribution in ngc 253 may be truncated at a radius of only 10 kpc , compared to the much larger dark halo distributions observed in our galaxy ( freeman 1996 ) and other large spirals ( zaritsky 1997 ) . the reduced chi square for the models presented in fig . 7@xmath61 and 7@xmath60 are 5.2 and 2.9 respectively . the latter model has the highest statistical significance of all models , but the basic assumptions are unlikely to be physical . a flat rotation curve given by equation [ arctan ] does not fit the measurements adequately . the reduced chi square is 6.3 and , more significantly , 10 of 12 points outside of @xmath59 miss the curve by more than 2@xmath38 . but the inclusion of a truncation radius lowers the reduced chi square to 3.9 and the curve now passes through all the outer points ( fig . 7@xmath51 ) . the most striking feature of the spectra presented in figs . 4 and 5 is the inferred high / ratios . in order to ensure that the whitelight calibration adequately removed the filter response over the field ( @xmath64 ) , we examined the line ratios in hii regions over the inner disk . where we have two interference rings that fall close together on the sky one from , the other from [ nii ] the hii regions appear weak in [ nii ] compared to , indicative of the order of magnitude difference expected for inner disk hii regions . if the outer hi gas has sub - solar abundances ( pagel 1989 ; diaz 1989 ) , the enhanced line ratio must arise from one or more excitation processes . however , to date , numerical models involving a single ionizing source have not managed to produce such an enhancement ( e.g. sokolowski 1994 ) . thus , we separate our discussion of ionization and excitation wherever possible . almost certainly , the anomalous ratio is indicative of a higher local electron temperature for which there are various mechanisms . in solar - abundance hii regions , this ratio rarely exceeds 0.15 ( evans & dopita 1985 ) : to substantially increase this ratio requires that we selectively heat the electrons without producing . the emission measure from the surface of a cloud embedded in a bath of ionizing radiation gives a direct gauge , independent of distance , of the intensity of the ambient radiation field beyond the lyman continuum ( lyc ) edge ( hogan & weymann 1987 ) . this assumes that the covering fraction _ seen by the ionizing photons _ is known and that there are sufficient gas atoms to soak up the incident ionizing photons . at electron temperatures of 10@xmath109k , collisional ionization processes are negligible . perpendicular to the surface , from equation [ ion_balance ] , the column recombination rate in equilibrium must equal the incident ionizing photon flux , @xmath110 , where is the rate at which lyman continuum photons arrive at a planar cloud surface ( phot @xmath4 s@xmath11 ) , and is the column density of ionized hydrogen . the emission measure produced by this ionizing flux is @xmath111 where @xmath112 ( bland - hawthorn & maloney 1997 , hereafter bm ) . for an optically thin cloud in an isotropic radiation field , the solid angle from which radiation is received is @xmath113 , while for one - sided illumination , @xmath114 . however , can be anisotropic and @xmath115 can be considerably less than @xmath116 . the measured surface brightness values for the emission lines in figs . 4 and 5 are as follows : for the line , 90@xmath565 mr ( @xmath117 ) , 46@xmath565 mr ( @xmath118 ) ; for the line , 81@xmath566 mr ( @xmath62 ) , 41@xmath568 mr ( @xmath119 ) . if the underlying hi is optically thick to the lyman continuum , the lines require an ionizing flux of @xmath120 and @xmath121 . the present upper limit on the metagalactic ionizing flux ( @xmath63.1 ) is @xmath122 ( @xmath30 ) which sets an upper limit on the one - sided ionizing flux at the surface of an optically thick hi sheet of @xmath123 , thereby ruling out the metagalactic background as the dominant ionizing source . the combined ionizing radiation from hot gas and galaxies within the sculptor group is also much too weak . sciama ( 1995 ) predicts an order of magnitude higher ionizing flux from decaying neutrinos which could conceivably account for much of the flux . however , the narrow energy bandpass of decaying neutrinos ( sciama 1996 ) at the lyman limit can not produce , and therefore requires an additional source of ionization . the most recent summary of the macho project indicates that as much as half of the dark matter in the galaxy out to the lmc is made up of solar mass objects . there is wide disagreement on whether the missing mass could comprise a halo population of white dwarfs ( adams & laughlin 1996 ; kawaler 1996 ; chabrier , segretain & mera 1996 ) . but most plausible models invoke a population which produces essentially no uv flux today . this is supported by high redshift observations of the precursor halo population ( charlot & silk 1995 ) . unlike ellipticals , spiral galaxy disks tend not to `` grow '' when photographic plates are stacked or amplified to reveal light at very faint levels ( malin 1983 ) . but a photographically amplified plate of ngc 253 ( malin 1981 ) reveals a faint blue disk extending to the limits of our outermost detection ( @xmath124 ) . following a suggestion from g.b . field , we now consider whether these could comprise an extended white dwarf population ( including the precursor population of central stars in planetary nebulae ) young enough to produce a significant uv disk flux . such a population was originally broached by lyon ( 1975 ) and bania & lyon ( 1980 ) . in fig . 8 , we consider whether white dwarfs are able to account for the diffuse line emission in concert with the faint stellar continuum at the hi edge in ngc 253 ( barred arrow ) . we present the expected emission measure versus @xmath50 surface brightness for an isothermal population of white dwarfs as a function of the disk surface density . the contribution to both of these quantities from white dwarfs in the solar neighbourhood is also shown ( filled symbols ) . the luminosity function is taken from liebert , dahn & monet ( 1988 ) and the luminosity@xmath7temperature conversion from wood ( 1990 ) . there are two basic problems in using white dwarfs to explain the disk ionization . in the solar neighbourhood , the combined effect of all white dwarfs could produce @xmath125 which fails to explain the reynolds layer emission by an order of magnitude or more , in agreement with nordgren , cordes & terzian ( 1992 ) . if we assume a comparable star formation history in the outer reaches of ngc 253 , this model could conceivably be used to explain the measured emission . this is particularly attractive in that the high temperatures of the white dwarf population produce elevated / ratios ( sokolowski & bland - hawthorn 1991 ) . however , if the luminosity function resembles the solar neighbourhood , the combined disk surface brightness at @xmath50 from the cooler white dwarfs is very much brighter than observed , particularly when one includes the contribution of early main sequence stars . alternatively , attempting to explain the low emission measure in terms of 2000 white dwarfs kpc@xmath37 at 45,000 k leads to a timing problem . we have measured an age - temperature relation from wood ( 1990 ) for which @xmath126 stars hotter than 40,000 k have left the asymptotic giant branch less than 10@xmath127 years ago which would require a highly contrived star formation history throughout the disk . ryan has suggested the possibility of hot horizontal branch stars , in particular , an isothermal population with temperatures around 15,000 k ( lee 1993 ) . if we assume comparable line blanketing with white dwarfs , fig . 8 illustrates that such a population will always produce too much blue light for the required ionizing flux , irrespective of the uncertain sizes and surface densities of both objects . the characteristic `` temperature '' of the ionizing source needs to be sufficiently high to fall within the strict @xmath50 band limit . interestingly , this is just what is needed to significantly enhance the / ratio as photons with higher mean energies produce higher electron temperatures . _ ram pressure heating . _ mathis ( 1986 ) has stressed the problems associated with shocks as a source of ionization and heating for the reynolds layer , particularly the near - uniformity of the emission . an interesting possibility is ram pressure heating as ngc 253 moves through an external medium where the disk is inclined at some angle to its direction of motion . the wake of ngc 7421 ( ryder 1996 ) suggests that such a process can take place . to ionize a column of 10@xmath3 atoms @xmath4 requires shock speeds close to 130 ( dopita & sutherland 1996 ) . a medium with pre - shock density @xmath128 @xmath33 moving through an ambient medium with the relatively high density @xmath129 @xmath33 at 400 could produce the necessary ram pressure . the post - shock temperature at a shock velocity of 130 is @xmath130 which is a factor of ten higher than the upper limit from the doppler parameter ( 22 ) of the observed line profiles . for and to have been collisionally ionized , the gas must have had time to cool . but groups with low velocity dispersions , like the sculptor group , tend to have very little intracluster medium ( fadda 1996 ) . therefore , the galaxy would need to be running into a large , external gas cloud for the shock model to remain plausible . there exists a wide range of ` exotic ' models for preferentially heating the electron population including turbulence - driven mhd wave heating ( raymond 1990 ) , and mixing layers driven by bulk flows ( slavin , shull & begelman 1993 ) . bulk flows and turbulence in the disk are thought to be maintained primarily by energy injection from star formation . the star formation rates at the hi edge are likely to be very small . an alternative possibility is that the outer disk is being rained on by galactic fountain material driven by the inner starburst ( e.g. benjamin & shapiro 1993 ) . it is unclear at present whether fountain models can generate the necessary heating through shocks . ferland & mushotsky ( 1984 ) have shown that under rather special conditions , low energy cosmic ray electrons can selectively heat ions and electrons through coulomb repulsion . however , sciama ( 1996 ) emphasizes the difficulties involved in cosmic rays penetrating gas clouds and these do not appear to be an important agent . the inner disk of ngc 253 hosts a young , stellar population producing copious amounts of ionizing photons . it is plausible that the outer disk sees this radiation either through dust scattering or through warps in the outer parts . for rayleigh scattering , we have repeated the calculations of jura ( 1980 ) using up - to - date dielectric phase functions ( martin & rouleau 1990 ; draine & lee 1984 ) and assuming a standard grain mixture and distribution ( mathis , rumpl & nordsieck 1977 ) . we adopt the henyey - greenstein ( 1941 ) phase function which relates the asymmetry parameter @xmath81 to photon energy . the expected ionizing flux at the hi edge is much less than 10@xmath131 due primarily to the highly forward - scattering behaviour ( @xmath132 ) of the grains at increasing energy towards the lyman limit . while there is some uncertainty as to the fraction of ionizing photons which escape a normal spiral galaxy , two independent lines of argument suggest that roughly 5% of ionizing photons escape the galaxy ( domgrgen & mathis 1994 ; bm ) . the absolute @xmath50 magnitude of ngc 253 is comparable to that inferred for the galaxy and thus , if the disk opacity is comparable in both cases , the same models in bm apply here . bothun ( personal communication ) suspects that the disk of ngc 253 may be more opaque than the milky way , in which case our predicted fluxes should be considered upper limits . to explain the flux at both outer positions ( 9 and 12 kpc ) requires an integral sign warp originating at 7 kpc reaching 20 at 10 kpc . it is difficult to rule this out on the basis of the hi data . at face value , fig . 8 of pcvg indicates such a warp starts at 7 kpc reaching @xmath1210 at 9 kpc . in the next section , we explore this model in more detail . after sokolowski ( 1994 ) , we attempt to simulate conditions at the hi edges of spiral galaxies ionized by the stellar radiation field from the central disk . this requires that the outer sw edge of ngc 253 is warped ( @xmath133 ) or flares up and therefore sees the central regions . we use the cloudy code ( ferland 1991 ) to ionize plane parallel slabs with cosmic abundances ( grevesse & anders 1989 ) . the physical state and emission spectrum of a low - density photoionized gas with a given composition are fixed by two parameters @xmath7 the shape of the ionizing continuum and the ionization parameter ( tarter , tucker & salpeter 1969 ) . for the ionization parameter , @xmath134 ( the ratio of ionizing photons to nucleons ) , we adopt @xmath135 where @xmath51 is the speed of light . in fig . 10 , the results for a range of emission - line diagnostics are presented as a function @xmath134 . the radiation - bounded models are truncated when the electron temperature @xmath136 falls below 4000k ; all models assume @xmath53 1 @xmath33 . the unabsorbed radiation field is a composite taking the form @xmath137 where @xmath138 is the current star formation rate , @xmath139 is the initial mass function , and @xmath140 is the main - sequence lifetime . the ionizing field is dominated by stars with @xmath141 ; we specify the upper mass cut - off to be @xmath142 . the stellar atmospheres are from mihalas ( 1972 ) and kurucz ( 1979 ) ; the solar metallicity evolutionary tracks are from maeder ( 1990 ) . for the massive star imf , we adopt @xmath143 which falls between the salpeter ( 1955 ) and the miller - scalo ( 1979 ) models . irrespective of the _ shape _ of the ionizing continuum , a dilute radiation field leads to enhanced emission from low ionization emission lines ( ferland & netzer 1983 ; halpern & steiner 1983 ) . but a general hardening of the ionizing field also produces the same trend ( e.g. sokolowski & bland - hawthorn 1991 ) . indeed , if the edges of spirals are ionized by the central stellar disk , the ionizing field is expected to be both dilute and hardened . there are at least two phenomena which serve to harden the local ionizing radiation field : interstellar opacity and metal depletion ( i.e. refractory elements ) onto grains . we assume that the ionizing photons have leaked from the young inner disk and that the ionizing spectrum has been hardened by intervening absorption ( @xmath144 perpendicular to the disk ) . we adopt the atomic photoionization cross - section for which the average interstellar opacity varies as @xmath145 above 13.6ev ( cruddace 1974 ) . in quiescent galaxies , the outer hi disk is expected to have sub - solar metallicities ( molla 1996 ) . but the ongoing nuclear starburst in ngc 253 ( beck & beckwith 1984 ; antonucci & ulvestad 1988 ) could conceivably enrich the outer disk through long - range galactic fountains ( corbelli & salpeter 1988 ) . on balance , we adopt solar abundances ( grevesse & anders 1989 ) modified by the known depletion rates for cold gas ( cowie & songaila 1986 ; jenkins 1987 ; savage & massa 1987 ) . the formation of grains depletes primarily ca , fe and si which in turn suppresses the dominant coolants from the singly - ionized stages of these atoms ( ferland 1992 ) . for example , at @xmath146 , 158@xmath8 m and 35@xmath8 m account for one quarter of the total nebular cooling . while c , n and o are not strongly depleted , the forbidden line emission from , , and is greatly enhanced by the increased temperature . in particular , mathis ( 1986 ) finds that the / ratio increases as @xmath147 . therefore , a 20% increase in the local electron temperature can effectively double this line ratio . we neglect grain heating and cooling processes ( reynolds & cox 1992 ) as these are only important in the high @xmath134 limit ( baldwin 1991 ; ferland 1992 ) . complete removal of the grain population from the depleted gas causes line strengths to change by no more than 5% ( shields 1992 ) . _ photoionization models . _ in fig . 9 , we show the detailed ionization and thermal structure within the gaseous slab . , and are closely coupled with slight differences due to charge exchange reactions . , and are therefore also closely linked , except that divergence can occur at the front of the slab if the radiation field is sufficiently strong to produce a higher state of ionization in n or o. the soft photons are soaked up and fully ionize the front of the slab ; the harder photons propagate further into the gas and set up a partially ionized zone . here , the electron temperature increases by 25% which boosts the emissivities of the , , and lines . the high energy photons are essential : cutting off the continuum just below the ionization potential of nii ( 29.6 ev ) produces / line ratios no higher than 0.25 . 10 shows the dependence of five important line diagnostics on both ionization parameter and the gas column density . the range of column densities shown are relevant to spiral edges . the line ratio / peaks at @xmath148 with a fivefold enhancement when compared with hii regions . in the warped disk model above , we expect @xmath149 . only the high end of this range is sufficient to explain the recombination emission . for a gas disk with a thickness of 1 kpc , @xmath53 0.1 implies @xmath150 @xmath33 , such that for a fully ionized gas , @xmath151 . the high end of this range is indeed where the / ratio peaks in fig . the observed ratio ( / @xmath12 1 ) implies / ( @xmath152 ) is a factor of two higher than the peak of the curve in fig . while this model may have some application to the hi edges of ngc 253 , a full explanation of the gas excitation requires an additional heat source . our attempt to explain the enhanced / ratios with the dilute , hardened radiation field is made much more difficult if the particle column density seen from the nucleus is much less than @xmath153 atoms @xmath4 . it is noteworthy that @xmath154 is produced deep within the slab compared to and . the relative strength of these lines exhibits a complex interdependence on the shape of the ionizing spectrum , the ionization parameter and , most crucially , the column density of the ionized gas . fig . 10 illustrates that , in fact , all emission lines exhibit some dependence on the gas column density . rand ( 1997 ) emphasizes the importance of the @xmath155/ and the / ratios , taken together , for constraining the he ionizing fraction and the ionizing spectrum . domgrgen & mathis ( 1994 ) predict a high he ionizing fraction if the enhanced emission is due to a dilute , hardened radiation field . a direct measurement of the kinetic temperature is possible from a detection of the @xmath156 line ( osterbrock 1989 ) . but these @xmath157 band lines are expected to be an order of magnitude fainter than which puts them at the limit of detectability ( btvs ) . finally , we summarize the steps that were taken to achieve the enhanced low ionization line ratios in fig . 10 . if / @xmath53 0.3 is typical for a high @xmath134 model ( e.g. hii region ) , a dilute radiation field ( @xmath158 ) in the neighbourhood of hot stars can double this ratio ( e.g. mathis 1986 ) . hardening the dilute radiation field with interstellar absorption can produce line ratios closer to unity ( e.g. sokolowski 1991 ) . an additional 50% increase is achieved with the known gas phase depletions of refractory elements . further enhancement of this ratio ( for a fixed abundance ) requires that we selectively heat the electrons without producing a higher ionization state of nitrogen . we have succeeded in detecting ionized gas at and beyond the hi cut - off radius in the nearby spiral galaxy , ngc 253 . this galaxy is a member of a small ensemble , the sculptor group , with a low internal dispersion and little or no associated hot medium . the detected emission measures of the and lines are sufficiently strong that it is unlikely the source of the ionization is the metagalactic uv background . the strength of the line with respect to the line argues for an enhanced electron temperature at large galactic radius compared with the inner hii regions . the dominant ionization mechanism is suspected to be due to hot young stars in the inner regions which see the warped outer hi disk . we present a composite ionization model which may have some application to the hi edges of ngc 253 , but a full explanation requires additional heat sources . the kinematic measurements confirm that the rotation curve is still rising at and beyond the hi edge . in some respects , since the hi disk ceases to be detectable at only 1.2 , this is not an ideal object for testing the original proposal of bochkarev & sunyaev ( 1977 ) . but there are a few objects which subtend a large solid angle on the sky and have hi disks extending to beyond 2 : these are the focus of subsequent papers . jbh wishes to thank oxford university for a visiting fellowship and for their hospitality during the preparation of this manuscript . we acknowledge extended dialogues with dennis sciama , jon weisheit and james binney . we are indebted to walter dehnen for his insights , and for carrying out a least - squares fit to our data . both an anonymous referee and the scientific editor , g.d . bothun , made substantive comments which improved the presentation of this manuscript . keith taylor and sylvain veilleux assisted in some of the observations and the etalon was loaned to us by brent tully . we acknowledge comments from george field communicated by dennis sciama . jbh wishes to thank jim sokolowski for permission to reproduce unpublished work arising from extensive collaborations . adams , f.c . & laughlin , g. 1996 , apj , 468 , 586 antonucci , r.r.j . & ulvestad , j.s . 1988 , apj , 330 , l97 arp , h. 1985 , aj , 90 , 1012 bahcall , j.n . 1991 , apj , 377 , l5 baldwin , j. 1991 , apj , 374 , 580 beck , s.c . & beckwith , s.v . 1984 , mnras , 207 , 671 benjamin , r.a . & shapiro , p.r . 1993 , in the evolution of galaxies and their environment , nasa ames publ . , p. 338 bica , e. & alloin , d. 1987 , a&a , 70 , 281 bland - hawthorn , j. 1997 , pasa , in press bland - hawthorn , j. , ekers , r.d . , van breugel , w. , koekemoer , a. & taylor , k. 1995 , apj , 447 , l77 bland - hawthorn , j. & maloney , p.r . 1997 , apj , in press bland - hawthorn , j. , taylor , k. , veilleux , s. & shopbell , p.l . 1994 , apj , 437 , l95 bochkarev , n.g . & sunyaev , r.a . 1977 , soviet astr . 21 , 542 ( originally 1975 , azh , 54 , 957 ) boggs , p.t . , byrd , r.h . & schnabel , r.b . 1987 , siam j. sci . & stat . computing , 8(6 ) , 1052 bosma , a. 1978 , university of groningen thesis . bowyer , s. 1991 , araa , 29 , 59 carignan , c. 1985 , apj , 299 , 59 carignan , c. 1991 , proc . workshop on dark matter , space telescope sci . carignan , c. & freeman , k.c . 1985 , 294 , 494 chabrier , g. , segretain , l. & mera , d. 1996 , apj , 468 , l21 charlot , s. & silk , j. 1995 , apj , 445 , 124 corbelli , e. & salpeter , e.e . 1988 , apj , 326 , 551 corbelli , e. , schneider , s.e . & salpeter , e.e . 1989 , aj , 97 , 390 cowie , l.l . & songaila , a. 1986 , araa , 24 , 499 cruddace , r. 1974 , apj , 187 , 497 dehnen , w. & binney , j. 1997 , mnras , in press diaz , a.i . 1989 , in evolutionary phenomena in galaxies , eds . beckman & b.e.j . pagel ( cambridge : cambridge university press ) , 377 domggen , h. & mathis , j.s . 1994 , apj , 428 , 647 dopita , m.a . & sutherland , r. 1996 , apjs , 102 , 161 dove , j.b . & shull , j.m . 1994 , apj , 423 , 196 draine , b.t . & lee , h.m . 1984 , apj , 285 , 89 evans , n. & dopita , m.a . 1985 , apjs , 58 , 125 fadda 1996 , apj , 473 , 670 ferland , g. & mushotsky , r. 1984 , apj , 286 , 42 ferland , g. & netzer , h. 1983 , apj , 264 , 105 ferland , g. 1991 , cloudy code , osu internal report , 91 - 01 ferland , g. 1992 , apj , 389 , l63 freeman , k.c . 1993 , in physics of nearby galaxies : nature or nurture , ed trinh thuan ( paris : editions frontieres ) , p 201 . freeman , k.c . 1996 , in unsolved problems of the milky way , ed l. blitz and p. teuben ( dordrecht : kluwer ) , p 645 . grevesse , n. & anders , e. 1989 , in cosmic abundances of matter , ( new york : aip ) , p. 1 haardt , f. & madau , p. 1996 , apj , 461 , 20 halpern , j.p . & steiner , j.e . 1983 , apj , 269 , l37 haynes , m.p . & roberts , m.s . 1979 , apj , 227 , 767 henyey , l.g . & greenstein , j.l . 1941 , apj , 93 , 70 henry , r.c . 1991 , araa , 29 , 89 hogan , c.j . & weymann , r.j . 1987 , mnras , 225 , 1p jenkins , e.b . 1987 , in interstellar processes , eds . hollenbach & h.a . thronson ( dordrecht : reidel ) , 533 jura , m. 1980 , apj , 241 , 965 kawaler , s. 1996 , apj , 467 , l61 kamphuis , j. 1993 , phd , groningen kondratyev , k. ya . 1969 , in radiation in the atmosphere , academic press , new york & london koribalski , b. , whiteoak , j.b . & houghton , s. 1995 , pasa , 12 , 20 kurucz , r.l . 1979 , apjs , 40 , 1 lee , y .- w . 1993 , in the globular cluster - galaxy connection , eds . smith & j.p . brodie , asp . 48 , 142 liebert , j. , dahn , c.c . & monet , d.g . 1988 , apj , 332 , 891 lyon , j.g . 1975 , apj , 201 , 168 bania , t.m . & lyon , j.g . 1980 , apj , 239 , 173 maeder , a. 1990 , a&as , 84 , 139 malin , d.f . 1981 , j. phot . , 29 , 199 malin , d.f . 1983 , in astronomy with schmidt - type telescopes , ed . m capaccioli , iau coll . 78 , 57 ( reidel : dordrecht ) maloney , p. 1993 , apj , 414 , 41 martin , p.g . & rouleau , f. 1990 , preprint mathewson , d.s . & ford , v.l . 1984 , in structure & evolution of the magellanic clouds ( reidel : dordrecht ) , iau symp . 108 , 125 mathewson , d.s . , cleary , m.n . & murray , j.d . 1975 , apj , 195 , l97 mathis , j. 1986 , apj , 301 , 423 mathis , j. , rumpl , w. & nordsieck , k.h . 1977 , apj , 217 , 425 mihalas , d. 1972 , non - lte model atmospheres for b & o stars ( ncar - tn / str-76 ) miller , g.e . & scalo , j.m . 1979 , apjs , 41 , 513 molla , m. , ferrini , f. & diaz , a.i . 1996 , apj , 466 , 668 nordgren , t.e . , cordes , j.m . & terzian , y. 1992 , aj , 104 , 1465 olling , r.p . 1995 , 110 , 591 osterbrock , d.e . & martel , a. 1992 , pasp , 104 , 76 osterbrock , d.e . 1989 , astrophysics of gaseous nebulae & active galactic nuclei , univ . books pagel , b.e.j . 1989 , in evolutionary phenomena in galaxies , eds . beckman & b.e.j . pagel ( cambridge : cambridge university press ) , 201 puche , d. & carignan , c. 1991 , apj , 378 , 487 puche , d. , carignan , c. & van gorkom , j. 1991 , aj , 101 , 456 rand , r. 1997 , apj , 474 , 129 raymond , j. 1990 , apj , 365 , 387 reynolds , r.j . & cox , d.p . 1992 , apj , 400 , l33 rots , a.h . 1975 , a&a , 45 , 43 ryder , s. 1996 , pasa , 14 , 81 ryu , d. , olive , k.a . & silk , j. 1990 , apj , 353 , 81 salucci , p. & frenck , c.s . 1989 , mnras , 237 , 247 salpeter , e.e . 1955 , apj , 121 , 161 sargent , w.l.w . & steidel , c.c . 1990 , in baryonic dark matter , eds . d. lynden - bell & g. gilmore ( dordrecht : kluwer ) , 223 savage , b.d . & massa , d. 1987 , apj , 314 , 380 schommer , r. 1993 , aj , 105 , 97 sciama , d.w . 1995 , mnras , 276 , l1 sciama , d.w . 1996 , modern cosmology & the dark matter problem ( cambridge : cambridge univ . press ) shields , j.c . 1992 , apj , 339 , l27 slavin , j.d . , shull , j.m . & begelman , m.c . 1993 , apj , 407 , 83 smith , g.p . 1963 , bull . neth . , 17 , 203 sokolowski , j. 1994 , preprint ( unpubl . ) sokolowski , j. & bland - hawthorn , j. 1991 , pasp , 103 , 911 spyromilio , j. 1995 , mnras , 277 , l59 staveley - smith , l. 1990 , apj , 364 , 23 sutherland , r.s . & dopita , m.a . 1993 , apjs , 88 , 253 tarter , c.b . , tucker , w.h . & salpeter , e.e . 1969 , apj , 156 , 943 van albada , t.s . , bahcall , j.n . , begeman , k. & sancisi , r. 1985 , apj , 295 , 305 van gorkom , j. 1993 , in environment & evolution of galaxies , eds . shull & h.a . thronson ( kluwer : dordrecht ) , p. 345 veilleux , s. 1988 , phd thesis , u. c. santa cruz vogel , s.n . , weymann , r. , rauch , m. & hamilton , t. 1995 , apj , 441 , 162 wakker , b.p . 1991 , aa&as , 90 , 495 wood , p. 1990 , jrasc , 84 , 150 yelle , r.v . & roesler , f. 1985 , jgr , 90 , 7568 zaritsky , d. , smith , r. , frenk , c. , & white , s.d.m . 1997 , apj , 478 , 39 * fig . 1 . * [ b_four ] taurus-2 5 fields ( sw1 , sw2 , sw3 , sw4 ) superimposed on a @xmath50 image of ngc 253 . the blue line occurs at smaller galaxian radii compared with the h@xmath0 line in the same field . there are four distinct fields with two emission lines occurring in each . at the central position , the etalon was used at two different spacings which meant that we obtained line detections at four discrete positions . * 2 . * [ hi_four ] same as fig . 1 except that the taurus-2 fields have been superimposed onto a deep atnf hi map ( koribalski , whiteoak & houghton 1995 ) . the outermost hi contour corresponds to a column density of @xmath159 @xmath4 . * [ zoom ] a magnified image of part of the sw3 field in figs . 1 and 2 . the field of view is rotated so that the vertical axis lies parallel to the galaxy major axis . clumpy and h@xmath0 emission is clearly visible . what is not easily rendered is the diffuse emission between and at large off - axis angles to the major axis . notice the extremely faint galaxy continuum ( @xmath160 mag arcsec@xmath37 ) over much of the field . the bright arcs are atmospheric oh lines . * 4 . * [ spec1 ] the emission - line spectrum at the hi edge ( field sw3 in figs . 1 and 2 ) compared with the off - field spectrum . the difference of these spectra is shown below . remarkably , the line has a surface brightness comparable to the h@xmath0 surface brightness , as compared with solar - abundance hii regions where the ratio is an order of magnitude smaller . this result holds after removing the dense knots in fig . 3 . the azimuthally averaged galaxy continuum underlies the spectrum and corresponds to roughly @xmath161 23 mag arcsec@xmath37 below falling to 25 mag arcsec@xmath37 below h@xmath0 . the equivalent widths of the lines are 0.60aa and 1.8 respectively . * [ spec2 ] the emission - line spectrum for the field beyond the hi edge ( field sw4 in figs . 1 and 2 ) . as we did not get a matching sky exposure , the spectra are shown from binning over two different opening angles . the top spectrum results from binning over 25 about the major axis from the optical axis ; the lower spectrum arises from binning over the entire field . the difference of these spectra is shown below . the emission is clearly seen in the top spectrum and , indeed , in a high contrast image . this line is completely washed out in the lower spectrum . the h@xmath0 emission is just visible in the high contrast image , and does not begin to appear until we sum over the entire field . the difference has been taken after weighting the top spectrum for its lower signal to noise ratio . again , we see a very faint galaxy continuum spectrum which is roughly @xmath161 26 mag arcsec@xmath37 after azimuthal binning below the line dropping to @xmath161 27 mag arcsec@xmath37 below the h@xmath0 line . the equivalent widths of the lines are 7.8 and 9.9 respectively . * fig . 6 . * [ v_observed ] kinematic measurements along the major axis of ngc 253 deduced from vla hi , taurus-2 and lines . in @xmath61 , the original rotation curve of pcvg has been projected onto the sky with their inferred orbit inclination at each radius . three sets of measurements are shown : these represent the approaching side ( open circles ) , receding side ( squares ) , and from fitting to the full velocity field ( filled circles ) . in @xmath60 , we show pcvg s original curves where the data have been deprojected with the measured inclination at each radius . in @xmath51 , the solid line is the same curve as the squares in @xmath61 . the filled squares are the taurus-2 measurements ; the circles are the measurements . * fig . 7 . * [ v_model ] representative fits to the pcvg and taurus-2 data using a 3-component mass model ( disk , halo , gas ) . the contribution from the hi surface density ( dotted line ) is the same in all models . in @xmath61 , we adopt an exponential disk ( short - dashed line ) and choose an isothermal sphere for the spherical halo ( long - dashed line ) using the numerical approach developed by carignan ( 1985 ) . in @xmath60 , we have used the disk - halo models of dehnen & binney ( 1997 ) : the least - squares fit was carried out by w. dehnen . in @xmath51 , we have truncated the halo at a radius of 10 kpc to demonstrate a possible explanation for the last measured point . * fig . 8 . * [ white_dwarf ] the predicted emission measure versus @xmath50 surface brightness for an isothermal population of white dwarfs as a function of surface density . the solid lines are isotherms for white dwarf temperatures in the range @xmath162 k. the dashed lines are isochoric lines indicating variations in white dwarf surface density in steps of 0.5 in dex starting at 1 pc@xmath37 at the top . the arrow indicates the average emission measure and disk surface brightness upper limit at the edge of the hi disk . the filled circles are the expected contribution to the @xmath50 band flux and emission measure for the white dwarf population in the solar neighbourhood ( assuming that all photons are absorbed ) . the galactic disk is assumed to be 600 pc thick and the white dwarf radius is assumed to be 0.14 r@xmath163 . * fig . 9 . * [ sok1 ] the lower panel shows the relative ionization fraction of important neutrals and ions , where the front of the gas slab ( @xmath164 ) is ionized by a dilute radiation field ( @xmath165 ) . the middle panel shows the temperature and mean electron density structure within the slab . the top panel shows the normalized line emissivities and illustrates the relative dependence on electron temperature , electron density , and ionization fraction . * fig . 10 . * [ sok2 ] the dependence of five important line diagnostics on the strength of the radiation field and hydrogen column density . the solid lines are radiation bounded models where all ionizing photons are soaked up by the gas . also shown are matter bounded models where the slab has been truncated at 1 , 2 , 5 , 10 and 20 @xmath39 10@xmath166 @xmath4 . the observed line ratio / is a factor of 2.98 smaller than /(veilleux 1988 ) . 9 gives some appreciation for the distinct trends shown by different lines ( e.g. vs. ) .

through matches with the sloan digital sky survey ( sdss ) catalogue we identify the location of various families of astronomical objects in _ wise _ colour space . we identify reliable indicators that separate galactic / local from extragalactic sources and concentrate here on the objects in our galaxy and its closest satellites . we develop colour and magnitude criteria that are based only on _ wise _ data to select asymptotic giant branch ( agb ) stars with circumstellar dust shells , and separate them into o - rich and c - rich classes . with these criteria we produce an all - sky map for the count ratio of the two populations . the map reveals differences between the galactic disc , the magellanic clouds and the sgr dwarf spheroidal galaxy , as well as a radial gradient in the large magellanic cloud ( lmc ) disc . we find that the c : o number ratio for dusty agb stars increases with distance from the lmc centre about twice as fast as measured for near - ir selected samples of early agb stars . detailed radiative transfer models show that _ wise _ colours are well explained by the emission of centrally heated dusty shells where the dust has standard properties of interstellar medium ( ism ) grains . the segregation of different classes of objects in _ wise _ colour space arises from differences in properties of the dust shells : those around young stellar objects have uniform density distributions while in evolved stars they have steep radial profiles . [ firstpage ] radiative transfer stars : agb and post - agb galaxy : structure magellanic clouds infrared : general infrared : stars .

1405.7966

infrared emission from radiatively heated dust carries important information about the distribution of material around the heating source . this information can be deciphered either through detailed imaging and analysis of the spectral energy distribution ( sed ) of individual objects , or by studying the sed properties ( such as colours ) of statistically large populations of sources . the pioneering _ infrared astronomical satellite _ ( _ iras _ ; launched in 1983 ) all - sky survey ushered in the era of modern highly successful astronomical surveys @xcite . the _ iras _ survey provided unprecedented opportunity to classify the infrared properties of about 350000 astronomical objects using a homogeneous data set obtained with a single facility ( e.g. @xcite ) . thanks to sensitivity improvements of up to three orders of magnitude compared to _ iras _ , the recent _ wide - field infrared survey explorer _ ( _ wise _ ; launched in 2010 ) all - sky survey detected about 560 million objects @xcite . _ wise _ represents the next major step in surveying and understanding of the infrared sky . for example , data from the _ iras _ point source catalogue ( psc ) showed that certain galactic objects tend to cluster in well - defined regions of _ iras_colour colour ( cc ) diagrams , and the same clustering is expected in the _ wise _ photometric system but for a significantly larger sample that probes a much larger volume of the galaxy . furthermore , and perhaps more importantly , _ wise _ data are deep enough for its catalogues to contain a very large number of extragalactic sources , and they too are clustered in well - defined regions of _ wise _ cc and colour - magnitude ( cm ) diagrams . although _ iras _ , too , detected some extragalactic objects ( for a review see @xcite ) , _ wise_can be considered a true ir counterpart to modern optical surveys of extragalactic sources , such as the sloan digital sky survey ( sdss ; @xcite ) . 0.1 in [ fig : bands ] _ wise _ mapped the sky at 3.4 , 4.6 , 12 , and 22 with an angular resolution of 6.1 , 6.4 , 6.5 and 12.0 arcsec , respectively . [ fig : bands ] shows the _ wise _ bandpasses and their comparison to other major infrared surveys . throughout this paper we will refer to the filter names as ` w1 filter ' ( and w2 , w3 , and w4 filters ) . the source sed @xmath0 is averaged in each band by the instrument s transmission function @xmath1 , where @xmath2 . this yields the in - band fluxes @xmath3 . _ wise_catalogue provides magnitudes calibrated in the vega system using an empirical fit to the vega spectrum ( see equation 2 in * ? ? ? thus the calibrated in - band fluxes are @xmath4 ` colour ' is the magnitude difference between two measured fluxes of an astronomical source , e.g. , @xmath5 = 2.5\log(f_2/f_1)$ ] for bands 1 and 2 . we will also use the short - hand notation for such colour . _ wise _ attained 5@xmath6 point - source sensitivities better than 0.068 , 0.098 , 0.86 and 5.4 mjy in the through bands , respectively , which correspond to vega - based magnitudes 16.83 , 15.6 , 11.32 and 8.0 ] . these survey sensitivity limits apply in unconfused regions on the ecliptic ; they improve towards the ecliptic poles due to denser coverage and a lower zodiacal background . photometric measurements of sources brighter than approximately 8.1 , 6.7 , 3.8 and -0.4 mag ( vega ) are affected by saturation of the detectors . the astrometric precision for sources brighter than snr=40 is about 0.2 arcsec . the _ wise _ all - sky release comprises all data taken during the full cryogenic mission phase ( 2010 january 7 to 2010 august 6 ) . it includes an atlas of 18240 images , a source catalogue with positional and photometric data for over 563 million objects , and also an explanatory supplement . unlike _ iras _ catalogues which reported fluxes in jansky , _ wise _ measurements are reported on the vega magnitude scale . although this difference makes a direct comparison with some _ iras_-based studies more difficult , we follow already published work based on _ wise _ data and use unaltered catalogued vega - based magnitudes . a more detailed discussion can be found in @xcite and at the _ wise_data release website . @xcite have demonstrated that _ wise _ detected all the main families of extragalactic sources , including star - forming ( sf ) galaxies , galaxies with active nuclei ( agn ) and quasars ( qso ) . their positions in a representative _ wise _ cc diagram are shown schematically in the left panel of fig . [ fig : compare - to - wise ] , and the challenge to theory is to come up with an explanation as to why different types of objects segregate the way they do . similar structuring has been observed in _ iras_-based cc diagrams ; its details were explained by @xcite . the colours of a ` naked ' radiation source correspond to a point in colour space ; since _ wise _ colours are normalized to vega , colours of blackbody emission are close to 0 . embedding the source in a dust shell , the radiation is reprocessed to longer wavelengths and the colours are shifted to the red . the larger is the dust optical thickness , the redder are the colours . therefore , increasing the shell optical depth while keeping all its other properties constant produces a track in cc diagrams with distance from the origin along the track increasing with optical depth . the right panel of fig . [ fig : compare - to - wise ] shows for illustration three tracks produced by spherical ism - dust shells with power - law radial density distributions of varying steepness . the tracks for uniform density and for @xmath7 density falloff nicely outline the boundaries of the region occupied by stars and extragalactic sources in the _ wise _ cc diagram . similarly , the _ wise_colours produced by clumpy torus models @xcite , shown as grey - scale number density distribution , match the region of qso and seyferts . our aim is to present in this paper a more detailed analysis of _ wise _ colours of sources in the galaxy and its satellites , and in a companion paper for agn and quasars ( ) . we begin this first paper by identifying the _ wise _ colours of families of astronomical sources through matching with the sdsscatalogue . the matching criteria and their results are described in section [ sec : sdss ] . in particular , almost 40 million _ wise_sources are matched with galactic stars , and in section [ sec : bright - objects ] we concentrate on the nature of these stellar sources . we proceed to calculate model colour tracks for the _ wise _ photometric system which can be used to interpret the observed clustering of different stellar populations in the cc diagrams . the models and their comparison to _ wise _ data are described in detail in section [ sec : models ] . our results are summarized and discussed in section [ sec : summary ] . to enable a quantitative identification of source populations and subsequent comparison with our models , we use a sub - sample of _ wise_sources with sdss - based identifications for five types of object : quasars , galaxies classified as agn , sf and luminous red galaxies ( lrg ; essentially elliptical galaxies ) , and stars ( optically unresolved objects not in the quasar sample ) . we first employ positional matching within a 1.0 arcsec radius of the _ wise _ all - sky data release catalogue , using procedures described in @xcite and @xcite : we match with sdss data release 7 catalogue for samples with spectroscopy , and data release 8 for unresolved sources in imaging data . the 1 arcsec cutoff matching radius is well justified . [ fig : matching ] shows the distribution of matching radii between _ wise _ and sdss for our sample of stars : the median and standard deviation are @xmath8 arcsec . for a sample of lrgs and star forming galaxies ( not shown ) it is @xmath9 arcsec . after positional matching , we enforce quality cuts for _ wise _ photometry , as described further below . [ fig : color - mag - scatter ] shows three representative _ wise _ cm diagrams for the five sub - samples selected as follows . * _ quasars : _ starting with 98047 sources from the sdss catalogue of quasars @xcite that have a _ wise _ source closer than 1 arcsec , we further require that all four _ wise _ magnitudes are brighter than 5@xmath6 limits and have a reliably measured uncertainty , yielding a sample of 14795 quasars . * _ star forming and agn galaxies : _ galaxies with sdss spectra are selected from the nyu vagc catalogue @xcite . galaxies with spectral emission lines are further classified as sf and agn galaxies using the baldwin - phillips - terlevich diagram @xcite . there are 12379 agn and 180329 sf galaxies within 1 arcsec of a _ wise _ source . the cuts for their _ wise _ photometry are the same as for quasars , yielding samples of 4509 agn and 38092 sf galaxies . * _ lrgs : _ luminous red galaxies are adopted from a catalogue constructed for cosmological studies by @xcite . there are 66093 galaxies from that catalogue within 1 arcsec of a _ wise _ source . requiring at least a 5@xmath6 detection in each of the four _ wise _ bands leaves only 70 sources because of faint and magnitudes . as a result , for lrgs we enforce the 5@xmath6 cuts only in the and bands . this yields a sample of 47989 galaxies , whose c[2,3 ] and c[3,4 ] colours are , of course , _ not reliable_. to emphasize this point , the 10@xmath6 limits for the and bands are shown as solid lines in the two corresponding panels in fig . [ fig : color - mag - scatter ] . * _ stars : _ sdss data release 8 contains 119 million unresolved objects within 1 arcsec of a _ wise _ source . to ensure high reliability of sdss star / galaxy separation we limit the sample to @xmath10 . the majority of these stars are on the main sequence , and sample both disc and halo populations ( see * ? ? ? * for detailed discussion ) . similarly to the case of lrgs , we apply cuts sequentially . we first require @xmath11 15.8 and @xmath11 14.8 ( 10@xmath6 limits instead of 5@xmath6 because the sample is already fairly large ) , and @xmath12 < 0.8 $ ] ( to avoid contamination by faint unrecognized quasars ; see below ) , yielding 39 million sources ( bottom left panel in fig . [ fig : color - mag - scatter ] ) . applying the 10@xmath6 cuts also to and bands reduces drastically the sample size , yielding three million sources after the @xmath13 < 10.5 $ ] cut ( bottom centre panel ) , and @xmath1480000 sources when cuts in all four bands ( @xmath15 < 7.2 $ ] ) are enforced ( bottom right panel ) . the final sample sizes for all object classes and their mean _ wise_colours are listed in table [ tab : wisecolors ] . the main conclusion derived from this analysis of sdss-_wise _ matches is that subsamples of _ wise sources with @xmath11 11 must be dominated by galactic objects _ , with the increasing fraction of extragalactic objects at fainter magnitudes . .median and interquartile - based standard deviation , , of _ wise _ colours for the five main source classes . for selection details , see section [ sec : sdss ] . [ cols="<,>,^,^,^",options="header " , ] why do objects segregate in colour space and what causes the particular segregation patterns of different families of _ wise_sources ? these questions were already addressed by @xcite in the context of _ iras _ sources . the dust radiative transfer problem possesses general scaling properties , as discussed in detail by @xcite . there are only two input quantities whose magnitudes matter : the dust optical depth at some fiducial wavelength and its temperature at some position in the source . all other input is defined by dimensionless , normalized profiles that describe ( 1 ) the spectral shapes of the dust absorption and scattering coefficients , ( 2 ) the spectral shape of the external heating radiation , and ( 3 ) the dust spatial distribution . physical dimensions , e.g. luminosity and linear sizes , are irrelevant . scaling applies to arbitrary geometries and implies that _ different dusty objects are expected to segregate in cc diagrams primarily because their dust shells are different , not because they have different central sources_. this behaviour is displayed by the objects in the _ iras _ psc and is similarly expected for _ wise _ colours . we now describe our basic modelling framework for dusty stellar environments and then discuss comparisons of models with data in more detail . all modelling was performed with our public code dusty@xcite . we employ two families of dust grains . our ` standard ism ' dust contains a standard galactic mix of 53 per cent silicates , with optical constants from the @xcite ` cold dust ' for silicate , and 47 per cent graphite , with optical properties from @xcite . the other family represents pure - composition dust typically found in agb stars . dust around c - rich stars comprises amorphous carbon , whose properties we take from @xcite , while for o - rich stars we take the warm silicates from @xcite . we assumed spherical grains and considered both single - size dust grains and models with mrn grain - size distribution @xcite . dusty handles dust distributions through the composite grain approximation by averaging over the chemical composition and size distribution . this approach was shown to reproduce reasonably accurately the results from exact treatment of grain mixtures ( see e.g. * ? ? ? the extinction profiles used in the calculations are shown in fig . [ fig : xsections ] . . the shell optical depth is = 30 at visual , its relative thickness is @xmath16 and its temperature on the inner boundary is either @xmath17 k ( solid lines ) or @xmath18 k ( dashed ) . the density profiles are power - law @xmath19 with @xmath20 = 0 and 2 , as marked . the shapes of the _ wise _ filter response functions are outlined in light grey . ] with dusty we produced models of spherical dust shells around central sources ; the code calculates exact solutions for the dust temperature profiles , emerging flux and other properties of interest . as noted above , the only input parameters whose magnitudes matter are the dust temperature on the shell inner boundary and the shell overall optical depth at a given wavelength ; all other model input describes dimensionless , normalized profiles . we computed a range of dusty shell models with overall optical depth at visual from = 0.1 to 200 and dust temperature on the inner boundary from @xmath21 = 300 k to 1200 k. the spectral shape of the central source was taken as a blackbody with temperature @xmath22 k , 5000 k and 10000 k. for the radial density distribution of the spherical dusty shell we used a power - law @xmath19 with @xmath20 = 0 , 1 and 2 . in addition , for the agb shells we used the radial density profiles determined from a solution of the full equation of motion of radiatively driven dusty winds , obtained with the rdw option of dusty ( see * ? ? ? the shell relative thickness , ratio of outer to inner radius @xmath23 , was taken as 100 . resulting seds for some representative cases are shown in fig . [ fig : spectra - amcsil ] . as noted above ( section [ sec : naked_stars ] ) , a naked star corresponds to a point near the origin of the _ wise _ colour space . surrounding the star with a dust shell will shift the colours to the red , moving the source away from the origin . given the shell properties and varying only its overall optical depth , the dust temperature profile remains unchanged as long as the shell remains optically thin . therefore the colours remain the same ; all optically thin dusty shells with the same properties correspond to a single point in the colour space . increasing the optical depth further , the shell temperature profile starts to change when it enters the optically thick regime , leading to deviation of the source colours away from the optically - thin point towards redder colours . this results in a track in the 3d _ wise _ colour space , with position along the track determined by the shell overall optical depth . projections on the 2d colour planes produce tracks in cc diagrams , like those shown in the right panel of fig . [ fig : compare - to - wise ] . this figure shows that the tracks for @xmath20 = 0 and 1 shells of ism dust trace the boundaries of the region occupied by extragalactic sources in the cc diagram . this behaviour is in agreement with the findings by @xcite and @xcite that the deep silicate absorption features observed in ulirg spectra require burial of luminous central sources in very optically thick ( @xmath24 ) dusty shells with flat ( @xmath20 = 01 ) radial density distributions . the figure also shows that stars occupy a region much closer to the colour origin so that the optical thicknesses of their dust shells are generally much smaller . we now discuss in detail the _ wise _ colours of galactic sources . tabulations of the colour tracks used in figures displayed below are provided in table [ tab : tabulations ] , where we also list the model in - band fluxes , useful for studies which depend on source luminosity . k. the shell density profile is a power - law @xmath19 with @xmath25 , as marked . each track starts at the colours of the dust - free blackbody and the optical depth increases along the track from @xmath26 to @xmath27 . the dust is taken as ism type with mrn size distribution . in each panel different tracks correspond to dust temperature @xmath21 on the shell inner boundary of 300 k ( solid lines ) , 600 k ( dashed lines ) and 1200 k ( dotted lines ) . ] @xcite found that _ iras _ colours of ysos could be explained by the colour tracks of dust shells with flat ( @xmath28 ) density distributions . [ fig : d - colors ] explores in detail the _ wise_colours expected from @xmath20 = 0 and 1 radial profiles for spherical shells of ism dust with mrn size distribution and optical depths up to @xmath29 , heated by a central 10000 k blackbody . models were produced for dust temperatures on the shell inner boundary varying from @xmath30 k to @xmath18 k. this figure provides a zoom - in on the smaller- region of the tracks for shallow density distributions that were shown in fig . [ fig : compare - to - wise ] . its various plots show the range of possibilities for colour tracks covering the likely region of parameter space for dust shells around ysos . a characteristic feature of flat density distributions is that even a small increase in optical depth can significantly enhance long wavelength emission because it places more dust at large radial distances with low temperatures . for this reason , the @xmath31 portions of the tracks are parallel to the colour axis with longer wavelengths while maintaining a value of roughly 0 for the other axis with the ` warmer ' colour . deviations from this behaviour increase with @xmath21 and are mostly noticeable in the diagram , especially for the @xmath20 = 1 tracks . ; each panel is numbered at its top right corner . overplotted are model colour tracks of spherical dusty shells with standard ism dust grains and uniform density profiles ( @xmath20 = 0 ) . the shells surround central stars with vega - like emission . the dust temperature on the shell inner boundary is either 600 k ( red tracks ) or 1200 k ( green ) . position along the track depends only on the dust optical depth ; a few values of the @xmath32-band optical depth are marked . the tracks delineate the plume of ysos . ] [ fig : plume_yso ] shows the comparison with observations . its three panels reproduce the projections on to 2d colour planes of the 3d colour distribution of bright _ wise _ sources that were shown previously in fig . [ fig : star - plumes ] , with model tracks plotted over the source distribution . the models producing these tracks are the same as the @xmath20 = 0 models shown in fig . [ fig : d - colors ] except for the spectral shape of the heating source : instead of a pure blackbody they use the empirical vega fit employed for normalization by @xcite ( see also section [ sec : naked_stars ] ) . these tracks follow rather well the long , narrow plume of stars along the and axes at @xmath33 in panels 1 and 2 of fig . [ fig : plume_yso ] and along the axis at @xmath33 in panel 0 . this plume has been previously identified as the locus of ysos among bright _ wise _ sources ( section [ sec : envelopes ] ) . in most colour - colour combinations , little differentiates the tracks for different values of @xmath21 , the dust temperature on the inner boundary . the tracks diverge significantly only in the panels involving the colour combination , where only the @xmath21 = 1200 k track passes through the blobs roughly clustered around ( 2.5,2 ) in panel 0 and ( 2.2,0.7 ) in panel 2 . it is not clear , though , that these blobs give a reliable indication of thermal dust emission because these sources could be dominated by pah emission covered by the broad filter ( see section [ sec : agb ] ) . a reliable determination of the nature of these sources would require in - depth , detailed modelling of individual objects . the small range of optical depths spanned by the main plume of sf sources implies that _ wise _ has uncovered a large number of yso with small amounts of dust around them while giving little indication of such objects covered by deep layers of dust . this is surprising because heavy dust obscuration is expected in many ysos . indeed , the sf regions identified by _ iras _ were characterized by large optical depths ( 110 ; see fig . 3 in @xcite ) . these differences reflect the starkly different sensitivity limits of the two facilities . _ iras _ required for detection strong ir emission , which precluded sources with optically thin dusty shells . however , optically thick yso shells , which were a prerequisite for _ iras_detection , produce _ wise _ colour differences that exceed 4 . such large differentials are eliminated by the cuts we place on saturation limits , which remove sources too bright at short wavelengths , and on signal - to - noise ratio , which remove sources too weak at the long wavelength bands . as is evident from fig . [ fig : compare - to - wise ] , the colour of optically thick @xmath20 = 0 objects is larger than 4 , entering the domain populated by extragalactic sources which we deliberately sought to avoid in this analysis . relaxing our selection criteria should therefore reveal a heavily obscured yso population mixed with the extragalactic sources . to test this hypothesis we identified in our basic selections those sources that are moderately bright in and are matched by high optical depths of our model colour tracks . these would most likely be ysos with high shell optical depths . such a population , compatible with the previous _ iras _ detections , indeed exists in the _ wise _ data base . it can be found approximately within @xmath34 \lesssim 11 $ ] , @xmath35 \lesssim 1.4 $ ] , and @xmath36 \lesssim 5.7 $ ] . all sources with are shown in the cc scatter diagram in fig . [ fig : yso_extended ] ( left panel ) ; all other selection criteria from fig . [ fig : plume_yso ] apply , except that the constraint on galactic latitude was dropped . the cc diagram contains 29892 sources while the blob outlined with a blue rectangle contains 6192 objects , presumably ysos with high dust optical depths . indeed , the centre of the selection cube , @xmath37 = 10.5 $ ] , @xmath12 = 0.9 $ ] , @xmath38 = 4.85 $ ] , implies that @xmath13 = 4.75 $ ] ( vega ) . with this magnitude we can use @xmath39 to compute the in - band flux density @xmath40 , where @xmath41 is the zero - magnitude flux density in jansky . it is given as 31.674 jy for band 3 in ( * ? ? ? * see their table 1 ) . thus our magnitude @xmath42 corresponds to @xmath43 jy at 12 ( ) while the faint limit of the _ iras _ faint source catalogue at 12 is 0.2 jy . therefore , our findings are compatible with _ iras _ discovering only sources in the blue box but not those with lower 12 fluxes . also plotted in the left panel of fig . [ fig : yso_extended ] are the same model colour tracks from panel 1 of fig . [ fig : plume_yso ] , but over an expanded colour range . the green track ( @xmath44 k ) shows a turn - over at @xmath38 \sim 5.5 $ ] , and all its values between 0.025 and 9.4 stay perfectly within the blue yso rectangle . the right panel of fig . [ fig : yso_extended ] shows the all - sky distribution of the @xmath45 sources from the yso plume . the distribution is extremely heavily concentrated towards the galactic plane , a preferred location to find local sf regions . this is further evidence that our conclusions are consistent . the choice of dusty shell models for ysos can be debated because some young stars , e.g. t tauri stars , are known to be surrounded by dusty disks . @xcite showed that a flared disk produces an sed identical to that of an equivalent optically thin spherical dust shell . the latter is thus a good proxy for the more realistic disk model . there is no way to distinguish between them , or any combination of the two , with flux measurements alone ; imaging is necessary for that . from equations 7 and 8 in section 2.1 of @xcite , a constant density shell ( as applied in our yso models ) is equivalent to a disk with linear flaring . the very small optical depths we find in the wise data of ysos may be also interpreted as the difference in flaring angle across the disk ( equation 7 ) , but it is difficult to decide without imaging . \le 11 $ ] in a _ wise _ cc diagram . all other cuts are as in fig . [ fig : plume_yso ] , except that there is no constraint on galactic latitude . the green and red dusty model colour tracks are as in fig . [ fig : plume_yso ] . the blue rectangle outlines the expected colour region of ysos with high dust obscuration . right : all - sky distribution of the sources in the blue rectangle ( in mollweide projection ) , with the grey - scale normalized to peak source count density . it is heavily concentrated towards the galactic plane , where local sf regions are predominantly found . ] radiatively driven winds ( rdw ) have been shown to well describe infrared emission from dusty envelopes around agb stars ( see , e.g. : * ? ? ? * ; * ? ? ? * and references therein ) . their circumstellar dust distribution approximately follows a power - law @xmath46 . dusty can calculate rdw models exactly and in fig . [ fig : rdw ] we show rdw model colour tracks for a central source at @xmath22 k and a @xmath47 shell . tracks are shown for two temperatures at the inner dust boundary , @xmath17 k and @xmath18 k. agb stars are classified as o - rich or c - rich stars ( see section [ sec : agb ] ) and models were produced with two corresponding types of dust : warm silicates from @xcite ( sil - ow , left panels ) , and amorphous carbon ( amc , right panels ) . both single - size grains and an mrn size distribution were used , as indicated in the legend . as seen from fig . [ fig : rdw ] , the effect of different grain sizes on the colour tracks is negligible compared to varying the dust temperature . = 2500 k. left panels are for silicate dust ( sil - ow ) , right panels for amorphous carbon ( amc ) . the dust temperature @xmath21 at the wind origin is taken as 600 k ( solid lines ) and 1200 k ( dashed lines ) . model tracks for various grain sizes and for the mrn size distribution are colour coded as marked . the overall optical depth of the shell increases along each track from = 0 to = 30 . ] amorphous carbon absorbs more in the and bands , while silicates absorb more in the and bands ( see fig . [ fig : xsections ] ) . therefore , given two circumstellar shells with the same overall optical depth but different chemical composition , the silicate shell is expected to be redder in the c[2,3 ] colour and bluer in the c[1,2 ] colour . as illustrated in fig . [ fig : rdw ] , the _ wise _ colour diagrams can separate sources with steep density profile and predominantly silicate dust ( e.g. typical agb stars ) from those with carbonaceous dust ( e.g. carbon stars ) , for the same dust sublimation temperature . how do the rdw model tracks compare to agb colours measured by _ wise _ ? first , we can fix the stellar temperature @xmath48 at 2500 k and the ratio of inner - to - outer radius of the shells at 100 , since both are quite typical values for agb stars . we also investigated other @xmath49 , but they did not agree as well with the data ( the allowed temperature range is only a few hundred k wide ) . the two most important free model parameters are dust interface temperature , @xmath21 , and shell optical depth , . this two - dimensional space of model parameters maps to _ wise _ colour space ( and vice versa ) . into model parameter constraints . left for c - rich agb stars , right for o - rich stars . each equation translates into curved areas in the ( @xmath21 , ) parameter space . the excluded areas are shaded ( see legend for the colour / equation code ) . only the white fraction of each panel contains permissible model parameter combinations . see text for details . ] we have demonstrated in section [ sec : agb ] that agb stars can be selected using only _ wise_photometry . equations in appendix [ app : agb - selection ] are the mathematical definitions of our selection criteria . the corresponding loci of agb stars are visualized in fig . [ fig : agb - selection ] with red and blue lines . these equations also define lower and upper bounds of several _ wise _ linear colour combinations . for instance , equation can be written as ( where , ) . with given dust properties ( for example , amorphous carbon dust ) , we can compute the value of this linear colour combination , @xmath50 , for any dusty model parameter combination ( @xmath21 , ) . then the colour selection criterion demands that this value be within @xmath51 $ ] . therefore , the interval boundaries are simply contours in the parameter plane , and are plotted as solid lines in fig . [ fig : agb - analysis - exclusion ] ( for amorphous carbon dust in the left panel , for o - rich star models in the right panel ) . the legends relate the equation numbers to colours used for plotting . the areas outside of the interval are excluded by the selection criterion , and are shown shaded in the figure . each of the colour - selection equations translates to such parameter - exclusion areas . the complement of their union is the only portion of the parameter space with permissible combinations . it is shown in white in fig . [ fig : agb - analysis - exclusion ] . since our c - rich star colour selection volume is confined quite tightly ( cf . [ fig : agb - selection ] ) , the allowed parameter space for dusty models is also well - constrained . the dust shells must have interface temperatures not lower than @xmath52 k and the visual optical depths are in the range @xmath53 . for o - rich star models the selection in cc space was not as tight ; consequently the permissible parameter ranges are less well - constrained . our results seem to indicate that c - rich agb stars tend to have higher maximal optical depths than o - rich agbs , and that they lack the very lowest values . in contrast , o - rich agb models extend to smaller optical depths , and values larger than a few are excluded . this conclusion is supported by samples analysed in @xcite ; see their figs 5 and 6 , with data from @xcite , @xcite and @xcite ( we caution the reader that those data sets may be affected by unknown selection effects ) . we emphasize that our results should be viewed qualitatively and note that the parameter boundaries are ` soft ' because they depend on the exact colour selection criteria . the distribution of c - rich and o - rich agb stars in the _ wise _ colour space , as previously selected in section [ sec : agb ] and appendix [ app : agb - selection ] , is shown in fig . [ fig : agb - analysis - cc ] . overplotted are fiducial rdw model colour tracks , for which the temperature at the inner boundary is @xmath54 k ( solid lines ) or @xmath55 k ( dotted ) , in accordance with the parameter ranges found above . all tracks are computed for a shell thickness of 100 , and with a stellar temperature of @xmath56 k , as explained before . panels 0002 ( left side ) show our c - rich star selection and colour tracks computed with amorphous carbon dust , while panels 1012 ( right ) show our o - rich star selection and tracks with ` warm ' circumstellar silicates from @xcite . the model tracks do not match all agb plumes in all three orthogonal projections simultaneously , but the range spanned by models of these dust interface temperatures brackets well the observed _ wise _ colours of agb stars . the @xmath54 k rdw model with amorphous carbon dust ( blue solid line ) follows the plume of c - rich stars reasonably well in all three two - dimensional projections of the _ wise _ colour space ( panels 0002 ) . although the agreement is not perfect , it is likely that these 0.20.3 mag large offsets could be rectified with minor modifications of the dust opacity . we have not attempted a detailed study because such small changes could also be explained by deviations from spherical symmetry , non - uniform dust distribution , variations in photospheric temperature as the star ascends the asymptotic giant branch sequence , and possibly other effects . these degeneracies can not be broken with only three _ wise _ colours . the red tracks in panels 1012 of fig . [ fig : agb - analysis - cc ] correspond to a fiducial model for o - rich stars with standard silicate dust opacity and @xmath21 = 800 k and 500 k. models in this temperature range bracket the 3d distribution of o - rich agbs in the _ wise _ colour volume . it is remarkable that most of the extension of this o - rich star plume can be covered with shell optical depths below a value of unity . with detections for about 560 million objects , _ wise _ represents a major step forward in the surveying and understanding of the infrared sky . its combination of up to 1000 times higher sensitivity than previous ir surveys , and a relatively high angular resolution , enables studies of source samples with unprecedented statistical significance . we have computed colour tracks in the _ wise _ photometric system for two families of models : smooth spherical shells and clumpy dust torus models , and decided to study local / galactic populations in this paper , and the extragalactic sky in nikutta et al . ( in preparation ) . the model tracks provide guidance for the observed distribution of sources in _ wise _ cc diagrams , such as illustrated in the left panel in fig . [ fig : compare - to - wise ] . the main results of our computations are shown in the right panel in fig . [ fig : compare - to - wise ] . the model - based colour tracks outline reasonably well the distribution of sources expected to be associated with dust emission ( i.e. all source types except t dwarfs and stars without dusty shell ) . the location of dust - free stars coincides very well with the optically thin point of all dust - shell models irrespective of the dust composition . deviations from the zero - point are caused by _ wise _ filters and being sensitive to the blackbody peaks of the coolest stars , thus producing colours other than those for the rayleigh jeans slope . in the other extreme , at high optical depths , the dusty model tracks of ism dust shells with shallow radial density distributions overlap with those cc areas populated by ultra - luminous infrared galaxies ( ulirgs ) , which are known to be deeply shrouded in dust @xcite . we find that _ wise _ photometry alone is sufficient to reliably classify c - rich and o - rich asymptotic giant branch stars with dusty shells ( section [ sec : agb ] ) . with the large population of agb stars in the _ wise _ catalogue we were able to define magnitude and colour selection criteria that distinguish between the two classes . it is necessary to regard the distribution of _ wise _ colours as a three - dimensional arrangement in colour colour colour space . the traditional two - dimensional orthogonal projections are very helpful , but not sufficient . with the ability to cleanly separate the local populations of c - rich and o - rich agb stars we computed the most detailed map of the c : o - rich star count ratio for the milky way and its satellites ( see section [ sec : co - ratio ] ) . we find a strong spatial gradient away from the galactic plane , with the magellanic clouds showing a c : o ratio about 1000 times higher than the galactic bulge . we also clearly uncover the sagittarius dwarf spheroidal galaxy in the map . we find that the c : o dusty agb star ratio increases with distance from the lmc centre about twice as fast as measured for near - ir selected samples of early agb stars . detailed modelling of these measurements has the potential to shed new light on the formation and evolution of lmc . dusty model tracks for shells with rdws follow the identified agb plumes well . by mapping the previously found colour selection criteria to model parameters , specifically inner dust shell temperature @xmath21 and shell optical depth , we found observational constraints on these model parameters . they seem to indicate slight differences in the physical properties of c - rich and o - rich agb shells ( see section [ sec : agb - analysis ] ) . spherical dusty shell models with flat radial density distributions generate _ wise _ colours compatible with ysos . yso samples selected using _ iras _ and _ wise _ surveys have very little overlap due to vastly different imaging depths : while _ iras _ was sensitive to bright sources with relatively high visual optical depths , _ wise _ detects much fainter ysos with relatively small optical depths ( see section [ sec : yso ] ) . to investigate further how well models describe the _ wise _ colours of ir sources , sufficiently large samples need to be studied object - by - object , with sed fitting for each source over a large wavelength range . the _ spitzer _ sage survey @xcite , for instance , has been previously used for such a task . @xcite matched the ir _ spitzer _ spectra with optical and near - ir data to produce broad - band seds for more than 30000 agb stars in the lmc , and fitted them with _ grams _ models @xcite . thanks to _ wise _ , a similar study can be now undertaken for agb stars in the galaxy . clumpy models of agn dust torus emission , with their more complicated geometry and dust distributions , are much better than simple spherical shells in explaining the general location of qsos and other agns in the _ wise _ cc diagram . the location of the highest clumpy model density aligns almost perfectly with the centre of the circular region identified as ` seyferts ' by @xcite in the left panel of fig . [ fig : compare - to - wise ] . a detailed study of the _ wise_colours of agns will be presented in the companion paper ( ) . [ [ acknowledgements ] ] acknowledgements : + + + + + + + + + + + + + + + + + we are grateful to maria - rosa cioni for illuminating discussions and for making her data available to us . we reproduced fig . 12 from @xcite as part of our fig . [ fig : compare - to - wise ] by permission of the aas and the authors . we thank chao - wei tsai for providing us a high - quality version . we thank both referees for their helpful suggestions which improved the paper . we express our gratitude to xavier koenig for reminding us of the disk / shell dichotomy of yso models ( see section [ sec : yso ] ) , and to james davenport for his insightful post - submission comments on the variability information in the wise database . we are also thankful to patricia arvalo and ron wilhelm for helpful discussions . rn acknowledges support by the alma - conicyt fund , project no . 31110001 , and by fondecyt grant i thanks the hungarian academy of sciences for its support through the distinguished guest professor grant no . i and nhw acknowledge support by nsf grant ast-1008784 to the university of washington . me acknowledges nasa support and the award of an npp senior fellowship from orau , which supported a sabbatical leave at jpl / caltech where much of this work was completed . this publication makes use of data products from the wide - field infrared survey explorer , which is a joint project of the university of california , los angeles , and the jet propulsion laboratory / california institute of technology , funded by the national aeronautics and space administration . funding for sdss - iii has been provided by the alfred p. sloan foundation , the participating institutions , the national science foundation , and the u.s . department of energy office of science . the sdss - iii web site is http://www.sdss3.org/. sdss - iii is managed by the astrophysical research consortium for the participating institutions of the sdss - iii collaboration including the university of arizona , the brazilian participation group , brookhaven national laboratory , carnegie mellon university , university of florida , the french participation group , the german participation group , harvard university , the instituto de astrofisica de canarias , the michigan state / notre dame / jina participation group , johns hopkins university , lawrence berkeley national laboratory , max planck institute for astrophysics , max planck institute for extraterrestrial physics , new mexico state university , new york university , ohio state university , pennsylvania state university , university of portsmouth , princeton university , the spanish participation group , university of tokyo , university of utah , vanderbilt university , university of virginia , university of washington , and yale university . this research has made use of the ned which is operated by the jet propulsion laboratory , california institute of technology , under contract with the national aeronautics and space administration . some of the results in this paper have been derived using the healpix / healpy package @xcite . , j. a. , phillips , m. m. , & terlevich , r. 1981 , , 93 , 5 , p. , & demers , s. 2011 , in asp conf . ser . , vol . 445 , why galaxies care about agb stars ii : shining examples and common inhabitants , ed . f. kerschbaum , t. lebzelter , & r. f. wing , 479 , m. r. , et al . 2005 , , 129 , 2562 , a. d. , et al . 2007 , , 655 , 212 , m. l. , et al . 2013 , , 774 , 83 , m .- l. 2009 , , 506 , 1137 , m .- l. , & habing , h. j. 2003 , , ( ch03 ) , 402 , 133 , k. r. , et al . 2007 , , 134 , 2398 , j. j. , et al . 2012 , , 198 , 6 , b. t. 2003 , , 598 , 1017 , a. , & rowan - robinson , m. 1994 , , 266 , 212 , m. p. , van dyk , s. d. , & price , s. d. 2001 , , 122 , 1844 , m. , & ivezi , . 2001 , , 327 , 403 , m. w. , abedigamba , o. p. , & whitelock , p. a. 2010 , , 408 , l76 , l. , et al . 2010 , , 724 , 1030 , k. m. , hivon , e. , banday , a. j. , wandelt , b. d. , hansen , f. k. , reinecke , m. , & bartelmann , m. 2005 , , 622 , 759 , m. 1988 , grain optical properties , technical report , y. , et al . 2008 , , 60 , 435 , . , beers , t. c. , & juri , m. 2012 , , 50 , 251 , . , connolly , a. , vanderplas , j. , & gray , a. 2014 , statistics , data mining and machine learning in astronomy ( princeton university press ) , . , & elitzur , m. 1997 , , 287 , 799 , . , & elitzur , m. 2000 , , 534 , l93 , . , & elitzur , m. 2010 , , 404 , 1415 , . , nenkova , m. , & elitzur , m. 1999 , arxiv : astro - ph/9910475 , . , et al . 2007 , , 134 , 973 , t. , ivezi , . , & knapp , g. r. 2002 , , 337 , 749 , t. h. , et al . 2011 , , 735 , 112 , a. , van loon , j. t. , khosroshahi , h. , & mirtorabi , m. t. 2013 , , 432 , 2824 , e. a. , et al . 2010 , , 710 , 1444 , n. a. , sirocky , m. m. , hao , l. , spoon , h. w. w. , marshall , j. a. , elitzur , m. , & houck , j. r. 2007 , , 654 , l45 , p. , girardi , l. , bressan , a. , groenewegen , m. a. t. , silva , l. , & granato , g. l. 2008 , , 482 , 883 , j. s. , rumpl , w. , & nordsieck , k. h. 1977 , , 217 , 425 , m. , et al . 2006 , , 132 , 2268 , j. , feast , m. , whitelock , p. , olivier , e. , matsunaga , n. , & da costa , g. 2008 , , 385 , 1045 , m. , sirocky , m. m. , ivezi , . , & elitzur , m. 2008a , apj , 685 , 147 , m. , sirocky , m. m. , nikutta , r. , ivezi , . , & elitzur , m. 2008b , , 685 , 160 , g. , et al . 1984 , , 278 , l1 , s. , & weinberg , m. d. 2000 , , 542 , 804 , r. , hunt - walker , n. , nenkova , m. , ivezi , . , & elitzur , m. 2013 , , to be submitted ( paper ii ) , m. , et al . 2006 , , 370 , 1677 , f. m. , et al . 1986 , , 65 , 607 , h. , eriksson , k. , gustafsson , b. , & carlstroem , u. 1993 , , 87 , 305 , v. , henning , t. , & mathis , j. s. 1992 , , 261 , 567 , e. , spoon , h. w. w. , & tielens , a. g. g. m. 2004 , , 613 , 986 , a. m. s. , & yates , j. a. 1998 , irish astronomical journal , 25 , 7 , g. t. , et al . 2003 , , 126 , 1131 , d. , srinivasan , s. , sargent , b. , & meixner , m. 2012 , , 753 , 71 , b. a. , srinivasan , s. , & meixner , m. 2011 , , 728 , 93 , j. d. , norris , j. p. , jackson , b. , & chiang , j. 2013 , , 764 , 167 , d. p. , et al . 2010 , , 139 , 2360 , m. m. , levenson , n. a. , elitzur , m. , spoon , h. w. w. , & armus , l. 2008 , , 678 , 729 , v. , et al . 2004 , , 615 , l141 , b. t. , neugebauer , g. , & houck , j. r. 1987 , , 25 , 187 , s. , sargent , b. a. , & meixner , m. 2011 , , 532 , a54 , n. r. , et al . 1999 , , 346 , 843 , x. , & wang , z .- x . 2013 , research in astronomy and astrophysics , 13 , 323 , r. p. 2001 , , ( mc2 ) , 122 , 1827 , r. p. , & cioni , m .- r . l. 2001 , , ( mc1 ) , 122 , 1807 , w. e. c. j. , & habing , h. j. 1988 , , 194 , 125 , j. , connolly , a. , ivezi , . , & gray , a. 2012 , in conference on intelligent data understanding ( cidu ) , 47 , u. p. , et al . 2009 , , 137 , 3139 , d. , ivezi , . , miroshnichenko , a. s. , & elitzur , m. 2003 , , 346 , 1151 , p. a. , menzies , j. w. , feast , m. w. , matsunaga , n. , tanab , t. , & ita , y. 2009 , , 394 , 795 , e. l. , et al . 2010 , , 140 , 1868 , d. g. , et al . 2000 , , 120 , 1579 , k. 1995 , , 445 , 872 , a. a. , et al . 2006 , , 370 , 1961 in order to develop selection criteria for agb stars , we follow @xcite and analyse the _ wise _ colour distribution of known agb stars listed in the simbad data base . aided by their results , our procedure is simpler . they first selected about 76 million sources from the full _ wise _ all - sky source catalogue with detections in , and bands ( and free of contamination and confusion flags in all four bands ) that have a unique 2mass source within 3 arcsec . then they positionally matched this sample to objects listed in the simbad data base within 2 arcsec ( except in the case of high proper - motion stars which were matched within 4 arcsec ) , and further restricted the sample to 20 degrees from the galactic plane , yielding a sample of 3.2 million sources . this sample includes agb stars , as well as all other sources known to simbad and detected by _ wise _ , which satisfy the selection criteria . we begin by obtaining a sample of stars from simbad whose main identifier was consistent with agb stars ( agb star , c star , s star , oh / ir star , or mira - type variable ) , resulting in 32556 objects . the positions of these stars were matched to the positions of objects in the _ wise _ all - sky source catalogue within 2 arcsec , rejecting sources with more than one association . the resulting sample of 24942 stars is further reduced by rejecting all sources with contamination and confusion flags , as well as removing sources outside of saturation limits ( bands through ) and 5@xmath6 faint limits ( see section [ sec : intro ] ) , yielding the final sample of 6171 objects with presumably both reliable _ wise _ photometry and reliable simbad classifications . we did not enforce any cut on galactic coordinates . the distribution of this sample in representative cc and cm diagrams constructed with 2mass and _ wise _ photometry is shown in fig . [ fig : simbad ] . as known from previous work , most agb stars have 2mass colour @xmath57 . most of them also show clear _ wise _ colour excess , @xmath12 > 0 $ ] and @xmath38 > 0 $ ] , as well as @xmath58>0 $ ] . the two magnitudes limits discernible from the source distribution in the top right panel are due to adopted faint limit @xmath13 < 11.32 $ ] and saturation limit @xmath37 > 8.1 $ ] . the latter is responsible for a significant difference between the data distribution in our c[2,3 ] versus c[1,2 ] diagram ( the bottom left panel ) and the diagram shown in fig . 4 from @xcite . their version displays c stars below the dashed line in our bottom left panel , and oh / ir stars above the dashed line . based on this separation , they proposed the dashed line as a separator between these two classes of agb stars . however , in our version , all the stars below the line are absent , as well as all oh / ir stars , and the region above the line is dominated by c stars , and not by oh / ir stars ! it turns out that these differences are caused by rejecting stars brighter than saturation limits - when we relax this criterion , we obtain a data distribution reminiscent of fig . 4 from @xcite . this is shown in fig . [ fig : saturated ] , similar to fig . [ fig : simbad ] , but now includes formally saturated stars . all saturated stars , with @xmath118.1 and @xmath116.7 , are shown with red dots , while the black dots show unsaturated sources . notice in the lower left panel that almost all saturated stars fall below the dashed line that @xcite have devised to separate c - rich from o - rich agb stars . they did not mention saturation limits in their work . we believe that sub - standard _ wise_photometry for saturated agb stars has induced unreliable source distribution in fig . 4 from @xcite . the band is most likely to blame , as evidenced by fig . 8 ( in section vi.3.c.i.4 ) of the _ wise _ explanatory supplement ( see footnote [ foot : explsuppl ] ) , where the bias for saturated sources is most pronounced in . with saturation limits enforced , the distribution of c stars , by and large from lmc and smc , clearly stands out in fig . [ fig : simbad ] ( see the bottom right panel ) . we define the 3d colour space @xmath59 $ ] , @xmath60 $ ] , @xmath61 $ ] . the box that outlines a high concentration of c - rich agb stars is defined by @xmath62 in the c[2,3 ] versus c[1,2 ] diagram shown in the bottom left panel of fig . [ fig : simbad ] , o - rich stars with silicate dust are seen as a low - density cloud towards the left from the c - rich star box . further insight in the distribution of o - rich and c - rich agb stars in _ wise_cc diagrams is based on complete _ wise _ samples and is discussed in section [ sec : agb ] . based on this analysis our samples of agb stars for fig . [ fig : agb - selection ] are selected as follows . the initial criteria are : @xmath63/\w[2]/\w[3]/\w[4 ] < 16.83/15.60/11.32/8.0 & 5\sigma \mbox{-limits}\\ \w[1]/\w[2]/\w[3]/\w[4 ] > 8.1/6.7/3.8/{-}0.4 & \mbox{saturation limits}\\ \w[1]<11,\ \w[2]<10 & \mbox{only bright sources}\\ \mbox{snr1 \ & 2 } > 5,\ \mbox{snr3 \ & 4 } > 10 & \mbox{signal - to - noise ratio } \end{array}\ ] ] the regions with @xmath64 deg were also excluded because ysos tend to cluster in the galactic plane @xcite , and also because of the uncontrolled / unknown levels of galactic extinction at these latitudes . in the @xmath65 colour space defined above we also require @xmath66 , @xmath67 , excluding the long narrow plume of stars with constant - density dust shells . this so - selected initial agb sample consists of 8835 sources . to select o - rich agb stars from this initial sample we used the colour - colour diagrams 35 in fig . [ fig : agb - selection ] as a guideline . we defined a wedge - shaped volume bounded by these criteria : @xmath68 the selected o - rich star sample has 1608 members . the c - rich stars , visible as a strongly elongated plume in fig . [ fig : agb - selection ] , were selected by carving out a 3d elongated volume around the plume . it is the intersection of the following two slabs : @xmath69 note that ( [ eq : c1 ] ) is the inverse of ( [ eq : n1 ] ) , and that we abandon ( [ eq : n2 ] ) . the c - rich star sample contains 1383 sources . for the c : o - rich star ratio map in fig . [ fig : co - ratio ] we have relaxed the @xmath64 deg requirement , and have selected c - rich and o - rich star candidates through the cc volumes and magnitudes defined above . the preselected sample contains 90114 sources , of which 14969 end up in the o - rich star group . the size of the c - rich star selection grows to 1629 objects . to obtain a de - projected map of the lmc disc , for our analysis of the radial gradient of the c : o - rich star ratio in section [ sec : co - ratio ] , we follow the procedure outlined in two papers on the structure of the magellanic clouds , and . we begin with the right ascension and declination angles @xmath70 and @xmath71 of our colour and magnitude selected c - rich and o - rich agb stars ( given in degrees or radians ) . we further assume as the origin of this equatorial coordinate system @xmath72 , as defined in section 4.1 in . this origin corresponds to the centre of the density iso - contours of the _ denis _ survey , but is slightly offset w.r.t . the standard locus of the lmc centre [ e.g. @xmath73 given in the ned data base ] . from equation ( 1 ) of we obtain the formula for transformed angular coordinate @xmath74 @xmath75 merging equations 2 and 3 from the same paper we get the other transformed coordinate @xmath76 @xmath77 then defines in equations 2 the de - projected coordinates @xmath78 and @xmath79 as linear distances ( in kpc ) from the lmc centre , in the plane of its disc @xmath80 \label{eq : yp } y ' & = \frac{d_0\ \sin\rho\ \sin(\phi-\theta_{\rm far } ) } { \cos i\ \cos\rho\ -\ \sin i\ \sin\rho\ \sin(\phi-\theta_{\rm far})},\end{aligned}\ ] ] where @xmath81 is the distance from earth to the lmc , 51 kpc , and is the inclination of the lmc disc plane with respect to our line of sight . @xmath82 , where @xmath83 is the position angle ( pa ) of the `` line of nodes '' , the intersection of the lmc disc plane and the plane of the sky . see the discussion in as to why @xmath84 differs from the pa of the lmc semi - major axis . finally rotate the ( @xmath78,@xmath79 ) coordinates by the angle @xmath85 to align them with another set of coordinates in their paper . for our work , this step is not necessary , but we execute it to obtain a map oriented identically to @xmath86 y '' = x ' \sin\theta_{\rm far}\ + \ y ' \cos\theta_{\rm far}.\end{aligned}\ ] ] these are the coordinates we plot in the lower left panel of our fig . [ fig : co - ratio - lmc ] . we select all c - rich and o - rich stars that fall within a @xmath87 kpc circle . their counts are , respectively , and . equation averages the observed sed of a source or model with the instrument s total transmission curves , and calibrates the in - band fluxes to an empirical fit for vega @xcite . recall that the _ wise _ colour between two bands 1 and 2 is defined as @xmath88 & = 2.5\log\frac{f_2}{f_1 } = 2.5 \left(\log \frac{f_2^{\rm src}}{f_1^{\rm src } } + \log \frac{f_1^{\rm vega}}{f_2^{\rm vega } } \right),\\ & = ( \c[1,2])^{\rm src } - ( \c[1,2])^{\rm vega}.\end{aligned}\ ] ] to compute colours , the absolute scale of the sed is irrelevant , because all scales cancel out . however , to allow luminosity - dependent studies using our models , we list in table [ tab : tabulations ] , together with the vega - calibrated _ wise _ colours , the unnormalised in - band fluxes @xmath89 of the dusty models used in figs [ fig : compare - to - wise ] , [ fig : d - colors][fig : rdw ] , and [ fig : agb - analysis - cc ] . the pure vega in - band fluxes , using the aforementioned empirical fit , are constants . their values are : for bands @xmath90 14 , in the same units as the models . the dusty model seds are normalized with the bolometric flux , thus for comparison with observational data the reported @xmath89 values must be multiplied with the true bolometric flux of the source . the entire tabulations are available as an electronic supplement file ` model_tables.dat ` to this paper . all physical properties of the models are listed in the header of each table , and are tabulated in the electronic file . they are : the shell radial optical depth at visual ( 0.55 ) , the dust chemistry , grain size distribution , the photospheric temperature @xmath48 of the central star , the radial density distribution ( either a power - law @xmath19 , or rdw ) , and the dust temperature @xmath21 at the inner radius of the shell . all models have radial shell thickness @xmath91 . llccccccc & & c[1,2 ] & c[2,3 ] & c[3,4 ] & @xmath92 & @xmath93 & @xmath94 & @xmath95 + & & [ 3.4 - 4.6 ] & [ 4.6 - 12 ] & [ 12 - 22 ] & & & & + & & + + # 0 & & 7 & 8 & 9 & 10 & 11 & 12 & 13 + 0.000e[1.0]@xmath960 & & 5.780e[1.0]@xmath973 & 1.541e[1.0]@xmath972 & 1.303e[1.0]@xmath972 & 2.102e[1.0]@xmath973 & 1.114e[1.0]@xmath973 & 2.640e[1.0]@xmath974 & 1.779e[1.0]@xmath975 + 1.000e[1.0]@xmath974 & & 6.098e[1.0]@xmath973 & 2.228e[1.0]@xmath971 & 1.769e[1.0]@xmath960 & 2.102e[1.0]@xmath973 & 1.114e[1.0]@xmath973 & 3.196e[1.0]@xmath974 & 1.085e[1.0]@xmath974 + & & +

we prove that the generator of the renormalization group of potts models on hierarchical lattices can be represented by a rational map acting on a finite - dimensional product of complex projective spaces . in this framework we can also consider models with an applied external magnetic field and multiple - spin interactions . we use recent results regarding iteration of rational maps in several complex variables to show that , for some class of hierarchical lattices , lee - yang and fisher zeros belong to the unstable set of the renormalization map .

0708.0616

potts models on hierarchical lattices have been introduced in 1979 by berker and ostlund @xcite as an interpretation of migdal - kadanoff models , defined in 1975 @xcite in order to approximate classical spin models on @xmath0 . later , in 1981 , griffiths and kaufman @xcite provided a rigorous definition of hierarchical lattices and studied some examples in detail . one of such examples , the diamond hierarchical lattice , was later considered in a paper by derrida , de seze and itzykson @xcite , who showed that the generator of the renormalization group ( see e.g. @xcite ) could be written as a rational map acting on the riemann sphere ; as a consequence , the fisher set of the model coincides with the julia ( i.e. unstable ) set of the renormalization group map . later , similar results were established to study other specific lattices ( e.g @xcite ) or to introduce coupling with an external magnetic field in a similar dynamical framework ( e.g. @xcite ) . + in this paper we generalize the result of @xcite to all hierarchical lattices , i.e. we prove that the generator of the renormalization group of a potts model on a hierarchical lattice can be represented by a rational map acting on a complex multiprojective space ( sections [ hierarchical ] and [ dynproj ] ) . the general approach that we introduce , not only allows to describe all models on hierarchical lattice that have already been studied , but it also provides an extremely natural way to deal with an external magnetic field ( ) . the study of the dynamics obtained by iteration of a rational map in several complex variables is a quite recent research subject and , as such , it is still quite incomplete . nevertheless , recent results by dinh - sibony @xcite allow us to prove that , at least for some class of hierarchical lattices , lee - yang and fisher sets are a subset of the julia set of the renormalization map ( ) . this paper features two technical appendices that give the basic mathematical background needed to understand the statements in the main part and provide references for the interested reader . a number of examples of potts models on hierarchical lattices are presented in @xcite , where it is shown how to obtain both exact and numerical results by using the general methods developed in this paper . in order to state our result in full generality , we need to provide formal definitions and notations for the objects we will use in the paper . in spite of the technical nature of such definitions , they are indeed quite natural and , most importantly , they will lead to a very simple proof of the result . _ hierarchical lattices _ ( in short _ _ hl__s ) are lattices that are left invariant by a given coarse - graining operation . the most famous example is provided by the diamond hierarchical lattice @xcite which is obtained by iterating the substitution which replaces an edge with four edges linking the original vertices with two new ( internal ) vertices ( see , e.g. @xcite , figure 1 or @xcite , figure 1 ) . our goal is to extend this procedure so as to be able to consider more general cases . to this purpose , we are going to define hierarchical lattices as limits of sequences of finite objects obtained iterating a _ decoration _ procedure , which is going to be dual to the coarse - graining operation . the finite objects we consider are a generalization of graph called _ hypergraph _ ( see e.g. @xcite ) ; hypergraphs have been briefly considered in @xcite ( see section v ) under the name of `` generalized graphs '' and they were used for defining hierarchical lattices with multiple spin interactions . in fact , hypergraphs differ from graphs in the sense that edges ( sometimes also called _ hyperedges _ or _ links _ ) are allowed to connect an arbitrary number of vertices . hereby follows the standard definition . a hypergraph @xmath1 is defined by a set @xmath2 of vertices and a set @xmath3 of edges that are finite ordered non - empty subsets of @xmath2 ; the same vertex can not appear more than once in an edge . given an edge @xmath4 , we define _ rank _ of @xmath4 its cardinality @xmath5 as a subset of @xmath2 . if all edges have the same rank @xmath6 , the hypergraph is said to be @xmath7uniform and @xmath6 is said to be the _ order _ of the hypergraph . + given a hypergraph @xmath8 , a _ partial hypergraph _ @xmath9 is defined as a hypergraph such that @xmath10 and @xmath11 . from the physical point of view , edges will connect spins that are coupled to each other ; notice that the definition only takes into account edges of finite rank as we do not consider interactions of infinite range . we _ do not _ assume that either @xmath2 or @xmath3 are finite . + it will be necessary to associate different properties ( e.g. interactions ) to edges of the same rank ; such properties will be indexed by a ( at most ) countable index set @xmath12 ( the set of _ types _ ) that will be common among all hypergraphs ; the notion of _ structured hypergraph _ takes into account this additional piece of information . in order to define it , we first need to introduce the notion of _ partition _ of a hypergraph @xmath8 _ into uniform partial hypergraphs _ @xmath13 , where the _ rank _ @xmath14 and the _ type _ @xmath15 have been fixed . this partition is obtained as follows : we define @xmath16 to be the set of all edges of @xmath1 with rank equal to @xmath6 and type equal to @xmath17 . one of course has : @xmath18 the edge set @xmath3 of the original hypergraph will be the disjoint union @xmath19 and , denoting with @xmath13 the uniform partial hypergraph @xmath20 , we have @xmath21 note that since each @xmath13 is @xmath6-uniform , each element of @xmath16 is an ordered @xmath7tuple of vertices . the space of all pairs ( rank , type ) is called @xmath22 and we denote its elements by greek letters e.g. @xmath23 . + we can now define a _ structured hypergraph _ @xmath24 as a hypergraph @xmath1 along with a partition into uniform partial hypergraphs . the sets @xmath25 will be called _ partial edge sets _ of @xmath24 . we define the _ multiorder _ of @xmath24 to be the set @xmath26 . when @xmath27 is _ finite _ , then @xmath24 is said to be a _ finitely structured hypergraph _ ; if @xmath28 has only one element ( i.e. @xmath29 ) , then @xmath24 is said to be @xmath30uniform and @xmath31 will be called _ order _ of the structured hypergraph . given a structured hypergraph @xmath32 , it is convenient to consider each uniform partial hypergraph @xmath33 also as a @xmath30uniform structured hypergraph @xmath34 , so that we can write @xmath35 . ) , width=264 ] @xmath36 e_{(1,1)}&=&\left\{(v_1),(v_2)\right\}&\mathrm{rank\ 1\ type\ 1}\\ e_{(2,1)}&=&\left\{(v_2,v_4),(v_3,v_4)\right\}&\mathrm{rank\ 2\ type\ 1}\\ e_{(2,2)}&=&\left\{(v_5,v_7)\right\}&\mathrm{rank\ 2\ type\ 2}\\ e_{(3,1)}&=&\left\{(v_1,v_2,v_3)\right\}&\mathrm{rank\ 3\ type\ 1}\\ e_{(4,1)}&=&\left\{(v_4,v_5,v_6,v_7)\right\ } & \mathrm{rank\ 4\ type\ 1}\\ \end{array}\ ] ] @xmath37 [ hypergraph ] for convenience of notation , in the remaining of this section we will consider only _ finitely structured hypergraphs _ ; all the statements can easily be generalized to the infinite case . let @xmath24 be a structured hypergraph @xmath38 and let @xmath39 be a map @xmath39 from the set @xmath2 to another set @xmath40 such that the restriction of @xmath39 to every edge @xmath41 is injective ; we call such an @xmath39 a _ locally injective map_. given a locally injective map @xmath39 , for all @xmath42 we can induce a map @xmath43 from each @xmath25 to the set of ordered @xmath6-tuples of @xmath40 as follows : @xmath44 by local injectivity , @xmath45 can be regarded as an edge set on @xmath40 and we can define ( with a slight abuse of notation ) @xmath46 as the _ structured hypergraph induced by _ @xmath39 . the decoration procedure we want to define ( that will be dual to the coarse - graining operation ) will consist in gluing a fixed structured hypergraph to each edge of a given rank and type of another structured hypergraph . in order to do so , we need to mark the vertices which will be used in the gluing process : structured hypergraphs with marked vertices will be called _ decorated edges _ ( see e.g. figure [ spiderweb ] ) . let @xmath42 . a _ decorated @xmath30edge _ @xmath47 ( of rank @xmath6 and type @xmath17 ) is a structured hypergraph @xmath24 with @xmath6 marked vertices . marking vertices amounts to choose an additional _ ordered _ @xmath7tuple of vertices ; borrowing the terminology from @xcite , section v , marked vertices will be called _ external ( or surface ) vertices _ and vertices that are not external will be called _ internal ( or core ) vertices_. @xmath48 a decorated @xmath30edge is said to be uniform if the underlying hypergraph @xmath24 is @xmath30uniform . + @xmath49 e_{(3,1)}&=&\left\{(v_1,v_4,v_3),(v_2,v_4,v_1),(v_3,v_4,v_2)\right\ } \end{array}\ ] ] @xmath50 [ spiderweb ] decorated edges can be physically regarded as the inner structure of an edge of a given rank and type . notice moreover that the value of @xmath17 is not taken into consideration in the definition of a general decorated edge ; it will , however , play a role in what follows . it is easy ( see [ sum ] ) to introduce a natural notion of sum on decorated @xmath30edges ; the attempt to define a natural multiplication operation leads to a fundamental operation on a structured hypergraph @xmath24 which will be called _ decoration_. it amounts to substituting edges of rank @xmath6 and type @xmath17 in a hypergraph with given decorated edges of the same rank and type . + let @xmath42 be fixed , @xmath51 be an @xmath30uniform structured hypergraph and @xmath52 be a decorated @xmath30edge . the _ product _ of @xmath24 with @xmath47 is the structured hypergraph given by the following procedure : each edge @xmath41 is removed from @xmath24 and replaced by a copy of @xmath47 , with surface vertices of @xmath47 identified to the vertices of @xmath4 , respecting their ordering . the partition in uniform partial hypergraphs for the resulting hypergraph will be the one induced by the partition of @xmath47 ; the resulting structured hypergraph will be denoted by @xmath53 : more formally , let @xmath54 . if we define the collapsing map @xmath55 as follows : @xmath56@xmath57 then the edge sets are given by : @xmath58 and the resulting structured hypergraph will be @xmath59 given a structured hypergraph @xmath24 , one can multiply simultaneously and independently each @xmath30uniform partial hypergraph of the partition @xmath60 with a decorated @xmath30edge @xmath61 . + we define the _ identity _ decorated @xmath30edge to be the uniform decorated @xmath30edge with @xmath6 surface vertices , no core vertices and only one @xmath30edge ( of rank @xmath6 and type @xmath17 ) connecting the surface vertices with the correct ordering . we define a _ decoration _ @xmath63 as a choice of decorated edges @xmath64 , such that only finitely many @xmath61 are different from @xmath65 . then @xmath63 acts on a structured hypergraph @xmath24 as follows : @xmath66 notice that if we choose @xmath67 for all @xmath68 we have the identity operation @xmath69 . for notational convenience we will explicitly write as subscripts of @xmath63 only the non - trivial decorated edges involved in the decoration procedure . moreover , it is clear that any decoration of a finitely structured hypergraph with decorated edges that are themselves finitely structured will yield a finitely structured hypergraph . note that we can define a decoration operation in the class of decorated @xmath30edges by applying the decoration to the underlying structured hypergraph and keeping the same external vertices . this last remark allows us to define the composition of decorations in the following natural way : let @xmath70 and @xmath71 be two decorations ; then we define the composite decoration as : @xmath72 using the decoration procedure , we introduce a partial ordering in the class of structured hypergraphs . we say that @xmath73 if there exists a decoration procedure @xmath63 such that @xmath74 . + we now fix a a decoration operation @xmath63 and a finite initial hypergraph @xmath75 ; decorating @xmath75 will yield @xmath76 . if we iterate the action of @xmath63 ( see e.g. ) there can be two cases : either at some point the decoration operation acts trivially on the obtained hypergraph because we run out of edge to decorate , or not . the former case corresponds to _ finitely renormalizable lattices _ ; in the latter case the infinite lattice @xmath77 obtained as the inductive limit of the decoration procedure is called a _ hierarchical lattice_. with the decorated edge of , width=302 ] in this setting it is now clear that the decoration operation is dual to the coarse - graining process that amounts to gluing the original edges back in the place of the corresponding decorated edges . moreover , notice that this is only one particular way to construct an infinite lattice using decorations ; for instance it would be interesting to study thermodynamical properties of an infinite lattice obtained by fixing two ( or more ) decorations and then choosing one or the other at random to define the sequence @xmath78 ( i.e. a random walk on decorations ) . + in all subsequent sections we will only deal with finitely structured hypergraphs , therefore , without risk of confusion , we will drop the words `` finitely structured '' and use just the word `` hypergraph '' . we will consider potts models on hierarchical lattices ; hamiltonians will be obtained by summing over all edges a local interaction that depends only on the states of the spins belonging to the edge , i.e. a nearest - neighbours interaction . it is worthwhile to notice that , since edges of hypergraphs may connect an arbitrary number of vertices , such interactions are not restricted to pair interactions ; this flexibility turns out to be useful as , for instance , it allows at the same time to deal with external magnetic fields ( by considering edges of rank 1 ) or to study the more complicated interactions that arise renormalizing a pair interaction . + let @xmath79 be the number of potts states of the model ; for a given hypergraph @xmath80 , a _ configuration _ @xmath81 is a map from @xmath2 to @xmath82 . in order to associate an energy to each configuration we , first need to fix the nearest - neighbours interactions : this amounts , for each edge set @xmath83 , to fix the energy contribution of the configuration of the @xmath6 spins connected by such edges , i.e. to fix @xmath84 complex numbers . such numbers will be denoted by @xmath85 , where @xmath86 and @xmath87 is a multi - index ranging over @xmath88 . the total energy associated to a configuration @xmath81 is therefore easily expressed in terms of such @xmath89 : @xmath90 the associated partition function is : @xmath91 where @xmath92 ; define now the _ boltzmann weights _ as : @xmath93 in such coordinates , each term @xmath94 is a monomial of degree given by the number of edges in the hypergraph . if we fix a partial edge set @xmath95 , the degree of the polynomial in the variables @xmath96 is given by the number of edges in @xmath95 . thus , @xmath97 is a _ homogeneous polynomial _ that is _ separately _ homogeneous in @xmath98 for all fixed @xmath31 . as decorated edges of rank @xmath6 are hypergraphs with @xmath6 marked vertices , it is natural to consider the _ conditional _ partition functions of a decorated edge , for which we specify the @xmath6 states @xmath99 of the external vertices @xmath100 and restrict the sum to configurations satisfying the condition : @xmath101 once more , these are homogeneous and separately homogeneous polynomials in @xmath98 of fixed degree , independent of the choice of the external states . it is easy to check that the identity edge @xmath65 gives the trivial conditional partition function @xmath102 . conditional partition functions provide a natural way to connect the partition function of a hypergraph and the partition function of its image under decorations . consider a decorated @xmath30edge @xmath52 . we define the renormalization map @xmath103 as given in coordinates by the conditional partition functions : @xmath104 consider a @xmath30uniform hypergraph @xmath105 ; the partition function of @xmath105 is the polynomial @xmath106 . if we multiply @xmath105 with @xmath47 we obtain a hypergraph @xmath107 whose partition function is the polynomial @xmath108 . the fundamental property of the partition function of a product is that it is obtained by composing the original partition function with the renormalization map , i.e. we claim that : @xmath109 in fact , one can rewrite the sum over configurations involved in the partition function of @xmath107 by _ first _ summing over the configurations of vertices that belong to @xmath105 as well , _ then _ over configurations of all vertices that have been generated by decorating each edge of @xmath105 . in this way it is straightforward to see that @xmath110 is obtained by substituting each occurrence of @xmath111 in @xmath97 with @xmath112 . + a decoration @xmath63 amounts to a choice , for all @xmath68 , of a decorated @xmath30edge @xmath61 such that only finitely many @xmath61 are different from the identity . to each @xmath61 we can associate its renormalization map : @xmath113 and finally we can define the renormalization map @xmath114 as the juxtaposition of the maps @xmath115 , i.e. : @xmath116 where @xmath117 is the natural projection . + now consider the case of general hypergraphs ; let be a structured hypergraph ; its partition function is a polynomial @xmath118 . let ; the partition function of @xmath119 is a polynomial . again the claim is : @xmath120 and it follows by applying the previous argument to each element of the partition into partial uniform hypergraphs . + the relation between the decoration operation @xmath63 and the renormalization map @xmath114 is contravariant , i.e. : @xmath121 in fact , the renormalization operation is covariant to the coarse - graining operation dual to the decoration procedure . moreover , notice that the domain of the renormalization map @xmath114 is the infinite dimensional space of all interactions ; however , since @xmath63 acts as the identity on all but finitely many edge sets , @xmath114 acts non - trivially on a finite dimensional space only . if we have a hierarchical lattice @xmath122 generated by the iteration of decoration procedure @xmath63 , then @xmath114 can be iterated on the space of boltzmann weights of @xmath122 and this space will be a finite dimensional complex vector space . as we will see later , the dynamics of @xmath114 will reflect thermodynamical properties of the potts model on @xmath122 . when defining the interactions @xmath123 , we can choose the zero of energy for each edge set independently and arbitrarily . this freedom is reflected by the fact that the physics of the system will not change if we apply the map @xmath124 or , equivalently , , for an arbitrary choice of @xmath125 . this elementary observation allows us to establish an equivalence relation on each space of boltzmann weights @xmath126 i.e. : @xmath127 equivalent boltzmann weights will give identical physical systems . if we take the quotient of @xmath128 with respect to this equivalence relation , we obtain a _ projective space _ . thus , the quotient of the space of all boltzmann weights with respect to all such equivalence relations is a product of projective spaces , i.e. a _ multiprojective space _ , that will be called _ dynamical space _ and will be denoted by @xmath129 . given @xmath130 , the dynamical space associated to @xmath105 will be the finite dimensional multiprojective space : @xmath131 notice that , if we have a @xmath30uniform hypergraph @xmath24 , @xmath42 , the dynamical space @xmath132 is a standard complex projective space of dimension @xmath133 . hereafter the boltzmann weights @xmath134 will be considered to belong to the dynamical space and they will be denoted by @xmath135 $ ] . natural coordinates on the resulting projective space are _ homogeneous coordinates _ of which we recall the definition in [ appa ] . + notice that the renormalization map is a well - defined rational map on the dynamical space , since each coordinate given by a separately homogeneous polynomial . moreover , the dynamical space of a hierarchical lattice is finite dimensional and invariant under the renormalization map . this means that at most a finite number of new interactions will be generated by the renormalization procedure ; in this sense , potts models on hierarchical lattices are completely renormalizable . the approach we just presented is particularly convenient for studying the dynamics of the renormalization map , as the dynamical space has now been compactified in a natural way . all homogeneous thermodynamical quantities ( e.g. susceptibility ) can still be defined using variables in the dynamical space , but in order to define inhomogeneous quantities ( such as free energy ) we need to fix a zero of energy i.e. to consider variables belonging to the linear ( not the projective ) spaces . + we will now look for invariant ( projective ) subspaces of the dynamical space ; studying the dynamics of the renormalization map in such smaller subspaces is both interesting , as they correspond to special physically symmetric configurations , and convenient , as a map on a lower dimensional space is generally easier to study . + we are going to consider two different symmetries of the dynamical space : the first one is generated by @xmath136 , the group of permutations of @xmath137 ; the second symmetry is generated by the groups @xmath138 , where each @xmath139 is the group of permutations of vertices of edges belonging to @xmath25 . + the group @xmath136 acts on the dynamical space in the following natural way : let @xmath140 ; for all @xmath31 we denote by @xmath141 the map @xmath142 defined as follows : @xmath143\right)= \left[z^{{\alpha_{}}}_{us_1\cdots us_{r}}\right].\ ] ] with a slight abuse of notation we denote by @xmath141 also the map that acts on an arbitrary product @xmath144 by applying @xmath141 to each factor @xmath145 . the following proposition can be regarded as a general statement about the fact that if we perform the renormalization of a system with no external magnetic field , then the renormalized system will have no external magnetic field . more precisely : for all @xmath47 , the action of @xmath136 commutes with @xmath146 . [ white ] by definition , each component of @xmath146 is a conditional partition function ; let us consider the partition function associated to the choice of a multi - index @xmath87 : @xmath147\right)= \sum_{\begin{subarray}{c}\sigma\in s^v s.t . \cr \sigma(\mathrm{ext})=i\end{subarray}}\exp(-\beta{\mathscr{h}}^{\mathscr{e}}(\sigma)).\ ] ] given an element @xmath140 , we can write its action after the renormalization map : @xmath148 since the sum is over all the configuration space we can as well sum over @xmath149 , so that : @xmath150\right){\doteqdot}\big({\mathscr{r}}^{\mathscr{e}}u^*\big)_i.\ ] ] in all cases of interest , we will consider the action of a subgroup @xmath151 of @xmath136 that is either going to be the whole group @xmath136 ( no external magnetic field : all states are considered equal ) or @xmath152 ( external magnetic field : one state is special , all others are equal ) . consider the subset of @xmath129 of points fixed by the action of @xmath151 ; then , proposition [ white ] states that this subset is _ invariant _ under @xmath114 . this subset will turn out to be a lower dimensional multiprojective space naturally embedded in @xmath129 . we will provide this embedding shortly , but first we need to describe the action of the other symmetry group . + each group @xmath153 acts on the dynamical space in a natural way as well : let @xmath154 . we denote by @xmath155 the map @xmath156\mapsto\left[z^{{\alpha_{}}}_{s_{v1}\cdots s_{vr}}\right]$ ] on the dynamical space . given a decorated @xmath30edge @xmath47 , @xmath146 does not necessarily commute with the action of @xmath139 , since @xmath47 may have some internal structure that could break the symmetry . this amounts to say that if we renormalize a completely @xmath139-symmetric interaction we can possibly obtain a renormalized interaction that is _ not _ @xmath139 symmetric . in fact , given a subgroup @xmath157 of @xmath139 we say that a decorated @xmath30edge @xmath47 is @xmath158symmetric if @xmath146 commutes with the action of @xmath157 . most of the times , we will consider decorations @xmath63 that are completely symmetric , i.e. such that all decorated edges @xmath61 are @xmath159symmetric . in such cases the space of interactions fixed by the action of the whole group is again invariant under @xmath114 and we can focus on the action of the renormalization group on this smaller submanifold that is again going to be an embedded multiprojective space . + we are now going to present , for each @xmath31 , a decomposition of @xmath126 into subspaces that are invariant under @xmath136 ; we will then select a _ fixed _ vector in each of such subspaces and the set of such vectors will ultimately form a basis for the linear subspace of fixed vectors , that projected on @xmath160 will give an embedded projective space . the same decomposition , applied to each factor of @xmath129 , will give an embedded multiprojective space . the same idea will then be used to find the appropriate multiprojective space in the case of @xmath152 , i.e. of an external magnetic field . + we first need to classify basic invariant subspaces ; in order to do so we need to define a variation of young tableaux : a _ young diagram _ represents a way to write a natural number @xmath6 as the sum of @xmath161 naturals @xmath162 . it is pictured as @xmath6 boxes arranged in @xmath161 rows as in the following example : @xmath163 a ( generalized ) _ young tableau _ is a young diagram in which we fill the boxes with numbers from 1 to @xmath6 according to the rule that numbers on the same row are increasing from left to right and numbers on the first column of rows of equal length are increasing from top to bottom , for example : @xmath164 this is _ not _ the usual definition of young tableaux involved in the classification of representation of the permutation group : in fact , for this purpose , each column would be ordered so as to be increasing from top to bottom as well . the definition we presented is , however , exactly what we need to classify basic invariant subspaces . + for each @xmath42 , numbers from 1 to @xmath6 are associated to the corresponding spin of each @xmath7tuple of vertices belonging to the edge set @xmath25 ; to each young tableau with @xmath6 boxes and at most @xmath165 rows we associate the invariant subspace given by the following constraints : spins belonging to the same row have to be in the same state ; spins belonging to different rows have to be in different states . in the case of completely symmetric decorations we can do the same with young diagrams , as we can forget about the ordering of the spins . for each invariant subspace there exists a one - dimensional space on which the permutations act trivially , that is the subspace generated by the sum of all base vectors ; such vector will be denoted by @xmath166 with the corresponding young tableau as a subscript ; the direct sum of all such fixed spaces is obviously fixed by the permutation groups and it projects onto a projective space on @xmath160 . [ ezexample ] consider the case @xmath167 . the complex space of boltzmann weights @xmath126 is a linear space of complex dimension 27 and it will have as a basis : @xmath168 all possible young tableaux according to our definition , with the corresponding invariant subspaces are : @xmath169 \young(12,3)&\rightarrow & \langle e_{112},e_{113},e_{221},e_{223},e_{331},e_{332}\rangle\\[15pt ] \young(13,2)&\rightarrow & \langle e_{121},e_{131},e_{212},e_{232},e_{313},e_{323}\rangle\\[15pt ] \young(23,1)&\rightarrow & \langle e_{211},e_{311},e_{122},e_{322},e_{133},e_{233}\rangle\\[15pt ] \young(1,2,3)&\rightarrow & \langle e_{123},e_{132},e_{213},e_{231},e_{312},e_{321}\rangle \end{array}\ ] ] where we denote by @xmath170 the @xmath171dimensional complex vector subspace of @xmath172 obtained by taking @xmath173linear combinations of the vectors @xmath174 . the complex 1-dimensional fixed subspace associated to each tableau is generated by the sum of the corresponding base vectors . @xmath175 passing to the quotient , this subspace of complex dimension 5 will therefore project down on @xmath176 as an embedded @xmath177 . + in the completely symmetric case , we can use young diagrams instead of young tableaux , obtaining a yet lower dimensional subspace , as the three subspaces corresponding to the young diagram @xmath178 are now part of the same subspace . passing to the quotient we thus obtain an embedded @xmath179 . in the case of external magnetic field we will need to consider special young diagrams and tableaux with a privileged row that do not mix under permutations with the others . this leads to even more complicated young tableaux ; in the following example we will consider completely symmetric decorations , so we can just use marked young diagrams : case @xmath180 , @xmath181 . we will consider state 1 as the special ( magnetic ) one . a natural basis for the complex space is : @xmath182 all possible marked young diagrams , with the corresponding invariant subspaces are : @xmath183{7.3pt}{7.3pt}}}\vspace{-8pt}}{{\textcolor{gray}{\rule[-1.3pt]{7.3pt}{7.3pt}}}\vspace{-8pt}})&\rightarrow&\langle e_{11}\rangle\\[5pt ] \young({{\textcolor{gray}{\rule[-1.3pt]{7.3pt}{7.3pt}}}\vspace{-8pt}},\hfil)&\rightarrow & \langle e_{12},e_{13},e_{21},e_{31}\rangle\\[15pt ] \yng(2)&\rightarrow & \langle e_{22},e_{33}\rangle\\[15pt ] \yng(1,1)&\rightarrow & \langle e_{23},e_{32}\rangle\\[15pt ] \end{array}\ ] ] the projective space associated to this symmetry is therefore a @xmath184 . in the previous section we presented the structure of the space @xmath129 on which the renormalization map acts ; the space @xmath129 contains all multiple - spin interactions that can possibly be generated by the renormalization procedure and , as such , it is the natural space to consider for studying the dynamics of the renormalization map . however , from the physical point of view , we are usually interested in a restricted set of interactions , given , for instance , by pair - interaction between spins and coupling with an external magnetic field . + following the reasoning in the previous section , we expect that this space , that we call _ physical space _ and we denote by @xmath185 , can be given a natural structure of a product @xmath186 . in fact , defining a pair interaction amounts to assign a certain energy @xmath187 to two neighbouring spins that are in the same state ( parallel ) and energy @xmath188 to the configuration for which they are in different states ( antiparallel ) . the two values @xmath187 and @xmath188 are affected by the arbitrary choice of zero of energy , thus , once more , we can define an equivalence relation on boltzmann weights @xmath189 . equivalence classes are given in homogeneous coordinates by @xmath190=[z : w]\in{\mathbb{p}}^1 $ ] . the coupling with an external magnetic field can be treated in the same way : on a potts model we choose a special state to be coupled to the field with energy @xmath191{2.4pt}{2.4pt}}}})}$ ] while all other states will have energy @xmath192 ; these values are again affected by the choice of zero of energy so that we have another projective pair on boltzmann weights , that we denote in the usual homogeneous coordinates by @xmath193{2.4pt}{2.4pt}}}})}:h_{\yng(1)}]$ ] . + for a given hierarchical lattice , one has to define how the physical space @xmath185 is mapped into the dynamical space @xmath129 . we will now present a canonical ( and natural ) way to embed the magnetic field variables in @xmath129 . let @xmath77 be a hierarchical lattice defined by iterating a decoration procedure @xmath63 on an initial hypergraph @xmath75 . we introduce in @xmath194 an auxiliary space of 1-interactions @xmath195 , given by the magnetic field variables @xmath193{2.4pt}{2.4pt}}}})}:h_{\yng(1)}]$ ] ; let @xmath196 . for each decorated edge @xmath47 of the decoration @xmath63 , we attach to each _ core _ vertex one 1-edge corresponding to the magnetic field variables ; the auxiliary 1-edges will be decorated with the identity edge ; let the resulting decoration be @xmath197 . finally , attach to each vertex of @xmath75 an 1-edge corresponding to the magnetic field variables ; let the resulting hypergraph be @xmath198 and let @xmath199 be the hierarchical lattice generated by iteration of the decoration @xmath197 on @xmath198 . it is easy to check that @xmath199 will have _ one _ auxiliary edge attached to each vertex , therefore the magnetic field variables will induce a genuine coupling with an external magnetic field . it is important to note that since the auxiliary edges are not decorated , the external magnetic field variables will act as parameters of the renormalization map instead of being genuine dynamical variables . recall that , in case of a magnetic field , one also has to take into account also a restricted symmetry of the states , as shown in the following example [ campo ] let us consider 2-interactions with a magnetic field . the dynamical space is @xmath200 ; with homogeneous coordinates given by : @xmath201{2.4pt}{2.4pt}}}}{{\textcolor{gray}{\rule[-0.4pt]{2.4pt}{2.4pt}}}})}:z_{\young({{\textcolor{gray}{\rule[-0.4pt]{2.4pt}{2.4pt}}}},\hfil)}:z_{\yng(2)}:z_{\yng(1,1)}\right],\left[h_{\young({{\textcolor{gray}{\rule[-0.4pt]{2.4pt}{2.4pt}}}})}:h_{\yng(1)}\right]\ ] ] the natural embedding is : @xmath202,\left[h_{\young({{\textcolor{gray}{\rule[-0.4pt]{2.4pt}{2.4pt}}}})}:h_{\yng(1)}\right]\mapsto\left[z_{\young({{\textcolor{gray}{\rule[-0.4pt]{2.4pt}{2.4pt}}}}{{\textcolor{gray}{\rule[-0.4pt]{2.4pt}{2.4pt}}}})}=z : z_{\young({{\textcolor{gray}{\rule[-0.4pt]{2.4pt}{2.4pt}}}},\hfil)}=w : z_{\yng(2)}=z : z_{\yng(1,1)}=w\right],\left[h_{\young({{\textcolor{gray}{\rule[-0.4pt]{2.4pt}{2.4pt}}}})}:h_{\yng(1)}\right]\ ] ] the situation for the pair - interaction variables is quite different , as we can not define a canonical embedding of the pair - interaction variables as we did for magnetic field variables . in fact , the embedding depends on the particular hierarchical lattice we want to consider . in the following examples we present a number of cases . the easiest situation is given by a model on a completely symmetric 2-uniform hypergraph ( i.e. a standard graph ) . in this case one maps directly the physical @xmath195 in the dynamical @xmath195 with the identity map : @xmath203\mapsto\left[z_{\yng(2)}=z : z_{\yng(1,1)}=w\right].\ ] ] [ trivial ] [ tri ] consider a model on a completely symmetric 3-uniform hypergraph without external magnetic field and @xmath204 . as stated in the previous section , example [ ezexample ] , the dynamical space is a @xmath179 . suppose we want to put 2-spin interactions along each side of the triangle . this is a way to embed the projective pair @xmath205 $ ] in the dynamical space : @xmath203\mapsto\left[z_{\yng(3)}=z^3:z_{\yng(2,1)}=zw^2:z_{\yng(1,1,1)}=w^3\right].\ ] ] in fact , if all three spins are in the same state we have three parallel pairs i.e. @xmath206 ; if two spins are in the same state and the third one is in a different state , then we have one parallel pair and a two antiparallel pairs i.e. @xmath207 ; finally if all three spins are in different states , then all pairs will be antiparallel i.e. @xmath208 . + notice that with the embedding defined in example [ tri ] , each side will be counted as many times as the number of 3-edges that share that side . sometimes this is undesirable , since such number can vary from side to side . in such cases one can add to the decorated edge some auxiliary 2-edges that will not be decorated ( exactly as we did in the case of magnetic field variables ) and that will be the edges carrying the physical pair - interaction . this formally adds to the dynamical space a new @xmath195 factor ; again , since the auxiliary 2-edges are not decorated , interactions belonging to this @xmath195 will be considered as a parameter of the renormalization map . consider the decorated edge in ; at the @xmath209-th iteration each side of the original tetrahedron will be shared by @xmath210 4-edges . if we want to avoid counting such multiplicities , we need to attach to the decoration four additional auxiliary 2-edges , namely the four sides that are _ inside _ the tetrahedron . these 2-edges will not be decorated , but they will be the ones carrying the pair interaction of the physical space as in example [ trivial ] ; the dynamical variables associated to such edges will therefore act as parameters in the renormalization map . the dynamical space will be given by @xmath211 and the embedding in this case is : @xmath212\mapsto\left[z_{\yng(4)}=1:z_{\yng(3,1)}=1:z_{\yng(2,2)}=1:z_{\yng(2,1,1)}=1:z_{\yng(1,1,1,1)}=1\right],\left[z_{\yng(2)}=z : z_{\yng(1,1)}=w\right].\ ] ] [ ingiuria ] although the embedding of example [ ingiuria ] is constant in the @xmath177 factor of @xmath129 , the renormalization will create 4-edge interactions that will be carried by variables belonging to this factor . + as the examples suggest , the physical space @xmath185 is mapped into the dynamical space ( possibly after extending @xmath129 with new auxiliary interactions ) as a submanifold ; in general , this submanifold is not preserved by the dynamics of the renormalization map . this amounts to the well - known fact that the renormalization of pair interactions introduce , in general , new multiple - spin interactions . in any case , once we obtain all thermodynamical functions in the ( possibly extended ) dynamical space , it is easy to restrict to the physical space to obtain thermodynamical functions in _ relevant _ coordinates . as we showed , the generator of the renormalization map for hierarchical lattices can be represented by a rational map on a complex multiprojective space . we refer the interested reader to the appendices for a minimal technical introduction on the subject of iteration of such maps . the key result we are going to use is that a rational map comes quite naturally associated with a so - called _ green current _ , that can be thought as a differential form with distributional coefficients with support on the unstable set of the map . such current is the limit under pull - back of the standard khler form if the map satisfies two properties called _ dominance _ and _ algebraic stability_. our goal is to show a connection between the green current of the renormalization map and the non - analyticity locus of the free energy of the hierarchical lattice generated by the corresponding decoration . to prove such connection we use results that so far are only available for rational maps acting on projective spaces ( not multi - projective spaces ) ; for this reason , in what follows , we will consider only uniform hypergraphs and decorations , for which the renormalization map is acting on a projective space , although everything ( but theorem [ dinhsib ] ! ) holds true in the more general setting . + let us fix a @xmath30uniform decorated edge @xmath47 and let @xmath63 be the decoration induced by @xmath47 . the renormalization map is @xmath213 ; let @xmath214 denote the algebraic degree of @xmath114 , i.e. the degree of the polynomials we obtain lifting the map to @xmath126 . + fix now a @xmath30uniform hypergraph @xmath215 and consider the zero set of the partition function @xmath216 of the @xmath209 times decorated hypergraph @xmath217 . by , this set is just the @xmath209-th preimage of the zero set of @xmath218 under the renormalization map . such zero set is a codimension 1 algebraic variety that we will denote @xmath219 . if we consider the ( normalized ) current of integration @xmath220 $ ] on the variety @xmath219 we can express its relation to the current of integration @xmath221 $ ] on the zeros @xmath222 associated to @xmath215 in the following way : @xmath223=\frac{1}{d^n}\left(\left({\mathscr{r}}^{\mathscr{d}}\right)^n\right)^*[ly_0].\ ] ] recall that the number of edges of the hypergraph @xmath217 is @xmath224 times the number of edges of @xmath215 ; as @xmath225 , the free energy per edge of @xmath217 is : @xmath226 the last formula shows that the free energy @xmath227 is just the pluripotential of the current supported on the zero locus of the polynomial @xmath228 . in the limit @xmath229 the support of this current coincides with the lee - yang @xcite and fisher zero locus of the model on the hierarchical lattice @xmath122 . results for this kind of limits have been found by brolin @xcite , lyubich @xcite for in the 80s , by favre - jonnson @xcite for holomorphic maps of in 2003 . very recently dinh and sibony proved the following let @xmath230 a holomorphic map of degree @xmath214 on the projective space of complex dimension @xmath161 , @xmath231 its green current . there exists a completely invariant proper analytic subset @xmath3 such that if @xmath157 is a hypersurface of degree @xmath232 in @xmath233 which does not contain any component of @xmath3 , then @xmath234 \to st\ ] ] where @xmath235 $ ] is the current of integration on @xmath157 . [ dinhsib ] the maximal completely invariant proper subset @xmath236 has been found ( @xcite ) to be a finite union of linear subspaces and bounds have been found for the maximal number of components of codimension 1 ( @xcite ) that can not be more than @xmath237 ( sharp ) and for codimension 2 ( @xcite ) that is less than @xmath238 ( possibly not sharp ) . + we recall ( see [ appa ] for details ) that , while rational maps on @xmath239 are automatically holomorphic , this is not true in general for rational maps in higher dimensional spaces ; in fact , holomorphic maps are such that the so - called _ indeterminacy set _ is empty . from the physical point of view , the indeterminacy set contains all boltzmann weights that can not be renormalized , i.e. such that applying the renormalization map to them gives all boltzmann weights equal to 0 . although renormalization maps are not in general holomorphic , their restrictions on symmetrical interaction submanifold ( see ) usually are . moreover , the requirement of being holomorphic is a technical assumption that can possibly be removed using a more careful definition of the green current . the connection is nevertheless interesting and it is worthwhile to try to understand how properties of the decoration are related to regularity properties of the corresponding renormalization map . as summarized in the appendix , we need the map to enjoy two main properties in order for the green current to be at least defined : _ dominance _ and _ algebraic stability_. + the dominance property states that the jacobian determinant of the map should not be identically zero . it is therefore very easy to check if a particular renormalization map enjoy this property ; nevertheless it is interesting to point out that , in general , decoration that present some degeneracies will correspond to non - dominant maps . we now give two examples of such degenerate decorations : as a first example consider a decoration such that the renormalization map is _ invariant _ under permutations of @xmath139 ; a 2-decoration suffices to illustrate the fact : @xmath240 this implies that the image of the map is an algebraic subvariety , that in turn implies that the map is not dominant . this degeneracy is in some sense _ removable _ as it can be ruled out by naturally restricting the map to the invariant variety which corresponds to @xmath159invariant interactions . as a second example consider the following uniform decorated edge : @xmath241 in this case the 3-spin interactions can be expressed in terms of 2-spin interactions . clearly the map will not be surjective on the space of 3-spin interactions as it will provide just interactions that can be described by @xmath242edges , which in turn form a subvariety of codimension 1 . in such cases one should again restrict to the appropriate space of interactions to obtain a dominant map . the other regularity condition we have to check is _ algebraic stability _ ; this property is much harder to verify than the dominance condition . in fact , algebraic stability is related to the growth of the degrees of iterates of the renormalization map . it may happen that iterating the map we obtain factors that are common to all coordinates and which therefore have to be simplified ; this operation lowers the degree of the map . in the maps studied so far , common factors do appear , but only in the definition of the map ( i.e. the first iteration ) ; we believe that once we simplify common factors which are possibly present at the first iteration , the renormalization map should be algebraically stable . also , from a mathematical point of view , it would be quite important to prove algebraic stability for such maps , or at least to find conditions in terms of the decorations in order to ensure that this property holds . in fact , a characterization of algebraically stable maps is still lacking ; for instance , it is not yet known how to build nontrivial maps that are a priori algebraically stable . to give an example of the appearance of common factors we consider the model shown in figure [ cayley ] . the model can be given by a non - uniform decoration ; in this decoration we have two different kinds of one - dimensional edges ( dotted and solid in the picture ) . the resulting graph is also known as the cayley graph of the free group on 2 generators . if we are in the case without an external magnetic field , the renormalization map associated to the decoration has common factors . removing them corresponds to pruning all the branches of the tree and leaving a one - dimensional chain ; this equivalence was observed long ago in @xcite . this is the physical meaning to the idea of factoring out common factors in such a model although one probably can not always give such a physical interpretation to the mathematical operation . the authors are indebted to c. favre , s. caracciolo and to anonymous referees for their most useful comments and suggestions . in this appendix we define a natural sum operation on decorated edges . let @xmath42 and let @xmath243 and @xmath244 be two decorated @xmath31-edges : @xmath245 we define their sum @xmath246 to be the decorated @xmath30edge obtained by taking the disjoint union of the respective vertex and edge sets and then identifying surface vertices . the partition of the resulting edge set will be given by the union of the partitions of the summands : more formally let @xmath247 we define the collapsing map : @xmath248@xmath249 then the edge sets are given by : @xmath250 and the sum decorated edge will be : @xmath251 we can define the _ decorated @xmath30edge as the decorated @xmath31-edge with @xmath6 surface vertices , no core vertices and no edges ; we consider the zero decorated edge to be uniform . @xmath252 clearly the zero decorated edge is the null element of the sum operation . it is straightforward to check that conditional partition function @xmath253 of the zero decorated edge @xmath254 is constant . also , it is easy to check that given @xmath243 and @xmath244 two uniform decorated @xmath30edges we have that the renormalization map induced by the sum @xmath255 is given by the following expression @xmath256 where on the right hand side the product is defined coordinate - wise . in this appendix we give some basic notions about pluripotential theory which are useful in the study of the dynamics of the rg action . we refer the interested reader to the appropriate sections of @xcite for a more in - depth introduction . + let @xmath257 be a smooth manifold and @xmath258 the vector space of smooth real - valued functions with compact support on @xmath257 , endowed with the usual compact - open topology . the space of _ distributions _ @xmath259 is the vector space of continuous linear functional on @xmath258 endowed with the usual weak topology . + let @xmath260 be the laplace operator in @xmath261 ( as the 2-dimensional real euclidean space ) ; given a measure @xmath262 we define its _ potential _ as the distributional solution of the equation @xmath263 . functions that are local potentials of a positive measure @xmath262 are called _ subharmonic _ and are characterized as follows : let @xmath264 be an open domain of @xmath261 . an upper semi - continuous function @xmath265 is _ subharmonic _ if it is not identically equal to @xmath266 and it enjoys the subaverage property i.e. for all @xmath267 , for all @xmath268 such that the closed disk of center @xmath269 and radius @xmath6 is contained in @xmath264 , we have @xmath270 for example if @xmath39 is an holomorphic function then @xmath271 is subharmonic and @xmath272 is supported on the zeroes of @xmath39 . + in the multidimensional setting we will need to use _ currents _ and _ plurisubharmonic functions _ rather than distributions and subharmonic functions . we will now introduce the appropriate definitions . + let @xmath273 be the vector space of smooth differential @xmath274-forms with compact support endowed with the compact - open topology . a current @xmath137 of dimension @xmath274 is a continuous linear functional on @xmath273 ; the space of @xmath274-currents will be denoted as @xmath275 and will be given the weak topology . for example , since one can associate the dirac delta to a point , one can associate a @xmath274-current to any @xmath274-dimensional submanifold @xmath276 of @xmath257 by integrating @xmath274-forms over @xmath276 . operations on forms as exterior product with other forms and the exterior differential operator can act by duality on the space of currents as well:@xmath277 as a dual object to forms , a current @xmath137 can naturally be pushed forward by a map @xmath39 , provided that the restriction of @xmath39 to the support of @xmath137 is proper ( i.e. the preimage of compact sets is compact ) . moreover , if @xmath39 is a proper submersion one can define a push - forward operation for forms and therefore one can define a pull - back for currents . if the manifold has a complex structure we should distinguish between the holomorphic and antiholomorphic part of a form . a complex differential form of bidegree @xmath278 can be written as : @xmath279 a @xmath280-form is said to be _ positive _ if for all complex submanifold @xmath281 of dimension @xmath274 , its restriction to @xmath281 is a nonnegative volume form ; @xmath282currents are defined by duality and a @xmath283current is said to be positive if it evaluates as a positive number on any positive @xmath283form . + along with the exterior holomorphic @xmath284 and antiholomorphic @xmath285 differentiation we can define two real operators @xmath286 and @xmath287 . the second order operator @xmath288 is going to replace the laplacian operator in the multidimensional setting . we are now left to introduce the analogous of subharmonic functions . let @xmath264 be an open subset of @xmath289 . an upper semi - continuous function @xmath290 is _ plurisubharmonic _ ( in short _ psh _ ) in @xmath264 if it is not identically equal to @xmath266 and it enjoys the subaverage property when restricted to any 1-dimensional disk i.e. for all @xmath267 and for all @xmath291 such that the one - dimensional complex disk @xmath292 ( where @xmath293 is the closed unit disk in @xmath261 ) is contained in @xmath264 one has @xmath294 the space of psh functions enjoys an important compactness property : let @xmath295 be a sequence of plurisubharmonic functions on a domain @xmath296 . assume that for all compacts @xmath297 the sequence is dominated by a psh function . then either @xmath298 on all compact subsets of @xmath264 or there exists a subsequence @xmath299 which converges in @xmath300 to a psh function . a function @xmath301 is a.e . equal to a psh function the @xmath302current @xmath303 is positive ; conversely if @xmath137 is a positive closed @xmath302current , there exists a psh function @xmath304 such that @xmath304 is a local potential of @xmath137 . consider the complex vector space @xmath305 modulo the action of the multiplicative group @xmath306 by scalar multiplications . the resulting space is a complex manifold of dimension @xmath209 called projective space @xmath307 . the natural coordinates on the projective space are the so called homogeneous coordinates : @xmath308{\doteqdot}\pi(z_0,z_1,\cdots , z_n)\ ] ] where @xmath55 is the projection map that defines the quotient . @xmath307 comes naturally endowed with a standard khler form @xmath309 given by the relation @xmath310 . + a rational map of degree @xmath214 over @xmath307 is a map of the form : @xmath311\mapsto [ p_0:p_1:\cdots : p_n]\ ] ] where @xmath312s are homogeneous polynomials of degree @xmath214 with no nonzero common factors . the map @xmath39 can be lifted to a polynomial map @xmath313 on the complex space up to nonzero multiplicative factors . a rational map on @xmath307 is said _ dominant _ if given any lift @xmath313 , its jacobian determinant does not vanish identically . the set of dominant maps of degree @xmath214 will be denoted by @xmath314 . one then defines the _ indeterminacy set _ @xmath315 . + roughly speaking @xmath87 is a _ bad _ set for the dynamics and _ good _ maps are such that @xmath87 is small . the space @xmath316 of maps such that @xmath317 is defined as the space of holomorphic maps . in most applications a weaker condition on @xmath318 suffices : suppose there is no integer @xmath209 and no codimension 1 hypersurface @xmath2 such that @xmath319 ; then @xmath39 is said to be _ algebraically stable _ as the latter condition is equivalent to require that the degree of @xmath320 is @xmath224 . + a rational map @xmath39 acts on the space of positive closed @xmath302currents by pull - back i.e. given a potential @xmath304 of a current @xmath137 ( i.e. @xmath321 ) , @xmath322 is defined by the relation @xmath323 . such action is continuous provided that @xmath39 is dominant . an important result is the following let @xmath324 @xmath325 algebraically stable . then the sequence @xmath326 converges to a closed positive @xmath302current @xmath231 such that @xmath327 . @xmath231 is called the _ green current _ of @xmath39 . a potential of @xmath231 is called _ green function_. the support of the green current can be partially understood in a purely topological setting ; in fact , let us define the _ stable _ ( or fatou ) set of the map as follows : @xmath328 @xmath329 is called _ julia set _ of @xmath39 and is the _ unstable _ set for the dynamics ; this set always contains the support of the green current ( see @xcite ) , that therefore assume a definite topological meaning . + a multiprojective space is just a product of @xmath274 projective spaces ; rational maps on such spaces are those that are lifted to separately homogeneous polynomials . the notion of degree becomes that of multi - degree , that is a square integer matrix of dimension @xmath274 . studying the dynamics of rational maps on such spaces is more complicated and very few results have been proved so far ( @xcite ) , but among these there is the existence of the green current for algebraically stable dominant maps . berker a n and ostlund s 1979 renormalisation - group calculations of finite systems : order parameter and specific heat for epitaxial ordering _ _ * 12 * , 4961 - 75 migdal a a 1975 recurrence equations in gauge field theory _ jetp _ * 69 * , 810 - 22 migdal a a 1975 phase transitions in gauge and spin - lattice systems _ jetp _ * 69 * , 1457 - 67 kadanoff l p 1976 notes on migdal s recursion formulae _ _ * 100 * , 359 - 94 griffiths r b and kaufman m 1981 exactly soluble ising models on hierarchical lattices _ _ b _ _ * 24 * , 496 - 98 griffiths r b and kaufman m 1982 spin systems on hierarchical lattices . introduction and thermodynamic limit _ _ b _ _ * 26 * , 5022 - 32 griffiths r b and kaufman m 1984 spin systems on hierarchical lattices . ii . some examples of soluble models _ _ b _ _ * 30 * , 244 - 49 derrida b , de seze l and itzykson c 1983 fractal structure of zeroes in hierarchical models _ journal of statistical physics _ * 33 * , 559 - 69 fisher m e 1998 renormalization group theory : its basis and formulation in statistical physics _ _ * 70 * , no . 2 , 653 - 81 wilson k g 1983 the renormalization group and critical phenomena _ _ * 55 * , no . 3 , 583 - 600 gefen y , aharony a , shapir y and mandelbrot b b 1984 phase transitions on fractals : ii . sierpinski gaskets _ _ * 17 * , 435 - 44 burioni r , cassi d and donetti l 1999 lee - yang zeros and the ising model on the sierpinski gasket _ _ * 32 * , 5017 - 27 bleher p m and lyubich m yu 1991 julia sets and complex singularities in hierarchical ising models _ commun . phys . _ * 141 * , 453 - 74 dinh t - c and sibony n 2008 equidistribution towards the green current for holomorphic maps _ arxiv _ * * , math/0609686v3 de simoi j 2008 potts models on hierarchical lattices and renormalization group dynamics ii : examples and numerical results _ arxiv _ * * , 0711.4610v2 berge c 1974 hypergraph seminar _ lecture notes in mathematics _ * 411 * , yang c n and lee t d 1952 statistical theory of equations of state and phase transitions : theory of condensation _ _ * 87 * , 404 - 9 yang c n and lee t d 1952 statistical theory of equations of state and phase transitions : lattice gas and ising model _ _ * 87 * , 410 - 9 brolin h 1965 invariant sets under iteration of rational functions _ ark . mat . _ * 6 * , 103 - 14 lyubich m yu 1983 entropy properties of rational endomorphism of the riemann sphere _ erg . th . & dyn . sys . _ * 3 * , 351 - 85 favre c and jonsson m 2003 brolin s theorem for curves in two complex dimensions _ ann . fourier , grenoble _ * 53 * , 5 , 1461 - 501 briend j - y , cantat s and shishikura m 2004 linearity of the exceptional set for maps of @xmath233 _ math . ann . _ * 330 * , 39 - 43 fornaess j and sibony n 1994 complex dynamics in higher dimension i _ asterisque _ * 222 * , 201 - 31 amerik e and campana f 2005 exceptional points of an endomorphism of the projective plane _ math z. _ * 249 * , 741 - 54 eggarter t p 1974 cayley trees , the ising problem , and the thermodynamic limit _ _ b _ _ * 9 * , 2989 - 92 briend j - y and duval j 2001 deux caractrisations de la mesure dquilibre dun endomorphisme de @xmath330 _ publ . hautes tudes sci . _ * 93 * , 145 - 59 favre c and guedj v 2001 dynamique des applications rationelles des espaces multiprojectifs _ indiana univ . j. _ * 50 * , 2 , 881 - 934 smillie j and buzzard g t 1997 complex dynamics in several variables _ flavours of geometry _ * 31 * , 117 - 50 sibony n 1999 dynamique des applications rationelles de @xmath233 _ panoramas & synthses _ * 8 * , 97 - 185

clique - width is a graph invariant that has been widely studied in combinatorics and computer science . however , computing the clique - width of a graph is an intricate problem , the exact clique - width is not known even for very small graphs . we present a new method for computing the clique - width of graphs based on an encoding to propositional satisfiability ( sat ) which is then evaluated by a sat solver . our encoding is based on a reformulation of clique - width in terms of partitions that utilizes an efficient encoding of cardinality constraints . our sat - based method is the first to discover the exact clique - width of various small graphs , including famous graphs from the literature as well as random graphs of various density . with our method we determined the smallest graphs that require a small pre - described clique - width .

1304.5498

clique - width is a fundamental graph invariant that has been widely studied in combinatorics and computer science . clique - width measures in a certain sense the `` complexity '' of a graph . it is defined via a graph construction process involving four operations where only a limited number of vertex labels are available ; vertices that share the same label at a certain point of the construction process must be treated uniformly in subsequent steps . this graph composition mechanism was first considered by courcelle , engelfriet , and rozenberg @xcite and has since then been an important topic in combinatorics and computer science . graphs of small clique - width have advantageous algorithmic properties . algorithmic meta - theorems show that large classes of @xmath0-hard optimization problems and # p - hard counting problems can be solved in _ linear time _ on classes of graphs of bounded clique - width @xcite . similar results hold for the graph invariant _ treewidth _ , however , clique - width is more general in the sense that graphs of small treewidth also have small clique - width , but there are graphs of small clique - width but arbitrarily high treewidth @xcite . unlike treewidth , dense graphs ( e.g. , cliques ) can also have small clique - width . all these algorithms for graphs of small clique - width require that a certificate for the graph having small clique - width is provided . however , it seems that computing the certificate , or just deciding whether the clique - width of a graph is bounded by a given number , is a very intricate combinatorial problem . more precisely , given a graph @xmath1 and an integer @xmath2 , deciding whether the clique - width of @xmath1 is at most @xmath2 is @xmath0complete @xcite . even worse , the clique - width of a graph with @xmath3 vertices of degree greater than 2 can not be approximated by a polynomial - time algorithm with an absolute error guarantee of @xmath4 unless @xmath5 , where @xmath6 @xcite . in fact , it is even unknown whether graphs of clique - width at most @xmath7 can be recognized in polynomial time @xcite . there are approximation algorithms with an exponential error that , for fixed @xmath2 , compute @xmath8expressions for graphs of clique - width at most @xmath2 in polynomial time ( where @xmath9 by @xcite , and @xmath10 by @xcite ) . because of this intricacy of this graph invariant , the exact clique - width is not known even for very small graphs . [ [ clique - width - via - sat . ] ] clique - width via sat . + + + + + + + + + + + + + + + + + + + + + we present a new method for determining the clique - width based on a sophisticated sat encoding which entails the following ideas : 1 . _ reformulation_. the conventional construction method for determining the clique - width of a graph consists of many steps . in the worst case , the number of steps is quadratic in the number of vertices . translating this construction method into sat would result in large instances , even for small graphs . we reformulated the problem in such a way that the number of steps is less than the number of vertices . the alternative construction method allows us to compute the clique - width of much larger graphs . representative encoding_. applying the frequently - used direct encoding @xcite on the reformulation results in instances that have no arc consistency @xcite , i.e. , unit propagation may find conflicts much later than required . we developed the representative encoding that is compact and realizes arc consistency . [ [ experimental - results . ] ] experimental results . + + + + + + + + + + + + + + + + + + + + + the implementation of our method allows us for the first time to determine the exact clique - width of various graphs , including famous graphs known from the literature , as well as random graphs of various density . 1 . _ clique - width of small random graphs_. we determined experimentally how the clique - width of random graphs depends on the density . the clique - width is small for dense and sparse graphs and reaches its maximum for edge - probability @xmath11 . the larger @xmath3 , the steeper the increase towards @xmath11 . these results complement the asymptotic results of lee et al . _ smallest graphs of certain clique - width_. in general it is not known how many vertices are required to form a graph of a certain clique - width . we provide these numbers for clique - width @xmath12 . in fact , we could compute the total number of connected graphs ( modulo isomorphism ) with a certain clique - width with up to 10 vertices . for instance , there are only 7 connected graphs with 8 vertices and clique - width 5 ( modulo isomorphism ) , and no graphs with 9 vertices and clique - width 6 . there are 68 graphs with 10 vertices and clique - width 6 . the smallest one has 18 edges . clique - width of famous named graphs_. over the last 50 years , researchers in graph theory have considered a large number of special graphs . these special graphs have been used as counterexamples for conjectures or for showing the tightness of combinatorial results . we considered several prominent graphs from the literature and computed their exact clique - width . these results may be of interest for people working in combinatorics and graph theory . [ [ related - work . ] ] related work . + + + + + + + + + + + + + we are not aware of any implemented algorithms that compute the clique - width exactly or heuristically . however , algorithms have been implemented that compute upper bounds on other width - based graph invariants , including _ treewidth _ @xcite , _ branchwidth _ @xcite , _ boolean - width _ @xcite , and _ rank - width _ @xcite . samer and veith @xcite proposed a sat encoding for the exact computation of treewidth . boolean - width and rank - width can be used to approximate clique - width , however , the error can be exponential in the clique - width ; in contrast , treewidth and branchwidth can be arbitrarily far from the clique - width , hence the approximation error is unbounded @xcite . our sat encoding is based on a new characterization of clique - width that is based on partitions instead of labels . a similar partition - based characterization of clique - width , has been proposed by heggernes et al . there are two main differences to our reformulation . firstly , our characterization of clique - width uses three individual properties that can be easily expressed by clauses . secondly , our characterization admits the `` parallel '' processing of several parts of the graph that are later joined together . we consider propositional formulas in conjunctive normal form ( _ cnf formulas _ , for short ) , which are conjunctions of clauses , where a clause is a disjunction of literals , and a literal is a propositional variable or a negated propositional variables . a cnf formula is _ satisfiable _ if its variables can be assigned true or false , such that each clause contains either a variable set to true or a negated variable set to false . the satisfiability problem ( sat ) asks whether a given formula is satisfiable . all graphs considered are finite , undirected , and without self - loops . we denote a graph @xmath1 by an ordered pair @xmath13 of its set of vertices and its set of edges , respectively . an edge between vertices @xmath14 and @xmath15 is denoted @xmath16 or equivalently @xmath17 . for basic terminology on graphs we refer to a standard text book @xcite . let @xmath2 be a positive integer . a @xmath2-_graph _ is a graph whose vertices are labeled by integers from @xmath18 . we consider an arbitrary graph as a @xmath2graph with all vertices labeled by @xmath19 . we call the @xmath2-graph consisting of exactly one vertex @xmath15 ( say , labeled by @xmath20 ) an _ initial _ @xmath2-graph and denote it by @xmath21 . the _ clique - width _ of a graph @xmath1 is the smallest integer @xmath2 such that @xmath1 can be constructed from initial @xmath2-graphs by means of repeated application of the following three operations disjoint union ( denoted by @xmath22 ) ; 2 . relabeling : changing all labels @xmath20 to @xmath23 ( denoted by @xmath24 ) ; 3 . edge insertion : connecting all vertices labeled by @xmath20 with all vertices labeled by @xmath25 ( denoted by @xmath26 or @xmath27 ) ; already existing edges are not doubled . a construction of a @xmath2graph using the above operations can be represented by an algebraic term composed of @xmath22 , @xmath28 , and @xmath26 ( @xmath29 , and @xmath30 ) . such a term is called a _ @xmath2expression _ defining @xmath1 . thus , the clique - width of a graph @xmath1 is the smallest integer @xmath2 such that @xmath1 can be defined by a @xmath2expression . [ ex : p4 ] the graph @xmath31 is defined by the @xmath32expression @xmath33 hence @xmath34 . in fact , one can show that @xmath35 it has no @xmath36expression , and thus @xmath37 @xcite . as partitions play an important role in our reformulation of clique - width , we recall some basic terminology . a _ partition _ of a set @xmath38 is a set @xmath39 of nonempty subsets of @xmath38 such that any two sets in @xmath39 are disjoint and @xmath38 is the union of all sets in @xmath39 . the elements of @xmath39 are called _ equivalence classes_. let @xmath40 be partitions of @xmath38 . then @xmath41 is a _ refinement _ of @xmath39 if for any two elements @xmath42 that are in the same equivalence class of @xmath41 are also in the same equivalence class of @xmath39 ( this entails the case @xmath43 ) . initially , we developed a sat encoding of clique - width based on @xmath2-expressions . even after several optimization steps , this encoding was only able to determine the clique - width of graphs consisting of at most 8 vertices . we therefore developed a new encoding based on a reformulation of clique - width which does not use @xmath2expressions . in this section we explain this reformulation , in the next section we will discuss how it can be encoded into sat efficiently . consider a finite set @xmath44 , the _ universe_. a _ template _ @xmath45 consists of two partitions @xmath46 and @xmath47 of @xmath44 . we call the equivalence classes in @xmath46 the _ components _ of @xmath45 and the equivalence classes in @xmath47 the _ groups _ of @xmath45 . for some intuition about these concepts , imagine that components represent induced subgraphs and that groups represent sets of vertices in some component with the same label in a @xmath2-expression . a _ derivation _ of length @xmath48 is a finite sequence @xmath49 satisfying the following conditions . 1 . @xmath50 and @xmath51 . 2 . @xmath52 is a refinement of @xmath53 , @xmath54 . 3 . @xmath55 is a refinement of @xmath53 , @xmath56 . 4 . @xmath57 is a refinement of @xmath52 , @xmath58 . we would like to note that d1 and d2 together imply that @xmath59 . thus , in the first template @xmath60 all equivalence classes ( groups and components ) are singletons , and when we progress through the derivation , some of these sets are merged , until all components are merged into a single component in the last template @xmath61 . the _ width _ of a component @xmath62 is the number of groups @xmath63 such that . the width of a template is the maximum width over its components , and the width of a derivation is the maximum width over its templates . @xmath2derivation _ is a derivation of width at most @xmath2 . a derivation @xmath49 is a derivation _ of _ a graph @xmath64 if @xmath44 is the universe of the derivation and the following three conditions hold for all @xmath65 . _ edge property _ : for any two vertices @xmath66 such that @xmath67 , if @xmath68 are in the same group in @xmath69 , then @xmath68 are in the same component in @xmath70 . _ neighborhood property _ : for any three vertices @xmath71 such that @xmath67 and @xmath72 , if @xmath73 are in the same group in @xmath69 , then @xmath68 are in the same component in @xmath70 . _ path property _ : for any four vertices @xmath74 , such that @xmath75 and @xmath76 , if @xmath77 are in the same group in @xmath69 and @xmath73 are in the same group in @xmath69 , then @xmath68 are in the same component in @xmath70 . the neighborhood property and the path property could be merged into a single property if we do not insist that all mentioned vertices are distinct . however , two separate properties provide a more compact sat encoding . the following example illustrates that a derivation can define more than one graph , in contrast to a @xmath2expression , which defines exactly one graph . [ ex : derivation ] consider the derivation @xmath78 with universe @xmath79 and @xmath80{lclclcl } { \text{cmp}}(t_0)&=&\{\{a\},\{b\},\{c\},\{d\}\ } , & \quad & { \text{grp}}(t_0)&= & \{\{a\},\{b\},\{c\},\{d\}\ } , \\ { \text{cmp}}(t_1)&=&\{\{a , b\},\{c\},\{d\}\ } , & \quad & { \text{grp}}(t_1)&=&\{\{a\},\{b\},\{c\},\{d\}\},\\ { \text{cmp}}(t_2)&=&\{\{a , b , c\},\{d\}\ } , & \quad & { \text{grp}}(t_2)&= & \{\{a\},\{b\},\{c\},\{d\}\ } , \\ { \text{cmp}}(t_3)&=&\{\{a , b , c , d\}\ } , & \quad & { \text{grp}}(t_3)&= & \{\{a , b\},\{c\},\{d\}\}. \end{array}$ ] the width of @xmath81 is @xmath32 . consider the graph @xmath82 . to see that @xmath81 is a @xmath32derivation of @xmath1 , we need to check the edge , neighborhood , and path properties . we observe that @xmath83 are the only two vertices such that @xmath84 and both vertices appear in the same group of some @xmath69 ( here , we have @xmath85 ) . to check the edge property , we only need to verify that @xmath83 are in the same component of @xmath86 , which is true . for the neighborhood property , the only relevant choice of three vertices is @xmath87 ( @xmath88 , @xmath89 , and @xmath83 in a group of @xmath90 ) . the neighborhood property requires that @xmath91 are in the same component in @xmath86 , which is the case . the path property is satisfied since there is no template in which two pairs of vertices belong to the same group , respectively . = [ circle , draw , inner sep=1.5pt , fill = gray ] ( 0,1 ) node[circle ] ( a ) node[above ] @xmath92 ( 1,1 ) node[circle ] ( b ) node[above ] @xmath93 ( 0,0 ) node[circle ] ( c ) node[below ] @xmath94 ( 1,0 ) node[circle ] ( d ) node[below ] @xmath95 ( b)(a)(c)(d ) ; ( 0,1 ) node[circle ] ( a ) node[above ] @xmath92 ( 1,1 ) node[circle ] ( b ) node[above ] @xmath93 ( 0,0 ) node[circle ] ( c ) node[below ] @xmath94 ( 1,0 ) node[circle ] ( d ) node[below ] @xmath95 ( a)(b)(c)(a ) ( c)(d ) ; ( 0,1 ) node[circle ] ( a ) node[above ] @xmath92 ( 1,1 ) node[circle ] ( b ) node[above ] @xmath93 ( 0,0 ) node[circle ] ( c ) node[below ] @xmath94 ( 1,0 ) node[circle ] ( d ) node[below ] @xmath95 ( a)(c)(d ) ( b)(c ) ; ( 0,1 ) node[circle ] ( a ) node[above ] @xmath92 ( 1,1 ) node[circle ] ( b ) node[above ] @xmath93 ( 0,0 ) node[circle ] ( c ) node[below ] @xmath94 ( 1,0 ) node[circle ] ( d ) node[below ] @xmath95 ( a)(c)(b)(d)(a ) ; ( 0,1 ) node[circle ] ( a ) node[above ] @xmath92 ( 1,1 ) node[circle ] ( b ) node[above ] @xmath93 ( 0,0 ) node[circle ] ( c ) node[below ] @xmath94 ( 1,0 ) node[circle ] ( d ) node[below ] @xmath95 ( a)(b)(d)(a)(c)(b)(a ) ; ( 0,1 ) node[circle ] ( a ) node[above ] @xmath92 ( 1,1 ) node[circle ] ( b ) node[above ] @xmath93 ( 0,0 ) node[circle ] ( c ) node[below ] @xmath94 ( 1,0 ) node[circle ] ( d ) node[below ] @xmath95 ( a)(b)(d)(a)(c)(b)(a ) ( c)(d ) ; similarly we can verify that @xmath81 is a derivation of the graph @xmath96 , @xmath97 , @xmath98 . in fact , for all connected graphs with four vertices , there exists an isomorphic graph that is defined by @xmath81 ( see figure [ fig : four ] ) . however , @xmath81 is not a derivation of the graph @xmath99 since the neighborhood property is violated : @xmath100 and @xmath101 , @xmath83 belong to the same group in @xmath90 , while @xmath102 do not belong to the same component in @xmath86 . we call a derivation @xmath103 to be _ strict _ if @xmath104 holds for all @xmath65 . [ lem : make - strict ] if @xmath1 has a @xmath2derivation , it has a strict @xmath2derivation . let @xmath49 be a @xmath2derivation of @xmath1 . assume there is some @xmath105 such that @xmath106 . if also @xmath107 , then @xmath108 , and we can safely remove @xmath70 and still have a @xmath2derivation of @xmath1 . hence assume @xmath109 . this implies that @xmath110 . if @xmath111 , then we can safely remove @xmath61 from the derivation and @xmath112 is clearly a @xmath2 derivation of @xmath1 . hence it remains to consider the case @xmath113 . we show that by dropping @xmath69 we get a sequence @xmath114 that is a @xmath2 derivation of @xmath1 . the new sequence @xmath115 is clearly a @xmath2derivation . it remains to verify that @xmath115 is a derivation of @xmath1 . the template @xmath116 is the only one where these properties might have been violated by the removal of @xmath69 . however , since all three properties impose a restriction on the set of components of the template preceding @xmath116 , and since @xmath106 , the properties are not effected by the deletion of @xmath69 . hence @xmath115 is indeed a @xmath2derivation of @xmath1 . by repeated application of the above shortening we can turn any @xmath2derivation into a strict @xmath2derivation . [ lem : strict - short ] every strict @xmath2derivation of a graph with @xmath3 vertices has length at most @xmath117 . let @xmath103 be a strict @xmath2derivation of a graph with @xmath3 vertices . since @xmath118 and @xmath119 , it follows that @xmath120 . in the proofs of the next two lemmas we need the following concept of a _ @xmath2expression tree _ , which is the parse tree of a @xmath2expression equipped with some additional information . let @xmath121 be a @xmath2expression for a graph @xmath64 . let @xmath122 be the parse tree of @xmath121 with root @xmath123 . that is , @xmath122 contains a node for each occurrence of an operation @xmath22 , @xmath124 , and @xmath26 in @xmath121 and for each initial @xmath2graph @xmath21 in @xmath121 ; the initial @xmath2graphs are the leaves of @xmath122 , and the other nodes have as children the nodes which represent the two subexpressions of the respective operation . consider a node @xmath125 of @xmath122 and let @xmath126 be the subexpression of @xmath121 whose parse tree is the subtree of @xmath122 rooted at @xmath125 . then @xmath125 is labeled with the @xmath2graph @xmath127 constructed by the @xmath2expression @xmath126 . thus the leaves of @xmath122 are labeled with initial @xmath2graphs and the root @xmath123 is labeled with a labeled version of @xmath1 . we call a non - leaf node of @xmath122 an @xmath22node , @xmath128node , or @xmath129node , according to the operation it represents . one @xmath22node of the parse tree can represent several directly subsequent @xmath22operations ( e.g. , the operation @xmath130 can be represented by a single node with three children ) . for technical reasons we will also allow @xmath22nodes with a single child . each @xmath2expression gives rise to a @xmath2expression tree where each @xmath22node has no @xmath22nodes as children , let us call such a @xmath2expression tree to be _ succinct_. evidently , @xmath2expressions and their ( succinct ) @xmath2expression trees can be effectively transformed into each other . [ lem : expression->derivation ] from a @xmath2expression of a graph @xmath1 we can obtain a @xmath2 derivation of @xmath1 in polynomial time . let @xmath121 be a @xmath2expression of @xmath64 and let @xmath122 be the corresponding succinct @xmath2expression tree with root @xmath123 . for a node @xmath131 let @xmath132 denote the number of @xmath22-nodes that appear on the path from @xmath123 to @xmath125 ; thus @xmath133 . we write @xmath134 and @xmath135 for the set of @xmath22nodes and the set of leaves of @xmath122 , respectively . we let @xmath136 . for @xmath137 we define @xmath138 and @xmath139 . we observe that for each @xmath140 and @xmath65 there is exactly one @xmath141 such that @xmath142 we define a derivation @xmath49 as follows . for @xmath143 we put @xmath144 and @xmath145 where @xmath146 denotes the partition of @xmath147 into sets of vertices that have the same label . by construction , @xmath81 is a derivation with universe @xmath44 . furthermore , since @xmath121 is a @xmath2 expression , @xmath148 for all nodes @xmath125 of @xmath122 . hence @xmath81 is a @xmath2derivation . it remains to show that @xmath81 is a @xmath2derivation of @xmath1 . let @xmath65 . to show that the _ edge property _ holds , consider two vertices @xmath66 such that @xmath67 and @xmath68 are in the same group in @xmath69 . assume to the contrary that @xmath68 belong to different components @xmath149 in @xmath70 . since @xmath68 are in the same group in @xmath69 , they are also in the same component of @xmath69 . hence there is an @xmath22node @xmath150 with @xmath151 . let @xmath152 be the children of @xmath125 with @xmath153 and @xmath154 . hence @xmath155 . however , since @xmath68 are in the same group in @xmath69 , this means that @xmath68 have the same label in @xmath127 . thus the edge @xmath16 can not be introduced by an @xmath128 operation , and so @xmath156 , a contradiction . hence the edge property holds . to show that the _ neighborhood property _ holds , consider three vertices @xmath71 such that @xmath67 , @xmath72 , and @xmath73 are in the same group of @xmath69 . assume to the contrary that @xmath68 are in different components of @xmath70 , say in components @xmath157 and @xmath158 , respectively . since @xmath73 are in the same group of @xmath69 , they are also in the same component @xmath159 of @xmath69 . let @xmath150 be the @xmath22node such that @xmath160 , and let @xmath152 be the children of @xmath125 with @xmath153 and @xmath154 . clearly @xmath161 , hence there must be an @xmath128node @xmath162 somewhere on the path between @xmath125 and @xmath123 where the edge @xmath16 is introduced . however , since @xmath15 and @xmath163 share the same label in @xmath127 , they share the same label in @xmath164 . consequently , the @xmath128operation that introduces the edge @xmath16 also introduces the edge @xmath165 . however , this contradicts the assumption that @xmath72 . hence the neighborhood property holds as well . to show that the _ path property _ holds , we proceed similarly . consider four vertices @xmath74 , such that @xmath166 and @xmath76 . assume that @xmath77 are in the same group in @xmath69 and @xmath73 are in the same group in @xmath69 . assume to the contrary that @xmath68 are in different components of @xmath70 , say in components @xmath157 and @xmath158 , respectively . above we have shown that the neighborhood property holds . hence we conclude that @xmath167 belong to the same component of @xmath70 , and @xmath168 belong to the same component of @xmath70 . since @xmath77 are in the same group in @xmath69 , they are also in the same component of @xmath69 , say in component @xmath159 . since @xmath167 belong to the same component of @xmath70 , they also belong to the same component of @xmath69 , thus @xmath169 . by a similar argument we conclude that @xmath170 . thus all four vertices @xmath171 belong to @xmath159 . let @xmath150 be the @xmath22node with @xmath172 , and let @xmath152 be the children of @xmath125 with @xmath153 and @xmath154 . clearly @xmath173 , hence there must be an @xmath128node @xmath162 somewhere on the path between @xmath125 and @xmath123 where the edge @xmath16 is introduced . however , since @xmath15 and @xmath163 share the same label in @xmath127 , and @xmath14 and @xmath174 share the same label in @xmath127 , this also holds in @xmath164 . hence the @xmath128 operation that introduces the edge @xmath16 also introduces the edge @xmath175 . however , this contradicts the assumption that @xmath176 . hence the path property holds as well . we conclude that @xmath81 is indeed a @xmath2derivation of @xmath1 . the above procedure for generating the @xmath2derivation can clearly be carried out in polynomial time . [ ex : expression->derivation ] consider the @xmath32expression @xmath121 for the graph @xmath35 of example [ ex : p4 ] . applying the procedure described in the proof of lemma [ lem : expression->derivation ] we obtain the @xmath32 derivation @xmath81 of example [ ex : derivation ] . [ lem : derivation->expression ] from a @xmath2derivation of a graph @xmath1 we can obtain a @xmath2 expression of @xmath1 in polynomial time . let @xmath49 be a @xmath2derivation of @xmath64 . using the construction of the proof of lemma [ lem : make - strict ] we can obtain a strict @xmath2derivation of @xmath1 from any given @xmath2derivation of @xmath1 . hence we may assume , w.l.o.g . , that @xmath81 is strict . let @xmath177 . we are going to construct in polynomial time a @xmath2 expression tree for @xmath1 , which can clearly be turned into a @xmath2 expression for @xmath1 in polynomial time . we proceed in three steps . first we construct a @xmath2expression tree @xmath178 that only contains @xmath22nodes and leaves . for each component @xmath179 of @xmath60 we introduce a leaf @xmath180 with label @xmath181 . for each @xmath65 and each component @xmath182 we introduce an @xmath22node @xmath183 . we add edges to @xmath178 such that @xmath184 is a child of @xmath183 if and only if @xmath185 . properties d1 and d3 of a derivation ensure that @xmath178 is a tree . note that @xmath178 is not necessarily succinct , and may contain @xmath22 nodes that have only one child . in the next step we add to @xmath178 certain @xmath129nodes to obtain the @xmath2expression tree @xmath186 . we visit the @xmath22nodes of @xmath178 in a depth - first ordering . let @xmath183 be the currently visited node . between @xmath183 and each child @xmath184 of @xmath183 we add at most @xmath2 @xmath129nodes ( so that the edge between @xmath183 and @xmath184 becomes a path ) such that afterwards @xmath183 has a child @xmath187 with @xmath188 . this is possible because of properties d2 and d4 of a derivation . as a final step , we add @xmath128nodes to @xmath186 and the @xmath2expression tree @xmath122 . let @xmath67 be an edge of @xmath1 . we show that there is an @xmath22node @xmath125 in @xmath186 above which we can add an @xmath128node @xmath162 ( @xmath125 is a child of @xmath162 ) which introduces edges including @xmath16 but does not introduce any edge not present in @xmath189 . let @xmath183 be the @xmath22node of @xmath186 with smallest @xmath20 such that @xmath190 . we write @xmath191 and @xmath192 and observe that @xmath182 . among the children of @xmath125 are two distinct nodes @xmath152 such that @xmath193 and @xmath194 . it follows that there are distinct components @xmath195 with @xmath196 and @xmath197 . by the edge property , @xmath14 and @xmath15 belong to different groups of @xmath69 , and so @xmath14 and @xmath15 have different labels in @xmath127 , say the labels @xmath198 and @xmath199 , respectively . we add an @xmath128node @xmath162 above @xmath125 representing the operation @xmath200 . this inserts the edge @xmath16 to @xmath127 . we need to show that @xmath200 does not add any edge that is not in @xmath189 . we show that for all pairs of vertices @xmath201 where @xmath202 has label @xmath198 and @xmath203 has label @xmath199 in @xmath127 , the edge @xmath204 is in @xmath189 . we consider four cases . case 1 : @xmath205 . trivially , @xmath206 . case 2 : @xmath207 . assume to the contrary that @xmath208 . since @xmath15 and @xmath203 have the same label in @xmath127 , they belong to the same group of @xmath69 . the neighborhood property implies that @xmath14 and @xmath15 belong to the same component of @xmath70 , a contradiction to the minimal choice of @xmath20 . hence @xmath209 . case 3 : @xmath210 . this case is symmetric to case 2 . case 4 : @xmath211 . assume to the contrary that @xmath212 . it follows by from cases 2 and 3 that @xmath213 . the path property implies that @xmath14 and @xmath15 belong to the same component of @xmath70 , a contradiction to the minimal choice of @xmath20 . hence @xmath209 . consequently , we can successively add @xmath128nodes to @xmath186 until all edges of @xmath189 are inserted , but no edge outside of @xmath189 . hence we obtain indeed a @xmath2expression tree for @xmath1 . this procedure for generating the @xmath2expression tree can clearly be carried out in polynomial time , hence the lemma follows . we note that we could have saved some @xmath129operations in the proof of lemma [ lem : derivation->expression ] . in particular the @xmath2expression produced may contain @xmath129operations where the number of different labels before and after the application of the @xmath129operation remains the same . it is easy to see that such a @xmath129operations can be omitted if we change labels of some initial @xmath2graphs accordingly . consider the derivation @xmath81 of graph @xmath1 in example [ ex : derivation ] . we construct a @xmath32expression of @xmath1 using the procedure as described in the proof of lemma [ lem : derivation->expression ] , however , to save space , we give the construction in terms of @xmath2expressions instead of @xmath2expression trees . first we obtain @xmath214 . next we insert @xmath129 operations to represent how the groups evolve through the derivation : @xmath215 . finally we add @xmath128 operations , and obtain @xmath216 . by lemma [ lem : strict - short ] we do not need to search for @xmath2derivations of length @xmath217 when the graph under consideration has @xmath3 vertices . the next lemma improves this bound to @xmath218 which provides a significant improvement for our sat encoding , especially if the graph under consideration has large clique - width . [ lem : strict - shorter ] let @xmath219 . if a graph with @xmath3 vertices has a @xmath2 derivation , then it has a @xmath2derivation of length @xmath218 . let @xmath220 be fixed . we define the _ @xmath2length _ of a derivation as the number of templates that contain at least one component of size larger than @xmath2 ( these templates form a suffix of the derivation ) . let @xmath221 be the largest @xmath2length of a strict derivation over a universe of size @xmath3 . before we show the lemma , we establish three claims . for these claims , the groups of the considered derivations are irrelevant and hence we will be ignored . _ claim 1 : _ @xmath222 . to show the claim , consider a strict derivation @xmath49 over a universe @xmath44 of size @xmath3 with @xmath2 length @xmath223 . we take a new element @xmath198 and form a strict derivation @xmath115 over the universe @xmath224 by adding the singleton @xmath225 to @xmath226 for @xmath227 and adding a new template @xmath228 with @xmath229 . the new derivation @xmath115 has @xmath2length @xmath230 . _ claim 2 : _ let @xmath49 be a strict derivation over a universe @xmath44 of size @xmath3 of @xmath2length @xmath221 . then , @xmath231 has exactly one component of size @xmath232 and all other components are singletons . we proceed to show the claim . let @xmath233 , and observe that @xmath23 is the largest index where all components of @xmath234 have size at most @xmath2 . let @xmath235 be the components of @xmath236 of size greater than @xmath19 such that @xmath237 . thus @xmath238 . we show that @xmath239 . assume to the contrary that @xmath240 . we pick some element @xmath241 , @xmath242 , and set @xmath243 . the derivation @xmath81 induces a strict derivation @xmath115 over the universe @xmath244 . observe that @xmath245 . evidently @xmath115 has the same @xmath2 length as @xmath81 , hence @xmath246 , a contradiction to claim 1 . hence @xmath239 , and @xmath157 is the only component in @xmath236 of size greater than @xmath2 , all other components of @xmath236 are singletons . we show that @xmath247 . we assume to the contrary that @xmath248 . we pick @xmath232 elements @xmath249 and set @xmath250 . similarly as above , we observe that @xmath81 induces a strict derivation @xmath251 over the universe @xmath252 , and that @xmath251 has the same @xmath2length as @xmath81 . since @xmath253 we have again a contradiction to claim 1 . hence claim 2 is established . _ claim 3 : _ @xmath254 . to see the claim , let @xmath49 be a strict derivation over a universe @xmath44 of size @xmath3 of @xmath2length @xmath221 . let @xmath233 . by claim 2 we know that @xmath236 has exactly one component of size @xmath232 and all other components are singletons ( hence there are @xmath255 singletons ) . we conclude that @xmath256 . since @xmath81 is strict , we have @xmath257 . thus @xmath258 , and the claim follows . we are now in the position to establish the statement of the lemma . let @xmath49 be a @xmath2 derivation of a graph @xmath64 with @xmath259 . by lemma [ lem : make - strict ] we may assume that @xmath81 is strict . let @xmath223 be the @xmath2length of @xmath81 and let @xmath260 . by claim 3 we know that @xmath261 . we define a new template @xmath262 with @xmath263 and @xmath264 , and we set @xmath265 . we claim that @xmath115 is a @xmath2derivation of @xmath1 . clearly @xmath115 is a derivation , but we need to check the edge , neighborhood , and path property for @xmath262 and @xmath236 in @xmath115 . the properties hold trivially for @xmath262 since all its groups are singletons . for @xmath236 the properties hold since @xmath262 has the same components as @xmath234 . thus @xmath115 is indeed a @xmath2derivation of @xmath1 . the length of @xmath115 is @xmath266 , hence the lemma follows . again , consider the derivation @xmath81 of example [ ex : derivation ] . @xmath81 defines @xmath35 which has clique - width 3 @xcite . according to lemma [ lem : strict - shorter ] , it should have a derivation of length @xmath267 . we can obtain such a derivation by removing @xmath268 from @xmath81 , which gives @xmath269 . by combining lemmas [ lem : expression->derivation ] , [ lem : derivation->expression ] , and [ lem : strict - shorter ] , we arrive at the main result of this section . [ pro : expression ] let @xmath270 . a graph @xmath1 with @xmath3 nodes has clique - width at most @xmath2 if and only if @xmath1 has a @xmath2derivation of length at most @xmath218 . let @xmath64 be graph , and @xmath271 an integer . we are going to construct a cnf formula @xmath272 that is satisfiable if and only if @xmath1 has a derivation of length @xmath48 . we assume that the vertices of @xmath1 are given in some arbitrary but fixed linear order @xmath273 . for any two distinct vertices @xmath14 and @xmath15 of @xmath1 and any @xmath274 we introduce a _ component variable _ @xmath275 . similarly , for any two distinct vertices @xmath14 and @xmath15 of @xmath1 with @xmath276 and any @xmath274 we introduce a _ group variable _ @xmath277 . intuitively , @xmath275 or @xmath277 are true if and only if @xmath14 and @xmath15 are in the same component or group , respectively , in the @xmath20th template of an implicitly represented derivation of @xmath1 . the formula @xmath272 is the conjunction of all the clauses described below . the following clauses represent the conditions d1d4 . @xmath278 + for @xmath279 , @xmath276 , @xmath227 . we further add clauses that ensure that the relations of being in the same group and of being in the same component are transitive . @xmath280 @xmath281 + for @xmath71 , @xmath282 , @xmath227 . in order to enforce the _ edge property _ we add the following clauses for any two vertices @xmath66 with @xmath276 , @xmath67 and @xmath65 : @xmath283 . further , to enforce the _ neighborhood property _ , we add for any three vertices @xmath71 with @xmath284 and @xmath285 and @xmath65 , the following clauses . @xmath286 finally , to enforce the _ path property _ we add for any four vertices @xmath171 , such that @xmath75 , and @xmath76 , @xmath287 and @xmath65 the following clauses : ( @xmath288 the following statement is a direct consequence of the above definitions . [ lem : der ] @xmath272 is satisfiable if and only if @xmath1 has a derivation of length @xmath48 . in this section , we describe how the formula @xmath272 can be extended to encode a derivation of width at most @xmath2 . ideally , one wants to encode that unit propagation results in a conflict on any assignment of component and group variables representing a derivation containing a component with more than @xmath2 groups . first we will describe the conventional direct encoding @xcite followed by our representative encoding . only the latter encoding realizes arc consistency @xcite . we introduce new boolean variables @xmath289 for @xmath140 , @xmath290 , and @xmath227 . the purpose is to assign each vertex for each template a group number between @xmath19 and @xmath2 . the intended meaning of a variable @xmath289 is that in @xmath69 , vertex @xmath15 has group number @xmath198 . let @xmath291 denote the formula obtained from @xmath272 by adding the following three sets of clauses . the first ensures that every vertex has at least one group number , the second ensures that every vertex has at most one group number , and the third ensures that two vertices of the same group share the same group number . @xmath292 for @xmath293 , @xmath227 , @xmath294 for @xmath295 , @xmath296 , @xmath227 , @xmath297 + for @xmath279 , @xmath287 , @xmath298 , @xmath299 . together with lemma [ lem : der ] this construction directly yields the following statement . let @xmath64 be graph and @xmath300 . then @xmath291 is satisfiable if and only if @xmath301 . [ example : unitprop ] let @xmath64 and @xmath302 . vertices @xmath303 in template @xmath69 , are in one component , but in different groups . hence the corresponding component variables are true , and the corresponding group variables are false . the clauses containing the variables @xmath304 with @xmath305 after removing falsified literals are : @xmath306 + @xmath307 these clauses can not be satisfied , yet unit propagation will not result in a conflict . therefore , a sat solver may not be able to cut off the current branch . to overcome the unit propagation problem of the direct encoding , as described in example [ example : unitprop ] , we propose the _ representative encoding _ which uses two types of variables . first , for each @xmath140 and @xmath65 we introduce a representative variable @xmath308 . this variable , if assigned to true , expresses that vertex @xmath15 is the representative of a group in template @xmath69 . in each group , only one vertex can be the representative and we choose to make the first vertex in the lexicographical ordering the representative . this results in the following clauses : @xmath309 for @xmath295 , @xmath227 additionally we introduce auxiliary variables to efficiently encode that the number of representative vertices in a component is at most @xmath2 . these auxiliary variables are based on the _ order encoding _ consider a ( non - boolean ) variable @xmath310 with domain @xmath311 , whose elements denote the group number of vertex @xmath15 in template @xmath69 . in the direct encoding , we used @xmath2 variables @xmath289 with @xmath312 . assigning @xmath313 in that encoding means @xmath314 . alternatively , we can use _ order variables _ @xmath315 with @xmath295 , @xmath316 , @xmath317 . assigning @xmath318 means @xmath319 . consequently , @xmath320 means @xmath321 . given an assignment to the order variables @xmath315 , one can easily construct the equivalent assignment to the variables in the direct encoding ( and the other way around ) . below is a visualization of the equivalence relation with @xmath322 . in the middle is a binary representation of each of the @xmath2 labels by concatenating the boolean values to the order variables . @xmath80{c@{\;\;}c@{\;\;}c@{\;\;}c@{\;\;}c@{\;\;}c@{\;\;}l } l_v = 1 & \leftrightarrow & l_{v,1,i } = { \tt 1 } & \leftrightarrow & { \tt 0000 } & \leftrightarrow & o^>_{v,1,i } = o^>_{v,2,i } = o^>_{v,3,i } = o^>_{v,4,i } = { \tt 0 } \\ l_v = 2 & \leftrightarrow & l_{v,2,i } = { \tt 1 } & \leftrightarrow & { \tt 1000 } & \leftrightarrow & o^>_{v,1,i } = { \tt 1 } , o^>_{v,2,i } = o^>_{v,3,i } = o^>_{v,4,i } = { \tt 0 } \\ l_v = 3 & \leftrightarrow & l_{v,3,i } = { \tt 1 } & \leftrightarrow & { \tt 1100 } & \leftrightarrow & o^>_{v,1,i } = o^>_{v,2,i } = { \tt 1 } , o^>_{v,3,i } = o^>_{v,4,i } = { \tt 0 } \\ l_v = 4 & \leftrightarrow & l_{v,4,i } = { \tt 1 } & \leftrightarrow & { \tt 1110 } & \leftrightarrow & o^>_{v,1,i } = o^>_{v,2,i } = o^>_{v,3,i } = { \tt 1 } , o^>_{v,4,i } = { \tt 0 } \\ l_v = 5 & \leftrightarrow & l_{v,5,i } = { \tt 1 } & \leftrightarrow & { \tt 1111 } & \leftrightarrow & o^>_{v,1,i } = o^>_{v,2,i } = o^>_{v,3,i } = o^>_{v,4,i } = { \tt 1 } \end{array}$ ] although our encoding is based on the variables from the order encoding , we use none of the associated clauses . we implemented the original order @xcite , which resulted in many long clauses and the performance was comparable to the direct encoding . instead , we combined the representative and order variables . our use of the order variables can be seen as the encoding of a sequential counter @xcite . we would like to point out that if @xmath14 and @xmath15 are both representative vertices in the same component of template @xmath69 and @xmath287 , then @xmath323 and @xmath324 must hold for some @xmath325 . consequently , @xmath326 ( vertex @xmath14 has not the highest group number in @xmath69 ) , ( vertex @xmath15 has not the lowest group number in @xmath69 ) , and : these constraints can be expressed by the following clauses . @xmath327for @xmath279 , @xmath287 , @xmath328 . consider a graph @xmath329 with @xmath330 and the representative encoding with @xmath331 . we will show that if @xmath14,@xmath15,@xmath163 , and @xmath174 are all in the same component and they are all representatives of their respective group numbers in template @xmath69 , then unit propagation will result in a conflict ( because there are four representatives and only three group numbers ) . observe that all corresponding component and representative variables are true . this example , with falsified literals removed , contains the clauses @xmath332 , @xmath333 , @xmath334 , @xmath335 , @xmath336 , @xmath337 , @xmath332 , @xmath338 , @xmath339 , @xmath340 , , @xmath337 , @xmath341 , @xmath342 , @xmath343 , @xmath344 , @xmath345 , @xmath346 . literals that are falsified by unit clauses are shown in bold . notice that @xmath347 is falsified , i.e. , a conflicting clause . both the direct and representative encoding require @xmath348 variables . the number of clauses depends on the set of edges . in worst case , the number of clauses can be @xmath349 due to the path condition . in this section we report the results we obtained by running our sat encoding on various classes of graphs . given a graph @xmath350 , we compute that @xmath1 has clique - width @xmath2 by determining for which value of @xmath2 it holds that @xmath351 is satisfiable and latexmath:[$f(g , k-1 , 2.2 @xcite to solve the encoded problems . glucose solved the hardest instances about twice as fast ( or more ) as other state - of - the - art solvers such as lingeling @xcite , minisat @xcite and clasp @xcite . we used a 4-core intel xeon cpu e31280 3.50ghz , 32 gb ram machine running ubuntu 10.04 . although the direct and representative encodings result in cnf formulas of almost equal size , there is a huge difference in costs to solve these instances . to determine the clique - width of the famous named graphs ( see below ) using the direct encoding takes about two to three orders of magnitude longer as compared to the representative encoding . for example , we can establish that the paley graph with 13 vertices has clique - width 9 within a few seconds using the representative encoding , while the solver requires over an hour using the direct encoding . because of the huge difference in speed , we discarded the use of the direct encoding in the remainder of this section . we noticed that upper bounds ( satisfiable formulas ) are obtained much faster than lower bounds ( unsatisfiable formulas ) . the reason is twofold . first , the whole search space needs to be explored for lower bounds , while for upper bounds , one can be lucky " and find a solution fast . second , due to our encoding , upper bound formulas are smaller ( due to a smaller @xmath48 ) which makes them easier . table [ tab : random20 ] shows this for a random graph with 20 vertices for the direct encoding and the representative encoding . .runtimes in seconds of the direct and representative encoding on a random graph with 20 vertices and 95 edges for different values of @xmath2 . up to @xmath353 the formulas are unsatisfiable , afterwards they are satisfiable . timeout ( to ) is 20,000 seconds . [ cols="<,^,^,^,^,^,^,^,^,^,^,^,^,^,^",options="header " , ] we have presented a sat approach to the exact computation of clique - width , based on a reformulation of clique - width and several techniques to speed up the search . this new approach allowed us to systematically compute the exact clique - width of various small graphs . we think that our results could be of relevance for theoretical investigations . for instance , knowing small vertex - minimal graphs of certain clique - width could be helpful for the design of discrete algorithms that recognize graphs of bounded clique - width . such graphs can also be useful as gadgets for a reduction to show that the recognition of graphs of clique - width 4 is @xmath0hard , which is still a long - standing open problem @xcite . furthermore , as discussed in section [ sect : intro ] , there are no heuristic algorithms to compute the clique - width directly , but heuristic algorithms for related parameters can be used to obtain upper bounds on the clique - width . our sat - based approach can be used to empirically evaluate how far heuristics are from the optimum , at least for small and medium - sized graphs . so far we have focused in our experiments on the exact clique - width , but for various applications it is sufficient to have good upper bounds . our results ( see table [ tab : random20 ] ) suggest that our approach can be scaled to medium - sized graphs for the computation of upper bounds . we also observed that for many graphs the upper bound of lemma [ lem : strict - shorter ] is not tight . thus , we expect that if we search for shorter derivations , which is significantly faster , this will yield optimal or close to optimal solutions in many cases . finally , we would like to mention that our sat - based approach is very flexible and open . it can easily be adapted to variants of clique - width , such as linear clique - width @xcite , @xmath354clique - width @xcite , or nlc - width @xcite . hence , our approach can be used for an empirical comparison of these parameters . the authors acknowledge the texas advanced computing center ( tacc ) at the university of texas at austin for providing grid resources that have contributed to the research results reported within this paper . gilles audemard and laurent simon . predicting learnt clauses quality in modern sat solvers . in _ proceedings of the 21st international jont conference on artifical intelligence _ , ijcai09 , pages 399404 , san francisco , ca , usa , 2009 . morgan kaufmann publishers inc . martin bey . fast algorithm for rank - width . in _ mathematical and engineering methods in computer science , 8th international doctoral workshop , memics 2012 , znojmo , czech republic , october 25 - 28 , 2012 , revised selected papers _ , volume 7721 of _ lecture notes in computer science _ , pages 8293 . springer verlag , 2013 . armin biere . lingeling and friends entering the sat challenge 2012 . in a. balint , a. belov , a. diepold , s. gerber , m. jrvisalo , and c. sinz , editors , _ solver and benchmark descriptions _ , volume b-2012 - 2 of _ department of computer science series of publications b. _ , pages 3334 . university of helsinki , 2012 . bruno courcelle , joost engelfriet , and grzegorz rozenberg . context - free handle - rewriting hypergraph grammars . in hartmut ehrig , hans - jrg kreowski , and grzegorz rozenberg , editors , _ graph - grammars and their application to computer science , 4th international workshop , bremen , germany , march 59 , 1990 , proceedings _ , volume 532 of _ lecture notes in computer science _ , pages 253268 , 1991 . p. alex dow and richard e. korf . best - first search for treewidth . in _ proceedings of the twenty - second aaai conference on artificial intelligence , july 22 - 26 , 2007 , vancouver , british columbia , canada _ , pages 11461151 . aaai press , 2007 . niklas en and niklas srensson . an extensible sat - solver . in enrico giunchiglia and armando tacchella , editors , _ theory and applications of satisfiability testing _ , volume 2919 of _ lecture notes in computer science _ , pages 502518 . springer berlin heidelberg , 2004 . m. gebser , b. kaufmann , a. neumann , and t. schaub . clasp : a conflict - driven answer set solver . in c. baral , g. brewka , and j. schlipf , editors , _ proceedings of the ninth international conference on logic programming and nonmonotonic reasoning ( lpnmr07 ) _ , volume 4483 of _ lecture notes in artificial intelligence _ , pages 260265 . springer - verlag , 2007 . vibhav gogate and rina dechter . a complete anytime algorithm for treewidth . in _ proceedings of the proceedings of the twentieth conference annual conference on uncertainty in artificial intelligence ( uai-04 ) _ , pages 201208 , arlington , virginia , 2004 . auai press . martin charles golumbic and udi rotics . on the clique - width of some perfect graph classes . , 11(3):423443 , 2000 . selected papers from the workshop on graph - theoretical aspects of computer science ( wg 99 ) , part 1 ( ascona ) . pinar heggernes , daniel meister , and udi rotics . computing the clique - width of large path powers in linear time via a new characterisation of clique - width . in alexander s. kulikov and nikolay k. vereshchagin , editors , _ computer science - theory and applications - 6th international computer science symposium in russia , csr 2011 , st . petersburg , russia , june 14 - 18 , 2011 . proceedings _ , volume 6651 of _ lecture notes in computer science _ , pages 233246 . springer verlag , 2011 . eivind magnus hvidevold , sadia sharmin , jan arne telle , and martin vatshelle . finding good decompositions for dynamic programming on dense graphs . in dniel marx and peter rossmanith , editors , _ parameterized and exact computation - 6th international symposium , ipec 2011 , saarbrcken , germany , september 6 - 8 , 2011 . revised selected papers _ , volume 7112 of _ lecture notes in computer science _ , pages 219231 . springer verlag , 2012 . hadi katebi , karem a. sakallah , and igor l. markov . conflict anticipation in the search for graph automorphisms . in nikolaj bjrner and andrei voronkov , editors , _ logic for programming , artificial intelligence , and reasoning - 18th international conference , lpar-18 , mrida , venezuela , march 11 - 15 , 2012 . proceedings _ , volume 7180 of _ lecture notes in computer science _ , pages 243257 . springer verlag , 2012 . marko samer and helmut veith . encoding treewidth into sat . in _ theory and applications of satisfiability testing - sat 2009 , 12th international conference , sat 2009 , swansea , uk , june 30 - july 3 , 2009 . proceedings _ , volume 5584 of _ lecture notes in computer science _ , pages 4550 . springer verlag , 2009 . carsten sinz . towards an optimal cnf encoding of boolean cardinality constraints . in peter van beek , editor , _ 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 verlag , 2005 . toby walsh . v csp . in r. dechter , editor , _ 6th international conferenc on principles and practice of constraint programming ( cp 2000 ) _ , volume 1894 of _ lecture notes in computer science _ , pages 441456 . springer verlag , 2000 .

optical and ultraviolet observations for the nearby type ii - plateau supernova ( sn iip ) 2013am in the nearby spiral galaxy m65 are presented in this paper . the early spectra are characterized by relatively narrow p - cygni features , with ejecta velocities much lower than observed in normal sne iip ( i.e. , km s vs. km in the middle of the plateau phase ) . moreover , prominent absorptions are also detected in sn 2013am at relatively early phases . these spectral features are reminiscent of those seen in the low - velocity and low - luminosity sn iip 2005cs . however , sn 2013am exhibits different photometric properties , having shorter plateau phases and brighter light - curve tails if compared to sn 2005cs . adopting=3.1 and a mean value of total reddening derived from the photometric and spectroscopic methods(i.e . ,.19 mag ) , we find that sn 2013am may have reached an absolute-band peak magnitude of mag , and produced ani mass ofm in the explosion . these parameters are close to those derived for sn 2008 in and sn 2009n which have been regarded as gap - filler " objects linking the faint sne iip to the normal ones . this indicates that some low - velocity sne iip may not necessarily result from the low - energetic explosions , and the low expansion velocities could be due to a lower metallicity of the progenitor stars , a larger envelope mass ejected in the explosion , or that is was observed at an angle that is away from the polar direction .

1409.3400

type ii - plateau supernovae ( sne iip ) represent a major group of core - collapse supernovae ( cc sne , see @xcite for a review ) , characterized by a nearly constant optical luminosity for about 3 - 4 months after the peak . special attentions have been paid to the sne iip showing low ejecta velocities ( hereafter lv sne iip ) because of their unusual properties in some aspects . the well - studied example include sn 1997d @xcite and sn 2005cs @xcite . these lv sne iip generally show a low - luminosity plateau lasting for about 100 days , followed by an under - luminous exponential tail at late times . their spectra show prominent but narrower p - cygni profiles . @xcite suggested that these lv objects are at the faint end of the luminosity distribution of sne iip . @xcite recently presented observations of five faint sne iip ( sne 1999gn , 2002gd , 2003z , 2004eg , and 2006ov ) . combining with the published sne iip with low luminosities ( e.g. , sn 1994n , sn 1997d , sn 1999br , sn 1999eu , sn 2001dc , @xcite , @xcite , @xcite ; sn 2005cs , @xcite ; sn 2008 in , @xcite ) , they found that all faint sne iip have ejecta velocities below @xmath01000 km s@xmath1 at the end of the plateau , and the @xmath6ni mass ejected in the explosion is smaller than 0.01 m@xmath8 . it is also suggested that those subluminous sne iip with low ejecta velocities probably originate from less energetic explosions of intermediate - mass stars , with an initial mass in the range of 10 - 15m@xmath8 . on the other hand , the ` gap - filler ' events like sn 2008 in @xcite and sn 2009n @xcite were recently reported , which showed a striking resemblance to the faint event sn 1997d but produced more @xmath6ni , i.e. , @xmath00.015 m@xmath8 for sn 2008 in and @xmath00.02 m@xmath8 sn 2009n . in this paper we present optical and ultraviolet ( uv ) data of another low - velocity object sn 2013am , obtained through an observational campaign that lasted for about one year on several telescopes . the spectral evolution of sn 2013am shows an overall similarity to that of sn 2005cs which is known for its low expansion velocity and low luminosity . observations and data reductions are described in section [ sect : obs ] , the uv and optical light and color curves are presented in section [ sect : lc ] , while the spectral evolution is shown in section [ sect : spe ] . in section 5 , we estimate the explosion parameters such as the absolute magnitudes , quasi - bolometric light curve , and the @xmath6ni mass ejected in the explosion . a brief discussion about diversities of sne iip is also given in section [ sect : discu ] . finally , a summary is given in section [ sect : con ] . -band image was taken on mar . 23 2013 by the 2.4-m telescope at lijiang observatory . the supernova and ten local reference stars are marked . north is up and east is left.,width=302 ] sn 2013am was discovered at roughly a magnitude of 15.6 mag on mar . 21.64 ut 2013 ( ut is used throughout this paper ) in an unfiltered image of the spiral galaxy m65 ( ngc 3623 ) taken using a 0.35-m f/11.0 schmidt - cassegrain telescope @xcite . its coordinates are ra = 11h18m56s.95 , dec = @xmath90349.4(j2000 ) . the sn is located 15 east and 102 south of the center of the host ( see figure [ < img > ] ) . this object was classified as a young type iip based on a spectrum obtained on mar . 22.84 2013 @xcite . the sn was also independently discovered by the palomar transient factory ( ptf ) on mar . 23.94 2013 ( iptf13aaz , @xcite ) . nothing was detected at this location in the ptf @xmath10-band image taken on mar . 19.34 2013 , with an upper limit of 21.36 mag . this allows us to estimate the explosion time to be around mar . 20.5 2013 ( jd = [email protected] ) . this epoch is thus assumed as the explosion date in this paper . our observations started on mar . 22.85 ut 2013 , @xmath0 + 2.4 days after explosion . the @xmath12 photometry of sn 2013am was obtained with the 2.4-m telescope at li - jiang observatory ( hereinafter ljt , see @xcite for a detailed description ) , yunnan observatories ( ynao ) , and the yunnan faint object spectrograph and camera ( yfosc , see @xcite for detailed descriptions ) . optical photometry was also obtained with the liverpool 2.0-m telescope ( hereinafter lt ) , which is located at la palma and is owned and operated by the astrophysics research institute of liverpool john moores university . data were also collected with the tsinghua - naoc 80-cm telescope ( hereinafter tnt , see wang et al . 2008 and huang et al . 2012 for detailed descriptions ) at xing - long station of national astronomical observatories ( naoc ) , china . all ccd images were reduced using the iraf standard procedure , including corrections for bias , flat field , and removal of cosmic rays . the standard point - spread - function ( psf ) fitting method from the iraf daophot package @xcite was used to measure the instrumental magnitudes for both the sn and local standard stars . these @xmath12 magnitudes were then converted to the standard johnson @xmath13 @xcite and cousins @xmath14 @xcite systems through transformations established by observing @xcite standards during photometric nights . the average value of the photometric zeropoints , determined on three photometric nights , was used to calibrate the local standard stars in the field of sn 2013am . table [ tab : photo_stand ] lists the standard @xmath12 magnitudes and the corresponding uncertainties of 10 local standard stars as labeled in figure [ < img > ] . the magnitudes of these stars were then used to transform the instrumental magnitudes of sn 2013am to those of the standard @xmath12 system . the @xmath15- , @xmath10- , and @xmath16-band magnitudes of the standard stars listed in table [ tab : photo_stand ] are from the sloan digital sky survey data release 10 ( sdss dr10 ) . these @xmath17-band magnitudes were converted into @xmath18 magnitudes in light of the relations given by @xcite and @xcite . note that those transformation relations were derived from normal stars , which may introduce systematic errors when applying to the photometry of supernovae . the final results of the @xmath12 magnitudes of sn 2013am are listed in table [ tab : gdpho ] , and the @xmath17-band magnitudes are listed in table [ tab : gri ] . ccccccccc[!th ] 1 & 16.75(04 ) & 15.42(02 ) & 14.08(03 ) & 13.28(03 ) & 12.59(03 ) & 14.83(01 ) & 13.66(01 ) & 14.51(01 ) + 2 & 19.49(04 ) & 18.74(03 ) & 17.10(02 ) & 16.08(03 ) & 14.88(02 ) & 18.03(01 ) & 16.56(01 ) & 15.66(01 ) + 3 & 14.64(03 ) & 14.47(03 ) & 13.78(03 ) & 13.39(04 ) & 13.04(02 ) & 14.07(01 ) & 13.64(01 ) & 13.51(01 ) + 4 & 18.84(04 ) & 18.46(02 ) & 17.70(03 ) & 17.25(04 ) & 16.78(04 ) & 18.05(01 ) & 17.49(01 ) & 17.27(01 ) + 5 & 18.33(05 ) & 18.27(04 ) & 17.52(03 ) & 17.08(05 ) & 16.68(04 ) & 17.84(01 ) & 17.38(01 ) & 17.19(01 ) + 6 & 15.58(03 ) & 15.35(02 ) & 14.54(04 ) & 14.07(05 ) & 13.60(03 ) & 14.93(01 ) & 14.36(01 ) & 14.14(01 ) + 7 & 14.06(04 ) & 14.17(04 ) & 13.58(03 ) & 13.23(02 ) & 12.90(02 ) & 13.81(01 ) & 13.48(01 ) & 13.39(01 ) + 8 & 17.17(05 ) & 16.02(02 ) & 14.58(02 ) & 13.74(03 ) & 12.94(02 ) & 15.33(01 ) & 14.04(01 ) & 13.52(01 ) + 9 & 14.67(05 ) & 14.71(03 ) & 14.14(02 ) & 13.82(04 ) & 13.54(03 ) & 14.37(01 ) & 14.05(01 ) & 13.97(01 ) + 10 & 18.70(03 ) & 18.02(03 ) & 16.87(02 ) & 16.30(03 ) & 15.72(02 ) & 17.44(01 ) & 16.56(01 ) & 16.24(01 ) [ tab : photo_stand ] sn 2013am was also observed with the uvot @xcite on board the @xmath19 satellite @xcite , spanning from @xmath20 days to @xmath21 days from the explosion . the photometric observations were performed in three uv filters ( @xmath22 , @xmath23 , and @xmath24 ) and three broadband optical filters ( @xmath25 , @xmath26 , and @xmath4 ) . these @xmath19 images were reduced using the uvotsource " program of the heasoft ( the high energy astrophysics software ) with the latest @xmath19 calibration database . aperture photometry was adopted in the imaging reduction as the psf profile is slightly dependent on the count rate of the source ; the photometric aperture and the background value were set according to @xcite . since the uvot is a photon - counting detector and suffers from coincidence loss ( c - loss ) for bright sources , the observed counts need corrections for such losses . this has now been automatically done by the uvotsource " program based on an empirical relation given by @xcite . as the instrumental response curves of the uvot optical filters do not follow exactly those of the johnson @xmath13 system , color - term corrections were further applied to the magnitudes @xcite . table [ tab : swiftpho ] lists the final uvot uv / optical magnitudes of sn 2013am . cccccccc 56373.85 & 2.35 & & 16.88(13 ) & 16.40(10 ) & 15.88(07 ) & 15.58(07 ) & lt + 56374.58 & 3.08 & 16.69(04 ) & 16.90(05 ) & 16.41(04 ) & 15.86(02 ) & 15.48(05 ) & ljt + 56375.74 & 4.24 & 16.80(08 ) & 16.90(07 ) & 16.38(06 ) & 15.78(07 ) & 15.41(05 ) & ljt + 56378.87 & 7.37 & & 17.08(14 ) & 16.29(11 ) & 15.67(09 ) & 15.31(09 ) & lt + 56379.65 & 8.15 & 17.07(07 ) & 16.94(06 ) & 16.37(03 ) & 15.73(03 ) & 15.31(07 ) & ljt + 56380.96 & 9.46 & & 17.07(09 ) & 16.31(07 ) & 15.64(02 ) & 15.24(02 ) & lt + 56381.69 & 10.19 & 17.13(06 ) & 17.00(06 ) & 16.36(04 ) & 15.67(07 ) & 15.18(07 ) & ljt + 56383.73 & 12.23 & 17.25(07 ) & 17.03(05 ) & 16.36(03 ) & 15.64(03 ) & 15.18(03 ) & ljt + 56383.87 & 12.37 & & 17.11(12 ) & 16.33(10 ) & 15.59(07 ) & 15.18(07 ) & lt + 56384.73 & 13.23 & 17.33(05 ) & 17.18(05 ) & 16.38(04 ) & 15.62(03 ) & 15.14(06 ) & ljt + 56391.58 & 20.08 & 17.96(06 ) & 17.50(05 ) & 16.41(04 ) & 15.61(04 ) & 15.08(07 ) & ljt + 56394.00 & 22.50 & & 17.89(16 ) & 16.49(13 ) & 15.71(07 ) & 15.17(07 ) & lt + 56395.91 & 24.41 & & 17.92(09 ) & 16.51(08 ) & 15.71(05 ) & 15.15(05 ) & lt + 56398.55 & 27.05 & & 17.70(06 ) & 16.48(04 ) & 15.65(05 ) & 15.03(06 ) & ljt + 56399.92 & 28.42 & & 17.96(11 ) & 16.50(08 ) & 15.59(02 ) & 14.91(02 ) & lt + 56409.90 & 38.40 & & 18.09(09 ) & 16.55(07 ) & 15.63(04 ) & 14.96(04 ) & lt + 56422.54 & 51.04 & & 18.39(09 ) & 16.51(04 ) & 15.63(04 ) & 14.89(03 ) & tnt + 56422.93 & 51.43 & & 18.33(10 ) & 16.61(08 ) & 15.64(06 ) & 14.91(06 ) & lt + 56424.55 & 53.05 & & 18.40(14 ) & 16.54(08 ) & 15.69(07 ) & 14.90(03 ) & tnt + 56430.88 & 59.38 & & 18.46(08 ) & 16.67(06 ) & 15.64(06 ) & 14.87(06 ) & lt + 56437.65 & 66.15 & & 18.62(06 ) & 16.65(03 ) & 15.73(03 ) & 14.90(04 ) & ljt + 56441.54 & 70.04 & & 18.65(11 ) & 16.61(07 ) & 15.66(04 ) & 14.92(05 ) & tnt + 56442.54 & 71.04 & & 18.71(13 ) & 16.63(09 ) & 15.60(07 ) & 14.90(07 ) & tnt + 56444.54 & 73.04 & & & 16.65(06 ) & 15.65(02 ) & 14.89(05 ) & tnt + 56446.52 & 75.02 & & & 16.67(03 ) & 15.66(03 ) & 14.94(03 ) & tnt + 56455.93 & 84.43 & & 18.77(11 ) & 16.88(09 ) & 15.87(05 ) & 15.11(05 ) & lt + 56456.54 & 85.04 & & 18.89(11 ) & 16.80(06 ) & 15.86(06 ) & 15.02(05 ) & ljt + 56462.91 & 91.41 & & 18.85(11 ) & 16.93(09 ) & 15.94(05 ) & 15.21(05 ) & lt + 56480.40 & 108.90 & & & 18.30(03 ) & 17.22(13 ) & 16.41(09 ) & ljt + 56588.26 & 216.76 & & & & 18.79(07 ) & 17.72(06 ) & lt + 56593.26 & 221.76 & & & & 18.84(08 ) & 17.78(07 ) & lt + 56602.89 & 231.39 & & & 20.64(05 ) & 19.12(15 ) & 17.81(08 ) & ljt + 56644.28 & 272.78 & & & & 19.15(20 ) & 18.19(20 ) & lt + 56653.24 & 281.74 & & & & 19.29(10 ) & 18.22(10 ) & lt + 56660.95 & 289.45 & & & 20.88(05 ) & 19.38(06 ) & 18.12(06 ) & ljt + 56665.85 & 294.35 & & & 20.91(13 ) & 19.41(08 ) & 18.14(07 ) & ljt + 56687.94 & 316.44 & & & 20.98(10 ) & 19.47(10 ) & 18.21(08 ) & ljt + 56715.74 & 344.24 & & & & 19.55(15 ) & 18.45(12 ) & ljt [ tab : gdpho ] ccccc 56373.85 & 2.35 & 16.70(07 ) & 16.14(05 ) & 16.02(03 ) + 56378.87 & 7.37 & 16.70(09 ) & 15.94(02 ) & 15.79(07 ) + 56380.96 & 9.46 & 16.70(06 ) & 15.97(01 ) & 15.78(01 ) + 56383.87 & 12.37 & 16.71(07 ) & 15.90(04 ) & 15.70(04 ) + 56394.00 & 22.50 & 17.11(10 ) & 16.07(04 ) & 15.74(04 ) + 56395.91 & 24.41 & 17.14(06 ) & 16.08(02 ) & 15.73(04 ) + 56399.92 & 28.42 & 17.16(07 ) & 16.06(01 ) & 15.60(01 ) + 56409.90 & 38.40 & 17.26(05 ) & 16.07(03 ) & 15.61(02 ) + 56422.93 & 51.43 & 17.43(06 ) & 16.05(03 ) & 15.54(04 ) + 56430.88 & 59.38 & 17.53(05 ) & 16.07(01 ) & 15.51(05 ) + 56455.93 & 84.43 & 17.80(07 ) & 16.25(03 ) & 15.70(03 ) + 56462.91 & 91.41 & 17.87(06 ) & 16.28(04 ) & 15.76(02 ) + 56588.26 & 216.76 & & 18.78(04 ) & 18.28(07 ) + 56593.26 & 221.76 & & 18.84(06 ) & 18.30(04 ) + 56644.28 & 272.78 & & 19.26(08 ) & 18.62(07 ) + 56653.24 & 281.74 & & 19.43(06 ) & 18.68(06 ) [ tab : gri ] cccccccc[!b ] 56373.82 & 2.32 & 17.80(20 ) & 18.30(20 ) & 17.30(20 ) & 16.70(20 ) & & + 56378.86 & 7.36 & 19.00(15 ) & 19.40(20 ) & 17.90(20 ) & 17.89(18 ) & 16.99(12 ) & 16.38(0.07 ) + 56380.26 & 8.75 & 19.26(18 ) & & 18.29(27 ) & 17.95(17 ) & 17.04(09 ) & 16.36(0.06 ) + 56380.63 & 9.13 & 19.40(13 ) & 20.03(12 ) & 18.48(31 ) & 18.03(20 ) & 17.03(10 ) & 16.37(0.05 ) + 56382.42 & 10.92 & 19.71(11 ) & 20.16(15 ) & 18.83(25 ) & 18.12(15 ) & 17.05(10 ) & 16.36(0.04 ) + 56384.46 & 12.96 & 19.53(22 ) & 20.72(20 ) & 19.09(30 ) & 18.25(21 ) & 17.03(09 ) & 16.38(0.06 ) + 56386.42 & 14.92 & 19.85(17 ) & & & 18.31(15 ) & 17.24(10 ) & 16.39(0.06 ) + 56386.76 & 15.26 & 19.71(17 ) & & & 18.43(20 ) & 17.28(09 ) & 16.40(0.05 ) + 56388.42 & 16.92 & & & & 18.65(13 ) & 17.33(08 ) & 16.38(0.05 ) + 56390.39 & 18.89 & 19.96(21 ) & & & 18.78(16 ) & 17.48(10 ) & 16.35(0.04 ) + 56400.69 & 29.18 & & & & 19.50(17 ) & 17.78(08 ) & 16.39(0.05 ) [ tab : swiftpho ] a total of 11 spectra were obtained with the yfosc and ljt , covering the wavelength from @xmath03400 to @xmath09100(with a resolution of @xmath018 ) . a spectrum taken with the low resolution spectrograph on the 3.58-m telescopio nazionale galileo ( hereinafter tng ) at la palma @xcite is also included in our analysis . these spectra were reduced using standard iraf routines and were flux - calibrated with spectrophotometric flux standard stars observed at a similar air mass on the same night . the continuum of the spectra were adjusted to match the @xmath18 photometry with a precision of 0.1 mag . the spectra were further corrected for the atmospheric absorption at the observing site , and telluric lines were also removed from the data . a journal of the spectroscopic observations is given in table [ tab : spec_log ] . cccccccc mar . 23.1 & 56374.10 & 2.60 & 2.72 & 3340 - 8080 & 1.18 & 600.00 & tng+lrs + mar . 23.6 & 56374.58 & 3.08 & 2.87 & 3500 - 9100 & 1.22 & 1200.00 & ljt+yfosc + mar . 24.8 & 56375.74 & 4.24 & 2.87 & 3490 - 9100 & 1.10 & 1200.00 & ljt+yfosc + mar . 28.8 & 56379.65 & 8.15 & 2.87 & 3510 - 9120 & 1.19 & 1200.00 & ljt+yfosc + mar . 30.7 & 56381.69 & 10.19 & 2.87 & 3520 - 9080 & 1.06 & 1200.00 & ljt+yfosc + apr . 01.8 & 56383.73 & 12.23 & 2.87 & 3480 - 9120 & 1.15 & 1200.00 & ljt+yfosc + apr . 02.8 & 56384.73 & 13.23 & 2.87 & 3510 - 9100 & 1.21 & 1200.00 & ljt+yfosc + apr . 09.6 & 56391.58 & 20.08 & 2.87 & 3490 - 9120 & 1.07 & 1200.00 & ljt+yfosc + apr . 16.6 & 56398.55 & 27.05 & 2.87 & 3520 - 9150 & 1.10 & 1200.00 & ljt+yfosc + apr . 22.6 & 56404.63 & 33.13 & 2.87 & 3510 - 9140 & 1.04 & 1200.00 & ljt+yfosc + may 25.7 & 56437.65 & 66.15 & 2.87 & 3510 - 9100 & 1.13 & 1200.00 & ljt+yfosc + jun . 11.5 & 56454.54 & 83.04 & 2.87 & 3520 - 9080 & 1.14 & 1200.00 & ljt+yfosc [ tab : spec_log ] figure [ < lc > ] shows the optical and uv light curves of sn 2013am . the light curves in the uv bands show a rapid evolution compared to those in the optical bands . the post - maximum magnitude declined at a rate of 10 - 30 mag @xmath27 in uv , while it remains nearly constant in the optical . detailed analysis of the light and color curves are presented in the following subsections . the optical light curves of sn2013am are shown in figure [ < lccomp > ] . overplotted are those of representative sne iip , including the normal objects sne 1999em ( @xcite ) and 2012a @xcite , the low - velocity one sn 2005cs ( @xcite ) , and the ` gap - filler ' events sne 2008 in @xcite and 2009n @xcite . the light curves of peculiar sn iip 1987a @xcite are also shown for comparison . applying a polynomial fit to the observed data , we found that sn 2013am reached at the @xmath26- and @xmath4- band maximum at @xmath0 + 2 and @xmath0 + 7 days after the explosion , with @xmath28 mag and @xmath29 mag , respectively . in the @xmath25 band , sn 2013am shows a post - maximum evolution similar to sn 2009n and sn 1999em , with a magnitude decline of about 1.2 mag in 20 days from the maximum light . in comparison , sn 2005cs and sn 2008 in have a magnitude decline of @xmath02.5 mag and @xmath02.0 mag , respectively , at similar phases . this difference seen in the @xmath25 band becomes marginally important in other bands . however , the light curves of different sne iip show significant scatter in the transition and radioactive - decay phases . for example , the magnitude decline from the end of the plateau to the tail phase ( i.e. , at @xmath30 + 150 days after explosion ) can vary from about 2.0 mag ( for sn 1999em ) to over 4.0 mag ( for sn 2005cs ) in the @xmath4 band . although sn 2013am and sn 2005cs have very similar spectroscopic features ( see discussions in section [ subsect : spev ] ) , the former has a tail that is @xmath01.0 mag brighter than the latter . a brighter tail indicates that more nickel is likely produced in the explosion . one can see that sn 2013am bears a close resemblance to the transition object sn 2009n in light of the entire shape of the light curves . although sn 2013am also shows some similarities to sn 1999em , but it has a shorter duration of plateau " phase . to examine the evolution of the late - time light curves , we also measured the decline rates in @xmath31 bands during the period @xmath0220 - 320 days ( figure [ < lccomp > ] ) . for sn 2013am , we obtained @xmath32 = 0.42@xmath33 , @xmath34 = [email protected] , and @xmath35 = 0.54@xmath36 mag/@xmath37 . these values are smaller than that of sn 2005cs derived at similar phases ( i.e. , @xmath32 = 0.65@xmath36 , @xmath34 = [email protected] , and @xmath35 = 0.78@xmath36 mag/@xmath37 , derived from the light curves of sn 2005cs in @xcite ) . note that the decay rates derived for these two sne iip are smaller than that expected from the co@xmath38fe decay ( i.e. , @xmath39 mag/@xmath37 ) especially in the @xmath4 band . this phenomenon was also observed in low - luminous , low - energy events @xcite . it implies that the tail luminosity of sn 2013am and sn 2005cs do not fall directly onto the radioactive tail , which might be due to contribution from additional radiation generated in the warmer inner ejecta propagating throughout the transparent cooler external layers @xcite . the deviation of the last two points in @xmath40- and @xmath41- band perhaps arises from the photometric errors due to the contamination by host - galaxy light . excluding the last two points , the decay rates become as @xmath34 = [email protected] and @xmath35 = 0.79@xmath42 mag/@xmath37 . the @xmath19 uv light curves of sn 2013am are presented in figure [ < uvlc > ] , together with those of sn 2005cs @xcite , sn 2008 in @xcite , sn 2009n @xcite , and sn 2012a @xcite . in the uvw2 band , sn 2013am shows a somewhat flatter evolution than sn 2005cs ; and it is unclear whether this trend holds in the uvm2 and uvm1 bands for the lack of data . after @xmath3010 days from the explosion , sn 2013am seems to show a relatively higher luminosity than the comparison sne . uv light curves of sn 2013am with sne , 2005cs , 2008 in , 2009n and 2012a . all the light curves were normalised to the peak and in uvw1 and uvw2 bands were shifted by @xmath43 and @xmath44 mag , respectively . see text for details.,width=309 ] figure [ < cc > ] shows the color curves of sn 2013am , corrected for the galactic ( @xmath45=0.02 ) and host - galaxy reddening ( @[email protected] mag ) that is derived in section [ subsect : extmag ] ) . the color curves of the comparison sne are also plotted . we found that the color curves of sn 2013am show an evolution that is similar to the comparison sne iip during the plateau phase . this similarity of early color curves has been also noticed for a large sn iip sample ( i.e. , faran et al . 2014 ) . it is notable that the color of sne iip evolve towards red more quickly in bluer bands at the early phase owing to that the luminosity at shorter wavelengths is decreasing more quickly . in the nebular phase , the @xmath47 color is found to be unusually red in sn 2013am and sn 2005cs . on the other hand , @xcite found that the decline rate of the @xmath41-band light curve correlates with the ejecta velocity at @xmath48 days after explosion . they proposed that high ejecta velocities may indicate a fast decrease in density , which results in a quicker release of radiation . the released radiation would heat the sn for a while , which consequently slows down the decrease of the photospheric temperature of some sne iip . in other words , the temperature of sne iip with slower ejecta velocity tends to be cooling quicker . therefore , the unusual red @xmath47 color of faint sne iip might be a typical feature of low - velocity ( or low - luminosity ) sne iip . @xcite suggested using the @xmath49 color as a better reddening indicator because it is expected to be less sensitive to metallicity effects . @xcite found that the @xmath50 $ ] color curves of sne iip tend to evolve in a similar way during the plateau phase , which might allow us to determine the host - galaxy reddening that sn 2013am suffered by comparing with the sn sample shown in figure [ < cc > ] . as we do not have the near infrared photometry of sn 2013am , we apply this method to the @xmath49 color curve and derive a reddening of e@xmath51 = [email protected] mag for sn 2013am by comparing with the un - reddened @xmath49 template from krisciunas et al . furthermore , @xcite suggest that the host reddening can be derived with a precision of @xmath52 = 0.2 mag from the @xmath49 color toward the end of the plateau , which follows as : @xmath53 , \ ] ] where the @xmath49 color corresponds to the color of a given sn at 30 days before the end of plateau , corrected for k - terms and galactic reddening ; and @xmath54 is related with @xmath55 ( e.g. , @xmath56=2.518 for @xmath55=3.1 ; @xcite ) . based on the @xmath49 color of sn 2013am measured at the appropriate phase , the host - galaxy reddening is estimated as @[email protected] mag for @xmath55 = 3.1 and @[email protected] mag for @xmath55=1.4 . @xcite assumed a correlation between the @xmath41-band luminosity and velocity of ( in km s@xmath1 ) at @xmath58 days : @xmath59,\ ] ] where @[email protected] mag , @xmath61=6.69@xmath62 , @xmath63= 0.53 , and @xmath64=1.36 ( corresponding to r@xmath65 = 3.1 , @xcite ) . based on an expanded sample of sne iip , @xcite allowed r@xmath66 to be a free parameter and obtained @[email protected] mag , @xmath61=4.6@xmath67 , and @xmath68 ( corresponding to @xmath2 = 1.5@xmath62 ) . however , these two methods yield a similar estimate of the extinction @[email protected] mag ( see also table [ tab : extin ] ) under the assumption that sn 2013am follows the velocity - luminosity relation . however , this assumption may not be valid ( see discussions in section [ subsect : diver ] ) . thus , the reddening derived from the above empirical correlation are not adopted in our analysis . the values of a@xmath65 determined from the velocity - luminosity relation for r@xmath65 = 3.1 and 1.4 were listed in table 6 . besides the @xmath49 color curve , the absorption strength of na i lines in the spectra provides an alternative diagnostic of the host - galaxy reddening . in sn 2013am , an obvious d absorption feature can be detected in the early spectra of sn 2013am , as shown in figure [ < tril > ] . the equivalent width ( @xmath70 ) of the d absorption measured from our low - resolution spectra is [email protected] . several empirical correlations between reddening and @xmath70 of d have been applied to estimate the reddening as listed in table [ tab : extin ] , which yield a mean value of @xmath46 = 0.52 @xmath110.24 mag . the large error indicates that the correlation of reddening with the absorption strength of na i d lines is not very reliable @xcite . combining the results from the photometric and spectroscopic methods , we obtain the host - galaxy reddening as @xmath46 = [email protected] mag for sn 2013am . cccc color curve & [email protected]&@xmath71 & [email protected] + color curve & [email protected]&@xmath72 & [email protected] + color curve & [email protected]&formula 1 & [email protected] + color curve & [email protected]&formula 1 & [email protected] + velocity - luminosity & [email protected]&@xmath73 & [email protected] + velocityluminosity & [email protected]&@xmath74 & [email protected] + d & & @xmath75 & [email protected] + d & & @xmath76 & [email protected] + d & & @xmath77 & [email protected] + d & & @xmath78 & 0.63 @xmath110.30 + [ tab : extin ] a total of 12 spectra of sn 2013am taken by tng and ljt are plotted in figure [ < whsp > ] , which covers phases from @xmath79 to @xmath80 days after explosion . the early - time spectra are characterized by asymmetric h@xmath61 and h@xmath81 emission with prominent p - cygni profile . a weak @xmath825876 emission is also visible in the earlier spectra . the red continuum suggests significant line - of - sight reddening towards sn 2013am . spectral comparisons between sn 2013am and five well - observed sne iip at four selected epochs are shown in figure [ < spcomp > ] . all spectra have been corrected for redshift and reddening . at @xmath20 days , all of the comparison sne show very blue continuum , with dominant line features of h balmer and @xmath825876 . the minor absorption feature on the blue side of @xmath825876 is likely due to @xmath825679 . this feature is similarly identified in sn 2005cs @xcite and might relate to nitrogen enrichment @xcite . a relatively strong absorption in the spectrum of sn 2013am near 5900 is due to the host d , which implies a significant line - of - sight reddening . a notable change in the @xmath83 days spectrum is that the h@xmath61 absorption at @xmath0 6400 is well developed . the minor absorption on the blue wing can be attributed to @xmath826355 rather than high velocity feature of h@xmath61 @xcite . at this phase , only sn 2013am and sn 2005cs show strong features ( h@xmath84k and ir triplet ) . whereas this feature is not detectable in normal sne iip by @xmath302 weeks after explosion . the fact that absorptions are only detected in sn 2013am and sn 2005cs at this phase may indicate a larger ca / o ratio for their progenitors , which in turn points to a smaller progenitor mass because the ca / o ratio is expected to decrease with the progenitor mass @xcite . ( 6563 ) and ir triplet ( 8542 ) ; and the red dashed line is for @xmath826497.,width=302 ] for sn 2013am , compared with those of sne 2005cs , 1999em , 2008 in , 2009n , and 2012a . bottom panel : the evolution of ejecta velocities of sn 2013am inferred from @xmath825169 , @xmath826246 , @xmath826355 , h@xmath61 , @xmath827773 and @xmath828662 in the spectra . , width=302 ] by @xmath85 month , the supernova enters the plateau phase and the spectrum becomes red as the decrease of the photospheric temperature . the absorption feature of ca ii ir triplet is well developed . in particular , numerous narrow metal lines ( , , , , , and ) emerge to dominate the spectrum ( see @xcite for a detailed line identification ) . at @xmath86 + 80 days , the supernova enters near the end of the plateau phase when the continuum spectrum becomes progressively redder . the p - cygni profile of h lines becomes much narrower at this phase , reaching at an expansion velocity of about 1000 km s@xmath1 . the narrow metal lines become more noticeable in sn 2013am and other comparison objects . in particular , the prominent absorption features of lines are visible in sn 2013am at @xmath82 5853 , 6142 , and 6497 . these ba ii lines are also strong in sn 2005cs and sn 2009n , while they are weak in sn 1999em and sn 2012a . in sn 2008 in , the strength of lines is medium among the comparison sample . @xcite suggested that these strong lines are not owing to the over - abundance of ba but it is probably a temperature effect based on the spectra modelling of faint sn iip 1997d . this might imply that a inverse relationship exist between the strength of and the photospheric temperature of sne iip @xcite . the three components of ir triplet can be also resolved ( see also figure [ < tril > ] ) in sne 2013am , 2005cs , and 2009n at this phase . in principle , formation of narrower spectral lines can relate with the slow moving ejecta and/or a narrower line - forming region . in this subsection we examine the ejecta velocity of sn 2013am through h and some metal lines . the location of the blueshifted absorption minimum was measured using both the gaussian fit routine and the direct measurement of the absorption minimum , and the results were averaged . the upper panel of figure [ < vha > ] shows the velocity inferred from h@xmath87 . as it can be seen , sn 2013am has a much lower ejecta velocity relative to the normal sne iip . at t@xmath0 + 50 days after explosion , the expansion velocity of sn 2013am is @xmath02000 km s@xmath1 , while the corresponding value for the normal sne iip like sn 1999em is 5000 - 6000 km s@xmath1 . a similarly low expansion velocity is also seen in sn 2005cs , with a velocity of @xmath02300 km s@xmath1 at t@xmath88 + 44 days . note that the expansion velocity of sn 2009n is similar to that of sn 1999em at @xmath89days but it decreases rapidly after that and becomes comparable to sn 2005cs at t @xmath90 100 days . the spectral features and the velocity evolution of sn 2013am are overall similar to those of sn 2005cs . we further examined the the ejecta velocity inferred from other lines of sn 2013am , as shown in the lower panel of figure [ < vha > ] . we notice that this velocity relates to the strength of the corresponding lines owing to that the stronger lines typically form at a larger radius @xcite . in the earlier phase , h@xmath61 is the strongest line and has the highest velocity ; while at later phase , ir triplet is the strongest feature and at the highest velocity . applying the tully - fisher distance ( @xmath91 = [email protected] mag ; @xcite ) and the reddening derived in section 3.4 to the observed b@xmath92 and v@xmath92band peak magnitudes , we obtain the absolute magnitudes at maximum light as m@xmath93 mag and @xmath94 mag when @xmath55=3.1 is adopted in the extinction corrections . on the other hand , a lower extinction ratio , i.e. , r@xmath65 = 1.4 - 1.5 , has been suggested for some sne ia ( i.e. , @xcite ) and even for sne iip ( e.g. , @xcite ) in light of the statistical study . assuming r@xmath65 = 1.4 for the extinction correction , the absolute magnitudes of sn 2013am drop to @[email protected] mag in @xmath26 and @xmath95 mag in @xmath4 , which are close to those of sn 2005cs but are much lower than the normal sne iip . @xcite found a relation between the absolute peak brightness ( @xmath96 ) and plateau phase decline rate ( @xmath97 , defined in the fig.4 of @xcite ) of sne iip , where the fainter sn iip shows slower decline rate at the plateau phase . this relation is best fitted in the @xmath10-band via : @xmath98=(-12.5\pm1.1)+m_{peak}[r](0.44\pm0.06).\ ] ] the decline rate of sn 2013am is @xmath99 = -5.84\pm0.20 $ ] derived from the measurements listed in table [ tab : gri ] , which yields m@xmath100 mag conforming to the result of @xmath55=1.4 ( i.e. , m@xmath101 mag ) . using the ubvri data presented in the previous sections , we computed the bolometric light curve of sn 2005cs . figure [ < bolo > ] shows the pseudo - bolometric ( @xmath12 ) light curve of sn 2013am , together with those of sne 1999em , 2005cs , 2008 in , 2009n , 2012a , and 1987a . as the late - time photometry was only available in @xmath31 bands for sn 2013am , we interpolate the missing @xmath25- and @xmath26-band flux from the well - observed sne iip such as sn 1999em and sn 1987a at some phases whenever it is necessary . this assumption is reasonable since different sne iip tend to show similar color evolution ( except in @xmath49 ) at late times as shown in figure [ < cc > ] . from the bolometric light curve , we find that sn 2013am reached a peak luminosity of @xmath102 erg s@xmath1 ) ( for @xmath55 = 3.1 ) at about 4 days after explosion , which is comparable to sn 2008 in and sn 2009n . while correcting for the extinction with @xmath55=1.4 yields a lower value @xmath103 erg s@xmath1 , similar to that of sn 2005cs . note that the bolometric light curve of sn 2013am declines more slowly relative to other comparison sne in the first 2 - 3 weeks after explosion , which might be related to the slower decline in the @xmath104 bands . on the other hand , the late - time evolution seems to be very similar to that of sn 2005cs , specially at period from @xmath0200 to @xmath0350 days after explosion ( see inset of figure [ < bolo > ] ) . the decline rate on the exponential tail is estimated as 0.72@xmath105 mag ( 100d)@xmath1 , similar to that of sn 2005cs ( i.e. , 0.75@xmath106 mag ( 100d)@xmath1 ) , both are smaller than that expected from @xmath6co@xmath38fe decay . the tail luminosity of the light curve can be used to estimate the mass of @xmath6ni ejected by the sn . comparing the tail luminosity of sn 2013am with that of sn 1987a via the relation follows as @xcite : @xmath107 from the above relation , we estimate that sn 2013am ejected @xmath108m@xmath8 of @xmath6ni if @xmath55 = 3.1 is used , which is larger than typical value for low - velocity sne iip ( i.e. , @xmath109ni ) @xmath110m@xmath8 , @xcite ) but is comparable to that ejected by sn 2008 in ( i.e. , @xmath109ni ) = 0.015 m@xmath8 , @xcite ) and sn 2009n ( i.e. , @xmath109ni ) = 0.02 m@xmath8 , @xcite ) . @xcite presented an alternative way to calculate the tail luminosity using only the @xmath4-band light curve . this estimation is free from the uncertainty caused by interpolating the missing flux in the @xmath25- and @xmath26- bands for sn 2013am , which follows as @xmath111 where @xmath112 is the tail luminosity in erg s@xmath1 , @xmath113 is the @xmath4-band magnitude at the tail phase , @xmath69 is the total extinction in the @xmath4 band , and @xmath114 is the distance in cm . the bolometric correction is estimated as bc = [email protected] @xcite . once the tail luminosity is determined , the @xmath6ni mass can be estimated via @xmath115m_{\sun}\ ] ] where @xmath116 is the phase after explosion . inserting the @xmath4-band tail magnitude , the distance , and the extinction of sn 2013am into the above two formula , we obtain an estimate of the @xmath6ni mass as @xmath117m@xmath8 , which is in line with the above estimate . ) light curve of sn 2013am with sne 2005cs , 2008 in , 2009n , 2012a , 1999em , and 1987a . the dashed line represents the slope of the co@xmath38fe decay . the error bars shown for sn 2013am include the uncertainties in extinction correction , distance modulus , and photometry.the insets shows the comparison of scaled sn 2013am and sn 2005cs at late time.,width=302 ] to get better understanding of the properties of sn 2013am , we further examined its location in the correlations among @xmath6ni mass , expansion velocity of the ejecta and @xmath4-band absolute magnitude in the middle of the plateau ( m@xmath118 ) . figure[<scni > ] shows such correlations for thirty four sne iip , including eleven low - luminosity or low - velocity objects from @xcite ; twenty normal sne iip from @xcite , and two transition events sn 2008 in and sn 2009n . sn 1999em and sn 2012a are also included in the plot . note that the velocity of normal sne iip were derived from @xmath825169 @xcite and the rest samples derived from @xmath826246 , both could indicate the photospheric velocity of sne iip . moreover , @xcite suggested that @xmath826246 is a better indicator of the photospheric velocity than @xmath825169 owing to the lower optical depth . on the other hand , @xcite also indicated that the velocity of @xmath826246 is very close to that of @xmath825169 ( i.e. , @xmath119 = 0.95 @xmath120 ) . therefore , the original measures of @xmath120 from @xcite are adopted at here without correction . as shown in figure 11 , the ejecta velocity of sne iip is largely scattered , ranging from @xmath01500 km s@xmath1 to @xmath08000 km s@xmath1 . significant scatter also exists in the plateau luminosity and the mass of @xmath6ni produced in the explosion . the wide range in luminosities and expansion velocities is clear manifestation of the great diversity of sne iip . nevertheless , one can see that sne iip with larger ejecta velocities generally produce more @xmath6ni and are brighter during the plateau phase , which is the basis for the use of sne iip as distance indicators ( e.g. , @xcite ) . in the @xmath6ni velocity space , sn 2013am is clearly separated from the low - velocity group and produced a nickel mass that is comparable to those of the gap - filler " objects like sn 2008 in and sn 2009n . the presence of these transition objects confirms the hypothesis of a continuous distribution in explosion parameters of sne iip . in the m@xmath118velocity space , however , the separation of sn 2013am from other low - velocity sne iip is not distinct , though it appears still brighter for r@xmath65 = 3.1 . moreover , we notice that the low - velocity sample of sne iip seems to show large scatter in synthesized nickel mass and plateau luminosity in comparison with those with higher ejecta velocities . more sample are needed to confirm wether the photometric diversity of sne iip increases at low ejecta velocities . on the other hand , the variations in the expansion velocities can be in principle related to the composition of the progenitor star , the envelope mass , and even the viewing angles to observe the explosions . at lower metallicity , the line - forming region will go deeper into the photosphere to have the same line opacity , and thus it results in smaller line velocities @xcite . therefore , it is possible that the progenitor of sn 2013am has lower metallicity compared to sn 2008 in or sn 2009 in considering that they have similar tail luminosities . moreover , the large envelope mass ejected in the explosion can also lead to smaller line velocity for the sne iip with the same explosion energy . on the other hand , the smaller ejecta velocity seen in sn 2013am might be also due to that it was viewed at an angle away from the polar direction in the explosion . this can be examined with the polarization observations made during the nebular phase . 6246 or @xmath825169 measured at @xmath121 days after explosion ( i.e. , @xmath122 ) vs. m(@xmath6ni ) . lower : @xmath122 vs. the @xmath4-band absolute magnitudes measured at @xmath121 days after explosion ( i.e. , @xmath123 ) . the filled star represents sn 2013am dereddened with @xmath55=3.1 , while the open star represents the case with @xmath2=1.4 . sne 2005cs , 2012a , and 1999em were plotted by opened signs for better display . the red lines represent the linear fit to the observed data.,width=302 ] we present the optical and ultraviolet observations of the low - velocity sn iip 2013am in nearby galaxy m65 . the spectra of sn 2013am are characterized by narrow p - cygni lines at lower velocities , and the overall evolution is very similar to that seen in the proto - type low - velocity ii - p supernova 2005cs . however , sn 2013am has shorter plateau " duration and brighter tail compared to sn 2005cs , and it shows close resemblance to the transition objects like sn 2009n and sn 2008 in . both photometric and spectroscopic methods suggest that sn 2013am suffered a significant reddening in the host galaxy ( i.e. , @xmath124 mag ) . after corrections for the extinction using the above reddening and @xmath2=3.1 , we find that sn 2013am has an absolute @xmath4-band magnitude of @xmath125 mag in the middle of the plateau , a @xmath12 bolometric luminosity of @xmath102 erg s@xmath1 ) around the maximum light , and an ejected @xmath6ni mass of @xmath7m@xmath8 . these values are comparable to those derived for the gap - filler " events sn 2009n and sn 2008 in , although the variation in r@xmath65 may cause additional uncertainties in these estimations . the higher luminosity of sn 2013am , compared with sn 2005cs , is also consistent with the report that sne iip with shorter plateau " durations tend to be more luminous @xcite . these results indicate that the velocity - luminosity relation may not hold for all sne iip . the ejecta velocity may be related to variations in metallicity of the progenitor star , the envelope mass , and even the viewing angles to observe the explosions . a better understanding of the properties of the progenitors and polarization observations would help make further distinguish between the above factors . we thank very much the anonymous referee for his / her constructive suggestions which helped to improve the paper a lot . we also acknowledge the support of the staff of the li - jiang 2.4-m telescope ( ljt ) , liverpool 2-m telescope ( lt ) , tsinghua - naoc 80-cm telescope ( tnt ) and 3.58-m telescopio nazionale galileo ( tng ) . funding for the ljt has been provided by chinese academe of science ( cas ) and the people s government of yunnan province . the lt is operated on the island of la palma by liverpool john moores university in the spanish observatorio del roque de los muchachos of the instituto de astrofisica de canarias with financial support from the uk science and technology facilities council . the tnt is owned by tsinghua university and jointly operated by tsinghua university and the national astronomical observatory of the chinese academy of sciences ( naoc ) . the tng is operated on the island of la palma by the fundacin galileo galilei of the inaf ( istituto nazionale di astrofisica ) . the data of uvot comes from the @xmath19 data center . we also very much thank a. pastorello , r. roy , k. takts , and l. tomasella for providing their spectra of sn 2005cs , sn 2008 in , sn 2009n , and sn 2012a . zhang is supported by the national natural science foundation of china ( nsfc , grant 11403096 ) . the work of x.f . wang is supported by the major state basic research development program ( 2013cb834903 ) , the nsfc ( grants 11073013 , 11178003 , 11325313 ) , tsinghua university initiative scientific research program , and the strategic priority research program the emergence of cosmological structures " of the chinese academy of sciences ( grant no . xdb09000000 ) . the work of j. m. bai is supported by the nsfc ( grants 11133006 , 11361140347 ) and the strategic priority research program the emergence of cosmological structures " of the chinese academy of sciences ( grant no . xdb09000000 ) . zhang is supported by nsfc ( grant 11203034 ) . wang is supported by nsfc ( grant 11303085 ) , and the western light youth project and the open research program of the key laboratory for research in galaxies and cosmology , cas . zhang is supported by nsfc ( grant 11203070 ) . wang is supported by nsfc ( grant 11103078 ) . anderson , j. , et al . 2014 , , 786 , 67 barbon r. , benetti s. , rosino l. , cappellaro e. , turatto m. , 1990 , , 237 , 79 benetti , s. , turatto , m. , balberg , s. , et al . 2001 , , 322 , 361 benetti , s. , tomasella , a. , pastorello , e. , et al . 2013 , cbet 3440 , 2 brown , p.j . , dessart , l. , holland , s.t . , 2007 , , 659 , 1488 brown , p.j . , breeveld , a. a. , holland , s , et al . 2014 , ap&ss , 301 cardelli et al . 1989 , apj , 345 , 245 catchpole , r.m . , 1987 , , 229 , 15 catchpole , r.m . , et al . 1988 , , 231 , 75 cousins , a. 1981 , south african astron . , 6 , 4 danziger i. j. , 1988 , snoy.conf , 3 dessart , l. , livne , e. , & waldman , r. 2010 , mnras , 408 , 827 elmhamdi , a. , danziger , i. j. , chugai , n. , 2003 , , 338 , 939 fan , y.f . , bai , j.m . , zhang , j.j . , et al . 2014 , research in astron . astrophys , in press faran , t. , poznanski , d. , filippenko , a.v . 2014 , arxiv:1404.0378 fukugita , m. , ichikawa , t. , gunn , j. e. , et al . 1996 , , 111 , 1748 gehrels , n. , chincarini , g. , giommi , p. et al . 2004 , , 611 , 1005 hamuy , m. , pinto , p.a . , maza , j. , et al . 2001 , , 558 , 615 hamuy , m. 2001 , ph.d . thesis , univ . arizona hamuy , m. 2003 , , 582 , 905 hatano k. , branch d. , fisher a. , millard j. , baron e. , 1999 , apjs , 121 , 233 houck , j. , & fransson , c. 1996 , , 456 , 811 hillier d.j . , & miller , d.l . 1998 , , 496 , 407 huang , f. , li , j.z . , wang , x.f . , et al . 2012 , research in astron . astrophys , 11 , 1585 jester , s. , schneider , d.p . , richards , g.t . , et al . 2005 , , 130 , 873 jordi , k. , et al . 2006 , , 460 , 339 johnson , h. , iriarte , b. , mitchell , r. , @xmath84 wisniewskj , w. 1966 , commun . lunar planet . lab . , 4 , 99 krisciunas , k. , hamuy , m. , suntzeff , n.b . , et al . 2009 , , 137 , 34 landolt , a. v. 1992 , aj , 104 , 340 lentz , e.j . , baron , e. , branch , d. , et al . 2000 , , 530 , 966l leonard , d.c . , filippenko , a.v . , gates , e.l . , 2002 , pasp , 114 , 35 maguire , k. , di carlo , e. , smartt , s.j . , 2010 , , 404 , 981 mazzali , p.a . , lucy , l.b . , @xmath84 butler , k. , 1992 , , 258 , 399 menzies , j.w . 1987 , , 227 , 39 nasonova , o. g. , de freitas pacheco , j. a. , @xmath84 karachentsev , i. d. 2011 , aa , 532 , 104 nugent , p. , sullivan , m. , ellis , r. , 2006 , , 645 , 841 olivares , f. , hamuy , m. , pignata , g. , et al . 2010 , , 715 , 833 pastorello , a . , zampieri , l. , turatto , m. , et al . 2004 , , 347,74 pastorello , a . , sauer , d. , taubenberger , s. , et al . 2006 , , 370 , 1752 pastorello , a. , valenti , s. , zampieri , l. , et al . 2009 , , 394 , 2266 poole , t. , breeveld , a. , page , m. , et al . 2008 , , 383 , 627 poznanski , d. , et al . 2009 , , 694 , 1067 poznanski , d. , ganeshalingam , m. , silverman , j.m . , @xmath84 filippenko , a.v . 2011 , , 415 , l81 pritchard , t.a . , roming , p. , brown , p. , et al . 2014 , , 787 , 157 roming , p. , kennedy , t. , mason , k. , et al . 2005 , space sci . rev . , 120 , 95 roy r. , et al . , 2011 , , 736 , 76 sanders , n.e . , soderberg , a.m. , gezari , s. , et al . 2014 , arxiv:1404.2004 schlegel , d. , finkbeiner , d. , and davis , m. , 1998 , , 500 , 525 smartt , s.j . 47 , 63 spiro , s. , pastorello , a. , pumo , m.l . , 2014 , , 439 , 2873 stetson , p. 1987 , , 99 , 191 sugano , m. , 2013 , cebt 3440 , 1 takts , k. , pumo , m. l. , elias - rosa , n. , et al . 2014 , , 438 , 368 tomasella , l. , cappellaro , e. , fraser , m. , et al . 2013 , , 434 , 1636 turatto , m. , mazzali , p.a . , young , t.r . , 1998 , , 498 , l129 turatto , m. , benetti , s. , & cappellaro , e. , 2003 , in from twilight to highlight : the physics of supernovae , hillebrandt , w. , leibundgut , b. , eds . , p.200 utrobin , v. p. 2007 , , 461 , 233 wang , x. , et al . 2008 , , 675 , 626 wang , x. , et al . 2009 , , 699 , l139 woosley s. e. , hartmann d. , pinto p. a.1989 , apj , 346 , 395 whitelock , p.a . , et al . 1988 , , 234 , 5 yaron , o. , gal - yam , a. , fox , o.d . , et al . 2013 , atel 4910 . zhang , j.j . y.f . , chang , l. , et al . , 2012 , astronomical research technology , 9,411 zhang , j.j , wang , x.f . , bai , j.m . , et al . , 2014 , , 148 , 1

we derive predictions for the nadir angle ( ) dependence of the ratio of the rates of thelike andlike multi - gev events measured in water - erenkov detectors in the case of 3-neutrino oscillations of the atmospheric ( ) and ( ) , driven by one neutrino mass squared difference , . this ratio is particularly sensitive to the earth matter effects in the atmospheric neutrino oscillations , and thus to the values of and , and being the neutrino mixing angle limited by the chooz and palo verde experiments and that responsible for the dominant atmospheric ( ) oscillations . it is also sensitive to the type of neutrino mass spectrum which can be with normal ( ) or with inverted ( ) hierarchy . we show that for , and at , the earth matter effects modify substantially the dependence of the ratio and in a way which can not be reproduced with and a different value of . for normal hierarchy the effects can be as large as for , can reach in the earth core bin , and might be observable . they are typically by smaller in the inverted hierarchy case . an observation of the earth matter effects in the nadir angle distribution of the ratio would clearly indicate that and . ref . sissa 28/03/ep ftuv-03 - 0513 ific/03 - 21 * atmospheric neutrino oscillations , and neutrino mass hierarchy * j. bernabu , sergio palomares - ruiz and s. t. petcov departamento de fsica terica and ific , universidad de valencia - csic , 46100 burjassot , valencia , spain + scuola internazionale superiore di studi avanzati and istituto nazionale di fisica nucleare , i-34014 trieste , italy

hep-ph0305152

0.6truecm the publication of the first results of the kamland experiment marks the beginning of a new era in the studies of neutrino mixing and oscillations - the era of high precision determination of the neutrino mixing parameters . the data obtained by the solar neutrino experiments homestake , kamiokande , sage , gallex / gno and super - kamiokande ( sk ) @xcite provided the first strong evidences for oscillations of flavour ( electron ) neutrinos . strong evidences for oscillations of the atmospheric @xmath31 ( @xmath32 ) neutrinos were obtained by the super - kamiokande ( sk ) experiment @xcite . the evidences for solar @xmath4 oscillations into active neutrinos @xmath33 , were significantly reinforced during the last two years i ) by the combined first data of the sno experiment @xcite , and the sk data @xcite , ii ) by the more recent sno neutral current data @xcite , and iii ) by the first results of the kamland @xcite experiment . the kamland data practically establishes @xcite , under the plausible assumption of cpt - invariance , the large mixing angle ( lma ) msw solution as unique solution of the solar neutrino problem . this result brings us , after more than 30 years of research , initiated by the pioneer works of b. pontecorvo @xcite and the experiment of r. davis et al . @xcite , very close to a complete understanding of the true cause of the solar neutrino problem . the interpretation of the solar and atmospheric neutrino , and of the kamland data in terms of neutrino oscillations requires the existence of 3-neutrino mixing in the weak charged lepton current ( see , e.g. , @xcite ) : @xmath34 here @xmath35 , @xmath36 , are the three left - handed flavor neutrino fields , @xmath37 is the left - handed field of the neutrino @xmath38 having a mass @xmath39 and @xmath40 is the pontecorvo - maki - nakagawa - sakata ( pmns ) neutrino mixing matrix @xcite , @xmath41 where we have used a standard parametrization of @xmath40 with the usual notations , @xmath42 , @xmath43 , and @xmath44 is the dirac cp - violation phase .. ] if we identify @xmath45 and @xmath46 with the neutrino mass squared differences which drive the solar and atmospheric neutrino oscillations , @xmath47 , @xmath48 , the data suggest that @xmath49 . in this case @xmath50 and @xmath12 , represent the neutrino mixing angles responsible for the solar and atmospheric neutrino oscillations , @xmath51 , @xmath52 , while @xmath11 is the angle limited by the data from the chooz and palo verde experiments @xcite . combined @xmath53 and @xmath54 oscillation analyses of the solar neutrino and kamland @xcite data , performed under the assumption of cpt - invariance which we will suppose to hold throughout this study , show @xcite that the data favor the lma msw solution with @xmath55 and @xmath56 . a second , statistically somewhat less favored lma solution ( lma ii ) exists at @xcite @xmath57 . the atmospheric neutrino data , as is well known , is best described @xcite in terms of dominant @xmath58 ( @xmath59 ) oscillations with @xmath60 and @xmath61 . the 90% c.l . allowed intervals of values of the two - neutrino oscillation parameters found in @xcite read @xmath62 and @xmath63 . according to the more recent combined analysis of the data from the sk and k2k experiments @xcite , one has @xmath64 . we will often use in our analysis as illustrative the values @xmath65 and @xmath66 . let us note that the atmospheric neutrino and k2k data do not allow one to determine the signs of @xmath67 , and of @xmath68 when @xmath69 . this implies that in the case of 3-neutrino mixing one can have @xmath15 or @xmath16 . the two possibilities correspond to two different types of neutrino mass spectrum : with normal hierarchy ( nh ) , @xmath70 , and with inverted hierarchy ( ih ) , @xmath71 . the fact that the sign of @xmath68 is not determined when @xmath69 implies that when , e.g. , @xmath72 , two values of @xmath10 are possible , @xmath73 . in what regards the mixing angle @xmath11 , a 3-@xmath74 oscillation analysis of the chooz data @xcite led to the conclusion that for @xmath75 , the limits on @xmath9 practically coincide with those derived in the 2-@xmath74 oscillation analysis in @xcite . a combined 3-@xmath74 oscillation analysis of the solar neutrino , chooz and the kamland data , performed under the assumption of @xmath76 ( see , e.g. , @xcite ) , showed @xcite that @xmath77 at 99.73% c.l . the authors of @xcite found the best - fit value of @xmath9 to lie in the interval @xmath78 . getting more precise information about the value of the mixing angle @xmath11 , determining the sign of @xmath67 , or the type of the neutrino mass spectrum ( with normal or inverted hierarchy ) , and measuring the value of @xmath10 with a higher precision is of fundamental importance for the progress in the studies of neutrino mixing . the mixing angle @xmath11 , or the absolute value of the element @xmath79 of the pmns matrix , @xmath80 , plays a very important role in the phenomenology of the 3-neutrino oscillations . it drives the sub - dominant @xmath81 ( @xmath82 ) oscillations of the atmospheric @xmath31 ( @xmath32 ) and @xmath4 ( @xmath5 ) @xcite . the value of @xmath11 controls also the @xmath83 , @xmath84 , @xmath85 and @xmath86 transitions in the long baseline neutrino oscillation experiments ( minos , cngs ) , and in the widely discussed very long baseline neutrino oscillation experiments at neutrino factories ( see , e.g. , @xcite ) . the magnitude of the t - violating and cp - violating probabilities in neutrino oscillations is directly proportional to @xmath87 ( see , e.g. , @xcite ) . thus , in the sub - dominant channels of interest to t- and cp - violation studies , the corresponding asymmetries become proportional to @xmath88 @xcite . the value of @xmath89 is thus of prime importance . if the neutrinos with definite mass are majorana particles ( see , e.g. , @xcite ) , the predicted value of the effective majorana mass parameter in neutrinoless double @xmath90decay depends strongly in the case of hierarchical neutrino mass spectrum on the value of @xmath9 ( see , e.g. , @xcite ) . the sign of @xmath67 determines , for instance , which of the transitions ( e.g. , of atmospheric neutrinos ) @xmath83 and @xmath85 , or @xmath84 and @xmath86 , can be enhanced by the earth matter effects @xcite . the predictions for the neutrino effective majorana mass in neutrinoless double @xmath90decay depend critically on the type of the neutrino mass spectrum ( normal or inverted hierarchical ) @xcite . the knowledge of the value of @xmath11 and of the sign of @xmath48 is crucial for the searches for the correct theory of neutrino masses and mixing as well . somewhat better limits on @xmath91 than the existing one can be obtained in the minos experiment @xcite . various options are being currently discussed ( experiments with off - axis neutrino beams , more precise reactor antineutrino and long base - line experiments , etc . , see , e.g. , @xcite ) of how to improve by at least an order of magnitude , i.e. , to values of @xmath92 or smaller , the sensitivity to @xmath9 . the sign of @xmath67 can be determined in very long base - line neutrino oscillation experiments at neutrino factories ( see , e.g. , @xcite ) , and , e.g , using combined data from long base - line oscillation experiments at the jhf facility and with off - axis neutrino beams @xcite . if the neutrinos with definite mass are majorana particles , it can be determined by measuring the effective neutrino majorana mass in neutrinoless double @xmath90decay experiments @xcite . under certain rather special conditions it might be determined also in experiments with reactor @xmath5 @xcite . in the present article we study possibilities to obtain information on the value of @xmath9 and on the sign of @xmath48 using the atmospheric neutrino data , accumulated by the sk experiment , and more generally , that can be provided by future water - erenkov detectors , like uno and hyper - kamiokande . we consider 3-neutrino oscillations of the atmospheric @xmath31 , @xmath32 , @xmath4 and @xmath5 under the condition latexmath:[${\mbox{$ \delta m^2_{\odot}$}}= \delta m^2_{21 } \ll suggested to hold by the current solar and atmospheric neutrino data . under the indicated condition , the expressions for the probabilities of @xmath83 ( @xmath85 ) and @xmath84 ( @xmath86 ) transitions contain @xmath10 as a factor , which determines their maximal values . depending on the sign of @xmath46 , the earth matter effects can resonantly enhance either the @xmath83 and @xmath85 , or the @xmath84 and @xmath94 transitions if @xmath95 . the effects of the enhancement can be substantial for @xmath96 . they are larger in the multi - gev @xmath97-like and @xmath2like samples of events and for atmospheric neutrinos with relatively large path length in the earth , crossing deeply the mantle @xcite or the mantle and the core @xcite , i.e. , for @xmath19 , where @xmath0 is the nadir angle characterizing the neutrino trajectory in the earth . the @xmath83 ( @xmath84 ) and @xmath85 ( @xmath86 ) transitions in the earth lead to the reduction of the rate of the multi - gev @xmath2like events and to the increase of the rate of the multi - gev @xmath3like events in the super - kamiokande ( or any other water - erenkov ) detector with respect to the case of absence of these transitions ( see , e.g. , @xcite ) . correspondingly , as observables which are sensitive to the earth matter effects , and thus to the value of @xmath9 and the sign of @xmath46 , as well as to @xmath10 , we consider the nadir - angle distributions of the ratios @xmath98 and @xmath99 , where @xmath100 and @xmath101 and are the multi - gev @xmath2like and @xmath97-like numbers of events ( or event rates ) in the case of 3-@xmath74 oscillations of the atmospheric @xmath4 , @xmath5 and @xmath31 , @xmath32 , and @xmath102 is the number of @xmath3like events in the case of absence of oscillations ( @xmath21 ) . the ratio of the energy and nadir angle integrated @xmath2like and @xmath3like events , @xmath1 , has been measured with a relatively high precision by the sk experiment @xcite . the systematic uncertainty , in particular , in the nadir angle dependence of the ratio @xmath1 can be smaller than those on the measured nadir angle distributions of @xmath2like and @xmath3like events , @xmath103 and @xmath104 . we obtain predictions for the nadir - angle distributions of @xmath98 and of @xmath99 both for neutrino mass spectra with normal ( @xmath15 ) and inverted ( @xmath16 ) hierarchy , @xmath105 , @xmath106 , @xmath107 and @xmath108 , and for @xmath109 . we compare the latter with the predicted nadir - angle distributions i ) of the ratio @xmath110 for the case the 3-neutrino oscillations taking place in vacuum , @xmath111 , and ii ) of the ratio @xmath112 , where @xmath113 is the predicted number of @xmath2like events in the case of 2-neutrino @xmath58 and @xmath59 oscillations of the atmospheric @xmath31 and @xmath32 . predictions for the different types of ratios indicated above of the suitably integrated nadir angle distributions of the @xmath2like and @xmath3like multi - gev events are also given . our results show , in particular , that for @xmath27 @xmath17 and @xmath15 , the effects of the earth matter enhanced @xmath83 and @xmath85 transitions of the atmospheric @xmath31 and @xmath4 , might be observable with the super - kamiokande detector . conversely , if the indicated effects are indeed observed in the super - kamiokande experiment , that would suggest in turn that @xmath17 , @xmath27 and that the neutrino mass spectrum is with normal hierarchy , @xmath15 . let us note finally that the earth matter effects have been widely investigated ( for a recent detailed study see , e.g. , ref . @xcite which contains also a rather complete list of references to earlier work on the subject ) . however , the study of the magnitude of the earth matter effects in the nadir angle distribution of the ratio of the multi - gev @xmath2like and @xmath3like events , measured in water - erenkov detectors , performed here overlaps very little with the earlier investigations . 0.6truecm in the present section we summarize the results on the oscillations of atmospheric neutrinos crossing the earth , which we use in our analysis . 0.6truecm the @xmath114 ( @xmath115 ) and @xmath116 ( @xmath117 ) oscillations of atmospheric neutrinos should exist and their effects could be observable if genuine three - flavour - neutrino mixing takes place in vacuum , i.e. , if @xmath118 , and if @xmath119 is sufficiently large @xcite ( see also , e.g. , @xcite ) . under the condition @xmath49 , which the neutrino mass squared differences determined by the existing atmospheric and solar neutrino and kamland data satisfy , the relevant three - neutrino @xmath114 ( @xmath115 ) and @xmath116 ( @xmath117 ) transition probabilities reduce effectively to a two - neutrino transition probability @xcite . with @xmath46 and @xmath120 playing the role of the relevant two - neutrino oscillation parameters . thus , searching for the effects of @xmath114 ( @xmath115 ) and @xmath116 ( @xmath117 ) transitions of atmospheric neutrinos , amplified by earth matter effects , can provide unique information , in particular , about the magnitude of @xmath9 . as is not difficult to show using the results of @xcite , the 3-neutrino oscillation probabilities of interest for atmospheric @xmath121 having energy @xmath122 and crossing the earth along a trajectory characterized by a nadir angle @xmath123 , have the following form ( see also , e.g. , @xcite ) : @xmath124 @xmath125 @xmath126 @xmath127 , \label{p3mumu}\ ] ] @xmath128 here @xmath129 is the probability of two - neutrino oscillations in the earth which coincides in form with , e.g. , the two - neutrino @xmath130 transition probability , @xmath131 , but describes @xmath132 transitions , where @xmath133 @xcite , and @xmath134 and @xmath135 are known phase and two - neutrino transition probability amplitude . analytic expressions for @xmath136 , @xmath134 and @xmath137 will be given later . using eqs . ( [ p3ee ] ) - ( [ p3mutau ] ) it is not difficult to convince oneself that the fluxes of atmospheric @xmath121 of energy @xmath122 , which reach the detector after crossing the earth along a given trajectory specified by the value of @xmath123 , @xmath138 , are given by the following expressions in the case of the three - neutrino oscillations under discussion @xcite : @xmath139~p_{2\nu}\right ) , \label{phie}\ ] ] @xmath140~p_{2\nu } - 2c^2_{23}s^2_{23}~\left [ 1 - re~ ( e^{-i\kappa } a_{2\nu}(\nu_{\tau } \rightarrow \nu_{\tau } ) ) \right ] \right ) ~ , \label{phimu}\ ] ] where @xmath141 is the @xmath142 flux in the absence of neutrino oscillations and @xmath143 the interpretation of the sk atmospheric neutrino data in terms of @xmath58 oscillations requires the parameter @xmath144 to lie approximately in the interval ( 0.30 - 0.70 ) , with 0.5 being the statistically preferred value . for the predicted ratio @xmath145 of the atmospheric @xmath31 and @xmath4 fluxes for i ) the earth core crossing and ii ) only mantle crossing neutrinos , having trajectories for which @xmath146 , one has @xcite : @xmath147 for the neutrinos giving contribution to the sub - gev samples of super - kamiokande events , and @xmath148 for those giving the main contribution to the multi - gev samples . if @xmath149 and @xmath150 , we have @xmath151 and the possible effects of the @xmath114 and @xmath116 transitions on the @xmath4 and @xmath31 fluxes , and correspondingly on the sub - gev @xmath3like sample of events , would be strongly suppressed even if these transitions were maximally enhanced by the earth matter effects . for the multi - gev neutrinos we have @xmath152 for @xmath153 . the factor @xmath154 , for instance , amplifies the effect of the @xmath114 transitions in the @xmath3like sample for @xmath155 , for which @xmath156 @xcite . the same conclusions are valid for the effects of oscillations on the fluxes of , and event rates due to , atmospheric antineutrinos : @xmath5 and @xmath32 . actually , the formulae for anti - neutrino fluxes and oscillation probabilities are analogous to those for neutrinos : they can be obtained formally from eqs . ( [ p3ee ] ) - ( [ r ] ) by replacing the neutrino related quantities - probabilities , @xmath134 , @xmath157 and fluxes , with the corresponding quantities for antineutrinos : @xmath158 , @xmath159 , @xmath160 , @xmath161 , @xmath162 and @xmath163 . equations ( [ p3ee ] ) - ( [ p3mumu ] ) , ( [ phie ] ) - ( [ phimu ] ) and the similar equations for antineutrinos imply that in the case under study the effects of the @xmath114 , @xmath115 , and @xmath116 , @xmath117 , oscillations i ) increase with the increase of @xmath144 and are maximal for the largest allowed value of @xmath144 , ii ) should be considerably larger in the multi - gev samples of events than in the sub - gev samples , iii ) in the case of the multi - gev samples , they lead to an increase of the rate of @xmath97-like events and to a slight decrease of the @xmath2like event rate . this discussion suggests that the quantity most sensitive to the effects of the oscillations of interest should be the ratio of the @xmath2like and @xmath3like multi - gev events ( or event rates ) , @xmath110 . the magnitude of the effects we are interested in depends also on the 2-neutrino oscillation probabilities , @xmath164 and @xmath165 . in the case of oscillations in vacuum we have @xmath166 . given the existing limits on @xmath119 , the probabilities @xmath136 and @xmath167 can not be large if the oscillations take place in vacuum . however , @xmath136 or @xmath167 can be strongly enhanced by the earth matter effects . the latter depend on the earth density profile and we will discuss it next briefly . 0.6truecm as is well - known , the earth density distribution in the existing earth models is assumed to be spherically symmetric and there are two major density structures - the core and the mantle , and a certain number of substructures ( shells or layers ) . the core radius and the depth of the mantle are known with a rather good precision and these data are incorporated in the earth models . according to the stacey 1977 and the more recent prem models @xcite , which are widely used in the calculations of the probabilities of neutrino oscillations in the earth , the core has a radius @xmath168 km , the earth mantle depth is approximately @xmath169 km , and the earth radius is @xmath170 km . the mean values of the matter densities of the core and of the mantle read , respectively : @xmath171 and @xmath172 . all the interesting features of the atmospheric neutrino oscillations in the earth can be understood quantitatively in the framework of the two - layer model of the earth density distribution @xcite . the density profile of the earth in the two - layer model is assumed to consist of two structures - the mantle and the core , having different densities , @xmath173 and @xmath174 , and different electron fraction numbers , @xmath175 and @xmath176 , none of which however vary within a given structure . the densities @xmath173 and @xmath174 in the case of interest should be considered as mean effective densities along the neutrino trajectories , and they vary somewhat with the change of the trajectory @xcite : @xmath177 and @xmath178 . in the prem and stacey models one has for @xmath179 : @xmath180 and @xmath181 . for the electron fraction numbers in the mantle and in the core , where @xmath104 , @xmath182 and @xmath183 are the electron , proton and neutron number densities in matter , respectively . ] one can use the standard values @xcite @xmath184 and @xmath185 . numerical calculations show @xcite ( see also @xcite ) that , e.g. , the @xmath85 oscillation probability of interest , calculated within the two - layer model of the earth with @xmath186 and @xmath187 for a given neutrino trajectory determined using the prem ( or the stacey ) model , reproduces with a remarkably high precision the corresponding probability calculated by solving numerically the relevant system of evolution equations with the much more sophisticated earth density profile of the prem ( or stacey ) model . the resonance densities in the mantle and in the core can be obtained from the expressions @xmath211 @xmath212 being the nucleon mass , by using the specific values of @xmath213 in the mantle and in the core . we have @xmath214 because @xmath185 and @xmath184 @xcite ( see also @xcite ) . obviously , @xmath215 . the phase @xmath134 and the probability amplitude @xmath137 which appear in eq . ( [ phimu ] ) for the flux of atmospheric @xmath31 in the case of three - flavour neutrino mixing and hierarchy between the neutrino mass squared differences and therefore can play important role in the interpretation of the , e.g. , super - kamiokande atmospheric neutrino data , have the following form in the two - layer model of the earth density distribution @xcite : @xmath216 - { \delta m^2_{21}\over{2e}}x\cos2\theta_{12 } , \label{kappa}\ ] ] @xmath217 \cos^2\theta'_{m}~ \label{a2nu}\ ] ] @xmath218 where from eq . ( 1 ) in ref . @xcite by formally setting @xmath219 while keeping @xmath220 and @xmath221 arbitrary , and then interchanging @xmath222 ( @xmath223 ) and @xmath224 ( @xmath225 ) . ] @xmath226 . the expressions for @xmath167 , @xmath227 and @xmath228 can be obtained from the corresponding expressions for neutrinos by replacing @xmath229 with ( -@xmath229 ) in eqs . ( [ de ] ) and ( [ thetam ] ) . 0.6truecm in the two - layer model , the oscillations of atmospheric neutrinos crossing only the earth mantle ( but not the earth core ) , correspond to oscillations in matter with constant density . the relevant expressions for @xmath136 , @xmath134 and @xmath157 follow from eqs . ( [ p2nu ] ) , ( [ kappa ] ) and ( [ a2nu ] ) by setting @xmath230 and using @xmath231 . the expressions for @xmath136 ( @xmath167 ) and @xmath137 ( @xmath232 ) have the standard well - known form . in the case under study @xmath11 plays the role of a two - neutrino vacuum mixing angle in the probabilities @xmath136 and @xmath167 . since @xmath77 , we have @xmath233 . consequently , the earth matter effects can resonantly enhance @xmath136 for @xmath15 and @xmath167 if @xmath16 @xcite . due to the difference of cross sections for neutrinos and antineutrinos , approximately 2/3 of the total rate of the @xmath2like and @xmath3like multi - gev atmospheric neutrino events in the sk ( and in any other water - erenkov ) detector , i.e. , @xmath234 and @xmath235 , are due to neutrinos @xmath31 and @xmath4 , respectively , while the remaining @xmath236 of the multi - gev event rates , i.e. , @xmath237 and @xmath238 , are produced by antineutrinos @xmath32 and @xmath5 . this implies that the earth matter effects in the multi - gev samples of @xmath2like and @xmath3like events will be larger if @xmath15 , i.e. , if the neutrino mass spectrum is with normal hierarchy , than if @xmath16 and the spectrum is with inverted hierarchy . thus , the ratio @xmath1 of the multi - gev @xmath2like and @xmath3like event rates measured in the sk experiment is sensitive , in principle , to the type of the neutrino mass spectrum . consider for definiteness the case of @xmath15 . it follows from eqs . ( [ phie ] ) and ( [ phimu ] ) that the oscillation effects of interest will be maximal if @xmath239 . the latter is possible provided i ) the well - known resonance condition @xcite , leading to @xmath240 , is fulfilled , and ii ) @xmath241 . given the values of @xmath186 and @xmath175 , the first condition determines the neutrino energy at which @xmath136 can be enhanced : @xmath242~ ( { \rm \bar{n}_e^{man}[n_a cm^{-3}]})^{-1 } \cos2\theta_{13}~{\rm gev } , \label{eres}\ ] ] where @xmath243 $ ] is the value of @xmath46 in units of @xmath244 and @xmath245 $ ] is the electron number density , @xmath246 , in units of @xmath247 , @xmath248 being the avogadro number . if the first condition is satisfied , the second determines the length of the path of the neutrinos in the mantle for which one can have @xmath239 : @xmath249}~2x'[10^{4}~{\rm km}]~ , \label{xman}\ ] ] where @xmath205 is in units of @xmath250 . taking @xmath251 , @xmath252 and @xmath253 one finds from eq . ( [ eres ] ) : @xmath254 . the width of the resonance in @xmath122 , 2@xmath255 , is determined , as is well - known , by @xmath256 : @xmath257 . for @xmath258 , the resonance is relatively wide in the neutrino energy : @xmath259 . equation ( [ xman ] ) implies that for @xmath260 and @xmath261 , one can have @xmath239 only if @xmath262 . it follows from the above simple analysis @xcite that the earth matter effects can amplify @xmath136 significantly when the neutrinos cross only the mantle i ) for @xmath263 gev , i.e. , in the multi - gev range of neutrino energies , and ii ) only for sufficiently long neutrino paths in the mantle , i.e. , for @xmath19 . the magnitude of the matter effects in the ratio @xmath1 of interest increases with increasing @xmath9 . the same results , eqs . ( [ eres ] ) and ( [ xman ] ) , and conclusions are valid for the antineutrino oscillation probability @xmath167 in the case of @xmath16 . as a consequence , the ideal situation for distinguishing the type of mass hierarchy would be a detector with charge discrimination . 0.6truecm in this case @xmath136 , @xmath134 and @xmath137 are given by eqs . ( [ p2nu ] ) , ( [ kappa ] ) and ( [ a2nu ] ) . in the discussion which follows we will assume for concretness that @xmath15 , consider the probability @xmath136 and the transitions of neutrinos . if @xmath16 , the results we will briefly discuss below will be valid for the probability @xmath167 and the transitions of antineutrinos . for @xmath77 and @xmath15 , we can have @xmath239 _ only _ due to the effect of maximal constructive interference between the amplitudes of the the @xmath132 transitions in the earth mantle and in the earth core @xcite . the effect differs from the msw one @xcite . the _ mantle - core enhancement effect _ is caused by the existence ( for a given neutrino trajectory through the earth core ) of _ points of resonance - like total neutrino conversion _ , @xmath264 , in the corresponding space of neutrino oscillation parameters @xcite . the location of these points determines the regions where @xmath136 is large , @xmath265 . these regions vary slowly with the nadir angle , they are remarkably wide in the nadir angle and are rather wide in the neutrino energy @xcite , so that the transitions of interest produce noticeable effects in the ratio @xmath1 : we have , e.g. , @xmath266 for the values of @xmath9 of interest . the resonance - like total neutrino conversion due to the mantle - core enhancement effect takes place for a given neutrino trajectory through the earth core if the following two conditions are satisfied @xcite : @xmath267 where the signs are correlated and @xmath268 . as was shown in @xcite , conditions ( [ nolr1 ] ) are fulfilled for the @xmath114 and @xmath116 transitions of the earth core - crossing atmospheric neutrinos . a rather complete set of values of @xmath269 and @xmath119 for which both conditions in ( [ nolr1 ] ) hold and @xmath264 for neutrino trajectories with nadir angle @xmath270 was also found in @xcite . for @xmath77 , there are two sets of values of @xmath271 and @xmath9 for which eq . ( [ nolr1 ] ) is fulfilled and @xmath264 . these two solutions of eq . ( [ nolr1 ] ) occur for , e.g. , @xmath272 , at 1 ) @xmath273 , @xmath274 , and at 2 ) @xmath275 , @xmath276 ( see table 2 in ref . the first solution corresponds to @xmath277 , @xmath278 and and @xmath279 . for @xmath280 , the total neutrino conversion occurs in the case of the first solution at @xmath281 . the atmospheric @xmath282 and @xmath31 with these energies contribute to the multi - gev samples of @xmath3like and @xmath2like events in the sk experiment . the values of @xmath119 at which the second solution takes place are marginally allowed . if @xmath280 , one has @xmath264 for this solution for a given @xmath0 in the interval @xmath283 at @xmath122 lying in the interval @xmath284 . the above discussion suggests , in particular , that the effects of the mantle - core ( nolr ) enhancement of @xmath136 ( or @xmath167 ) in the ratios @xmath1 and @xmath99 increase rapidly with @xmath9 as long as @xmath285 , and should exhibit a rather weak dependence on @xmath9 for @xmath286 . if 3-neutrino oscillations of atmospheric neutrinos take place , the magnitude of the matter effects in the multi - gev @xmath3like and @xmath2like event samples , produced by neutrinos crossing the earth core , should be larger than in the event samples due to neutrinos crossing only the earth mantle ( but not the core ) . this is a consequence of the fact that in the energy range of interest the atmospheric neutrino fluxes decrease rather rapidly with energy - approximately as @xmath287 , while the neutrino interaction cross section rises only linearly with @xmath122 , and that the maximum of @xmath136 ( or @xmath167 ) due to the nolr takes place at approximately two times smaller energies than that due to the msw effect for neutrinos crossing only the earth mantle ( e.g. , at @xmath281 and @xmath288 , respectively , for @xmath280 ) . 0.6truecm the results of our analysis are summarized graphically in figs . 1 - 5 . we have used in the calculations the predictions for the nadir angle and energy distributions of the atmospheric neutrino fluxes given in @xcite . in this analysis , we only consider deep inelastic scattering ( dis ) cross sections and we make use of the grv94 parton distributions given in @xcite . the dependence on @xmath289 of the ratios of the multi - gev @xmath2 and @xmath3 like events ( or event rates ) , integrated over the neutrino energy in the interval @xmath290 gev , in the cases i ) of two - neutrino @xmath58 and @xmath59 oscillations in vacuum and no @xmath4 and @xmath5 oscillations , @xmath112 ( solid lines ) , ii ) three - neutrino oscillations in vacuum of @xmath31 , @xmath32 , @xmath4 and @xmath5 , @xmath291 ( dash - dotted lines ) , iii ) three - neutrino oscillations of @xmath31 , @xmath32 @xmath4 and @xmath5 in the earth and neutrino mass spectrum with normal hierarchy @xmath105 ( dashed lines ) , or with inverted hierarchy , @xmath106 ( dotted lines ) . the results shown are for @xmath292 , @xmath293 , and @xmath294 . , height=687 ] the predicted dependences on @xmath289 of the ratios of the multi - gev @xmath2 and @xmath3 like events ( or event rates ) , integrated over the neutrino energy from the interval @xmath290 gev , in the case i ) of two - neutrino @xmath58 and @xmath59 oscillations in vacuum and no @xmath4 and @xmath5 oscillations , @xmath112 , ii ) three - neutrino oscillations in vacuum of @xmath31 , @xmath32 , @xmath4 and @xmath5 , @xmath291 , iii ) three - neutrino oscillations of @xmath31 , @xmath32 , @xmath4 and @xmath5 in the earth in the cases of neutrino mass spectrum with normal hierarchy @xmath105 , and with inverted hierarchy , @xmath106 , for @xmath109 , @xmath295 and @xmath292 are shown in fig . 1 . as @xmath289 increases from 0 to @xmath296 , the neutrino path length in the earth increases from 0 to @xmath297 km . the @xmath13 ( @xmath14 ) oscillations , which for @xmath298 proceed in the earth essentially as in vacuum , fully develop . for @xmath299 and @xmath300 gev , for example , the maximum of the @xmath13 ( @xmath14 ) oscillation probability occurs for @xmath301 , or @xmath302 km . for @xmath298 and @xmath292 , the oscillations involving the atmospheric @xmath4 ( @xmath303 , @xmath304 ) and @xmath5 ( @xmath305 , @xmath306 ) with energies in the multi - gev range , @xmath307 gev , are suppressed . if @xmath15 , for instance , the earth matter effects suppress the antineutrino oscillation probability @xmath167 , but can enhance the neutrino mixing in matter , or @xmath308 . however , since the neutrino path in the earth mantle is relatively short one has @xmath309 , and correspondingly @xmath310 . thus , for @xmath298 , all four types of ratios we consider , @xmath112 , @xmath291 , @xmath105 and @xmath106 , practically coincide and exhibit the same dependence on @xmath289 : they decrease as @xmath289 increases from 0 , reaching a minimum at @xmath301 , and begin to rise as @xmath289 increases further . this behavior is clearly seen in figs . 1 and 2 . at @xmath19 , the earth matter effects in the oscillations of the atmospheric @xmath31 , @xmath32 @xmath4 and @xmath5 , can generate noticeable differences between @xmath112 ( or @xmath291 ) and @xmath311 , as well as between @xmath105 and @xmath106 . for @xmath312 and @xmath313 ( upper left panel in fig . 1 ) , we have at @xmath314 ( neutrinos crossing only the earth mantle ) : @xmath315 . for the earth - core - crossing atmospheric neutrinos , @xmath316 , the mantle - core interference effect ( or nolr ) suppresses the ratios @xmath317 with respect to @xmath112 ( or @xmath291 ) : at @xmath318 the relative averaged difference between @xmath112 and @xmath105 is and @xmath105 is defined as @xmath319 . ] @xmath320 , while the difference between @xmath105 and @xmath106 is rather small ( upper right panel in fig . 1 ) . at @xmath27 , the differences between @xmath112 and @xmath317 become noticeable already at @xmath19 . they increase with the increasing of @xmath10 and/or @xmath119 , and are maximal for @xmath321 and @xmath318 . the dependence on @xmath10 is particularly strong . the deviations from the vacuum oscillation ratio @xmath291 increase with the increasing of @xmath19 as well ; they are maximal for the earth - core - crossing neutrinos , @xmath316 . for @xmath322 and @xmath295 , the relative difference between @xmath112 and @xmath105 in the interval @xmath323 is approximately @xmath324 ; for @xmath325 and @xmath295 , it is @xmath326 . the same differences are larger in the earth core interval @xmath327 due to the mantle - core enhancement ( nolr ) @xcite , reaching on average values of @xmath328 for @xmath322 and @xmath295 , and @xmath329 for @xmath325 and the same two values of @xmath119 . for @xmath330 , and @xmath295 , the relative difference between @xmath112 and the ratio @xmath1 in the case of ih neutrino mass spectrum , @xmath106 , in the interval @xmath331 has a mean value of approximately @xmath332 , while if @xmath325 it is approximately @xmath333 . it is larger in the earth core bin , @xmath327 , reaching approximately @xmath334 for @xmath325 and @xmath318 . the magnitude of the difference between @xmath112 and @xmath311 exhibits a relatively strong dependence on the minimal value of the neutrino energy @xmath122 from the integration interval , @xmath335 , and a rather mild dependence on the maximal @xmath122 in the interval , @xmath336 . with the increase of @xmath335 , the relative magnitude of the contributions to the energy - integrated event rates of interest , coming from the energy interval in which the matter effects are significant , also increases , leading to larger difference between @xmath112 and @xmath311 . this is illustrated in fig . 2 , where the ratios of interest are shown as functions of @xmath289 for @xmath292 , @xmath330 , @xmath337 , and i ) @xmath338 gev , @xmath339 gev ( left panel ) , and ii ) @xmath340 gev , @xmath341 gev ( right panel ) . increasing @xmath335 ( @xmath336 ) while keeping @xmath336 ( @xmath335 ) intact would lead to the decreasing ( increasing ) of the statistics in the samples of @xmath2like and @xmath3like events of interest . 2 illustrates , the relative differences between i ) @xmath112 and @xmath105 and ii ) @xmath112 and @xmath106 , increase noticeably in the interval @xmath342 with the increase of @xmath335 , being almost constant for @xmath338 gev and reaching the values of approximately @xmath343 and @xmath344 , respectively . and @xmath105 is @xmath345 . ] for @xmath327 , the differences under discussion exhibit a characteristic oscillatory pattern . these differences have a completely different behavior as functions of @xmath289 if @xmath336 is increased to 100 gev , keeping @xmath340 gev ( fig . 2 , right panel ) : they increase approximately linearly with @xmath289 , starting from 0 at @xmath346 , and having at @xmath347 the values i ) @xmath348 and ii ) @xmath349 , respectively . the same as in fig . 1 but for the ratios of the @xmath2 and @xmath3 like events ( or event rates ) , integrated respectively over the neutrino energy in the intervals @xmath350 gev ( left panel ) , and @xmath351 gev ( right panel ) , and for @xmath292 , @xmath330 and @xmath318 . , height=272 ] and @xmath3 like events ( or event rates ) , integrated over the neutrino energy in the interval @xmath290 gev and over the nadir angle in the interval corresponding to @xmath352 , as functions i ) of @xmath119 for @xmath292 ( left panels ) , and ii ) of @xmath353 for @xmath318 ( right panels ) : @xmath112 ( solid lines ) , @xmath291 ( dash - dotted lines ) , @xmath105 ( dashed lines ) and @xmath106 ( dotted lines ) . the results shown are obtained for @xmath293 . , height=687 ] in fig . 3 we show the predictions for the different ratios of @xmath2like and @xmath3like event rates we consider , integrated over the neutrino energy in the interval @xmath354 gev and over @xmath289 in the interval @xmath355 , i ) as functions of @xmath119 for @xmath292 ( left panels ) , and ii ) as functions of @xmath353 for @xmath318 ( right panel ) . in each case the results presented are for three values of @xmath356 . the differences between the ratios of the integrated @xmath2like and @xmath3like event rates of interest , i.e. , i ) between @xmath112 and @xmath105 and ii ) between @xmath112 and @xmath106 , increase rather rapidly as @xmath119 increases from 0 to @xmath357 , while the increase is slower in the interval @xmath358 . at @xmath357 , the relative differences between @xmath112 and i ) @xmath105 and ii ) @xmath106 in the case under discussion are respectively @xmath359 and @xmath360 for @xmath330 ; for @xmath361 they are considerably larger , @xmath362 and @xmath359 . if @xmath363 , the same two differences for @xmath364 read @xmath365 and @xmath366 , respectively . it follows from fig . 3 , right panel , that differences between the integrated ratios @xmath112 ( or @xmath291 ) and @xmath105 , and @xmath106 and @xmath105 , are maximal for values of @xmath353 lying in the interval @xmath367 , which are favored by the current atmospheric neutrino data . of @xmath3like multi - gev events ( or event rates ) , for @xmath4 and @xmath5 taking part in 3-neutrino oscillations in the earth ( @xmath368 ) , and @xmath4 and @xmath5 not taking part in the oscillations ( @xmath369 ) , as a function of @xmath370 for @xmath371 ( dashed lines ) , and for neutrino mass spectrum with normal hierarchy ( upper solid or dashed lines ) and with inverted hierarchy ( lower solid or dashed lines ) . the results shown are for @xmath372 ( right panel ) , and for @xmath373 . see text for details . , height=272 ] i ) on @xmath374 for @xmath373 ( left panels ) , and ii ) on @xmath375 for @xmath376 ( right panels ) , in the cases of neutrino mass spectrum with normal hierarchy ( solid lines ) and inverted hierarchy ( dashed lines ) . in this figure @xmath368 and @xmath369 are the multi - gev e - like events ( or event rates ) for @xmath4 and @xmath5 taking part in 3-neutrino oscillations in the earth and @xmath4 and @xmath5 not taking part in the oscillations , respectively , integrated over the nadir angle @xmath0 in the intervals corresponding to a ) @xmath377 - neutrinos crossing the earth core ( upper panels ) , and b ) @xmath378 ( lower panels ) . the results shown are for @xmath312 ( lower doubly thick solid or dashed lines ) ; 0.50 ( middle thin solid or dashed lines ) ; 0.64 ( upper doubly thick solid or dashed lines ) . , height=453 ] in figs . 4 and 5 we present results just for the multi - gev e - like event rate . the earth matter effects are largest in the oscillations of the atmospheric @xmath4 or @xmath5 . the dependence of the ratio @xmath99 , @xmath101 and @xmath102 being the numbers of multi - gev @xmath97-like events ( or event rates ) predicted in the cases of 3-@xmath74 oscillations of @xmath4 , @xmath5 , and of absence of oscillations ( @xmath21 ) , on @xmath289 is shown in fig . 4 for @xmath364 and for @xmath379 . the deviation of @xmath99 from 1 would signal that @xmath4 and @xmath5 take part in oscillations and that @xmath380 . we can have @xmath381 only if a substantial fraction of the atmospheric @xmath31 ( or @xmath32 ) oscillate into @xmath282 ( @xmath382 ) . as fig . 4 shows , the ratio @xmath99 increases with the increasing of @xmath289 and can be significantly greater than 1 for @xmath19 . at @xmath383 , for instance , we have for @xmath330 and @xmath295 in the case of nh neutrino mass spectrum @xmath384 , while for @xmath361 one finds @xmath385 . for ih neutrino mass spectrum the ratio of interest , @xmath386 , is smaller and the corresponding values read : @xmath387 and @xmath388 , respectively . for the earth - core - crossing neutrinos both @xmath389 and @xmath386 are larger due to the nolr effect , and for @xmath361 and @xmath337 reach the values @xmath390 and @xmath391 . for given @xmath10 and and @xmath9 , the maximum of the ratio of the multi - gev e - like event rates of interest @xmath101 and @xmath392 , integrated over @xmath0 in the intervals corresponding to a ) @xmath393 ( earth core bin ) and b ) @xmath394 , occurs again for values of @xmath395 ( fig . 5 , right panels ) , favored by the existing atmospheric neutrino data . for @xmath292 , @xmath396 and @xmath361 , the two types of @xmath20 integrated ratios can be as large as : case a ) @xmath397 , and case b ) @xmath398 ( fig . 5 , left and right panels ) . 0.6truecm in the present article we have studied the possibility to obtain evidences for earth matter enhanced 3-neutrino oscillations of the atmospheric neutrinos involving , in particular , the @xmath4 ( or @xmath5 ) , from the analysis of the @xmath2like and @xmath3like multi - gev event data accumulated by the sk experiment , or that can be provided by future water - erenkov detectors . such evidences would give also important quantitative information on the values of @xmath9 and @xmath10 and on the sign of @xmath48 . we have considered 3-neutrino oscillations of the atmospheric @xmath31 , @xmath32 , @xmath4 and @xmath5 , assuming that the inequality latexmath:[${\mbox{$ \delta m^2_{\odot}$}}= \delta m^2_{21 } \ll suggested by the current solar and atmospheric neutrino data . depending on the sign of @xmath46 , the earth matter effects in this case can enhance either the @xmath83 and @xmath85 , or the @xmath84 and @xmath94 transitions if @xmath95 . the effects of the enhancement can be substantial for @xmath96 and @xmath400 . they are largest in the multi - gev @xmath97-like and @xmath2like samples of events and for atmospheric neutrinos with relatively large path length in the earth , crossing deeply the mantle or the mantle and the core , i.e. , for @xmath19 , where @xmath0 is the nadir angle characterizing the neutrino trajectory in the earth . as observables which are particularly sensitive to the earth matter effects , and thus to the values of @xmath9 and @xmath10 , and to the sign of @xmath46 , we have considered the nadir - angle distributions of the ratios @xmath98 and @xmath99 , where @xmath100 and @xmath101 are the multi - gev @xmath2like and @xmath97-like numbers of events ( or event rates ) in the case of 3-@xmath74 oscillations of the atmospheric @xmath4 , @xmath5 and @xmath31 , @xmath32 , and @xmath102 is the number of @xmath3like events in the case of absence of oscillations ( @xmath21 ) . as is well - known , the ratio of the energy and nadir angle integrated @xmath2like and @xmath3like events , @xmath1 , has been measured with a relatively high precision by the sk experiment @xcite . we have obtained predictions for the nadir - angle distributions of @xmath98 and of @xmath99 both for neutrino mass spectra with normal ( @xmath15 ) and inverted ( @xmath16 ) hierarchy , @xmath105 , @xmath106 , @xmath107 and @xmath401 , and for @xmath109 . we compared the latter with the predicted nadir - angle distributions i ) of the ratio @xmath110 for the case the 3-neutrino oscillations taking place in vacuum , @xmath111 , and ii ) of the ratio @xmath112 , the predicted number of @xmath2like and @xmath3like events in the case of 2-neutrino @xmath58 and @xmath59 oscillations of the atmospheric @xmath31 and @xmath32 , and @xmath282 and @xmath382 not taking part in the oscillations ( @xmath21 ) . the dependence of the nadir angle distributions of @xmath105 , @xmath106 , @xmath111 and @xmath112 on the minimal and maximal neutrino energies in the energy integration interval has also been studied . predictions for these four different types of ratios of the suitably integrated over @xmath0 nadir angle distributions of the @xmath2like and @xmath3like multi - gev events were also derived . our results are presented graphically in figs . 1 - 5 . we find that for @xmath322 and @xmath402 , the relative difference between @xmath112 and @xmath105 in the interval @xmath323 is approximately @xmath320 . if @xmath325 and @xmath318 , the same difference is approximately @xmath403 ( fig . the relative difference between @xmath112 and @xmath105 is larger in the earth core interval @xmath327 due to the mantle - core enhancement ( nolr ) @xcite , reaching on average values of @xmath403 for @xmath322 and @xmath402 , and of @xmath404 for @xmath325 and @xmath318 ( fig . 1 ) . in the case of neutrino mass spectrum with inverted hierarchy , the relative difference between @xmath112 and @xmath106 in the interval @xmath331 has a mean value of approximately @xmath405 , for @xmath361 and @xmath318 . it reaches approximately @xmath334 in the earth core bin , @xmath327 , for these values of @xmath406 and @xmath119 ( fig . 1 ) . the magnitude of the difference between @xmath112 and @xmath311 exhibits a relatively strong dependence on the minimal value of the neutrino energy @xmath122 from the integration interval , @xmath335 , and a rather mild dependence on the maximal @xmath122 in the integration interval , @xmath336 ( fig . 2 ) . increasing @xmath335 from 2 gev to 4 gev and keeping @xmath339 gev , leads for @xmath292 , @xmath330 and @xmath337 to a considerably larger relative difference between @xmath112 and @xmath311 in the interval @xmath342 , which reaches approximately @xmath407 ( fig . this difference is larger also in the earth core bin , @xmath327 . we have also found that the differences between the nadir angle and energy integrated ratios @xmath112 ( or @xmath291 ) and @xmath105 , and @xmath106 and @xmath105 , are maximal for values of @xmath353 lying in the interval @xmath367 , which are favored by the current atmospheric neutrino data ( fig . the same conclusion is valid for the integrated ratio of @xmath3like events @xmath99 ( fig . we have also shown that in the case of the 3-neutrino oscillations of the atmospheric neutrinos considered , the ratio @xmath99 increases with the increasing of @xmath289 and can be significantly greater than 1 for @xmath19 . at @xmath383 , for instance , we have for @xmath330 and @xmath318 in the case of nh neutrino mass spectrum @xmath408 , while for @xmath361 one finds @xmath409 ( fig . 4 ) . for ih neutrino mass spectrum the ratio of interest , @xmath386 , is smaller : @xmath410 . for the earth - core - crossing neutrinos both @xmath389 and @xmath386 are enhanced due to the nolr effect , and for @xmath361 and @xmath337 reach the values @xmath411 and @xmath391 ( fig . 4 ) . it follows from our results that the earth matter effects in the nadir angle distribution of the ratio of the multi - gev @xmath2like and @xmath3like atmospheric neutrino events ( or event rates ) , measured in the super - kamiokande ( or any future water - erenkov ) experiment , might be observable if the atmospheric neutrinos , including the @xmath4 and @xmath5 , take part in 3-neutrino oscillations and @xmath9 and @xmath10 are sufficiently large . the observation of relatively large earth matter effects ( 20% - 30% ) at @xmath19 would clearly indicate that @xmath17 , @xmath27 , and would suggest that the neutrino mass spectrum is with normal hierarchy , @xmath412 however , distinguishing statistically between the neutrino mass spectrum with normal and inverted hierarchy requires a high precision measurement of the nadir angle distribution of the multi - gev ratio @xmath1 and is a rather challenging task . this work is supported in part by the italian miur and infn under the programs `` fenome - nologia delle interazioni fondamentali '' and `` fisica astroparticellare '' , by the u.s . national science foundation under grant no . phy99 - 07949 ( s.t.p . ) , by the spanish grant fpa2002 - 00612 of the mct ( j.b . and s.p .- r . ) and by the spanish mecd for a fpu fellowship ( s.p .- r . ) . s.t.p . would like to thank the organizers of the program on `` neutrinos : data , cosmos and the planck scale '' at kitp , univ . of california at santa barbara , where part of the work on the present study was done , for kind hospitality , and m. freund for discussions at the initial stage of this work . s.p .- r . would like to thank the elementary particle physics sector of sissa , trieste , italy , for kind hospitality during part of this study . b. t. cleveland _ et al . _ , astrophys . j. * 496 * ( 1998 ) 505 ; y. fukuda _ et al . _ , phys . * 77 * ( 1996 ) 1683 ; v. gavrin , nucl . * 91 * ( 2001 ) 36 ; w. hampel _ et al . _ , phys . b * 447 * ( 1999 ) 127 ; m. altmann _ et al . _ , phys . b * 490 * ( 2000 ) 16 . super - kamiokande collaboration , y. fukuda _ et al . _ , phys . lett . * 86 * ( 2001 ) 5651 and 5656 . super - kamiokande collaboration , m. shiozawa , talk given at the int . conf . on neutrino physics and astrophysics `` neutrino02 '' , may 25 - 30 , 2002 , munich , germany . sno collaboration , q. r. ahmad _ et al . _ , phys . lett . * 89 * ( 2002 ) 011302 and 011301 . kamland collaboration , k. eguchi _ et al . _ , phys . * 90 * ( 2003 ) 021802 . b. pontecorvo , chalk river lab . report pd205 , 1946 ; zh . eksp . * 53 * ( 1967 ) 1717 . r. davis , d. s. harmer and k. c. hoffman , phys . lett . * 20 * , 1205 ( 1968 ) ; acta physica acad . sci . hung . * 29 * suppl . 4 , 371 ( 1970 ) ; r. davis , proc . of the neutrino ` 72' int . conference , balatonfured , hungary , june 1972 ( eds . a. frenkel and g. marx , omkdk - technoinform , budapest , 1972 ) , p. 5 . s. m. bilenky , c. giunti and w. grimus , prog . * 43 * ( 1999 ) 1 . b. pontecorvo , zh . eksp . * 33 * ( 1957 ) 549 and * 34 * ( 1958 ) 247 ; z. maki , m. nakagawa and s. sakata , prog . * 28 * ( 1962 ) 870 . s. m. bilenky _ et al . _ , phys . b*94 * ( 1980 ) 495 ; m. doi _ et al . _ , phys . * b102 * ( 1981 ) 323 ; j. bernabu and p. pascual , nucl . b * 228 * ( 1983 ) 21 ; p. langacker _ et al . _ , nucl . b * 282 * ( 1987 ) 589 . s. m. bilenky and s. t. petcov , rev . * 59 * ( 1987 ) 671 . m. apollonio _ et al . _ , phys . b * 466 * ( 1999 ) 415 . f. boehm _ et al . _ , phys . rev . lett . * 84 * ( 2000 ) 3764 and phys . d * 62 * ( 2000 ) 072002 . g. l. fogli _ et al . _ , phys . d * 67 * ( 2003 ) 073002 . v. barger and d. marfatia , phys . b * 555 * ( 2003 ) 144 ; m. maltoni , t. schwetz , and j. w. f. valle , hep - ph/0212129 ; a. bandyopadhyay _ et al . _ , phys . lett . b * 559 * ( 2003 ) 121 ; j. n. bahcall , m. c. gonzlez - garca and c. pea - garay , jhep * 0302 * ( 2003 ) 009 ; h. nunokawa , w. j. c. teves , and r. zukanovich funchal , hep - ph/0212202 ; p. aliani _ et al . _ , hep - ph/0212212 ; p. c. de holanda and a. y. smirnov , jcap * 0302 * ( 2003 ) 001 ; a. b. balantekin and h. yuksel , j. phys . g * 29 * ( 2003 ) 665 . s. m. bilenky , d. nicolo and s. t. petcov , phys . b * 538 * ( 2002 ) 77 . a. de rjula _ et al . _ , nucl . b * 168 * ( 1980 ) 54 . s. t. petcov , phys . b * 434 * ( 1998 ) 321 , ( e ) _ ibid . _ b * 444 * ( 1998 ) 584 . m. c. bauls , g. barenboim and j. bernabu , phys . b * 513 * , 391 ( 2001 ) . j. bernabu and s. palomares - ruiz , hep - ph/0112002 , and nucl . . proc . * 110 * , 339 ( 2002 ) , hep - ph/0201090 . p. i. krastev and s. t. petcov , phys . b * 205 * ( 1988 ) 84 . j. bernabu and m. c. bauls , nucl . * 87 * ( 2000 ) 315 , hep - ph/0003299 . s. m. bilenky , s. pascoli and s. t. petcov , phys . d * 64 * ( 2001 ) 053010 ; s. pascoli and s. t. petcov , phys . b * 544 * ( 2002 ) 239 ; s. pascoli , s. t. petcov and w. rodejohann , phys . b * 558 * ( 2003 ) 141 . l. wolfenstein , phys . d * 17 * , 2369 ( 1978 ) , and _ ibid . _ d * 20 * , 2634 ( 1979 ) . m. honda _ et al . _ , phys . d * 52 * , 4985 ( 1995 ) . v. agraval _ et al . _ , phys . d * 53 * , 1314 ( 1996 ) . g. fiorentini , v. a. naumov and f. l. villante , phys . b * 578 * ( 2000 ) 27 . f. d. stacey , _ physics of the earth , 2@xmath413 edition _ , john wiley and sons , london , new york , 1977 .

momentum relaxation can be built into many holographic models without sacrificing homogeneity of the bulk solution . in this paper we study two such models : one in which translational invariance is broken in the dual theory by spatially - dependent sources for massless scalar fields and another that features an additional neutral scalar field . we turn on a charged scalar field in order to explore the condensation of a charged scalar operator in the dual theories . after demonstrating that the relaxed superconductors we construct are thermodynamically relevant , we find that the finite dc electrical conductivity of the normal phase is replaced by a superfluid pole in the broken phase . moreover , when the normal phase possesses a drude behaviour at low frequencies , the optical conductivity of the broken phase at low frequencies can be described by a two - fluid model that is a sum of a drude peak and a superfluid pole , as was found recently for inhomogeneous holographic superconductors . we also study cases in which this low - frequency behavior does not hold . we find that the drude description is accurate when the retarded current - current correlator has a purely - dissipative pole that is well - separated from the rest of the excitations .

1412.6521

consider a system at finite charge density and temperature . if momentum is conserved then a non - zero overlap between the electric current and the momentum operator leads to an infinite dc electrical conductivity . this feature is indeed observed in the simplest holographic model of such a system : the reissner - nordstrm - ads ( rn ) black brane . a natural way to build more realistic holographic models that allow momentum to dissipate is to modify the rn geometry such that translational invariance is broken explicitly along the boundary directions . this can be achieved by introducing spatially - dependent sources for ( neutral ) matter fields in the bulk . the backreaction of such sources typically leads to an inhomogeneous solution . examples include a spatially - periodic massive neutral scalar @xcite or chemical potential @xcite . in these studies the dc conductivity is indeed finite , in good agreement with the field theory calculation of @xcite , and the delta function is resolved into a drude peak of the form @xmath0 however , the study of such setups is technically involved because one must solve non - linear pdes . it is thus desirable to have a simpler holographic model that incorporates dissipation . in this way one may hope to uncover , in a computationally simple way , universal properties of field theories with holographic duals in which momentum can relax . one can obtain such a model with a homogeneous bulk geometry by exploiting the presence of a symmetry in the matter sector . for example , a set of massless scalar fields that are linear in the boundary coordinates were added to the usual einstein - maxwell system in @xcite . the shift symmetry of the scalars leads to a homogeneous bulk stress tensor . we shall refer to such scalars as axions . moreover , by choosing the axion configuration in a particular way it is also possible to render the geometry isotropic , allowing for the existence of an analytical black brane solution . the finite dc conductivity can be expressed analytically and , in four bulk dimensions , turns out to be independent of temperature . at finite chemical potential , a transition from a coherent ( i.e. drude - like ) metallic behaviour to an incoherent one as the parameter that controls the amount of momentum relaxation is increased was identified in this model by @xcite . a similar transition was observed in the thermal conductivity of this system at zero chemical potential in @xcite , wherein this feature was nicely explained in terms of the structure of the quasi - normal modes ( qnm ) of the black brane . there it was observed that the black brane conducts coherently when there is a purely - dissipative mode parametrically separated from the rest of the qnm . this model can be generalized in a number of ways . one is to introduce a further neutral scalar field , this time with a potential and couplings to the axions and maxwell field . we shall refer to such a scalar as a dilaton . momentum - dissipating black brane solutions to such theories were constructed in @xcite and referred to as ` q - lattices ' . the dc conductivity of these solutions is again finite and determined in terms of the horizon properties of the background @xcite . moreover , @xcite identified metallic and insulating phases , the latter of which being characterized by a vanishing dc conductivity at zero temperature . in the metallic phase , drude behaviour is observed for a large range of temperatures , while in the insulating phase the low frequency region is not described by drude physics . metal - insulator transitions were observed in similar models in @xcite . see @xcite for further studies of such models and also @xcite for alternative generalizations . a conceptually distinct approach was considered in @xcite , wherein the theory of massive gravity @xcite was employed to remove translational invariance on the boundary as a result of breaking ( a subset of ) diffeomorphisms in the bulk . the resulting solutions possess a finite dc conductivity for which an analytical expression was provided in @xcite adapting a calculation in @xcite . in @xcite , it was argued that there is a qualitative connection between massive gravity and a perturbative inhomogeneous lattice generated by a neutral scalar with spatially modulated boundary conditions . more precisely , the authors found that , at leading order in the lattice strength , the equations of motion that govern the conductivity are those of massive gravity if the radially - dependent graviton mass is proportional to the square of the neutral scalar . see @xcite for further studies of massive gravity in this context . the simplicity of these models makes them amenable to generalization , allowing us to conveniently study different field theory phenomena in setups in which translational invariance is broken . the goal of this paper is to study superconductivity in the presence of momentum relaxation , focusing on the axion model and a particular axion - dilaton model . to do so , we shall generalize the by now canonical model of holographic superconductivity of @xcite , consisting of an einstein - maxwell - charged scalar theory , by coupling it to the neutral scalar sectors of @xcite and @xcite . the condensation of the charged order parameter is dual to the appearance of charged scalar hair on the black brane ( for scale invariant boundary conditions that set the source of the dual operator to zero ) as we lower the temperature below a critical value @xmath1 . superconducting inhomogeneous lattices were constructed in @xcite in an einstein - maxwell - charged scalar system with spatially - modulated chemical potential . below the critical temperature , the optical conductivity behaves at low frequencies as a two - fluid model consisting of a pole in the imaginary part plus a normal drude component : @xmath2 note that the appearance of a pole in the imaginary part demands there must be a delta function in the real part with strength set by @xmath3 in order to satisfy the kramers - kronig relations . as the temperature is decreased below @xmath1 , it was observed that the parameters @xmath3 and @xmath4 grow rapidly while @xmath5 quickly decreases . in addition , other simplified models of momentum relaxation have been utilized in the study of holographic superconductors . the condensation of a charged scalar was studied in a massive gravity theory in @xcite . where @xmath6 is the reference metric . this is different to the massive gravity theory whose optical conductivity coincides with that of @xcite . ] their results qualitatively agree with those in @xcite ; namely , they find a two - fluid model with a drude normal component . in @xcite , a charged scalar was coupled to the model of @xcite , which only has one neutral massless scalar . there it was observed that the critical temperature decreases as the scale of momentum dissipation increases and that a pseudo gap appears in the optical conductivity in the direction in which the breaking of translational invariance is introduced . see @xcite for work on the dynamical formation of a condensate in the same model . more recently , @xcite studied a holographic superconductor in an axion - dilaton model as we also do below , although with a different mass for the dilaton . there it was observed that the critical temperature is reduced by the presence of momentum relaxation and that breaking the @xmath7 from the metallic phase results in a two - fluid model behaviour with drude normal component . this is not so in the insulating phase , for which the normal component is not well described by a drude peak . see also @xcite for an alternative approach . in this paper we construct superconducting solutions and observe the transition from all the unbroken phases we have considered ; namely , the exact solution of @xcite in the axion model and one choice of metallic and insulating solutions in the axion - dilaton model of @xcite . crucially , we first demonstrate that the broken phases are indeed thermodynamically preferred , confirming their physical relevance . our main result is the observation of a zero - frequency pole in the imaginary part of the electrical conductivity , confirming that our solutions are indeed superconducting . moreover , in cases in which the normal phase possesses a drude behaviour at low frequencies ( namely , in the axion model with small momentum dissipation parameter and the metallic solution in the axion - dilaton model ) , the broken phase can be described as a two - fluid model whose low - frequency conductivity is given by . we find that this can be explained by the structure of the lowest qnms : drude peaks arise when there is a purely - dissipative mode that is well - separated from the rest of the excitations , resembling the picture of @xcite . when this ceases to be true , we observe departures from the drude behaviour in the real part of the conductivity and thus the first term in must be modified . we study the temperature dependence of the parameters in for the broken phase of the metal of the axion - dilaton model . as we lower the temperature from @xmath1 , we find that @xmath4 first decreases slowly and then increases quickly , @xmath5 is monotonically decreasing reaching a small value , and @xmath3 monotonically increases approaching a constant . we find that @xmath3 exhibits the same behaviour in the broken phase of the insulator and also in the broken phase studied in the axion model . we also check the ferrell - glover - tinkham ( fgt ) sum rule @xcite , which relates ` missing ' spectral weight to the strength of the zero - frequency pole . we verify that the sum rule holds for the broken metallic phase and the broken phase in the axion model , as was also demonstrated for the inhomogeneous lattice of @xcite . this paper is organized as follows . in section [ sec : axion ] we study the axion model coupled to a charged scalar . we describe a general class of solutions which implement momentum relaxation and discuss the calculation of the optical conductivity for such configurations . after a review of the normal phase and its conductivity , we move on to explore the broken phase . in section [ sec : axion_dilaton ] we also follow this structure for a particular axion - dilaton model and study both metallic and insulating states . we conclude with a discussion of our results and comments on extensions and more general models in section [ sec : discussion ] . both models we will study have the form @xmath8 in this section we focus on an axion model that has neutral sector @xcite @xmath9 \quad \textrm{where}\quad f = da\ ] ] the field equations admit ads@xmath10 with unit radius as a vacuum solution . we add to this action a charged sector given by @xmath11 \quad \textrm{where}\quad d_a \psi = \partial_a \psi - i q a_a \psi\ ] ] we will always choose this simple potential with @xmath12 . the charged scalar @xmath13 is dual to a scalar operator @xmath14 of dimension @xmath15 charged under a global @xmath7 . this model contains the minimal ingredients needed to construct a holographic superconductor in the presence of dissipation . our main goal here is to study the optical conductivity in the broken phase of this model . both the normal phase and broken phase are described by black brane solutions . we make the following isotropic ansatz for the bulk fields : @xmath16 note that the metric is isotropic and homogeneous but the full solution is not due to the presence of the spatial dependence in the axion fields @xmath17 . the parameter @xmath18 is the strength of the translational symmetry breaking in the dual theory , and setting @xmath19 restores translational invariance . in order to work at non - zero temperature in the field theory we demand a regular non - degenerate horizon at @xmath20 . this implies @xmath21 and that the temperature is given by @xmath22 we find the following asymptotic expansion near the ads@xmath10 boundary at @xmath23 : @xmath24 we consider the standard quantization of the charged scalar , in which @xmath25 is interpreted as the source and @xmath26 as the expectation value of @xmath14 . thus , a bulk solution with @xmath27 and @xmath28 corresponds to the broken phase in the dual theory . such solutions have been thoroughly investigated in the literature in the special case of @xmath19 , but in section [ sec : brokenaxion ] we turn on @xmath29 , allowing us to study the spontaneous breaking of the @xmath7 symmetry in the presence of dissipation . the remaining parameters in this expansion have the following interpretation in the dual theory : @xmath30 is the chemical potential , @xmath31 is the total charge density and @xmath32 is proportional to the total energy density . the phase space of the dual theory is parametrized by the dimensionless quantities @xmath33 and @xmath34 . in order to compute the optical conductivity , we consider fluctuations of the gauge field of the form @xmath35 . this couples to fluctuations of the metric and the axion fields , which we write as @xmath36 the equations that govern these fluctuations are @xmath37 note that the only difference when a charged scalar @xmath13 is turned on is an additional mass term for @xmath38 . near the conformal boundary , the linearized fields behave as @xmath39 the optical conductivity is defined via @xmath40 where @xmath41 is the retarded green s function for the current operator @xmath42 in the dual theory . in order to compute it , we need to turn on the source @xmath43 with all other sources set to zero . this can be achieved in a gauge - invariant way by requiring ( see e.g. @xcite ) @xmath44 in addition , to obtain the retarded correlator the linearized fields must satisfy ingoing boundary conditions at the horizon , which amounts to @xmath45 where the ellipses denote regular power series in @xmath46 . the leading order coefficients satisfy the relation @xmath47 imposing these conditions , we use a shooting method to read off the optical conductivity . as mentioned in the introduction , it will often be useful to study the quasi - normal frequencies of the black brane associated with such fluctuations . these are given by the poles in @xmath41 @xcite , which from are located at frequencies for which @xmath48 . before moving forward to discuss our results for the broken phase , first we review known results for both the dc and ac conductivity in the normal phase . we choose the following black brane solution as the normal phase @xcite : @xmath49 the parameter @xmath32 is determined by solving @xmath50 . the temperature of the black brane is simply @xmath51 it is clear that we recover the rn solution by setting @xmath52 . the dc conductivity in this phase was shown in @xcite to take the temperature - independent value @xmath53 the ac conductivity was studied in detail in @xcite . its real part interpolates between the dc value at zero frequency and unity at high frequency with a minimum at intermediate frequencies whose depth and location depends on @xmath33 and @xmath34 . the imaginary part has a peak at low frequencies then goes to zero as the frequency goes to zero . it was argued in @xcite that this behaviour is well - described by a slight modification of the drude form when @xmath34 is small but not when @xmath34 is large . indeed , we checked that when @xmath18 is much larger than both @xmath54 and @xmath30 , the conductivity is close to unity for all frequencies and there is clearly no coherent drude peak . next we study solutions with the charged scalar turned on . we construct these numerically using a shooting method . at all @xmath34 studied we find a black brane solution with charged scalar hair that exists below some critical @xmath55 . in figure [ fig : axion_tc ] we show this critical temperature as a function of @xmath34 for different values of the scalar charge @xmath56 . note that the curves become non - monotonic as @xmath56 is increased , in contrast to several studies mentioned in the introduction . from now on we fix @xmath57 . in figure [ fig : axion_condensate ] we show a family of condensate curves for different @xmath34 . for @xmath58 ( labelled from bottom to top).,scaledwidth=50.0% ] ( 0.3,0.4)(0,0 ) ( -5.7,2.5)(0,0)@xmath59 ( 0.2,0.2)(0,0)@xmath60 -1em -1em ( labelled from right to left).,scaledwidth=50.0% ] ( 0.3,0.4)(0,0 ) ( -5.7,2.5)(0,0)@xmath61 ( 0.2,0.2)(0,0)@xmath62 -1em in order to determine whether these solutions are thermodynamically relevant , we must compute their free energies . first we wick rotate to euclidean signature via @xmath63 the free energy density @xmath64 is given by the renormalised on - shell euclidean action via @xmath65 where @xmath66 and @xmath67 . using einstein s equations we can write the on - shell lagrangian as a total derivative : @xmath68 ' \nonumber \\ & = \beta v_2 \left [ \lim_{r\to\infty } ( 2\sqrt{g}\ , u\ , v ' + \alpha^2 r ) - \alpha^2 r_+\right]\end{aligned}\ ] ] where in the last line we have used @xmath69 . we must supplement the action with boundary terms in order to have a well - defined variational principle and also to obtain a finite @xmath64 @xcite . we find that is the induced metric on the timeline boundary at large @xmath70 with outward - pointing unit normal @xmath71 and @xmath72 is the extrinsic curvature with trace @xmath73 . ] @xmath74\ ] ] yields the result @xmath75 note that this formula is valid for both the normal and broken phases , but @xmath32 is found numerically for the latter . in figure [ fig : axion_free_energy ] we demonstrate that the broken phase has lower free energy density than the normal phase at the same @xmath34 and thus is thermodynamically preferred . ( labelled from top to bottom).,scaledwidth=50.0% ] ( 0.3,0.4)(0,0 ) ( -5.7,2.5)(0,0)@xmath76 ( 0.2,0.2)(0,0)@xmath62 -1em now we present our results for the optical conductivity in this phase . in figure [ fig : axion_sigma ] we show our results for @xmath77 for a range of temperatures below @xmath55 and in figure [ fig : axion_sigma_log ] we plot the same data on a log - log scale . two main features are apparent . firstly , the imaginary part of the conductivity diverges at the origin . by fitting the imaginary part at small frequencies we confirm the existence of a @xmath78 pole , which leads to a delta function at the origin in the real part via the kramers - kronig relations . 1em as a function of frequency for various temperatures . temperature decreases from @xmath79 down to @xmath80 , from top to bottom in the left plot and from bottom to top in the right.,title="fig:",scaledwidth=40.0% ] 2em as a function of frequency for various temperatures . temperature decreases from @xmath79 down to @xmath80 , from top to bottom in the left plot and from bottom to top in the right.,title="fig:",scaledwidth=40.0% ] ( 0.3,0.4)(0,0 ) ( -9.1,2.)(0,0)@xmath81 ( -4.6,2.)(0,0)@xmath82 ( 0.2,0.2)(0,0)@xmath83 -1em -0.5em 1em plotted on a log - log scale.,title="fig:",scaledwidth=40.0% ] 2em plotted on a log - log scale.,title="fig:",scaledwidth=40.0% ] ( 0.3,0.4)(0,0 ) ( -9.1,2.)(0,0)@xmath81 ( -0.8,2.)(0,0)@xmath82 ( 0.2,0.7)(0,0)@xmath83 -1.5em -0.5em in fact , at temperatures close to @xmath55 we find that the low frequency region of the conductivity can be well - approximated by a two - fluid model with drude normal component given by . specifically , we fit the real part of our data to the real part of this formula , then do the same for the imaginary counterparts over the same frequency range . for a temperature @xmath79 with @xmath84 we find that the squared sum of the residuals for these fits is @xmath85 or @xmath86 , respectively . we also find that the value of @xmath87 extracted from the real fit differs by that from the imaginary fit by @xmath88 and from the frequency of the lowest purely - dissipative qnm by @xmath89 . secondly , in the real part of the conductivity at low frequencies we see a departure from the normal phase value as the temperature is lowered . this is a result of spectral weight being transferred into the superfluid component of the fluid . the coefficient of the zero - frequency pole measures the superfluid density and is governed by the ferrell - glover - tinkham sum rule : @xmath90\ ] ] where @xmath91 is the conductivity of the normal phase at @xmath55 and @xmath92 is the conductivity of the broken phase at some @xmath93 . in figure [ fig : axion_sum_rule ] we demonstrate that , going to high enough frequencies , @xmath94 does indeed tend to the coefficient of the @xmath95 pole extracted from a fit of @xmath96 at low frequency . such agreement was also found for the inhomogeneous model studied in @xcite , where it is connected to the missing spectral weight observed in some cuprates . we discuss the interpretation of @xmath3 in section [ sec : discussion ] . where we choose a broken phase at @xmath97 . the dashed line is the value of @xmath98 extracted from a fit of @xmath99.,scaledwidth=50.0% ] ( 0.3,0.4)(0,0 ) ( -5.7,2.5)(0,0)@xmath100 ( 0.2,0.2)(0,0)@xmath83 -1em as the temperature is lowered , the peak at low frequencies in the imaginary part of the conductivity is overwhelmed by the superfluid pole . this is not in conflict with ; rather , it is simply a result of the interplay between the @xmath78 and @xmath101 components . however , one can see from figure [ fig : axion_sigma_log ] that for very low temperatures the local maximum in the real part at low frequencies also disappears . one might have expected to see the maximum reappear at lower frequencies but we think it is unlikely we have missed such a feature since our temperature steps are quite small . instead , an explanation for this feature is provided by studying the poles of the retarded green s function @xmath41 directly . in figure [ fig : axion_qnm ] we plot the imaginary part of the two poles closest to the axis for the broken phase as well as the drude pole for the normal phase . from the left - hand plot , we observe that the longest - lived purely imaginary pole in the broken phase is continuously connected to the drude pole of the normal phase , as one would expect . it is curious that the minimum of the normal phase curve occurs precisely at the critical temperature . for temperatures close to @xmath55 we see from the right - hand plot that the purely imaginary mode in the broken phase is isolated , which is why the fit to a drude form still works . however , as the temperature is lowered , a pair of propagating poles comes up from lower down in the complex plane and the drude pole is no longer isolated . thus the formula for the low frequency conductivity should be modified at low temperatures . note that this pair is not connected to the lowest propagating pair in the normal phase . 1em closest to the real axis as a function of temperature at @xmath102 . the black curve corresponds to the purely imaginary drude pole in the normal phase . for the broken phase we plot the purely imaginary pole in blue ( appearing in both plots ) and the ( imaginary part of the ) propagating pole in yellow.,title="fig:",scaledwidth=40.0% ] 2em closest to the real axis as a function of temperature at @xmath102 . the black curve corresponds to the purely imaginary drude pole in the normal phase . for the broken phase we plot the purely imaginary pole in blue ( appearing in both plots ) and the ( imaginary part of the ) propagating pole in yellow.,title="fig:",scaledwidth=40.0% ] ( 0.3,0.4)(0,0 ) ( -9.1,2.)(0,0)@xmath103 ( 0.2,2.3)(0,0)@xmath62 -1em in figure [ fig : axion_sigma_large_alpha ] we present the optical conductivity in the broken phase at the much larger value of @xmath104 . it is clear that there is no coherent drude - like peak , for any temperature lower than @xmath55 . 1em as a function of frequency for temperatures @xmath105 , labelled from top to bottom in the left plot and from bottom to top in the right.,title="fig:",scaledwidth=40.0% ] 3em as a function of frequency for temperatures @xmath105 , labelled from top to bottom in the left plot and from bottom to top in the right.,title="fig:",scaledwidth=40.0% ] ( 0.3,0.4)(0,0 ) ( -9.3,2.)(0,0)@xmath81 ( -4.5,2.)(0,0)@xmath82 ( 0.2,0.2)(0,0)@xmath83 -1em -1em in this section we study a model that has neutral sector @xcite @xmath106\ ] ] by parametrizing the complex scalar as @xmath107 we can re - write this theory as an axion - dilaton model . in comparison to the previous model we now only have a single axion , so we must build anisotropic normal phase solutions , which we have only been able to find numerically . we also have a neutral dilaton field that extends the phase space of solutions . we fix its mass to be @xmath108 . allowing for these differences , we proceed in a similar manner as for the previous model and add to this action the same charged sector @xmath109 as before . we consider the anisotropic ansatz @xmath110 and study black brane solutions with temperature @xmath111 the equations of motion admit the following asymptotic expansion near the ads@xmath10 boundary : @xmath112 we allow a source @xmath113 for the neutral scalar . this provides an extra direction to the phase space of the dual theory ( now parametrized by three dimensionless quantities @xmath33 , @xmath114 and @xmath115 ) , allowing for a very rich structure . in particular , by tuning these parameters one can exhibit transitions between metallic and insulating states , as was demonstrated in a similar model in @xcite . however , in this paper we restrict to two choices : @xmath116 describing a metal and @xmath117 describing an insulator . this characterization will be examined in the following section . as before , we are interested in solutions with @xmath27 and @xmath28 that correspond to turning on a charged condensate in these states . following @xcite , we compute the optical conductivity by considering coupled fluctuations of the gauge field , metric and the dilaton field of the form @xmath118 upon substitution into the equations of motion we obtain pdes involving real fields which depend on @xmath119 and @xmath70 . we then let @xmath120 for each field @xmath121 and obtain the following odes : @xmath122 as in section [ sec : axion ] , the only difference when a condensate scalar @xmath13 is turned on is an additional mass term for @xmath38 . ingoing boundary conditions at the horizon are completely analogous to that of the axion model see ( [ ingoing bc axion 1][ingoing bc axion 3 ] ) . at the boundary we have @xmath123 in terms of these coefficients , the gauge - invariant boundary condition required to calculate @xmath124 can be written as @xmath125 we extract the conductivity from fluctuations that satisfy ingoing boundary conditions at the horizon and in the uv . normal phase solutions @xmath126 were studied in @xcite . a formula for the dc conductivity in terms of the solution data was derived for a general class of axion - dilaton models in @xcite . in this particular model it reduces to @xmath127 an insulator has vanishing dc conductivity at zero temperature . in figure [ fig : axion_dilaton_sigma_dc ] we plot the dc conductivity as a function of temperature for our two points in phase space , confirming their metallic or insulating character . 2em ( 0.3,0.4)(0,0 ) ( -9,2.)(0,0)@xmath128 ( 0.2,0.2)(0,0)@xmath62 -1em -1em it is worth noting that the dc resistivity as a function of temperature in the metallic state is not monotonic but possesses a peak at finite temperature . this feature is also present in the model of @xcite and it is observed in some experimental heavy fermion setups . for the metal , the low frequency behaviour of the conductivity is well - approximated by a drude peak even at very low temperatures @xcite . in the insulating state , on the other hand , drude behaviour ceases to be a good approximation . in either case , equation gives the correct value of the dc conductivity . a full mapping of the phase diagram is lacking , and for now we will content ourselves with exploring a representative of the metallic and a representative of the insulating phases . we expect the drude / non - drude behaviour of the optical conductivity to be related to the structure of the qnm in analogy with @xcite , but we leave this interesting question for future work . next we study solutions with the charged scalar turned on . we construct these numerically using a shooting method . for the same two choices of @xmath129 we find black brane solutions with charged scalar hair that exist below some critical @xmath55 , corresponding to the broken phases of these two states . we will refer to the new phases as the broken metal and the broken insulator . in figure [ fig : axion_dilaton_condensate ] we show condensate curves for the two @xmath129 . ( 0.3,0.4)(0,0 ) ( -5.7,2.5)(0,0)@xmath61 ( 0.2,0.2)(0,0)@xmath62 -1em -1em the method for computing the free energy density is similar to that for the axion model . evaluating the euclidean on - shell action yields @xmath130 the renormalised action is given by @xmath131 and we eventually find @xmath132 this formula is valid for both the normal and broken phases . in figure [ fig : axion_dilaton_free_energy ] we demonstrate that the broken phase has lower free energy density than the normal phase at the same @xmath129 and thus is thermodynamically preferred . 2em ( 0.3,0.4)(0,0 ) ( -8.9,2.)(0,0)@xmath76 ( 0.2,0.2)(0,0)@xmath62 -1em -0.5em we present our results for the optical conductivity in the broken phase of the metal in figure [ fig : metal_sigma ] . 1em 1em to @xmath133 , from top to bottom in the left plot and from bottom to top in the right .,title="fig:",scaledwidth=40.0% ] 2em to @xmath133 , from top to bottom in the left plot and from bottom to top in the right .,title="fig:",scaledwidth=40.0% ] ( 0.3,0.4)(0,0 ) ( -9.1,2.)(0,0)@xmath81 ( -4.6,2.)(0,0)@xmath82 ( 0.2,0.2)(0,0)@xmath83 -1em in the range of temperatures we have considered , the low frequency region of the conductivity can be well - approximated by the two - fluid model with drude normal component . more specifically , for temperatures in the range @xmath134 , the value of @xmath135 obtained from the fit to the real part of the conductivity differs by no more than @xmath136 from the value obtained fitting the imaginary part , and by no more than @xmath137 from the value obtained from the characteristic time given by the lowest ( in this case , purely dissipative ) qnm . for lower temperatures , the pole in the imaginary part of the conductivity overwhelms the normal component , which makes it numerically harder to perform the fit to the imaginary part . however , the value of @xmath135 given by the fit to the real part still differs with the qnm value by only @xmath137 . for all the fits in the broken metallic case , the residuals do not exceed @xmath138 . we plot the parameters of the fit in figure [ fig : metal_fits ] . versus @xmath33 , the blue data is obtained from the fits and the yellow data comes from the lowest purely - dissipative qnm . we find excellent agreement between these two independent procedures.,title="fig:",scaledwidth=40.0% ] 2em versus @xmath33 , the blue data is obtained from the fits and the yellow data comes from the lowest purely - dissipative qnm . we find excellent agreement between these two independent procedures.,title="fig:",scaledwidth=40.0% ] 1em 1em versus @xmath33 , the blue data is obtained from the fits and the yellow data comes from the lowest purely - dissipative qnm . we find excellent agreement between these two independent procedures.,title="fig:",scaledwidth=40.0% ] ( 0.3,0.4)(0,0 ) ( -6.8,4.8)(0,0)@xmath135 ( -2.1,4.8)(0,0)@xmath139 ( 2.7,2.85)(0,0)@xmath62 ( -4.6,2.)(0,0)@xmath140 ( 0.2,0.15)(0,0)@xmath62 -1em the structure of the low frequency region can be qualitatively understood in terms of the qnm spectrum see figure [ fig : metal_qnm ] . we find that the lowest qnm is purely dissipative and well - separated from the other excitations , which , following the arguments of @xcite , accounts for the drude behaviour . moreover , the fact that the characteristic time obtained from the qnm calculation is in excellent agreement with the one obtained from the fits provides quantitative evidence in favour of this connection , in addition to a non - trivial cross check of our numerical procedure . 1em closest to the real axis as a function of temperature for the broken metal . left : we plot the purely - imaginary pole in blue and the ( imaginary part of the ) propagating pole in yellow . right : the absolute value of the difference between these two poles.,title="fig:",scaledwidth=40.0% ] 2em closest to the real axis as a function of temperature for the broken metal . left : we plot the purely - imaginary pole in blue and the ( imaginary part of the ) propagating pole in yellow . right : the absolute value of the difference between these two poles.,title="fig:",scaledwidth=40.0% ] ( 0.3,0.4)(0,0 ) ( -9.1,2.)(0,0)@xmath103 ( -4.4,2.)(0,0)@xmath141 ( 0.2,0.2)(0,0)@xmath62 -1em it is interesting to note that a pseudo gap forms as we lower the temperature , by which we mean that @xmath81 is very small for a range of frequencies but then the drude peak reappears at lower frequencies . this feature was also observed in @xcite , but not in the axion model as discussed in section [ sec : brokenaxion ] . we have also checked numerically that the fgt sum rule holds for the broken metal see figure [ fig : metal_fgt ] . once again , we need to go to high frequencies to obtain agreement . . the dashed line is the value of @xmath142 extracted from a fit of @xmath99.,scaledwidth=50.0% ] ( 0.3,0.4)(0,0 ) ( -5.7,2.5)(0,0)@xmath100 ( 0.2,0.2)(0,0)@xmath83 -1em we present our results for the optical conductivity in the broken phase of the insulator in figure [ fig : insulator_sigma ] . once again , we find a superconducting pole in the imaginary part of @xmath124 . however , in contrast to the broken metal , the two - fluid model does not give a good description for small @xmath101 . close to @xmath1 , the curves look qualitatively of the form but we find that the values obtained from fitting to the real and imaginary parts differ by more than @xmath143 . as we lower the temperature , the departure from is even more explicit since @xmath124 clearly has a qualitatively different structure . we argue that this is connected to the fact that , among the lowest - lying qnm , there is a purely - dissipative mode which is close in the complex plane to a propagating mode see figure [ fig : insulator_qnm ] . 1em down to @xmath144 , from top to bottom in the left plot and from bottom to top in the right.,title="fig:",scaledwidth=40.0% ] 2em down to @xmath144 , from top to bottom in the left plot and from bottom to top in the right.,title="fig:",scaledwidth=40.0% ] ( 0.3,0.4)(0,0 ) ( -9.1,2.)(0,0)@xmath81 ( -4.6,2.)(0,0)@xmath82 ( 0.2,0.2)(0,0)@xmath83 -1em 1em closest to the real axis as a function of temperature for the broken insulator . left : we plot the purely - imaginary pole in blue and the ( imaginary part of the ) propagating pole in yellow . right : the absolute value of the difference between these two poles , which is an order of magnitude smaller than for the broken metal see figure [ fig : metal_qnm].,title="fig:",scaledwidth=40.0% ] 2em closest to the real axis as a function of temperature for the broken insulator . left : we plot the purely - imaginary pole in blue and the ( imaginary part of the ) propagating pole in yellow . right : the absolute value of the difference between these two poles , which is an order of magnitude smaller than for the broken metal see figure [ fig : metal_qnm].,title="fig:",scaledwidth=40.0% ] ( 0.3,0.4)(0,0 ) ( -9.1,2.)(0,0)@xmath103 ( -4.4,2.)(0,0)@xmath141 ( 0.2,0.2)(0,0)@xmath62 -1em -1em in this paper we studied the condensation of a charged scalar operator in two holographic models that allow momentum to relax . we constructed black branes with charged scalar hair , corresponding to a broken phase , that are thermodynamically preferred over those without . we demonstrated that the optical conductivity in a broken phase could often be described by a two - fluid model with a normal drude component and a zero - frequency superfluid pole . in some cases , however , the normal component was non - drude , which was attributed to the lack of an isolated purely - dissipative qnm . it would be useful to study more points in the phase diagram , particularly in the axion - dilaton model , in order to obtain a complete picture . there are many similarities between our results and those found in other models , including the two - fluid model behavior at small frequencies , the rapid growth of the relaxation time as we lower the temperature , the pseudo gap in the broken metallic phase at low temperatures and the missing spectral weight manifested in the fgt sum rule . however , some differences are apparent . for example , whilst @xmath55 in the axion model decreases initially , we found that it increases for large enough values of @xmath34 , thus extending the results of @xcite . in addition , we find that the relaxation time in the broken metallic phase is not monotonic with the temperature , as opposed to the result presented in @xcite . also , unlike the inhomogeneous cases of @xcite and @xcite , we did not observe any resonances in the conductivity . as explained in @xcite , such resonances are due to coupling of the current perturbations to the sound mode in the stress tensor correlators . in our homogeneous setups , however , we consider spatially - independent perturbations that decouple from the sound mode , accounting for the absence of resonances in this case . it would be interesting to study the optical conductivity at finite momentum . next we discuss the interpretation of @xmath5 and @xmath3 appearing in . in @xcite , these parameters were identified with the density of the normal and superfluid components of the fluid , respectively . with this in mind , it is interesting to ask whether the system is fully condensed at zero temperature . proceeding naively and identifying @xmath3 with @xmath145 , where @xmath146 is indeed a density , in the axion model we find that @xmath147 tends to a constant that is less than unity at low temperatures . is the total charge density extracted from the fall - off of the gauge field in . ] this suggests that a normal component remains at zero temperature . furthermore , we observe that a greater normal component remains for larger @xmath34 . however , whilst this identification is tempting to make , one must be careful that the normalization of the superfluid density is correct . furthermore , the connection with the quantity that appeared as the superfluid density in the model of a holographic superfluid with finite supercurrent density in @xcite is not clear . we expect that one must compute the optical conductivity in that model in order to resolve this issue . if the normalization and interpretation can indeed be fixed , then the axion - dilaton models studied here may provide a laboratory with sufficiently many parameters available in order to test ` homes ' law a universal relation between the superfluid density at zero temperature and the dc conductivity in the normal phase at the critical temperature @xcite . this relation was first considered in a holographic context by @xcite . we leave these interesting open questions for further work . we conclude with a discussion of the generalizations that could be considered in the future . we have chosen two examples from a general family of bulk actions with neutral sector @xmath148 + \int x(\phi)\ , f\wedge f\ ] ] and charged sector @xmath149\ ] ] there is great freedom in this class of models . it would be worthwhile to work with a model that offers more control ; for instance , with couplings that yield a simple ir geometry at zero temperature . as a concrete example , it would be interesting to explore the superconducting phase transition in the model studied in @xcite in which the normal phase at low temperatures is governed by a novel non - trivial fixed point in the ir . in that model one could also study the effect on the broken phase when the normal phase is taken through a metal - insulator transition . moreover , alternative boundary conditions for the charged scalar could lead to additional effects , such as those described in @xcite . it would also be interesting to work directly with a more general model in order to understand which features are independent of the details of the model . it is a pleasure to thank francesco aprile , richard davison , blaise goutraux , elias kiritsis , kostas skenderis , marika taylor and benjamin withers for useful discussions . the work of ta was supported by the european research council under the european union s seventh framework programme ( erc grant agreement 307955 ) . sag was supported by national science foundation grant phy-13 - 13986 . ta thanks the physics department at the university of crete and the mathematical sciences department at the university of southampton for hospitality during the completion of this work . g. t. horowitz , j. e. santos , and d. tong , `` optical conductivity with holographic lattices , '' http://dx.doi.org/10.1007/jhep07(2012)168[_jhep_ * 1207 * ( 2012 ) 168 ] , http://arxiv.org/abs/1204.0519[arxiv:1204.0519 [ hep - th ] ] . g. t. horowitz , j. e. santos , and d. tong , `` further evidence for lattice - induced scaling , '' http://dx.doi.org/10.1007/jhep11(2012)102[_jhep_ * 1211 * ( 2012 ) 102 ] , http://arxiv.org/abs/1209.1098[arxiv:1209.1098 [ hep - th ] ] . y. ling , c. niu , j .- wu , and z .- y . xian , `` holographic lattice in einstein - maxwell - dilaton gravity , '' http://dx.doi.org/10.1007/jhep11(2013)006[_jhep_ * 1311 * ( 2013 ) 006 ] , http://arxiv.org/abs/1309.4580[arxiv:1309.4580 [ hep - th ] ] . a. donos and j. p. gauntlett , `` the thermoelectric properties of inhomogeneous holographic lattices , '' http://arxiv.org/abs/1409.6875[arxiv:1409.6875 [ hep - th ] ] . s. a. hartnoll and d. m. hofman , `` locally critical resistivities from umklapp scattering , '' http://dx.doi.org/10.1103/physrevlett.108.241601 [ _ phys.rev.lett . _ * 108 * ( 2012 ) 241601 ] , http://arxiv.org/abs/1201.3917[arxiv:1201.3917 [ hep - th ] ] . p. chesler , a. lucas , and s. sachdev , `` conformal field theories in a periodic potential : results from holography and field theory , '' http://dx.doi.org/10.1103/physrevd.89.026005[_phys.rev._ * d89 * ( 2014 ) 026005 ] , http://arxiv.org/abs/1308.0329[arxiv:1308.0329 [ hep - th ] ] . a. karch and a. obannon , `` metallic ads / cft , '' http://dx.doi.org/10.1088/1126-6708/2007/09/024[_jhep_ * 0709 * ( 2007 ) 024 ] , http://arxiv.org/abs/0705.3870[arxiv:0705.3870 [ hep - th ] ] . t. faulkner , n. iqbal , h. liu , j. mcgreevy , and d. vegh , `` from black holes to strange metals , '' http://arxiv.org/abs/1003.1728[arxiv:1003.1728 [ hep - th ] ] . s. a. hartnoll , j. polchinski , e. silverstein , and d. tong , `` towards strange metallic holography , '' http://dx.doi.org/10.1007/jhep04(2010)120 [ _ jhep _ * 1004 * ( 2010 ) 120 ] , http://arxiv.org/abs/0912.1061[arxiv:0912.1061 [ hep - th ] ] . n. iizuka , s. kachru , n. kundu , p. narayan , n. sircar , _ et al . _ , `` bianchi attractors : a classification of extremal black brane geometries , '' http://dx.doi.org/10.1007/jhep07(2012)193[_jhep_ * 1207 * ( 2012 ) 193 ] , http://arxiv.org/abs/1201.4861[arxiv:1201.4861 [ hep - th ] ] . a. donos and s. a. hartnoll , `` interaction - driven localization in holography , '' http://dx.doi.org/10.1038/nphys2701[_nature phys . _ * 9 * ( 2013 ) 649655 ] , http://arxiv.org/abs/1212.2998[arxiv:1212.2998 ] . t. andrade and b. withers , `` a simple holographic model of momentum relaxation , '' http://dx.doi.org/10.1007/jhep05(2014)101[_jhep_ * 1405 * ( 2014 ) 101 ] , http://arxiv.org/abs/1311.5157[arxiv:1311.5157 [ hep - th ] ] . y. bardoux , m. m. caldarelli , and c. charmousis , `` shaping black holes with free fields , '' http://dx.doi.org/10.1007/jhep05(2012)054[_jhep_ * 1205 * ( 2012 ) 054 ] , http://arxiv.org/abs/1202.4458[arxiv:1202.4458 [ hep - th ] ] . kim , k. k. kim , y. seo , and s .- j . sin , `` coherent / incoherent metal transition in a holographic model , '' http://arxiv.org/abs/1409.8346[arxiv:1409.8346 [ hep - th ] ] . s. a. hartnoll , `` theory of universal incoherent metallic transport , '' http://arxiv.org/abs/1405.3651[arxiv:1405.3651 [ cond-mat.str-el ] ] . r. a. davison and b. goutraux , `` momentum dissipation and effective theories of coherent and incoherent transport , '' http://arxiv.org/abs/1411.1062[arxiv:1411.1062 [ hep - th ] ] . a. donos and j. p. gauntlett , `` holographic q - lattices , '' http://dx.doi.org/10.1007/jhep04(2014)040[_jhep_ * 1404 * ( 2014 ) 040 ] , http://arxiv.org/abs/1311.3292[arxiv:1311.3292 [ hep - th ] ] . a. donos and j. p. gauntlett , `` novel metals and insulators from holography , '' http://dx.doi.org/10.1007/jhep06(2014)007[_jhep_ * 1406 * ( 2014 ) 007 ] , http://arxiv.org/abs/1401.5077[arxiv:1401.5077 [ hep - th ] ] . b. goutraux , `` charge transport in holography with momentum dissipation , '' http://dx.doi.org/10.1007/jhep04(2014)181[_jhep_ * 1404 * ( 2014 ) 181 ] , http://arxiv.org/abs/1401.5436[arxiv:1401.5436 [ hep - th ] ] . a. donos and j. p. gauntlett , `` thermoelectric dc conductivities from black hole horizons , '' http://dx.doi.org/10.1007/jhep11(2014)081[_jhep_ * 1411 * ( 2014 ) 081 ] , http://arxiv.org/abs/1406.4742[arxiv:1406.4742 [ hep - th ] ] . e. mefford and g. t. horowitz , `` simple holographic insulator , '' http://dx.doi.org/10.1103/physrevd.90.084042[_phys.rev._ * d90 * no . 8 , ( 2014 ) 084042 ] , http://arxiv.org/abs/1406.4188[arxiv:1406.4188 [ hep - th ] ] . m. blake and a. donos , `` quantum critical transport and the hall angle , '' http://arxiv.org/abs/1406.1659[arxiv:1406.1659 [ hep - th ] ] . m. taylor and w. woodhead , `` inhomogeneity simplified , '' http://arxiv.org/abs/1406.4870[arxiv:1406.4870 [ hep - th ] ] . l. cheng , x .- h . ge , and z .- y . sun , `` thermoelectric dc conductivities with momentum dissipation from higher derivative gravity , '' http://arxiv.org/abs/1411.5452[arxiv:1411.5452 [ hep - th ] ] . d. vegh , `` holography without translational symmetry , '' http://arxiv.org/abs/1301.0537[arxiv:1301.0537 [ hep - th ] ] . c. de rham , g. gabadadze , and a. j. tolley , `` resummation of massive gravity , '' http://dx.doi.org/10.1103/physrevlett.106.231101 [ _ phys.rev.lett . _ * 106 * ( 2011 ) 231101 ] , http://arxiv.org/abs/1011.1232[arxiv:1011.1232 [ hep - th ] ] . s. deser , k. izumi , y. ong , and a. waldron , `` problems of massive gravities , '' http://arxiv.org/abs/1410.2289[arxiv:1410.2289 [ hep - th ] ] . m. blake and d. tong , `` universal resistivity from holographic massive gravity , '' http://dx.doi.org/10.1103/physrevd.88.106004 [ _ phys.rev . * d88 * no . 10 , ( 2013 ) 106004 ] , http://arxiv.org/abs/1308.4970[arxiv:1308.4970 [ hep - th ] ] . n. iqbal and h. liu , `` universality of the hydrodynamic limit in ads / cft and the membrane paradigm , '' http://dx.doi.org/10.1103/physrevd.79.025023[_phys.rev._ * d79 * ( 2009 ) 025023 ] , http://arxiv.org/abs/0809.3808[arxiv:0809.3808 [ hep - th ] ] . m. blake , d. tong , and d. vegh , `` holographic lattices give the graviton an effective mass , '' http://dx.doi.org/10.1103/physrevlett.112.071602[_phys.rev.lett._ * 112 * no . 7 , ( 2014 ) 071602 ] , http://arxiv.org/abs/1310.3832[arxiv:1310.3832 [ hep - th ] ] . r. a. davison , `` momentum relaxation in holographic massive gravity , '' http://dx.doi.org/10.1103/physrevd.88.086003[_phys.rev._ * d88 * ( 2013 ) 086003 ] , http://arxiv.org/abs/1306.5792[arxiv:1306.5792 [ hep - th ] ] . a. amoretti , a. braggio , n. maggiore , n. magnoli , and d. musso , `` thermo - electric transport in gauge / gravity models with momentum dissipation , '' http://dx.doi.org/10.1007/jhep09(2014)160[_jhep_ * 1409 * ( 2014 ) 160 ] , http://arxiv.org/abs/1406.4134[arxiv:1406.4134 [ hep - th ] ] . a. amoretti , a. braggio , n. maggiore , n. magnoli , and d. musso , `` analytic dc thermo - electric conductivities in holography with massive gravitons , '' http://arxiv.org/abs/1407.0306[arxiv:1407.0306 [ hep - th ] ] . m. baggioli and o. pujolas , `` holographic polarons , the metal - insulator transition and massive gravity , '' http://arxiv.org/abs/1411.1003[arxiv:1411.1003 [ hep - th ] ] . a. adams , d. a. roberts , and o. saremi , `` hawking - page transition in holographic massive gravity , '' http://dx.doi.org/10.1103/physrevd.91.046003[_phys.rev._ * d91 * no . 4 , ( 2015 ) 046003 ] , http://arxiv.org/abs/1408.6560[arxiv:1408.6560 [ hep - th ] ] . s. a. hartnoll , c. p. herzog , and g. t. horowitz , `` building a holographic superconductor , '' http://dx.doi.org/10.1103/physrevlett.101.031601[_phys.rev.lett._ * 101 * ( 2008 ) 031601 ] , http://arxiv.org/abs/0803.3295[arxiv:0803.3295 [ hep - th ] ] . s. a. hartnoll , c. p. herzog , and g. t. horowitz , `` holographic superconductors , '' http://dx.doi.org/10.1088/1126-6708/2008/12/015[_jhep_ * 0812 * ( 2008 ) 015 ] , http://arxiv.org/abs/0810.1563[arxiv:0810.1563 [ hep - th ] ] . s. s. gubser , `` breaking an abelian gauge symmetry near a black hole horizon , '' http://dx.doi.org/10.1103/physrevd.78.065034 [ _ phys.rev . _ * d78 * ( 2008 ) 065034 ] , http://arxiv.org/abs/0801.2977[arxiv:0801.2977 [ hep - th ] ] . g. t. horowitz and j. e. santos , `` general relativity and the cuprates , '' http://arxiv.org/abs/1302.6586[arxiv:1302.6586 [ hep - th ] ] . h. b. zeng and j .- p . wu , `` holographic superconductors from the massive gravity , '' http://dx.doi.org/10.1103/physrevd.90.046001 [ _ phys.rev . _ * d90 * ( 2014 ) 046001 ] , http://arxiv.org/abs/1404.5321[arxiv:1404.5321 [ hep - th ] ] . koga , k. maeda , and k. tomoda , `` holographic superconductor model in a spatially anisotropic background , '' http://dx.doi.org/10.1103/physrevd.89.104024[_phys.rev._ * d89 * no . 10 , ( 2014 ) 104024 ] , http://arxiv.org/abs/1401.6501[arxiv:1401.6501 [ hep - th ] ] . d. mateos and d. trancanelli , `` the anisotropic n=4 super yang - mills plasma and its instabilities , '' http://dx.doi.org/10.1103/physrevlett.107.101601[_phys.rev.lett._ * 107 * ( 2011 ) 101601 ] , http://arxiv.org/abs/1105.3472[arxiv:1105.3472 [ hep - th ] ] . d. mateos and d. trancanelli , `` thermodynamics and instabilities of a strongly coupled anisotropic plasma , '' http://dx.doi.org/10.1007/jhep07(2011)054[_jhep_ * 1107 * ( 2011 ) 054 ] , http://arxiv.org/abs/1106.1637[arxiv:1106.1637 [ hep - th ] ] . x. bai , b .- h . lee , m. park , and k. sunly , `` dynamical condensation in a holographic superconductor model with anisotropy , '' http://dx.doi.org/10.1007/jhep09(2014)054[_jhep_ * 1409 * ( 2014 ) 054 ] , http://arxiv.org/abs/1405.1806[arxiv:1405.1806 [ hep - th ] ] . y. ling , p. liu , c. niu , j .- wu , and z .- y . xian , `` holographic superconductor on q - lattice , '' http://arxiv.org/abs/1410.6761[arxiv:1410.6761 [ hep - th ] ] . f. aprile and t. ishii , `` a simple holographic model of a charged lattice , '' http://dx.doi.org/10.1007/jhep10(2014)151[_jhep_ * 1410 * ( 2014 ) 151 ] , http://arxiv.org/abs/1406.7193[arxiv:1406.7193 [ hep - th ] ] . r. a. ferrell and r. e. glover , `` conductivity of superconducting films : a sum rule , '' http://dx.doi.org/10.1103/physrev.109.1398[_phys.rev._ * 109 * ( 1958 ) 13981399 ] . m. tinkham and r. ferrell , `` determination of the superconducting skin depth from the energy gap and sum rule , '' http://dx.doi.org/10.1103/physrevlett.2.331[_phys . lett . _ * 2 * ( 1959 ) 331333 ] . d. birmingham , i. sachs , and s. n. solodukhin , `` conformal field theory interpretation of black hole quasinormal modes , '' http://dx.doi.org/10.1103/physrevlett.88.151301[_phys.rev.lett._ * 88 * ( 2002 ) 151301 ] , http://arxiv.org/abs/hep-th/0112055[arxiv:hep-th/0112055 [ hep - th ] ] . d. t. son and a. o. starinets , `` minkowski space correlators in ads / cft correspondence : recipe and applications , '' http://dx.doi.org/10.1088/1126-6708/2002/09/042[_jhep_ * 0209 * ( 2002 ) 042 ] , http://arxiv.org/abs/hep-th/0205051[arxiv:hep-th/0205051 [ hep - th ] ] . m. henningson and k. skenderis , `` the holographic weyl anomaly , '' http://dx.doi.org/10.1088/1126-6708/1998/07/023[_jhep_ * 9807 * ( 1998 ) 023 ] , http://arxiv.org/abs/hep-th/9806087[arxiv:hep-th/9806087 [ hep - th ] ] . v. balasubramanian and p. kraus , `` a stress tensor for anti - de sitter gravity , '' http://dx.doi.org/10.1007/s002200050764 [ _ commun.math.phys . _ * 208 * ( 1999 ) 413428 ] , http://arxiv.org/abs/hep-th/9902121[arxiv:hep-th/9902121 [ hep - th ] ] . j. sonner and b. withers , `` a gravity derivation of the tisza - landau model in ads / cft , '' http://dx.doi.org/10.1103/physrevd.82.026001 [ _ phys.rev . _ * d82 * ( 2010 ) 026001 ] , http://arxiv.org/abs/1004.2707[arxiv:1004.2707 [ hep - th ] ] . c. homes , s. dordevic , m. strongin , d. bonn , r. liang , _ et al . _ , `` universal scaling relation in high - temperature superconductors , '' http://dx.doi.org/10.1038/nature02673[_nature_ * 430 * ( 2004 ) 539 ] , http://arxiv.org/abs/cond-mat/0404216[arxiv:cond-mat/0404216 [ cond - mat ] ] . c. homes , s. dordevic , t. valla , and m. strongin , `` scaling of the superfluid density in high - temperature superconductors , '' http://dx.doi.org/10.1103/physrevb.72.134517[_phys . b _ * 72 * ( 2005 ) 134517 ] . j. erdmenger , p. kerner , and s. muller , `` towards a holographic realization of homes law , '' http://dx.doi.org/10.1007/jhep10(2012)021[_jhep_ * 1210 * ( 2012 ) 021 ] , http://arxiv.org/abs/1206.5305[arxiv:1206.5305 [ hep - th ] ] . g. t. horowitz and m. m. roberts , `` holographic superconductors with various condensates , '' http://dx.doi.org/10.1103/physrevd.78.126008 [ _ phys.rev . _ * d78 * ( 2008 ) 126008 ] , http://arxiv.org/abs/0810.1077[arxiv:0810.1077 [ hep - th ] ] .

a possible interplay between the two terms of the general type - ii seesaw formula is exercised which leads to the generation of nonzero . the specific flavor structure of the model , guided by the symmetry and accompanied with the standard model singlet flavons , yields the conventional seesaw contribution to produce the tribimaximal lepton mixing which is further corrected by the presence of the triplet contribution to accommodate . we consider the cp symmetry to be spontaneously broken by the complex vacuum expectation value ( vev ) of a singlet field . while the magnitude of its complex vev is responsible for generating , its phase part induces the low energy cp violating phase ( ) and the cp violation required for leptogenesis . hence the triplet contribution , although sub - dominant , plays crucial role in providing a common source for non - zero , and cp - violation required for leptogenesis . we find that the recent hint for close to is somewhat favored in this set - up though it excludes the exact equality with . we also discuss the generation of lepton asymmetry in this scenario . * spontaneous cp violation in lepton - sector : a common origin for , dirac cp phase and leptogenesis * + biswajit karmakar , arunansu sil + _ indian institute of technology guwahati , 781039 assam , india _

1509.07090

the question whether there exists an underlying principle to understand the pattern of lepton mixing , which is quite different from the quark mixing , demands the study of neutrino mass matrix as well as the charged lepton one into a deeper level . the smallness of neutrino masses can be well understood by the seesaw mechanism in a natural way . i seesaw mechanism @xcite provides the simplest possibility by extending the standard model ( sm ) with three right - handed ( rh ) neutrinos . an introduction of discrete symmetries into it may reveal the flavor structure of the neutrino and charged lepton mass matrix . for example , a type - i seesaw in conjugation with @xmath8 explains the tribimaximal lepton mixing pattern ( tbm ) @xcite in presence of sm singlet flavon ( charged under @xmath8 ) fields which get vacuum expectation values ( vev)@xcite . however the original approach fails to accommodate the recent observation of non - zero @xmath0@xcite . in @xcite , we have shown that an extension of the altarelli - feruglio ( af ) model @xcite by one additional flavon field can be employed to have a nonzero @xmath0 consistent with the present experimental results . the set - up also constraints the two majorana phases involved in the lepton mixing matrix . the deviation of the tbm pattern is achieved through a deformation of the rh neutrino mass matrix compared to the original one . on the other hand , within the framework of a general type - ii seesaw mechanism ( where both rh neutrinos and @xmath2 triplet higgs are present ) , light neutrino mass depends upon comparative magnitude of the pure type - i ( mediated by heavy rh neutrinos ) and triplet contributions . this interplay is well studied in the literature @xcite . in recent years keeping in mind that @xmath0 is nonzero , efforts have been given to realize leptogenesis @xcite and linking it with @xmath0 in models based on type - ii seesaw@xcite . in this article , we focus on the generation of light neutrino mass matrix through a type - ii seesaw mechanism @xcite . the fields content of the sm is extended with three right - handed neutrinos , one @xmath2 triplet and a set of sm singlet flavon fields . a flavor symmetry @xmath1 is considered . the type - ii seesaw mechanism therefore consists of the conventional type - i seesaw contribution ( @xmath9 ) along with the triplet contribution ( @xmath10 ) to the neutrino mass matrix . here we find the type - i contribution alone can generate the tbm mixing pattern , where the charged lepton mass matrix is a diagonal one . then we have shown that the same flavor symmetry allows us to have a deviation from the conventional type - i contribution , triggered by the @xmath2 triplet s vev . we have found that this deviation is sufficient enough to keep @xmath0 at an acceptable level@xcite . we mostly consider the triplet contribution to the light neutrino mass is subdominat compared to the conventional type - i contribution . we further assume that apart from the flavons ( sm singlets charged under @xmath8 ) involved , there is a @xmath8 singlet ( as well as sm gauge singlet ) field @xmath3 , which gets a complex vacuum expectation value and thereby responsible for spontaneous cp violation@xcite and other extensions of standard model @xcite . ] at high scale @xcite . all other flavons have real vevs and all the couplings involved are considered to be real . it turns out that the magnitude of this complex vacuum expectation value of @xmath3 is responsible for the deviation of tbm by generating nonzero value of @xmath0 in the right ballpark . on the other hand , the phase associated with it generates the dirac cp violating phase in the lepton sector . so in a way , the triplet contribution provides a unified source for cp violation and nonzero @xmath0 . in lepton sector , the other possibilities where cp violation can take place , involves complex vev of higgs triplets@xcite , or when a higgs bi - doublet ( particularly in left - right models ) gets complex vev@xcite or in a mixed situation @xcite . however , we will concentrate in a situation where a scalar singlet @xmath3 present in the theory gets complex vev as in @xcite . we have also studied the lepton asymmetry production through the decay of the heavy triplet involved . the decay of the triplet into two leptons contributes to the asymmetry where the virtual rh neutrinos are involved in the loop . this process is effective when the triplet is lighter than all the rh neutrinos . it turns out that sufficient lepton asymmetry can be generated in this way . on the other hand if the triplet mass is heavier than the rh neutrino masses , the lightest rh neutrino may be responsible for producing lepton asymmetry where the virtual triplet is contributing in the one loop diagram . in @xcite , authors investigated a scenario where the triplet vevs are the sole contribution to the light neutrino mass and a single source of spontaneous cp violation was considered . there , it was shown that the low energy cp violating phase and the cp violation required for leptogenesis both are governed by the argument of the complex vev of that scalar field . the nonzero value of @xmath0 however followed from a perturbative deformation of the vev alignment of the flavons involved . here in our scenario , the tbm pattern is realized by the conventional type - i contribution . therefore in the tbm limit , @xmath0 is zero in our set - up . also there is no cp violating phase in this limit as all the flavons involved in @xmath9 are carrying real vevs , and hence no lepton asymmetry as well . now once the triplet contribution ( @xmath10 ) is switched on , not only the @xmath0 , but also the leptonic cp violation turn out to be nonzero . for generating lepton asymmetry , two triplets were essential in @xcite , while we could explain the lepton asymmetry by a single triplet along with the presence of rh neutrinos . in this case , the rh neutrinos are heavier compared to the mass of the triplet involved . the paper is organized as follows . in section [ sec2 ] , we provide the status of the neutrino mixing and the mass squared differences . then in section [ sec3 ] , we describe the set - up of the model followed by constraining the parameter space of the framework from neutrino masses and mixing in section [ nucons ] . in section [ sec : lep ] , we describe how one can obtain lepton asymmetry out of this construction . finally we conclude in section [ conlc ] . here we summarize the neutrino mixing parameters and their present status . the neutrino mass matrix @xmath11 , in general , can be diagonalized by the @xmath12 matrix ( in the basis where charged lepton mass matrix is diagonal ) as @xmath13 , where @xmath14 are the real mass eigenvalues for light neutrinos . the standard parametrization of the @xmath12 matrix @xcite is given by @xmath15 where @xmath16 , @xmath17 , the angles @xmath18 $ ] , @xmath19 $ ] is the cp - violating dirac phase while @xmath20 and @xmath21 are the two cp - violating majorana phases . the mixing angles @xmath22 , @xmath23 and the two mass - squared differences @xmath24 have been well measured at several neutrino oscillation experiments @xcite . recently the other mixing angle @xmath0 is also reported to be of sizable magnitude @xcite . very recently , we start to get hint for nonzero dirac cp phase@xcite . from the updated global analysis @xcite involving all the data from neutrino experiments , the 1@xmath25 and 3@xmath25 ranges of mixing angles and the mass - squared differences are mentioned ( nh and ih stand for the normal and inverted mass hierarchies respectively ) in table [ tab1 ] . the result by planck@xcite from the analysis of cosmic microwave background ( cmb ) also sets an upper limit on the sum of the three neutrino masses as given by , @xmath26 ev . the result from neutrinoless double beta decay by kamland - zen@xcite and exo-200 @xcite indicates a limit on the effective neutrino mass parameter @xmath27 as , @xmath28 ev at 90@xmath29 cl and @xmath30 ev at 90@xmath29 cl respectively . our starting point is the conventional type - i seesaw mechanism to explain the smallness of light neutrino masses which further predicts a tribimaximal mixing ( tbm ) pattern in the lepton sector . for this part , we use the original af model @xcite by introducing a discrete @xmath8 symmetry and @xmath8 triplet flavon fields @xmath31 along with a singlet @xmath32 field . of course three right handed neutrinos ( @xmath33 ) are also incorporated . in addition , we include a @xmath2 triplet field ( @xmath34 ) with hypercharge unity , the vev of which produces an additional contribution ( hereafter called the triplet contribution ) to the light neutrino mass . so our set - up basically involves a general type - ii seesaw , @xmath35 where @xmath36 is the typical type - i term and @xmath10 is the triplet contribution . to realize both , the relevant lagrangian for generation of @xmath11 can be written as , @xmath37 so that @xmath38 and @xmath39 , where @xmath40 and @xmath41 are the vevs of the triplet @xmath34 and sm higgs doublet ( @xmath42 ) respectively . @xmath43 and @xmath44 correspond to the yukawa matrices for the dirac mass and triplet terms respectively , the flavor structure of which are solely determined by the discrete symmetries imposed on the fields involved in the model . @xmath45 is the majorana mass of the rh neutrinos . in the following subsection , we discuss in detail how the flavor structure of @xmath46 and @xmath45 are generated with the flavon fields . a discrete symmetry @xmath47 is also present in our model and two other sm singlet fields @xmath48 and @xmath49 are introduced . these additional fields and the discrete symmetries considered play crucial role in realizing a typical structure of the triplet contribution to the light neutrino mass matrix as we will see below . among all these scalar fields present , only the @xmath3 field is assumed to have a complex vev while all other vevs are real . the framework is based on the sm gauge group extended with the @xmath1 symmetry . the field contents and charges under the symmetries imposed are provided in table [ teb2 ] . with the above fields content , the charged lepton lagrangian is described by , @xmath50 to the leading order , where @xmath51 is the cut - off scale of the theory and @xmath52 and @xmath53 are the respective coupling constants . terms in the first parenthesis represent products of two @xmath8 triplets , which further contracts with @xmath8 singlets @xmath54 , @xmath55 and @xmath56 corresponding to @xmath57 and @xmath58 respectively to make a true singlet under @xmath8 . once the flavons @xmath59 and @xmath60 get the vevs along a suitable direction as @xmath61 and @xmath62 respectively and @xmath60 are assumed here . we expect the minimization of the potential involving @xmath59 and @xmath60 can produce this by proper tuning of the parameters involved in the potential . however the very details of it are beyond the scope of this paper . ] , it leads to a diagonal mass matrix for charged leptons , once the higgs vev @xmath41 is inserted . below we will first summarize how the tbm mixing is achieved followed by the triplet contribution in the next subsection . the requirement of introducing sm singlet fields will be explained subsequently while discussing the flavor structure of neutrino mass matrix in detail . the relevant lagrangian for the type - i seesaw in the neutrino sector is given by , @xmath63 where @xmath64 and @xmath65 are the coupling constants . after the @xmath66 and @xmath59 fields get vevs and the electroweak vev @xmath41 is included , it yields the following flavor structure for dirac ( @xmath67 ) and majorana ( @xmath68 ) mass matrices , @xmath69 with @xmath70 . the @xmath8 multiplication rules that results to this flavor structure can be found in @xcite . therefore the contribution toward light neutrino mass that results from the type - i seesaw mechanism is found to be , @xmath71 note that this form of @xmath9 indicates that the corresponding diagonalizing matrix would be nothing but the tbm mixing matrix of the form @xcite @xmath72 as a characteristic of typical @xmath8 generated structure , the rh neutrinos mass matrix is as well diagonalized by the @xmath73 . in order to achieve the real and positive mass eigenvalues , the corresponding rotation @xmath74 is provided on @xmath45 as @xmath75 with @xmath76 once @xmath77 is considered . on the other hand for @xmath78 ; through @xmath79 itself , the real and positive eigenvalues of @xmath45 [ @xmath80 can be obtained . this would be useful when we will consider the decay of the rh neutrinos for leptogenesis in section [ sec : lep ] . the leading order lagrangian invariant under the symmetries imposed , that describes the triplet contribution to the light neutrino mass matrix ( @xmath10 ) , is given by , @xmath81 where @xmath82 and @xmath83 are the couplings involved . here @xmath48 develops a vev @xmath84 and the singlet @xmath3 is having a complex vev @xmath85 . as we have mentioned before , the vev of @xmath3 provides the unique source of cp violation as all other vevs and couplings are assumed to be real . cp is therefore assumed to be conserved in all the terms involved in the lagrangian . similar to @xcite , cp is spontaneously broken by the complex vev of the @xmath3 field . after plugging all these vevs , the above lagrangian in eq.([ld ] ) contributes to the following yukawa matrix for the triplet @xmath34 as given by , @xmath86 this specific structure follows from the @xmath8 charge assignments of various fields present in eq.([ld ] ) and is instrumental in providing nonzero @xmath0 as we will see shortly . before discussing the vev of the @xmath34 field , let us describe the complete scalar potential @xmath87 , including the triplet @xmath34 obeying the symmetries imposed , is given by , @xmath88 where @xmath89 ^ 2,\\\nonumber v_{sh}&=&\lambda_6 ( s^*s)h^{\dagger}h+\lambda_7 ( s^2+s^{*2})(h^{\dagger}h),\nonumber\\ v_{s\delta}&=&{\rm tr}(\delta^{\dagger}\delta ) [ \lambda_8 ( s^2+s^{*2})+\lambda_9 s^*s],\nonumber\\ v_{\delta h}&=&\lambda_{10}(h^{\dagger}h){\rm tr}(\delta^{\dagger}\delta ) + \lambda_{11}(h^{\dagger}\delta^{\dagger}\delta h ) + \left(-\frac{\mu}{\lambda}\widetilde{h}^t\delta\widetilde{h}\phi_{{\mbox{\tiny$s$}}}\phi_{{\mbox{\tiny$t$ } } } + h.c.\right).\end{aligned}\ ] ] the above potential contains several dimensionful ( denoted by @xmath90 ) and dimensionless parameters ( as @xmath91 ) , which are all considered to be real . similar to @xcite , here also it can be shown that the @xmath3 field gets a complex vev for a choice of parameters involved in @xmath92 as @xmath93 and @xmath94 . however contrary to @xcite , here we have only a single triplet field @xmath34 . once the @xmath31 get vevs , the last term of @xmath95 results into an effective @xmath96 interaction which would be important for leptogenesis . the vev of the triplet @xmath34 is obtained by minimizing the relevant terms . ] from @xmath87 after plugging the vevs of the flavons and is given by @xmath97 using eqs.([ydelta ] ) and ( [ d1d2 ] ) , the triplet contribution to the light neutrino mass matrix follows from the lagrangian @xmath98 as @xmath99 where @xmath100 note that only the triplet contribution ( @xmath101 ) involves the phase due to the involvement of @xmath102 in @xmath103 , while the entire type - i contribution @xmath9 remains real . therefore the term @xmath101 serves as the unique source of generating all the cp - violating phases involved in neutrino as well as in lepton mixing . this will be clear once we discuss the neutrino mixing in the subsequent section . now we can write down the entire contribution to the light neutrino mass as , @xmath104 in this section , we discuss how the neutrino masses and mixing can be obtained from the @xmath11 mentioned above . keeping in mind that @xmath9 can be diagonalized by @xmath73 , we first perform a rotation by @xmath73 on the explicit form of the light neutrino mass matrix obtained in eq.([m1p2 ] ) and the rotated @xmath11 is found to be @xmath105 here , we define the parameters @xmath106 and @xmath107 which are real and positive as part of the type - i contribution . we note that a further rotation by @xmath108 ( another unitary matrix ) in the 13 plane in required to diagonalize the light neutrino mass matrix , @xmath109 . with a form of @xmath108 as @xmath110 we have , @xmath111 , where @xmath112 are the real and positive eigenvalues and @xmath113 are the phases associated to these mass eigenvalues . we can therefore extract the neutrino mixing matrix @xmath114 as , @xmath115 where @xmath116 is the majorana phase matrix with @xmath117 and @xmath118 , one common phase being irrelevant . as the charged lepton mass matrix is a diagonal one , we can now compare this @xmath114 with the standard parametrization of lepton mixing matrix @xmath12 . the @xmath12 is therefore given by @xmath119 , where we need to multiply the @xmath114 matrix by a diagonal phase matrix @xmath120 @xcite from left so that the @xmath12 excluding the majorana phase matrix , can take the standard form where 22 and 33 elements are real as in eq.([upmns ] ) . hence we obtain the usual ( in @xmath8 models ) correlation @xcite between the angles and cp violating dirac phase @xmath4 as given by @xmath121 \label{ang22}. \end{aligned}\ ] ] the angle @xmath122 and phase @xmath123 associated with @xmath108 can now be linked with the parameters involved in @xmath11 . for this we first rewrite the triplet contribution @xmath101 as @xmath124 and define a parameter @xmath125 ( hence @xmath126 is real ) . this parameter indicates the relative size of the triplet contribution to the type - i contribution when @xmath127 . as @xmath108 diagonalizes the @xmath128 matrix , after some involved algebra , we finally get , @xmath129^{1/2 } } { \frac{2}{\beta \left ( 1-\alpha^2\right ) } + \cos{\phi_d } } ~~{\rm and } ~~ \tan{\psi } = ( \tan{\phi_d})/\alpha . \label{theta - psi}\ ] ] @xmath130 may take positive or negative value depending on the choices of @xmath131 as evident from the first relation in eq.([theta - psi ] ) . for @xmath132 , we find @xmath133 using @xmath134 and the second relation of eq.([theta - psi]).on the other hand for @xmath135 ; @xmath4 and @xmath123 are related by @xmath136 . therefore in both these cases we obtain @xmath137 and hence @xmath138 in our set - up , the source of this cp - violating dirac phase @xmath4 is through the phase @xmath139 associated with @xmath102 . note that @xmath140 is related with @xmath141 and @xmath142 as seen from eq.([delta - phid ] ) . now from the relation @xmath143 and using eq.([ydelta ] ) and ( [ modd ] ) , we obtain @xmath144 satisfying @xmath145 where @xmath82 and @xmath83 are the coupling involved in eq.([ld ] ) . as seen from eqs.([ang ] ) and ( [ theta - psi ] ) , we conclude that the @xmath12 parameters @xmath0 and @xmath4 depend on the model parameters @xmath142 , @xmath126 and @xmath144 . note that we expect terms @xmath146 and @xmath147 ( @xmath106 ) to be of similar order of magnitude as both originated from the tree level lagrangian ( see eqs.([l1 ] ) and ( [ mdmr ] ) ) . we categorize @xmath148 as case a , while @xmath149 is with case b. the other parameter @xmath126 basically represents the relative order of magnitude between the triplet contribution ( @xmath150 ) and the type - i contribution ( @xmath151 ) . our framework produces the tbm mixing pattern to be generated solely by type - i seesaw and triplet contribution is present mainly to correct for the angle @xmath0 which is small compared to the other mixing angles . therefore we consider that the triplet contribution is preferably the sub - dominant or at most comparable one . therefore we expect the parameter @xmath126 to be less than one . in the @xmath152 plane for various choices of @xmath4 as indicated inside the figure . left panel is for ( a ) @xmath148 and right panel is with ( b ) @xmath149 . , title="fig : " ] in the @xmath152 plane for various choices of @xmath4 as indicated inside the figure . left panel is for ( a ) @xmath148 and right panel is with ( b ) @xmath149 . , title="fig : " ] although we discuss what happens when @xmath153 in some cases , we will restrict ourselves with @xmath154 for the most of the analyses involved later in this work . in fig.[s13l ] left panel , we study the variation of @xmath142 and @xmath126 in order to achieve the best fit value of @xmath155 @xcite while different values of @xmath4 are considered . in producing these plots , we have replaced the @xmath144 dependence in terms of @xmath142 and @xmath4 by employing the second equation in eq.([theta - psi ] ) as @xmath156 . similarly in the right panel of fig.[s13l ] , contour plots for @xmath157 are depicted for @xmath149 with different values of @xmath4 . we find a typical contour plot for @xmath158 with a specific @xmath4 value coinsides with the one with other @xmath4 values obtained from @xmath159 . for example , one particular contour plot for @xmath160 is repeated for @xmath161 . diagonalizing @xmath162 in eq.([mnupl1 ] ) , the light neutrino masses turn out to be , @xmath163^{1/2},\label{m1}\\ m_2&=&k\left[1+\beta^2 - 2\beta\cos\phi_d\right]^{1/2},\label{m2 } \\ m_3&=&k\left[\left(\frac{\alpha}{\pm(1-\alpha^2 ) } + \frac{p}{k } \right)^2 + \left(\frac{q}{k}\right)^2\right]^{1/2}\label{m3},\end{aligned}\ ] ] where @xmath164 and @xmath165 are defined as , @xmath166 the ` @xmath167 ' sign in the expression of @xmath168 and @xmath169 is for @xmath148 ( case a ) where the ` @xmath170 ' sign is associated with @xmath149 ( case b ) . the majorana phases in @xmath171 ( see eq.([unu ] ) ) are found to be @xmath172\label{majo1}-\tan^{-1}\left [ \frac{\beta\sin\phi_d}{\beta\cos\phi_d-1 } \right]\label{majo2 } , \\ \alpha_{31}&=&\pi+\tan^{-1}\left [ \frac{q / k}{p / k \pm \frac{\alpha}{(\alpha^2 - 1 ) } } \right]\label{majo1}-\tan^{-1}\left [ \frac{q / k}{p / k \pm \frac{\alpha}{(1-\alpha^2 ) } } \right]\label{majo3}.\end{aligned}\ ] ] note that the redefined parameters @xmath173 and @xmath174 are functions of @xmath142 , @xmath126 and @xmath144 , while the mass eigenvalues @xmath175 , depend on @xmath176 as well . the parameters @xmath131 and @xmath144 can now be constrained by the neutrino oscillation data . to have a more concrete discussion , we consider the ratio , @xmath177 , defined by @xmath178 , with @xmath179 and @xmath180 considering normal hierarchy . following @xcite , the best fit values of @xmath181 and @xmath182 are used for our analysis . using eqs.([m1]-[m3 ] ) , we have an expression for @xmath177 as , @xmath183.\end{aligned}\ ] ] here also , ` @xmath167 ' corresponds to case a ( @xmath184 with @xmath148 ) and ` @xmath170 ' is for case b ( @xmath185 when @xmath149 ) . interestingly we note that @xmath177 depends on @xmath131 and @xmath144 . therefore using this expression of @xmath177 , we can now have a contour plot for @xmath186 @xcite in terms of @xmath142 and @xmath126 for specific choices of @xmath4 as we can replace the @xmath144 dependence in terms of @xmath142 and @xmath4 through eq.([delta - phid ] ) . for @xmath187 , this is shown in fig.[rl1 ] , left panel and a similar plot is made for @xmath188 in the right panel . although we argue that it is more natural to consider @xmath126 to be less than one , in this plot we allow larger values of @xmath126 as a completeness . with this , for @xmath148 ( case a ) we see the appearance of two separate contours of @xmath186 with @xmath189 , one is for @xmath190 and the other corresponds to @xmath153 . similar plots are obtained for @xmath191 as well . however these isolated contours become a connected one once the value of @xmath4 increases , e.g. at @xmath192 , it is shown in fig.2 , left panel . a similar pattern follows in case of @xmath193 case . below we discuss the predictions of our model for case a ( with @xmath187 ) and case b ( @xmath194 ) separately . are shown in the @xmath152 plane . here in the left panel ( with @xmath148 , case a ) red ( dotted ) , black ( dashed ) and blue ( continuous ) lines represent @xmath195 and @xmath196 respectively . similar contours are present for @xmath159 values of the cp violating dirac phase . in the right panel ( with @xmath149 , case b ) red ( dotted ) , black ( dashed ) and blue ( continuous ) lines represent @xmath197 and @xmath198 respectively . , title="fig : " ] are shown in the @xmath152 plane . here in the left panel ( with @xmath148 , case a ) red ( dotted ) , black ( dashed ) and blue ( continuous ) lines represent @xmath195 and @xmath196 respectively . similar contours are present for @xmath159 values of the cp violating dirac phase . in the right panel ( with @xmath149 , case b ) red ( dotted ) , black ( dashed ) and blue ( continuous ) lines represent @xmath197 and @xmath198 respectively . , title="fig : " ] note that we need to satisfy both the @xmath199 as well as the value of @xmath177 obtained from the neutrino oscillation experiments . for this reason , if we consider the two contour plots ( one for @xmath200 and the other for @xmath201 ) together , then their intersection ( denoted by ( @xmath142 , @xmath126 ) ) should indicate a simultaneous satisfaction of these experimental data for a specific choice of @xmath4 . this is exercised in fig.[2040 ] . in the left panel of fig.[2040 ] , contour plots of @xmath177 and @xmath199 are drawn in terms of @xmath142 and @xmath126 for two choices of @xmath4= @xmath202 and @xmath203 . and @xmath204 in the @xmath152 plane for various choices of @xmath4 . in the left panel , dotted and continuous lines represent @xmath205 and @xmath206 respectively . in the right panel , dashed and continuous lines represent contour plots for @xmath207 and @xmath208 respectively.,title="fig : " ] and @xmath204 in the @xmath152 plane for various choices of @xmath4 . in the left panel , dotted and continuous lines represent @xmath205 and @xmath206 respectively . in the right panel , dashed and continuous lines represent contour plots for @xmath207 and @xmath208 respectively.,title="fig : " ] we find that there is no such solution for @xmath209 which satisfy both @xmath177 and @xmath158 with @xmath210 in these cases . however there exists solution for @xmath142 very close to one with a pretty large value of @xmath126 as mentioned in table 3 . this solution as we expect is not a natural one , not only for a large value of @xmath126 , but also for its very fine tuned situation . note that @xmath142 requires to be sufficiently close ( and hence finely tuned ) to one in this case . this situation can be understood from the fact that @xmath126 being quite large ( @xmath211 ) , value of @xmath142 has to be adjusted enough ( see the involvement of the expression @xmath212 in eq.([rex ] ) ) so as to compete with the @xmath126 dependent terms to get @xmath213 . similarly variation of @xmath158 is very sharp with respect to @xmath142 ( when close to 1 ) for large @xmath126 . for example , a small change in @xmath142 values ( @xmath214 ) would induce a change in @xmath158 by an amount of @xmath215 near the intersection region . however the situation changes dramatically as we proceed for higher values of @xmath4 as can be seen from fig.[2040 ] , right panel . this figure is for two choices of @xmath4= @xmath216 and 75@xmath217 . we observe that with the increase of @xmath4 , the upper contour for @xmath177 is extended toward downward direction and the lower one is pushed up , thereby providing a greater chance to have an intersection with the @xmath199 contour . we also note that the portion of @xmath158 contour for @xmath187 prefers a region with relatively small value of @xmath218 as well . however a typical solution with both @xmath142 and @xmath154 appears when @xmath4 is closer to @xmath208 . with this @xmath4 , we could see the lower and upper contours open up to form a connected one and we can have a solution for @xmath219 . in this case , there is one more intersection between the @xmath177 and @xmath158 contours with @xmath220 as given by ( 0.77 , 0.93 ) . when @xmath4 approaches @xmath221 and up ( till @xmath222 ) we have have solutions with @xmath223 . & @xmath126&@xmath224(ev ) + @xmath225 & 0.99 & 28.26&0.0714 + @xmath226 & 0.99 & 20.94&0.0709 + @xmath227 & 0.98 & 11.16&0.0701 + @xmath228 & . [ tab3 ] @xmath131 values at the intersection points of the @xmath177 and @xmath229 contour plots are provided corresponding to different @xmath4 values . the sum of the light neutrino masses are also indicated in each case . [ cols="^ " , ] for @xmath4 s beyond @xmath230 ( when considered within @xmath222 ) , the solutions exhibit @xmath231 implying a fine tuned situation similar to case a. note that @xmath142 therefore falls in a narrow range @xmath232 in order to satisfy both @xmath233 and @xmath204 considering all @xmath4 values . in fig.([ag1p1 ] ) , left panel , we find the intersection and @xmath234 together in the @xmath235 plane . in the right panel , the intersection region is elaborated where 3@xmath25 regions of @xmath229 are depicted.,title="fig : " ] and @xmath234 together in the @xmath235 plane . in the right panel , the intersection region is elaborated where 3@xmath25 regions of @xmath229 are depicted.,title="fig : " ] is at ( @xmath236 ) for @xmath237 . considering this @xmath4 as a reference for discussion , we further include the @xmath238 range of @xmath199 in fig.[ag1p1 ] , right panel . we find @xmath142 to be varied between 1.35 and 1.39 while @xmath199 changes from the lower to the higher value , within 3@xmath25 limit . within this range , we predict for @xmath239 when @xmath149 ] individual light neutrino masses and their sum . here also we find normal hierarchy for them in fig.[mg1 ] . for different @xmath4-values , the @xmath241 ( corresponding to the best fit value of @xmath199 ) are provided in table [ ta1 ] . for showing the prediction of our model in terms of other quantities like @xmath27 and @xmath242 , the fig.[meefg1],left and right panels are included . considering all the @xmath4 values for which @xmath243 , we find @xmath244 to be within @xmath245 . when @xmath149 .,title="fig : " ] when @xmath149 .,title="fig : " ] in a general type - ii seesaw framework , leptogenesis can be successfully implemented through the decay of rh neutrinos @xcite or from the decay of the triplet(s ) involved @xcite or in a mixed scenario where both rh neutrino and the triplet(s ) contribute @xcite . in the present set - up , all the couplings involved in the pure type - i contribution are real and hence the neutrino yukawa matrices and the rh neutrino mass matrices do not include any cp violating phase . therefore cp asymmetry originated from the sole contribution of rh neutrinos is absent in our framework . as we have mentioned earlier , the source of cp violation is only present in the triplet contribution and that is through the vev of the @xmath3 field . however as it is known@xcite , a single @xmath2 triplet does not produce cp - asymmetry . therefore there are two remaining possibilities to generate successful lepton asymmetry @xcitein the present context ; ( i ) from the decay of the triplet where the one loop diagram involves the virtual rh neutrinos and ( ii ) from the decay of the rh neutrinos where the one loop contribution involves the virtual triplet running in the loop . provided the mass of the triplet is light compared to all the rh neutrinos ( @xmath246 ) , we consider option ( i ) . once the triplet is heavier than the rh neutrinos , we explore option ( ii ) . n3 ( 110,60 ) first we consider option ( i ) , @xmath185 when @xmath247 . at tree level the scalar triplet can decay either into leptons or into two higgs doublets , followed from the lagrangian in eq.([ld ] ) and ( [ sp ] ) . for @xmath248 , the one loop diagram involves the virtual rh neutrinos running in the loop as shown in fig . interference of the tree level and the one loop results in the asymmetry parameter @xcite @xmath249 } { \sum_{ij}|(y_{{\mbox{\tiny$\delta$}}})_{ij}|^2 m^2_{{\mbox{\tiny$\delta$}}}+|\eta|^2 } { \rm log}(1+m^2_{{\mbox{\tiny$\delta$}}}/m^2_{rk}).\end{aligned}\ ] ] here @xmath250 denote the flavor indices , @xmath251 in the basis where rh neutrino mass matrix is diagonal . @xmath44 , @xmath43 and expression of @xmath252 can be obtained from eqs.([mdmr]),([ydelta ] ) and ( [ d1d2 ] ) . masses of rh neutrinos can be expressed as @xmath253 therefore , the asymmetry parameter in our model is estimated to be @xcite @xmath254 } .\end{aligned}\ ] ] here we denote @xmath255 , where @xmath256 is considered to be the common vev of all flavons except @xmath3-field s vev @xmath257 . the associated phase @xmath139 is the only source of cp - violation here . the total decay width of the triplet @xmath34 ( for @xmath258 two leptons and @xmath258 two scalar doublets ) is given by @xmath259.\end{aligned}\ ] ] note that there are few parameters in eq.([ep1 ] ) , @xmath260 which already contributed in determining the mass and mixing for light neutrinos . also @xmath144 is related with @xmath139 by eq.([phidal ] ) . in the previous section , we have found solutions for @xmath261 that satisfy the best fit values of @xmath199 and @xmath177 for a specific choice of @xmath4 ( the reference values @xmath262 for @xmath187 and @xmath263 for @xmath194 ) . then we can find the values of @xmath176 and @xmath264 corresponding to that specific @xmath4 value . these set of @xmath265 produce correct order of neutrino mass and mixing as we have already seen . here to discuss the cp - asymmetry parameter @xmath266 , we therefore choose @xmath267 for @xmath148 and @xmath268 for @xmath149 . in the @xmath269 plane for @xmath187 . , title="fig : " ] in the @xmath269 plane for @xmath187 . , title="fig : " ] we further define @xmath270 where @xmath271 serves as a relative measure of the vevs . with this , the expression of @xmath266 takes the form @xmath272 } , \end{aligned}\ ] ] which is @xmath273 independent . the expression for @xmath264 as obtained from eq . [ modd ] can be written as @xmath274 using eq.([ep2 ] ) , we obtain the contour plot for @xmath275 with @xmath276 , @xmath277 and @xmath278 which are shown in fig . [ c1 ] , left panel . the electroweak vev is also inserted in the expression . in obtaining the plots we varied @xmath51 above the masses of rh neutrinos ( see eq.([mr ] ) ) . the variation of @xmath279 is also restricted from above by the condition that we work in regime ( i ) where @xmath280 . fig.[c1 ] is produced for a specific choice of @xmath281 which corresponds to the solution ( @xmath282 ) . the values of @xmath283 and @xmath176 corresponding to this set of ( @xmath284 ) are found to be 0.0068 ev and 0.06 ev respectively . we have chosen @xmath285 for the left panel of fig.[c1 ] . in order to keep @xmath286 , we find @xmath287 gev which produce the required amount of cp - asymmetry which in tern can generate enough lepton asymmetry . in the @xmath269 plane for @xmath194 . , title="fig : " ] in the @xmath269 plane for @xmath194 . , title="fig : " ] note that value of @xmath273 can be concluded from the expression of @xmath264 in eq.([dsq ] ) , for a choice of @xmath288 which produces a @xmath266 contour . this is because corresponding to a specific choice of @xmath4 value , @xmath289 is uniquely determined for the solution point @xmath209 . hence with fixed values of @xmath290 ( with the same values to have the @xmath266 contour ) , @xmath273 can be evaluated from @xmath289 for a chosen @xmath288 . it turns out that @xmath273 has a unique value for a specific @xmath266 for both the panels of fig.[c1 ] . for example , with @xmath291 , we need @xmath292 , while to have @xmath293 , @xmath273 required to be 0.36 . these @xmath273 values are provided in first bracket in each figure beside the @xmath266 value . the reason is the following . for the specified range of @xmath51 ( @xmath294 ) , it follows that the first bracketed term in the denominator of eq.([ep2 ] ) is almost negligible compared to the second term ( with the choice of @xmath290 as mentioned before ) and hence effectively @xmath295 therefore for a typical choice of @xmath266 , @xmath288 is almost fixed and then @xmath264 expression in eq.([eap ] ) tells that @xmath273 also is almost fixed . in the right panel of fig.[c1 ] , we take @xmath296 and draw the contours for @xmath266 while @xmath297 are fixed at their previous values considered for generating plots in the left panel . in this case , @xmath279 turns out to be @xmath298 gev . in the @xmath269 plane for @xmath148 ( left panel ) and @xmath149 ( right panel ) with small @xmath299 . , title="fig : " ] in the @xmath269 plane for @xmath148 ( left panel ) and @xmath149 ( right panel ) with small @xmath299 . , title="fig : " ] similarly contours for @xmath266 are drawn in fig.[c2 ] for @xmath149 case . correspondingly we have used solutions of ( @xmath300 ) and the value of @xmath301 ev and @xmath302 ev are taken for @xmath303 ( also for @xmath304 ) . we obtain somewhat lighter @xmath279 as correspond to the case with @xmath148 . in fig.[smu12 ] similar contour plots for @xmath266 are exercised with @xmath305 at some lower values , fixed at @xmath306 along with @xmath307 for both @xmath148 and @xmath149 . vs @xmath199 for @xmath148 ( left panel ) and @xmath149 ( right panel ) . , title="fig : " ] vs @xmath199 for @xmath148 ( left panel ) and @xmath149 ( right panel ) . , title="fig : " ] so overall we have found that enough @xmath266 can be created so as to achieve the required lepton asymmetry through @xmath308 with @xmath309 is the total number density of the triplet and @xmath310 is the efficiency factor . after converting it into baryon asymmetry by the sphaleron process , @xmath311 is given by @xmath312 . @xmath310 depends on the satisfaction of the out - of - equilibrium condition ( @xmath313 ) . being @xmath314 triplet , it also contains the gauge interactions . hence the scattering like @xmath315 sm particles can be crucial @xcite . in @xcite , it has been argued that even if the triplet mass ( @xmath279 ) is much below @xmath316 gev , the triplet leptogenesis mechanism considered here is not affected much by the gauge mediated scatterings . however the exact estimate of @xmath310 requires to solve the boltzmann equations in detail which is beyond the scope of the present work . however analysis toward evaluating @xmath310 in this sort of framework ( where a single triplet is present and rh neutrinos are in the loop for generating @xmath266 ) exits in @xcite . following @xcite , we note that with the effective type - ii mass @xmath317 ev , the efficiency @xmath310 is of the order of @xmath318 . in estimating ) as @xmath319 with the consideration @xmath320 . here @xmath321 and @xmath322 are corresponding branching ratio s of decay of the triplet into two leptons and two scalar doublets . ] @xmath323 , we have considered all the parameters in a range ( mentioned within fig . [ c1]-[c2 ] ) so as to produce @xmath266 of order @xmath324 as shown in fig . [ c1]-[c2 ] . now , using the approximated expression as given by eq.([eap ] ) we can obtain variation of @xmath266 against @xmath199 as given in fig . [ ccp ] . in doing so we have substituted @xmath325 from eq.([dsq ] ) in eq.([eap ] ) . then as discussed in the previous section , using solutions for @xmath284 for 3@xmath25 range of @xmath199 for fixed @xmath4 we have obtained fig.[ccp ] for both @xmath148 and @xmath149 . here the left panel is for @xmath326 and right panel in for @xmath327 . n5 ( 110,60 ) we now discuss the option ii , when rh neutrinos are lighter than @xmath279 . the contribution toward the cp - asymmetry parameter generated from the decay of the lightest neutrino is given by @xmath328 } { \sum_i | ( \hat{y}_{{\mbox{\tiny$d$}}})_{1i}|^2 } , \\ & = & -\frac{m_{{\mbox{\tiny$r$}}_1}}{2}\frac{1}{8\pi v^2}|d|\sin\phi_d , \\ \epsilon_{{\mbox{\tiny$n$}}_2}&=&m_{{\mbox{\tiny$r$}}_2}\frac{1}{8\pi v^2}|d|\sin\phi_d \hspace{0.5 cm } { \rm and } \hspace{.5 cm } \epsilon_{{\mbox{\tiny$n$}}_3}= \pm \frac{m_{{\mbox{\tiny$r$}}_3}}{2}\frac{1}{8\pi v^2}|d|\sin\phi_d .\end{aligned}\ ] ] where we have used @xmath329 from eq.([d ] ) . in the above , @xmath330 and @xmath331 sign stands for @xmath149 and @xmath148 cases respectively in computation of @xmath332 . note that in the present scenario the rh neutrino masses are not entirely hierarchical , rather they are closely placed . therefore the total baryon asymmetry from the decay of the three rh neutrinos is to be estimated as @xmath333 , where @xmath334 is the respective efficiency factor . it turns out that with the same @xmath334 for @xmath335 , @xmath336 as a result ( using @xmath337 from eq.([mr ] ) ) of the specific flavor structure considered . therefore it is expected that the lepton asymmetry would be suppressed in this case . also in this case @xmath338 , which can be obtained by considering smaller value of the yukawa coupling @xmath339 ( as to generate the required @xmath264 , specific values of @xmath340 are already chosen ) . this could reduce the individual @xmath341 . we conclude this contribution ( @xmath342 ) as a subdominant to @xmath266 . we have considered a flavor symmetric framework for generating light neutrino masses and mixing through type - ii seesaw mechanism . in realizing it , we have introduced three sm singlet rh neutrinos , one @xmath2 triplet and few flavon fields . the rh neutrinos contribute to the type - i term , which guided by the @xmath1 symmetry of the model produces a tbm mixing pattern . then we have shown that the typical flavor structure resulted from the model can generate nonzero @xmath0 . in this framework , all the couplings are considered to be real . the cp symmetry is violated spontaneously by the complex vev of a single sm singlet field , while other flavons have real vevs . interestingly this particular field is involved only in the pure type - ii term . hence the triplet contribution not only generates the @xmath0 , it is also responsible for providing dirac cp violating phase @xmath4 . therefore the model has the potential to predict @xmath4 in terms of the parameters involved in neutrino masses and mixing . we have therefore studied the parameter space of the set - up considering that the triplet contribution is subdominant or at most comparable to the type - i term . the model indicates the values of @xmath4 to be in the range @xmath343 , @xmath344 , @xmath345 , @xmath346 for @xmath148 and @xmath347 , @xmath348 , @xmath349 , @xmath350 for @xmath149 . however @xmath351 ( and hence @xmath352 ) is disfavored in our scenario as in that case no cp violation would be present . also @xmath353 are excluded here . these ranges can be tested in future neutrino experiments . we provide an estimate for the @xmath242 . the sum of the neutrino masses are also evaluated . it turns out that the scenario works with normal hierarchical masses of light neutrinos . we have also studied leptogenesis in this model . as the type - i contribution to the light neutrino mass does not involve any cp violating phase , rh neutrinos decay can not contribute to the lepton asymmetry in the conventional way . we have found the triplet decay with the virtual rh neutrino in the loop can produce enough lepton asymmetry . 00 p. minkowski , phys . b * 67 * , 421 ( 1977 ) . m. gell - mann , p. ramond and r. slansky , conf . c * 790927 * , 315 ( 1979 ) [ arxiv:1306.4669 [ hep - th ] ] . r. n. mohapatra and g. senjanovic , phys . lett . * 44 * , 912 ( 1980 ) . t. yanagida , prog . phys . * 64 * , 1103 ( 1980 ) . p. f. harrison , d. h. perkins and w. g. scott , phys . b * 458 * , 79 ( 1999 ) [ hep - ph/9904297 ] . e. ma , phys . d * 70 * , 031901 ( 2004 ) [ hep - ph/0404199 ] . g. altarelli and f. feruglio , nucl . b * 720 * , 64 ( 2005 ) [ hep - ph/0504165 ] . g. altarelli and f. feruglio , nucl . b * 741 * , 215 ( 2006 ) [ hep - ph/0512103 ] . y. abe _ et al . _ [ double - chooz collaboration ] , phys . * 108 * , 131801 ( 2012 ) [ arxiv:1112.6353 [ hep - ex ] ] . f. p. an _ et al . _ [ daya - bay collaboration ] , phys . lett . * 108 * , 171803 ( 2012 ) [ arxiv:1203.1669 [ hep - ex ] ] ; f. p. an _ et al . _ [ daya bay collaboration ] , arxiv:1406.6468 [ hep - ex ] . j. k. ahn _ [ reno collaboration ] , phys . lett . * 108 * , 191802 ( 2012 ) [ arxiv:1204.0626 [ hep - ex ] ] . k. abe _ et al . _ [ t2k collaboration ] , phys . rev . lett . * 112 * , 061802 ( 2014 ) [ arxiv:1311.4750 [ hep - ex ] ] . b. karmakar and a. sil , phys . d * 91 * , 013004 ( 2015 ) [ arxiv:1407.5826 [ hep - ph ] ] . a. s. joshipura and e. a. paschos , hep - ph/9906498 . a. s. joshipura , e. a. paschos and w. rodejohann , nucl . phys . b * 611 * , 227 ( 2001 ) [ hep - ph/0104228 ] . b. bajc , g. senjanovic and f. vissani , phys . lett . * 90 * , 051802 ( 2003 ) [ hep - ph/0210207 ] . s. antusch and s. f. king , nucl . b * 705 * , 239 ( 2005 ) [ hep - ph/0402121 ] . s. antusch and s. f. king , phys . b * 597 * , 199 ( 2004 ) [ hep - ph/0405093 ] . n. sahu and s. u. sankar , phys . d * 71 * , 013006 ( 2005 ) [ hep - ph/0406065 ] . w. rodejohann , phys . d * 70 * , 073010 ( 2004 ) [ hep - ph/0403236 ] . h. gu , h. zhang and s. zhou , phys . rev . d * 74 * , 076002 ( 2006 ) [ hep - ph/0606302 ] . a. abada , p. hosteins , f. x. josse - michaux and s. lavignac , nucl . b * 809 * , 183 ( 2009 ) [ arxiv:0808.2058 [ hep - ph ] ] . d. borah and m. k. das , phys . rev . d * 90 * , no . 1 , 015006 ( 2014 ) [ arxiv:1303.1758 [ hep - ph ] ] . d. borah , int . j. mod . a * 29 * , 1450108 ( 2014 ) [ arxiv:1403.7636 [ hep - ph ] ] . m. borah , d. borah , m. k. das and s. patra , phys . d * 90 * , no . 9 , 095020 ( 2014 ) [ arxiv:1408.3191 [ hep - ph ] ] . r. kalita , d. borah and m. k. das , nucl . b * 894 * , 307 ( 2015 ) [ arxiv:1412.8333 [ hep - ph ] ] . s. pramanick and a. raychaudhuri , arxiv:1508.02330 [ hep - ph ] . m. magg and c. wetterich , phys . b * 94 * , 61 ( 1980 ) . g. lazarides , q. shafi and c. wetterich , nucl . b * 181 * , 287 ( 1981 ) . r. n. mohapatra and g. senjanovic , phys . d * 23 * , 165 ( 1981 ) . j. schechter and j. w. f. valle , phys . d * 22 * , 2227 ( 1980 ) . f. capozzi , g. l. fogli , e. lisi , a. marrone , d. montanino and a. palazzo , phys . rev . d * 89 * , no . 9 , 093018 ( 2014 ) [ arxiv:1312.2878 [ hep - ph ] ] . m. c. gonzalez - garcia , m. maltoni and t. schwetz , jhep * 1411 * , 052 ( 2014 ) [ arxiv:1409.5439 [ hep - ph ] ] . d. v. forero , m. tortola and j. w. f. valle , phys . d * 90 * , no . 9 , 093006 ( 2014 ) [ arxiv:1405.7540 [ hep - ph ] ] . t. d. lee , phys . d * 8 * , 1226 ( 1973 ) . a. e. nelson , phys . b * 136 * , 387 ( 1984 ) . s. m. barr , phys . * 53 * , 329 ( 1984 ) . j. a. harvey , p. ramond and d. b. reiss , phys . b * 92 * , 309 ( 1980 ) . j. a. harvey , d. b. reiss and p. ramond , nucl . b * 199 * , 223 ( 1982 ) . g. c. branco , phys . d * 22 * , 2901 ( 1980 ) . l. bento and g. c. branco , phys . b * 245 * , 599 ( 1990 ) . l. bento , g. c. branco and p. a. parada , phys . b * 267 * , 95 ( 1991 ) . g. c. branco , p. a. parada and m. n. rebelo , hep - ph/0307119 . g. c. branco , r. gonzalez felipe , f. r. joaquim and h. serodio , phys . d * 86 * , 076008 ( 2012 ) [ arxiv:1203.2646 [ hep - ph ] ] . s. fukuda _ et al . _ [ super - kamiokande collaboration ] , phys . b * 539 * , 179 ( 2002 ) [ hep - ex/0205075 ] ; y. ashie _ et al . _ [ super - kamiokande collaboration ] , phys . rev . d * 71 * , 112005 ( 2005 ) [ hep - ex/0501064 ] . p. adamson _ et al . _ [ minos collaboration ] , phys . * 106 * , 181801 ( 2011 ) [ arxiv:1103.0340 [ hep - ex ] ] . t. araki _ et al . _ [ kamland collaboration ] , phys . * 94 * , 081801 ( 2005 ) [ hep - ex/0406035 ] . p. a. r. ade _ et al . _ [ planck collaboration ] , arxiv:1303.5076 [ astro-ph.co ] . k. asakura _ et al . _ [ kamland - zen collaboration ] , aip conf . * 1666 * , 170003 ( 2015 ) [ arxiv:1409.0077 [ physics.ins-det ] ] . j. b. albert _ et al . _ [ exo-200 collaboration ] , nature * 510 * , 229 ( 2014 ) [ arxiv:1402.6956 [ nucl - ex ] ] . t. hambye , m. raidal and a. strumia , phys . b * 632 * , 667 ( 2006 ) [ hep - ph/0510008 ] . g. c. branco , r. g. felipe and f. r. joaquim , rev . phys . * 84 * , 515 ( 2012 ) [ arxiv:1111.5332 [ hep - ph ] ] . d. aristizabal sierra , f. bazzocchi , i. de medeiros varzielas , l. merlo and s. morisi , nucl . b * 827 * , 34 ( 2010 ) [ arxiv:0908.0907 [ hep - ph ] ] . d. aristizabal sierra and i. de medeiros varzielas , fortsch . phys . * 61 * , 645 ( 2013 ) [ arxiv:1205.6134 [ hep - ph ] ] . e. k. akhmedov and w. rodejohann , jhep * 0806 * , 106 ( 2008 ) [ arxiv:0803.2417 [ hep - ph ] ] . d. aristizabal sierra , f. bazzocchi and i. de medeiros varzielas , nucl . b * 858 * , 196 ( 2012 ) [ arxiv:1112.1843 [ hep - ph ] ] . d. aristizabal sierra , m. dhen and t. hambye , jcap * 1408 * , 003 ( 2014 ) [ arxiv:1401.4347 [ hep - ph ] ] . t. hambye , new j. phys . * 14 * , 125014 ( 2012 ) [ arxiv:1212.2888 [ hep - ph ] ] . g. lazarides and q. shafi , phys . d * 58 * , 071702 ( 1998 ) [ hep - ph/9803397 ] . j. n. fry , k. a. olive and m. s. turner , phys . d * 22 * , 2977 ( 1980 ) . e. ma , s. sarkar and u. sarkar , phys . b * 458 * , 73 ( 1999 ) [ hep - ph/9812276 ] . t. hambye , hep - ph/0412053 . t. hambye , y. lin , a. notari , m. papucci and a. strumia , nucl . b * 695 * , 169 ( 2004 ) [ hep - ph/0312203 ] .

we explore the past evolution of saturn s moons using direct numerical integrations . we find that the past tethys - dione 3:2 orbital resonance predicted in standard models likely did not occur , implying that the system is less evolved than previously thought . on the other hand , the orbital inclinations of tethys , dione and rhea suggest that the system did cross the dione - rhea 5:3 resonance , which is closely followed by a tethys - dione secular resonance . a clear implication is that either the moons are significantly younger than the planet , or that their tidal evolution must be extremely slow ( ) . as an extremely slow - evolving system is incompatible with intense tidal heating of enceladus , we conclude that the moons interior to titan are not primordial , and we present a plausible scenario for the system s recent formation . we propose that the mid - sized moons re - accreted from a disk about 100 myr ago , during which time titan acquired its significant orbital eccentricity . we speculate that this disk has formed through orbital instability and massive collisions involving the previous generation of saturn s mid - sized moons . we identify the solar evection resonance perturbing a pair of mid - sized moons as the most likely trigger of such an instability . this scenario implies that most craters on the moons interior to titan must have been formed by planetocentric impactors .

1603.07071

one of the most important results of the cassini mission to saturn was the discovery of a high heat flow from the south polar region of enceladus @xcite . this heat flow is estimated at 10 - 15 gw @xcite , which is by an order of magnitude in excess of theoretical predictions assuming a dynamically steady state @xcite . a dynamical equilibrium means that enceladus maintains a constant orbital eccentricity , which corresponds to an equilibrium between excitation by the 2:1 orbital resonance with dione and damping by tidal dissipation within enceladus . when calculating this dynamical equilibrium , @xcite assumed that the strength of tidal dissipation within saturn ( which pushes enceladus into the orbital resonance ) can be quantified by a tidal quality factor @xmath1 , as suggested by @xcite . tidal @xmath1 corresponds to the strongest tidal response by saturn that is still consistent with mimas not evolving past its current location over 4.5 gyr @xcite . any larger tidal evolution rates ( corresponding to smaller tidal qs ) would place mimas within saturn s rings less than 4.5 gyr ago . .masses and mean orbital parameters of saturn s major moons and their trojan companions ( semimajor axes of trojans are marked by superscript `` t '' ) . all quantities were retrieved from the jet propulsion laboratory s solar system dynamics website ssd.jpl.nasa.gov on 04/13/15 . semimajor axes are scaled using saturn s equatorial radius of 60,268 km , and inclinations are measured relative to the relevant laplace planes . masses of the trojan moons ( in italics ) are estimates based on an assumed density of 500 kg m@xmath2 . eccentricities and inclinations shown here are `` free '' , and do not include the forced components ( see * ? ? ? * ) . in the cases of enceladus and hyperion , typical instantaneous eccentricities ( @xmath3 for enceladus and @xmath4 for hyperion ) are well in the excess of the free component , so the eccentricity shown is starred.[elem ] [ cols="<,^,^,^,^ " , ] there are other possible approaches to constraining the tidal response of saturn . one is direct observation of the moons tidal acceleration . secular expansion of satellite orbits with orbital periods longer than the planet s spin period is predicted by tidal theories @xcite but was previously thought to be too slow for detection . recently , @xcite re - analyzed a large number of historical observations of saturn s moons and found tidal recession corresponding to @xmath5 . more recent preliminary results using cassini - based and ground - based observations of the satellites apparently support the idea of such fast tidal evolution of saturn s moons @xcite . @xcite also point out that @xmath6 matches what is required for enceladus to be in tidal equilibrium and still produce the observed heat flux . however , such fast tidal rates would require the inner moons like mimas to have formed less than about 0.5 gyr ago ( as they would have otherwise migrated past their present positions , assuming the @xmath7 of saturn did not change over time ) . theoretical models of saturn s interior dissipation are also available , but they are not conclusive to this debate , as there is a basic disagreement about whether the dissipation happens in the core or the envelope @xcite . we can also constrain saturn s tidal response by theoretical modeling of present and past orbital resonances between different moons . low q ( i.e. a high rate of dissipation ) of saturn on the order of @xmath8 would make the apparent highly - evolved state of the titan - hyperion resonance a natural product of tidal evolution over the age of the solar system @xcite , whereas in the classical picture the resonance had to be primordial , as it would not have not changed significantly over 4.5 gyr @xcite . here we address the two most important orbital resonances that have likely happened in the past : the 3:2 mean - motion resonance ( mmr ) between tethys and dione , and the 5:3 mmr between dione and rhea . these resonances should have happened close in time about 1.1 gyr ago if @xmath1 @xcite , or about 100 myr ago if @xmath8 . the 3:2 tethys - dione resonance has previously been studied using semi - analytical methods , which found substantial excitation of the two moons orbital eccentricities , but ignored orbital inclinations @xcite . this likely eccentricity excitation was also suggested to be related to geological features on tethys that indicate past tidal heating @xcite . unlike eccentricities , which damp over time by tides , the moons orbital inclinations can not be easily erased once they are excited @xcite . therefore , it is essential to probe the effects of possible past resonances on the moons inclinations , using methods that account for the full dynamics of the system . in this work we use two different numerical integrators , simpl and complex . simpl is relatively efficient , but can not model close approaches between massive bodies and is applicable only to the situations where the moons orbits are well - separated . complex is capable of integrating close encounters between moons , but has a downside of being much slower than simpl . with a goal of studying the long - term tidal evolution of saturn s moons , we wrote a new symplectic integrator , simpl ( symplectic integrator for moons and planets ) . this is a combination of two levison - duncan - type integrators @xcite , one for the planets and one for the moons of one planet . this type of integrator is a democratic heliocentric " ( i.e. uses heliocentric coordinates and does not require strict ordering of planets ) version of the mixed - variable symplectic integrator of @xcite , which assumes bodies follow keplerian orbits and implements perturbations as discrete kicks " executed in cartesian coordinates . we adapted the specific algorithm from @xcite , who designed it to model the dynamics of planets around one component of a binary star . simpl directly includes mutual moon - moon interactions , as well as solar and planetary perturbations ( direct planetary perturbations can be switched off to speed up computations ; back reaction of satellites on planets is ignored ) . we also include the planet s oblateness , tides raised on the planet by the satellite ( as a simple tangential acceleration on the moon in the plane of its orbit , not changing the planet s rotation ) and satellite tides ( implemented as radial - only kicks ) , and artificial planetary or satellite migration ( to approximate interaction with protoplanetary or protosatellite disks ) . when calculating tidal accelerations , we assumed that the tidal quality factor @xmath7 of saturn was independent of the satellite s orbital frequency ( i.e. all the moons experienced the same @xmath7 for saturn ) . we extensively validated simpl against published results and analytical estimates , and found that all precessional motions and secular tidal effects are reproduced correctly . to integrate the orbits of moons like tethys and dione with orbital periods of 2 - 3 days , we used a 0.1 day ( @xmath9 yr ) integration time - step . this is the timestep we used in most simulations shown in this paper , unless otherwise noted . we always included titan in our simulations , and we usually included only jupiter and saturn in the heliocentric part of the integration . complex ( crossing orbit moon and planet extrapolator ) is a more basic type of symplectic integrator , in which the hamiltonian is not separated into keplerian and perturbation parts ( as in simpl ) but into kinetic and potential energy . the exact implementation has been taken from @xcite , and uses a fourth order algorithm . complex uses many of the same subroutines as simpl and the same input and output files , except that simpl s keplerian step has been replaced by a sequence of symplectic mappings in cartesian coordinates . the subroutines in simpl that implement perturbations from other bodies , the planet s figure and tidal deformations have been incorporated into the symplectic mapping , in the sub - step where the momenta are being modified due to potential energy . complex has no preference for keplerian orbits , and can therefore resolve close approaches between the moons naturally . the downside of complex is its much slower speed , requiring about 600 timesteps for the shortest orbital period . for the sake of simplicity , we decided to use the same routines to integrate the satellites and the planets in complex , although the planetary orbits could in principle be done more efficiently . overall , complex is about two orders of magnitude slower than simpl , and we used it only when the satellite system went through an instability . short collision times between moons ( centuries to millennia ) keep the relative inefficiency of complex from becoming a major issue . the switching between simpl and complex was done manually , meaning that we go back to the state of a simpl simulation just before the instability and continue the simulation using complex . while the timestep of complex was sufficiently small to resolve close approaches , it was still too large to reliably resolve collisions . therefore we shrunk the timestep by an additional factor of 20 when two moons were between 10 combined radii from each other . once again , relatively large collisional cross sections of the moons compared to the sizes of their orbits made this relatively inefficient method workable . note that this approach would not work for simulations of planets interacting with small bodies , where special treatment for the two bodies undergoing the close encounter @xcite makes more sense than slowing down the whole integration . the period ratio of dione and tethys is currently about 1.45 , so , as tethys evolves outward much faster than dione , these two moons should have crossed their mutual 3:2 resonance at some point in the past . in this section we use the symplectic numerical integrator simpl to explore the tethys - dione 3:2 resonance crossing . the benefit of the numerical approach is that all sub - resonances , including the inclination - related ones , are fully accounted for . figure [ td4c ] shows a typical evolution of eccentricities and inclinations of tethys and dione during the resonance crossing , assuming that both moons orbits were close to circular and equatorial before the resonant encounter . the moons are originally captured in a sub - resonance which leads to a monotonic growth of the inclination of tethys . once tethys s inclination grows to about 4@xmath10 , the resonance becomes chaotic and eccentricities and inclinations of both moons vary rapidly . eventually the resonant lock breaks , leaving both orbits significantly eccentric and inclined . while the eccentricities can damp due to eccentricity tides , the inclinations of saturn s mid - sized moons are not expected to evolve appreciably over the age of the solar system @xcite . the final inclination of dione is always substantial , typically comparable to a degree ( fig . [ scatter ] ) , while the observed inclination of dione is only 0.028@xmath10 ( table [ elem ] ) . we used @xmath11 for saturn in almost all the simulations is this paper , which is equivalent to @xmath12 , as saturn s love number is estimated to be @xmath13=0.341 @xcite for saturn ; this love number would require @xmath7 of saturn in our simulations to also be 10 - 20% larger ] . for the satellites , we used @xmath14 and @xmath15 in the simulation shown in figure [ td4c ] , which implies at least some melting , as expected when eccentricities reach a few percent @xcite . a more detailed model would have tidal parameters varying with tidal heating , but our integrator did not include that capability . instead , we explored a range of constant tidal parameters for the satellites . given that enceladus seems to currently have very low @xmath16 ( if enceladus s orbit is in equilibrium as proposed by * ? ? ? * ) , we also integrated the resonance crossing using @xmath17 ( figure [ td8b3 ] ) . despite strong tidal dissipation , the inclination of dione is still excited during the crossing . one can in principle imagine an even stronger tidal response from tethys or dione ( with @xmath18 ) , but these would cause tethys to move inwards due to satellite tides being faster than the outward migration due to saturn tides ( for eccentricities of the order @xmath19 ) . such a reversal of migration would effectively prohibit tethys from crossing the resonance with dione and reaching the present configuration ( cf . . [ tides ] ) . we found it impractical to directly model resonance crossing in the classical " case when q=18,000 for saturn , as the integrations would need to be extended to many hundreds of myr . however , chaotic excitation of both the eccentricity and inclination is expected to be even stronger when the resonance is crossed more slowly . if we assume a larger tidal q for saturn , a wider range of tidal responses from tethys would overwhelm saturn s tides and lead to divergent migration , making the resonance an impassable barrier for tethys . the ratio between the migration rate due to tidal dissipation within the satellite and the migration rate due to dissipation within the planet is : @xmath20 where subscripts sat " and pl " refer to the satellite and the planet , @xmath21 , @xmath22 and @xmath23 are respectively mass , radius and orbital eccentricity , while @xmath24 is the measure of tidal dissipation . for @xmath19 ( typical of eccentricities excited by the tethys - dione resonance ) and the size and mass of tethys , we get @xmath25=0.03 for outward migration to stop . if @xmath26 , then @xmath27 is necessary for tethys to continue outward migration . this limit is comparable to the current tidal response of enceladus inferred by @xcite by assuming an equilibrium . however , if @xmath28 , then @xmath29 for the migration to stay convergent , which likely excludes any major melting within tethys , despite significant eccentricity ( if @xmath14 then @xmath13 would have to be less than @xmath30 , indicating a relatively rigid body ) . without going into modeling the actual physical response of the satellite , we can just use the above considerations to exclude the parameter space with @xmath31 from our consideration , as those parameters would forever trap tethys on the inner side of the 3:2 resonance with dione . the only exception is @xmath32 , in which case the resonance was never crossed , regardless of satellite tides . however , this extremely slow evolution rate is about 40 times too slow to account for the heating of enceladus , and we prefer the interpretation that the resonance was never crossed because the moons are young . it is possible that the inclination of tethys was already excited before it encountered the resonance with dione . analytical considerations open a possibility that a pre - inclined tethys could have conceivably avoided capture into the inclination - type resonance evident in fig . [ td4c ] and the associated inclination increase . we tested this possibility using initial inclinations of 1@xmath10 and 0.3@xmath10 for tethys in multiple simulations . [ td5 ] and [ td7a ] each plot the results of one simulation with initial @xmath33 and @xmath34 for tethys , respectively . for initial @xmath34 resonance capture still happens , although it usually lasts a much shorter time than in fig . [ td4c ] , while for initial @xmath33 the moons enter the chaotic behavior at the very beginning of the resonance crossing . in all cases the resulting final inclination of dione is much higher than the observed one ( fig . [ scatter ] ) . we also conducted simulations that start with one moon s eccentricity being @xmath35 . while simulations with initially eccentric dione did lead to the onset of chaos earlier in the simulation , those with initially eccentric tethys show no difference from the circular case , chiefly because most of this initial inclination damps by the time chaos is triggered ( despite @xmath36 being used in this run ) . in all cases the inclinations of both tethys and dione are excited well in excess of the observed values ( fig . [ scatter ] ) . the numerical experiments shown in this section suggest that the inclinations of both tethys and dione are excessively excited by the crossing of their 3:2 resonance , regardless of the assumed initial conditions and tidal parameters . in the next section we will discuss possible resonances that could have affected dione and tethys later on , but they can not make a past 3:2 resonance crossing compatible with the present orbits . the solid line in fig . [ scatter ] plots the combinations of inclinations that have the observed out - of - plane angular momentum deficit , and is approximately @xmath37 for small @xmath23 and @xmath38 . ] ( amd ) , and they are all well short of the inclination amd seen among the simulation outcomes . capture into additional mean - motion resonances could only increase the amd of tethys and dione , while the secular resonance involving these two moons ( see next section ) would conserve amd associated with inclinations . likewise , chaotic interaction within a mean motion resonance is extremely unlikely to decrease the combined inclinations , but tends to redistribute inclinations on short timescales and increase amd over time . this applies to subsequent resonances with enceladus and mimas ; mimas does have a large inclination but small amd compared with tethys due to its small mass , and could not have absorbed out - of - plane amd from tethys and dione . the fact that both tethys and dione are always excessively inclined after their 3:2 mmr crossing is a strong constraint on their relative locations at formation , and we will discuss the implications in the following sections . in the relative chronology of resonance crossings in the saturnian system , the dione - rhea 5:3 mean - motion resonance would happen shortly after the 3:2 tethys - dione resonance . we have concluded in the previous section that the passage through 3:2 tethys - dione resonance did not happen , and here we explore the possibility that tethys formed outside the 3:2 resonance with dione and dione formed interior to the 5:3 resonance with rhea , and that dione and rhea have since crossed their 5:3 resonance . figure [ dr3 ] shows one simulation of the 5:3 dione - rhea mmr crossing . in the beginning dione is captured in the inclination - type resonance , which eventually breaks and a chaotic phase follows , during which the eccentricities and inclinations of both bodies are excited . this part of the evolution is expected , and while it could explain the excitation of rhea s inclination ( @xmath39 ) , it is seemingly at odds with the very low inclination of dione ( @xmath40 ) . while the eccentricities are also excited , they can be expected to damp later through satellite tides . however , at 24 myr we see a striking exchange of eccentricities and inclinations between tethys and dione . both the eccentricity and inclination of tethys get excited this way , while those of dione are reduced . we find that this unexpected dynamical event is caused by a previously unidentified tethys - dione secular resonance . the secular resonance seen in these simulations has the resonant argument @xmath41 , where @xmath42 and @xmath43 are the longitudes of the pericenter and the node , and the subscripts 1 and 2 refer to tethys and dione , respectively . the pattern of signs in the resonant argument indicates that in this resonance the eccentricity and inclination of each moon should be correlated , and anti - correlated with those of the other moon , consistent with numerical integrations showing tethys getting more inclined and eccentric , with dione s orbit becoming more circular and planar . the resonance appears to break when either the eccentricity or inclination of dione reaches zero , with the latter case being able to explaining the very low present inclination of dione . note that in fig . [ dr3 ] it is eccentricity , rather than inclination , that reaches zero first and breaks the secular resonance . the type of secular resonance seen here affecting tethys and dione has not , to our knowledge , been described in the literature . therefore , here we briefly discuss why it arises soon after the dione - rhea 5:3 mean motion resonance . the resonant argument @xmath44 is a combination of the arguments @xmath45 for each moon , which are typically very slowly varying for a regular satellite , as @xmath46 for low-@xmath23 , low-@xmath38 orbits chiefly perturbed by planetary oblateness . usually , the leading term in @xmath47 is @xmath48 @xcite , where @xmath49 is the standard oblateness moment and @xmath22 the radius of saturn , and @xmath50 and @xmath51 are the moon s semimajor axis and mean motion . this result does not depend on the definitions of orbital elements used ( geometric vs. osculating ) , which affect individual expressions for @xmath52 and @xmath53 but cancel out in their sum @xcite . this term currently leads to periods of about 5000 yr and 20,000 yr for @xmath45 of tethys and dione , making @xmath54 . clearly , another source of precession is necessary to bring these two moons into a secular resonance . precession due to oblateness is also affected by eccentricity and inclination . from @xcite , we have : @xmath55 where @xmath56 is the argument of pericenter , and @xmath23 and @xmath38 are the moon s orbital eccentricity and inclination ( to the planet s equator ) . for @xmath33 , we find that the quantity in parenthesis dependent on inclination is only @xmath57 , which for tethys would produce retrograde precession of @xmath58 with a period of 11,000 yr . this is a potentially significant term , but it would add a negative term to @xmath59 before the resonance is encountered , as it is dione , rather than tethys , that must be inclined before the resonance capture occurs . therefore this term can not be the cause of the resonance being encountered in the first place . however , inclination - dependent terms may be important for maintaining the resonance , as they will slow down the excess apsidal precession of tethys as it acquires inclination . since the secular resonance is encountered near dione s 5:3 resonance with rhea , it appears that near - resonant terms are important for establishing the secular commensurability . with this being the second - order resonance , six different resonant terms can affect the precession of dione . after some numerical testing , we have established that this additional precession is sensitively dependent on dione s distance from the outermost sub - resonance ( associated with the @xmath60 term ) and is also strongly dependent on dione s eccentricity . these terms appear to accelerate the precession of dione s longitude of pericenter , but do not affect the node as much , producing a significant change in the secular resonant argument @xmath44 . figure [ secres8 ] shows a simulation where dione and rhea are placed just outside their 5:3 resonance ( i.e. after an assumed 5:3 resonance passage ) , with tethys having a near - circular and planar orbit and dione having @xmath61 and @xmath62 ( rhea , not plotted , had initial @xmath63 and @xmath34 ) . about 1.4 myr into the simulation the secular resonance is established , and it lasts until about 8.5 myr , when the average inclination of dione drops below @xmath64 . while this is a somewhat artificial set - up ( we assume the outcome of the 5:3 dione - rhea resonance , without simulating it ) , it clearly shows the potential of the secular resonance to produce the current orbits of tethys and dione , with the right initial conditions . we find that resonance capture is almost certain ( assuming @xmath11 ) if the eccentricity of dione is larger than @xmath35 . the current low inclination of dione indicates that , if it was reduced by this secular resonance , the resonance had to break when the inclination of dione reached zero ( as opposed to the situation shown in fig . [ dr3 ] , where eccentricity reaches zero first ) . as @xmath65 in this secular resonance , and the angular momentum is conserved , this means that the initial @xmath66 . this gives us a requirement for post-5:3 resonance orbits of dione and rhea : @xmath67 , @xmath68 unconstrained , @xmath69 and @xmath70 ( current value ) . we assume low eccentricity and inclination for tethys at this time , as higher eccentricities of tethys often lead to greater chaos and sometimes overlap between the tethys - dione 3:2 and dione - rhea 5:3 resonances , which is long - lasting , hard to break and excites inclinations beyond the observed values . while the required initial conditions for the secular resonance fall within the range of stochastic outcomes of the 5:3 dione - rhea resonance , we could not find a combination of initial conditions and tidal parameters that lead to such a state with high probability . we find that the initial capture into the @xmath71 subresonance of the 5:3 mmr ( involving only the inclination of dione ) is certain for very low initial inclinations ( @xmath72 ) . this sub - resonance , however , tends to break only at @xmath73 , which is well above the value required for the secular resonance . an initial inclination @xmath74 leads to a jump through the the @xmath71 sub - resonance and capture into the the @xmath75 subresonance , in which both the inclinations of dione and rhea grow . this resonance usually breaks in the @xmath76 range , approaching the target inclination needed for the secular resonance , while the inclination of rhea reaches @xmath77 , approaching the current values . figure [ dr17a3 ] shows one simulation using initial @xmath78 that results in a very low inclination of dione and a substantial inclination of tethys after the secular resonance . figure [ scatter2 ] shows the outcome of several sets of simulations starting with @xmath78 , with all other eccentricities and inclinations ( except those of titan ) being close to zero . it is clear from the figure that the amount of excitation needed to explain the present system roughly matches the outcomes of the dione - rhea mmr . while no single simulation exactly reproduces the current orbits , very low inclinations of dione and significant inclinations of tethys are often produced through the secular resonance . while the mechanism involving the 5:3 dione - rhea resonance followed by the tethys - dione secular resonance is a very promising avenue for producing the observed inclinations , we were not able to achieve this outcome with any reasonable probability using simpl . one possible explanation is that the process is so stochastic that we can not expect to exactly reproduce the real history of the system . we think it is still too early to reach such a conclusion , and hope to explore this question in future work . the major limitations of our present approach include : 1 ) we assume that the tidal properties of the satellites are constant over time , despite varying eccentricities and tidal heating ; 2 ) we assume saturn s tidal @xmath7 is frequency - independent , which has implications for the exact timing of recent resonance crossings ; 3 ) while our integrator allows it , we did not explore different initial tidal parameters for the three moons ( i.e. we assumed they were all equally rigid or dissipative , which is likely to be wrong ) . a clear step forward would be incorporation of a geophysical model into the integrator that would adjust tidal parameters depending on the tidal heating and other processes the moon is subject to . the chief conclusion of this section is that the past 5:3 mean - motion resonance between dione and rhea likely did happen , unlike the 3:2 tethys - dione resonance discussed in the previous section . this gives us a very well - defined relative age of tethys , dione and rhea , which is related to their absolute age through the controversial tidal @xmath7 of saturn . as for the absolute ages , the classical @xmath1 @xcite and the fast " @xmath8 @xcite would place the formation of tethys , dione and rhea at about 1.1 gyr and 100 myr ago , respectively , with both dates being much younger than the age of the solar system . in the following sections , we will briefly explore the possible dynamical mechanisms that could have led to the late formation of the present set of mid - sized icy moons of saturn . as we are proposing that the moons of saturn ( out to rhea ) formed much later than the planet , we need to address the plausibility and basic physics of such an event . first of all , the flatness of saturn s regular satellite system indicates formation from a thin disk . since we think that tethys , dione and rhea all formed close to their present orbits , this disk would need to extend out to about 10 saturn radii and comprise the combined mass of the mid - sized moons . since the titan - hyperion resonance appears very old @xcite , we assume that titan ( and other more distant moons ) predated this protosatelite disk that tethys , dione and rhea formed from . this does not necessarily mean titan must be primordial , as it may have originated during another , more ancient satellite instability event @xcite , which is outside the focus of this paper . as saturn has youthful and dynamic rings surrounded by numerous small moons @xcite , we propose that the rings also formed in the same event as tethys , dione and rhea . this means that the protosatellite disk discussed here extended all the way inward to the roche limit ( possibly due to viscous spreading ) . unlike a primordial circumplanetary disk , this disk would be comprised exclusively of solids , save for short - lived vapor created in collisions . in this section we discuss the likely lifetime and fate of this disk , while in sections 7 and 8 we will address the origin of the disk . @xcite used numerical simulations to study the fate of ejecta in the saturnian system in order to constrain possible hazards for the cassini spacecraft . they found that the moons accrete any material initially found on crossing orbits in mere centuries ( also see * ? ? ? * ) , and theoretical predictions of their growth rate from debris are also on the order of millennia ( cf . * ; * ? ? ? we can use the expression in @xcite ( also see * ? ? ? * ) to calculate the coagulation time : @xmath79 where @xmath80 and @xmath81 are the volume and surface densities of a solid core and the disk , respectively , and @xmath82 is the gravitational focusing factor ( we will assume @xmath83 ) . if we estimate @xmath84 , @xmath85 and @xmath86 ( corresponding to a 4-day orbital period ) , we get @xmath87 therefore , it would take less than 1000 years to grow 1000 km satellites , in agreement with the numerical integrations of debris removal by @xcite . however there are serious problems with this approach . the first is that the present satellites have well - separated hill spheres and would therefore not sweep up all of the material between them . assuming a uniform disk , we can estimate the number ( @xmath51 ) and size ( @xmath88 ) of the satellites assuming that they are separated by 5 hill radii : @xmath89 so therefore @xmath90 where we have assumed @xmath91 . this makes the mass of a typical protosatellite @xmath92 kg , comparable to that of enceladus . these protosatellites would take significantly longer to merge into the larger moons we observe today . the mechanism for their destabilization and mergings would likely be chaotic interaction between the moons , many of which would likely be in overlapping mean - motion resonances with several others . additionally , there would be some fragmentation during their accretion and mergings , and rings of debris would likely occupy quasi - stable areas between neighboring moons cleared zones . interaction with this debris would also affect the protosatellites orbits , leading to orbit crossings and mergers . direct numerical modeling is clearly needed to explore the evolution of such a system . another caveat with accretion is outside perturbations , mainly from titan . as @xcite show , the orbits close to titan s 3:1 and 2:1 resonances are significantly more perturbed than those in between mid - sized moons . this excitation would inhibit accretion , at least in some locations , potentially leading to a relatively long - lived debris ring outside the orbit of rhea . eventually this disk would spread and disperse due to inter - particle scatterings and collisions . @xcite find that the timescale for the lifetime of planetary rings , based on numerical experiments , can be expressed as : @xmath93 where @xmath94 and @xmath95 are a particle s physical and hill radius , respectively . assuming the disk radius of @xmath96 m ( ten saturn radii ) and a density of 500 kg m@xmath2 for disk particles , @xmath97 and @xmath98=5 days , @xmath99 ^ 2}\ ] ] if we assume a surface density @xmath100 , the spreading time according to the above formula is about 2000 yr . after the surface density decreases by a factor of a few , the disk can last tens of kyr . at that point it would still contain a mass on the order of @xmath101 kg , making it possible to force protosatellites ( or satellite fragments ) of comparable mass to migrate as the disk is spreading . while the material on the inner edge would likely be accreted by rhea , the large gap between the inner system and titan would leave space for any protosatellites ( or fragments of disrupted satellites ) to migrate and potentially interact with titan through mean - motion resonances . the outer part of the disk , outside the 3:1 resonance with titan ( and particularly around the 2:1 resonance at 12.7 saturn radii ) , would experience strong forced eccentricity due to titan , and may endure as a debris disk for a much longer time @xcite . a satellitesimal ( or a fragment ) that is ejected outwards may be able to migrate by interacting with the spreading disk , and become trapped in a mean motion resonance with titan . the migration of such an object due to disk spreading could be much faster than that due to tidal migration and could excite titan s eccentricity to the observed value in a relatively short amount of time . in the simulation shown in figure [ mig6 ] , we introduced an artificial fast migration of a @xmath102 kg moon ( 0.3% of titan s mass ) starting around 13 saturn radii , just outside titan s 2:1 resonance . our hypothesis is that there would be a spreading ring of excited , non - accreting debris around the location of the resonance . the initial eccentricity of the moon was set to @xmath4 , as we would expect this moon s ( or fragment s ) orbit to be excited . the moon in question migrates out until it is caught in the 5:3 resonance with titan . in the first 1000 years after resonance capture , it is the eccentricity of the inner moon that grows , indicating a lindblad resonance ( involving the smaller body s eccentricity ) however , very soon the system settles into a co - rotation "- type resonance ( involving the larger body s eccentricity ) and stays in it for the rest of the simulation . this behavior is unexpected and is likely related to the relatively rapid rate of evolution , which stabilizes formally overlapping sub - resonances . in 50,000 years , the eccentricity of titan grows to about 0.05 , well above its observed value . eventually , the resonance would break as the eccentricities of both bodies become high ( the inner moon is indirectly affected by titan s eccentricity ) , or the migration slows down and long - term chaos takes over . figure [ mig5 ] shows a similar simulation involving a twice larger inner moon ( @xmath103 kg ) , and a similar capture into the 8:5 resonance with titan . after the resonance breaks , the inner moon collides with either titan or the disk . re - absorption of the protosatellite into the disk would have no consequences for titan or hyperion . however , close encounters followed by a collision with titan would perturb the orbits of both titan and hyperion . if the titan - hyperion resonance is several gyr old @xcite , its continued existence might be in conflict with any destructive events in recent history . we therefore used complex to run a set of 100 simulations of scatterings between titan , hyperion and a @xmath104 kg satellite . rogue " satellite was initially placed on an eccentric orbit with @xmath105 and @xmath106 , with each run having a different longitude of pericenter or a different initial longitude for the inner satellite . titan and hyperion were initially placed deep in the resonance , with titan on a circular orbit and hyperion having @xmath4 . in 85 out of one hundred 1000-year simulations using complex , titan and the inner moon collided , and in 76 of these titan and hyperion maintained a stable 4:3 resonance . the stability of the resonance was determined by integrating post - merger titan and hyperion for 10,000 years using simpl . the average orbits of titan and hyperion in the post - merger simpl simulations are shown in fig . [ shotgun ] . the eccentricity of titan is excited to about 0.01 - 0.02 on average ( about half of its present value ) , requiring a different excitation mechanism ( such as the mean - motion resonance described in this section ) to explain its present orbit . the main ( resonant ) component of hyperion s eccentricity stays close to 0.1 , with the other ( slow " ) component being proportional to titan s eccentricity . the resonant argument is excited , and the present libration semi - amplitude of @xmath107 is firmly within the ensemble of outcomes . we conclude that a resonant interaction with a moon as small as @xmath102 kg , migrating due to disk spreading , can excite titan s eccentricity to the current value . even if this moon subsequently impacts titan , the titan - hyperion mean - motion resonance is likely to survive this event . we numerically confirm resonance survival for a rogue moon " mass of @xmath104 kg , and it is certain that collisions with smaller moons would have an even less disruptive effect on the titan - hyperion resonance . an eccentricity excitation only about 100 myr ago would explain the surprisingly large free eccentricity of titan despite its likely dissipative surface and interior @xcite . in the previous sections we find that saturn s present - day moons interior to titan s orbit likely accreted much after the planet s formation , presumably from a disk , which covered the locations of the moons from mimas to rhea . as we discussed in section 5 , this disk would last only millennia , possibly somewhat longer in locations where titan s perturbations inhibit accretion . therefore we think that the disk was a temporary feature , and we propose that it originated in the destruction of a previous satellite system . while we argue that saturn s present mid - sized satellite system is relatively young , its expected lifetime is measured in gyr , for saturn , the next major resonance ( tethys - dione 4:3 mmr ) will not happen for at least 2 gyr . ] even in the case of fast tides " proposed by @xcite . therefore , we argue that the most long - term arrangement of the mass currently in the mid - sized moons of saturn is a system of satellites . as this previous system of satellites would have had the extent and total mass of the present mid - sized icy moon system , we will assume it also had most of its mass in the outermost two satellites , like dione and rhea today ( this architecture is also most stable against tidal evolution ) . is there a way of destabilizing a mid - sized satellite system ? we expect that any perturbations that can lead to large orbital eccentricities and moon - moon collisions would have to be resonant in nature . in this and the following sections , we will discuss several candidate resonances for destabilization of the putative previous - generation satellite system of saturn . assuming @xmath108 for saturn , a rhea - sized moon would evolve to about the present location of rhea over 4.5 gyr , while a somewhat smaller moon ( like tethys or dione ) would converge on positions very close to , but interior to , rhea s present orbit . this arrangement , with a rhea - sized moon close to rhea s present orbit , and a proto - dione " on a converging interior orbit , is the starting point for our studies of past dynamical instability . the simplest case would be a mean - motion resonance like that of tethys and dione described in section 3 . however , the relatively small masses of the inner moons make it impossible to excite their eccentricities to the high values required for orbit crossing ( at least @xmath109 ) . tidal evolution while the moons are captured in a mean - motion resonance can in principle lead to monotonic eccentricity growth , but tidal dissipation within the moons can easily limit eccentricity growth for realistic physical parameters . a chaotic resonance between a large and a much smaller moon could excite the smaller moon s eccentricity enough to cause a collision , such a mismatched collision is unlikely to lead to disruption of a dione or rhea - sized moon @xcite . resonances between a mid - sized satellite and titan have the potential of being more powerful , but are also unlikely to have been the cause of a past instability . we find that the 4:1 resonance with ( non - eccentric ) titan at about 8 saturn radii leads to temporary capture , after which the inner moon acquires an eccentricity of @xmath110 and @xmath34 . another major resonance close to the mid - sized moons is the 3:1 with titan , at 10 saturn radii . this resonance is strongly chaotic , making the inner moon significantly eccentric and plausibly leading to instability . however , this resonance overlaps with the 4:1 resonance with hyperion ( which is in the 3:4 resonance with titan ) . as hyperion is small and does not damp eccentricity , a rhea - sized moon in the 3:1 resonance with titan quickly destabilizes hyperion . due to the survival of hyperion we can exclude the 3:1 resonance with titan as the source of a past instability . another set of resonances in the region inhabited by the mid - sized satellites is the so - called evection - type resonances . these are commensurabilities between a satellite s apsidal precession and the sun s ( or another distant perturber s ) apparent planetocentric orbital period . given saturn s oblateness and heliocentric orbit , the strongest of these resonances happen for moons at about 8.3 saturn radii , which have an apsidal precession period equal to saturn s 29.5 year orbital period . additional harmonics with evection happen at 6.9 saturn radii ( precession is 1/2 of saturn s year ) and 7.3 saturn radii ( precession equals 2/3 of saturn s year ) . figure [ eve1 ] shows a moon encountering the main evection resonance assuming its tidal response stays limited ( @xmath111 for the satellite ) . the eccentricity of the moon is strongly excited , and so is the inclination . however , since this resonance affects only the eccentricity but not the mean motion of the satellite , and the satellite tides are too weak to reverse the evolution ( as we assumed @xmath15 ) , the satellite eventually passes the resonance without being destabilized ( as long as it has no very close neighbors ) . after crossing the resonance , both eccentricity and inclination are high . while eccentricity can damp , inclination will not , implying that any moon that passed through the evection resonance should have an inclination of several degrees . this is well in excess of the current inclination of rhea ( the only moon beyond the evection resonance ) , implying that an intact rhea with these tidal properties could not have formed close to the rings and migrated out to its present location , as suggested by @xcite . if we assume large - scale melting within the moon whose orbit is excited by the evection resonance , its orbital behavior changes . as shown in fig [ eve2 ] ( where we assumed a much larger @xmath112 for the moon ) , the evection resonance excites a large eccentricity , which then damps and in the process makes the moon migrate inward . after the eccentricity is damped , the outward migration continues and the cycle repeats . each next bounce " from the resonance will not be identical , as the inclination also evolves . we see no reason why this process can not continue for 1 gyr or more , trapping the satellite close to the original location of the evection resonance . the difference between the scenarios shown in figs . [ eve1 ] and [ eve2 ] depends only on the internal properties of the satellite . regardless of the outcome ( resonance crossing or long - term stalling of outward tidal evolution ) , the evection resonance has a potential of exciting satellite orbits to eccentricities of @xmath4 or more in very short amounts of time . before considering the effect of evection on two interacting satellites , we will briefly address some higher - order harmonics of evection . the evection resonance in the saturnian system is notably chaotic and tends to affect both eccentricity and inclination . this is in contrast to the evection resonance in the early earth - moon system , where it affects only eccentricity , and can lead to stable capture @xcite or a stable near - resonant equilibrium @xcite . we find that this complex behavior is caused by the evection resonance s overlap with its nearby harmonics , which affect inclination . the principal inclination harmonic has the resonant argument @xmath113 , where @xmath114 is the moon s longitude of ascending node , while @xmath115 is the node of the sun s apparent orbit measured relative to the moons laplace plane ( in practice , this angle is equivalent to the equinox of saturn ) . this harmonic can be seen as a combination of the annual equation and the principal secular perturbation in inclination @xcite . this argument s strength varies with saturn s eccentricity and therefore it can not lead to stable long - term capture , but it can induce significant variation of inclination , as seen in figs . [ eve1 ] and [ eve2 ] . in fig . [ eve2 ] , the inclination is stable after about 100 myr , indicating that the moon may be in a state where it is affected only by the principal evection resonance . the strongest among the 2:1 evection harmonics at 6.9 saturn radii involves not only the perturbed satellite and the sun , but also titan . this is due to its resonant argument @xmath116 , where @xmath42 designates longitude of pericenter and @xmath117 the mean longitude . this argument combines the basic evection term @xmath118 with the moon - titan secular term @xmath119 ( for a summary of different perturbation terms see * ? ? ? we find that mid - sized icy moons can get captured in this resonance , and that their further evolution through the resonance increases the eccentricity of titan . this resonance may seem like an attractive way of generating titan s eccentricity @xcite , but the resonance always breaks due to satellite tides when the smaller moon s eccentricity reaches a few percent . the increase in titan s eccentricity is always a factor of a few too small to explain the observed value , and this resonance never seems to lead to instability . the strongest 3:2 evection harmonic at 7.3 saturn radii has the resonant argument @xmath120 . this argument can be viewed as a combination of the evection and the annual equation @xmath121 @xcite . as its argument includes @xmath122 , the 3:2 evection harmonic term s strength is proportional to saturn s heliocentric eccentricity , which varies by an order of magnitude ( 0.01 - 0.09 ) during a 50,000 year secular cycle @xcite . this means that the resonance is weak during a low point in saturn s eccentricity , and our simulations show that a capture into this resonance breaks relatively fast , with the moon moving past the resonance with an eccentricity of a few percent and an inclination of a degree or two . we find that this is an unlikely candidate for instability . in order to produce a disk , we need a collision between two large moons , probably somewhat similar to dione and rhea . as the evection resonance is the most promising source of excitation , we envisage a scenario in which the outer moon is in or near the evection resonance distance , while the inner moon is somewhat closer to saturn . this configuration can arise either by moons migrating together in a mean - motion resonance , or by the inner moon catching up to the outer one . in general , similar - sized moons would tend to converge on close orbits due to tides , and the real dione and rhea ( mass ratio 1:2.4 ) are evolving toward a configuration more compact than their 4:3 mutual resonance . another possibility is that the outer moon is caught in a cycle of evection resonance bounces " ( fig . [ eve2 ] ) , while the inner moon is catching up . in fig . [ fduo11 ] we show an integration ( using simpl ) of convergent migration of a proto - dione " ( with a combined mass of dione and tethys ) and proto - rhea " ( with the mass of rhea ) into their mutual 4:3 resonance . once the outer moon enters the evection resonance , the satellites become dynamically coupled to saturn s heliocentric orbit and their eccentricities are ultimately excited to large values ( e=0.1 - 0.2 ) , leading to orbit - crossing and a collision . in this simulation , the 4:3 mean - motion resonance capture happens first , and only then do the moons enter the evection resonance . while the moons are in the 4:3 mean - motion resonance , they are also in the secular resonance @xmath123 ( i.e. apsidal anti - alignment ; fig . [ arg0 ] ) . due to the stabilizing effects of the secular resonance the moons are capable of simultaneously being in stable @xmath124 and @xmath125 sub - resonances of the 4:3 resonance , without the chaos usually accompanying resonance overlap ( i.e. the arguments @xmath126 and @xmath127 are both librating ) . this configuration is apparently stable until the outer satellite reaches the evection resonance . the next stage of their evolution , starting at about 24 myr , is chaotic and during this time the moons are not in either 4:3 or ( apsidal ) secular resonance . chaos is driven by both the evection resonance proper and its associated harmonics . while the main evection resonance @xmath128 and its inclination - related harmonic @xmath129 affect the outer satellite , the 2:1 evection harmonic @xmath130 affects the inner satellite ( @xmath43 is the longitude of the node , and the subscript t refers to titan ) . [ arg1 ] shows the 10 kyr interval 26 myr into the simulation , during which both the main evection argument and its 2:1 harmonic reverse the direction of circulation , and this behavior is correlated with the eccentricities of the moons ( which continually exchange angular momentum ) . the presence of the 2:1 evection harmonic at the location of the inner moon is a simple consequence that two moons around an oblate body have an apsidal precession period ratio of about two when they are close to the 4:3 resonance . since @xmath131 , @xmath132^{2/3}\bigr)^{7/2 } = 1.96 ; \label{num}\ ] ] therefore , if the outer moon s pericenter precesses at a rate close to saturn s mean motion , the inner moon would naturally precess at about twice that rate ( barring other complications ) , exposing itself to perturbations arising from 2:1 near - resonance with the heliocentric motion . the importance of the 2:1 harmonic depends on the eccentricity of titan , which was @xmath133 in this simulation ( i.e. one order of magnitude lower than observed ) . if titan efficiently damped any eccentricity it might have had before the last re - arrangement of the system , this resonance may have been much less prominent . in general , we find that the main evection resonance is the most important one for the eventual fate of the system . the two moons , while having large and rapidly varying eccentricities and inclinations , remain stable for several myr . eventually the pair enter a state in which the 2:1 harmonic is less important and the evection resonance dominates . just before instability , the outer moon gets a major eccentricity boost from the evection resonance ( fig . [ arg2 ] ) . soon after that close encounters begin and the two orbits are irreversibly coupled . at about 26.6 myr , as the orbits become crossing , we pass the simulation from simpl to complex . we find that the collision typically occurs in about 1000 years . this collision , at about 3 km / s , is just energetic enough to be erosive and can lead to the formation of a disk @xcite . we also achieved instability in a simulation where the moons are initially not in 4:3 resonance , but on orbits converging due to tidal evolution ( this time the masses are @xmath134 ) . in this simulation we placed the outer moon just interior to the evection resonance , so dramatic effects happen soon . however , given that a moon could be trapped at this distance by the evection resonance for a long time ( fig . [ eve2 ] ) , it is fully plausible to have the inner moon converge on the outer moon while the latter is trapped by the evection resonance . while the collision resulting from the simulation shown in fig . [ fduo11 ] is marginally energetic enough to disrupt the moons @xcite , in reality the moons inclinations may be higher by the time destabilization occurs . a previous residence in either the evection or 4:3 resonance , as well as crossing of any of the other evection harmonics or any other mean motion resonances ( with one another or with other satellites ) would lead to inclinations of several degrees before the beginning of the last resonance , increasing the likely collision velocity . we used direct numerical simulations to explore a past hypothetical 3:2 mean - motion resonance crossing between saturn s moons tethys and dione , as well as a 5:3 resonance crossing between dione and rhea . as the current tethys / dione and dione / rhea orbital period ratios are just above 2/3 and 3/5 respectively , and tidal recession from saturn is expected to be fastest for tethys and slowest for rhea , these resonances have generally been thought to have been crossed at some point in the past . we find that the 3:2 resonance crossing always leads to excessive excitation of the orbital inclinations of both tethys and dione , which is inconsistent with their current orbits . using the same approach , we conclude that the 5:3 resonance between dione and rhea , which closely follows the tethys - dione 3:2 resonance in the timeline of tidal evolution , most likely did happen . we find that the 5:3 dione - rhea resonance , immediately followed by a previously unknown tethys - dione secular resonance , is the most likely source of the current inclinations of tethys and rhea . we can therefore state that tethys and dione evolved tidally by only a modest amount over their lifetimes , which is only about a quarter of the tidal evolution envisaged in @xcite . there are two possible interpretations : either tidal evolution of saturn s moons has been very slow , or saturn s mid - sized moons are significantly younger than the solar system . while both interpretations are consistent with the lack of the past tethys - dione resonance , we favor the idea that the moons are young , possibly as young as 100 myr . young moons would be consistent with the astrometric analysis of @xcite , equilibrium tidal heating of enceladus @xcite and the evolved state of the titan - hyperion resonance @xcite . therefore , we propose that tethys , dione and rhea all formed in one large event about 100 myr ago . the trojan moons of tethys and dione that share their inclinations must have formed even more recently , after their passage through the secular resonance . mimas , enceladus and the rings could have formed at the same epoch as tethys , dione and rhea , or could be younger yet . tidal evolution of mimas and enceladus , and their current resonances with tethys and dione are a complex subject ( * ? ? ? * and references therein ) , and will be addressed in future work . our results require most craters on rhea and the moons interior to its orbit to be produced by planetocentric impactors . many authors have considered the cratering record of saturn s mid - sized moons to result from both heliocentric and planetocentric impactors , with the latter being especially important on mimas and enceladus @xcite . differences in crater size - frequency distributions ( sfds ) between moons do not contradict our scenario , as each moon would be primarily cratered by debris on nearby orbits . @xcite have shown that differences in surface gravity and impact velocity can produce a wide range of crater sfds for the moons orbiting a planet . impact speeds for planetocentric impactors onto the moons will be roughly an order of magnitude smaller than for comets from heliocentric orbit , so larger planetocentric impactors are needed to make a given crater . however , the required impactor sizes remain plausible , as they are still smaller than the crater sizes , even for basins in the saturn system . for instance , using keith holsapple s tool at http://keith.aa.washington.edu/craterdata/scaling/index.htm , we estimate that a planetocentric impactor with a diameter of 40 km impacting mimas at 3 km / s could procuce the herschel basin , which has a diameter of 130 km . planetocentric impactors are also fully compatible with more than one cratering episode : the relative youth of trojan companions and their apparent origin as debris from the larger moons imply at least one intense cratering episode following the resonance crossings described in this paper . the absence of a past 3:2 resonance between tethys and dione also rules out the simplest version of the scenario proposed by @xcite and @xcite , in which the moons formed sequentially out of past massive rings , and then migrated outward . that hypothesis implies that tethys had to catch up " with dione , invariably crossing the 3:2 resonance . our results do not exclude the possibility that the previous generation of moons did form this way , but were destroyed and re - accreted close to their present positions . other ultimate origins of the mid - sized satellite system , such as those proposed by @xcite and @xcite , or a late origin of titan @xcite , can certainly be reconciled with a recent episode of re - accretion . disruptions , followed by formation of a debris disk and re - accretion , erase much of the diagnostic information about the moons formation , and put few constraints on the ultimate origin of the system . using order - of - magnitude arguments , we conclude that the disk that gave rise to the present mid - sized moons can only be short - lived and must be a product of disruption of a previous generation of mid - sized moons . we explored different configurations of saturn s past satellite system in order to find a likely source of such an instability . the most likely trigger for a global instability is the evection ( semi - secular ) resonance between the largest past moon ( assumed to be rhea - sized ) and the sun , located just interior to the present orbit of rhea . if most of the current icy moons mass was concentrated in two moons on converging orbits ( somewhat similar to dione and rhea ) , with the outer one evolving into the evection resonance , we find that a large scale instability and mutual collisions are a common outcome . we also propose disk - driven migration of small moons ( at the edge of the resulting disk ) into orbital resonances with titan as a plausible source of titan s eccentricity . there are three possible avenues toward testing the hypothesis of the young saturnian mid - sized satellite system . while preliminary results appear to support fast migration @xcite , further work on the satellite astrometry using cassini data should be able to decisively confirm or falsify the findings of @xcite . second , better understanding of the geophysics of enceladus may be able to clarify if its energy source is indeed tidal and if the dynamical equilibrium hypothesis is appropriate . third , this hypothesis requires that the majority of craters on saturn s moons interior to titan ( many of which have heavily cratered surfaces ) must be the product of planetocentric impacts , as there would not have been enough time for them to accumulate sufficient numbers of cometary impacts @xcite . we hope that further work on crater and impactor population modeling will be able to put stronger constraints on impactor sources . allan , r. r. 1969 , , 74 , 497 alvarellos , j. l. , zahnle , k. j. , dobrovolskis , a. r. , & hamill , p. 2005 , , 178 , 104 asphaug , e. , agnor , c. b. , & williams , q. 2006 , , 439 , 155 asphaug , e. , & reufer , a. 2013 , , 223 , 544 bierhaus , e. b. , dones , l. , alvarellos , j. l. , & zahnle , k. 2012 , , 218 , 602 bierhaus , e. b. , dones , l. , & robbins , s. 2015 , aas / division for planetary sciences meeting abstracts , 47 , # 508.01 brouwer , d. , & clemence , g. m. 1961 , new york : academic press , 1961 burns , j. a. , & gladman , b. j. 1998 , , 46 , 1401 canup , r. m. 2010 , , 468 , 943 chambers , j. e. , quintana , e. v. , duncan , m. j. , & lissauer , j. j. 2002 , , 123 , 2884 charnoz , s. , crida , a. , castillo - rogez , j. c. , et al . 2011 , , 216 , 535 chen , e. m. a. , & nimmo , f. 2008 , , 35 , l19203 chen , e. m. a. , nimmo , f. , & glatzmaier , g. a. 2014 , , 229 , 11 chiang , e. , & laughlin , g. 2013 , , 431 , 3444 crida , a. , & charnoz , s. 2012 , science , 338 , 1196 uk , m. , & stewart , s. t. 2012 , science , 338 , 1047 uk , m. , dones , l. , & nesvorn , d. 2013 , arxiv:1311.6780 cuzzi , j. n. , burns , j. a. , charnoz , s. , et al . 2010 , science , 327 , 1470 danby , j. m. a. 1992 . fundamentals of celestial mechanics . richmond : willman - bell , 1992 , 2nd ed . dermott , s. f. 1971 , , 153 , 83 dermott , s. f. , & murray , c. d. 1981 , , 48 , 1 astronomy meeting 46 , # 304.01 . dones , l. , chapman , c. r. , mckinnon , w. b. , et al . 2009 , saturn from cassini - huygens , 613 forest , e. , & ruth , r. d. 1990 , physica d nonlinear phenomena , 43 , 105 gavrilov , s. v. , & zharkov , v. n. 1977 , , 32 , 443 goldreich , p. , lithwick , y. , & sari , r. 2004 , , 42 , 549 greenberg , r. 1973a , , 78 , 338 greenberg , r. 1973b , , 165 , 305 greenberg , r. 1981 , , 86 , 912 hamilton , d. p. 2013 , aas / division for planetary sciences meeting abstracts , 45 , # 302.01 howett , c. j. a. , spencer , j. r. , pearl , j. , & segura , m. 2011 , journal of geophysical research ( planets ) , 116 , 3003 howett , c. , spencer , j. , & verbiscer , a. 2014 , aas / division for planetary sciences meeting abstracts , 46 , # 405.02 kirchoff , m. r. , & schenk , p. 2010 , , 206 , 485 lainey , v. , karatekin , . , desmars , j. , et al . 2012 , , 752 , 14 lainey , v. , jacobson , r. a. , tajeddine , r. , et al . 2015 , arxiv:1510.05870 leinhardt , z. m. , & stewart , s. t. 2012 , , 745 , 79 levison , h. f. , & duncan , m. j. 1994 , , 108 , 18 meyer , j. , & wisdom , j. 2007 , , 188 , 535 meyer , j. , & wisdom , j. 2008 , , 193 , 213 murray , c. d. , & dermott , s. f. 1999 , solar system dynamics , cambridge u. press ogilvie , g. i. 2014 , , 52 , 171 peale , s. j. 1999 , , 37 , 533 porco , c. c. , helfenstein , p. , thomas , p. c. , et al . 2006 , science , 311 , 1393 remus , f. , mathis , s. , zahn , j .- p . , & lainey , v. 2015 , , 573 , a23 sagan , c. , & dermott , s. f. 1982 , , 300 , 731 sinclair , a. t. 1983 , iau colloq . 74 : dynamical trapping and evolution in the solar system , 106 , 19 sohl , f. , sears , w. d. , & lorenz , r. d. 1995 , , 115 , 278 storch , n. i. , & lai , d. 2015 , , 450 , 3952 touma , j. , & wisdom , j. 1998 , , 115 , 1653 wisdom , j. , & holman , m. 1991 , , 102 , 1528 wisdom , j. , & tian , z. 2015 , , 256 , 138 zhang , k. , & nimmo , f. 2012 , , 218 , 348

a two - dimensional nonlinear schrdinger lattice with nonlinear coupling , modelling a square array of weakly coupled linear optical waveguides embedded in a nonlinear kerr material , is studied . we find that despite a vanishing energy difference ( peierls - nabarro barrier ) of fundamental stationary modes the mobility of localized excitations is very poor . this is attributed to a large separation in parameter space of the bifurcation points of the involved stationary modes . at these points the stability of the fundamental modes is changed and an asymmetric intermediate solution appears that connects the points . the control of the power flow across the array when excited with plane waves is also addressed and shown to exhibit great flexibility that may lead to applications for power - coupling devices . in certain parameter regimes , the direction of a stable propagating plane - wave current is shown to be continuously tunable by amplitude variation ( with fixed phase gradient ) . more exotic effects of the nonlinear coupling terms like compact discrete breathers and vortices , and stationary complex modes with non - trivial phase relations are also briefly discussed . regimes of dynamical linear stability are found for all these types of solutions . , nonlinear coupling , peierls - nabarro potential , mobility , inversion of stability , modulational instability , power currents . 42.65.wi , 42.82.et , 63.20.pw , 05.60.-k

0805.3814

the use , in optics , of materials with nonlinear responses to external fields has over the past years provided a wealth of applications and prospects for all - optical communication , where light manipulates light itself @xcite . especially , nonlinear photonic structures endowed with a spatially periodic refractive index provide new means to control the flow of light in the form of localized wave packets , not possible in bulk media @xcite . the experimental observation of discrete localization in one and two dimensions , in arrays of coupled waveguides experiencing a cubic ( kerr ) nonlinearity @xcite or a saturable nonlinearity through the photovoltaic effect in a photorefractive medium @xcite as well as in optically induced nonlinear photonic lattices @xcite , shows significant progress in this direction . the theoretical foundation and predictions of these phenomena can , to good approximation , be made by models of the discrete nonlinear schrdinger ( dnls ) type @xcite derived within coupled - mode theory @xcite or the tight - binding approximation using wannier function expansion @xcite . the latter technique is also applicable for models of bose - einstein condensates trapped in deep optical lattices . in these contexts the issue of mobility of localized packets of energy is highly interesting as it provides insights as how to control optical signals , e.g. , for applications in multiport switching . as is well known @xcite , travelling localized solutions , or solitons , quite generally exist in continuum systems as a result of the competition between nonlinearity and dispersion . when the continuous symmetry is broken , as is the case when considering a nonlinear discrete system , the generic solution is instead a non - moving time - periodic one , called a discrete breather ( db ) or intrinsic localized mode ( see , e.g. , @xcite for a review ) . the presence of a lattice leads to pinning of localized solutions , as this underlying structure effectively introduces a periodic potential for travelling excitations due to the difference in energy between stationary solutions localized on a lattice site and solutions localized between lattice sites @xcite . this energy difference is referred to as a peierls - nabarro ( pn ) energy barrier and is the reason that narrow travelling excitations get trapped in a lattice . mobility of a highly localized stationary mode can be induced if an appropriate perturbation is added to overcome the potential barrier , but usually some energy is lost to radiation and the result is trapping @xcite . however , excitations that are broad , in comparison with the lattice spacing , can still propagate as the effects of the pn - barrier in this case is small and the dynamics is well approximated by the continuum limit of the discrete equation . this is provided when the continuum equation has stable localized travelling solutions , which , e.g. , is the case for the 1d cubic nonlinear schrdinger equation but not for the 2d ( cubic ) equation @xcite . notable exceptions from the pinning behaviour are integrable lattice equations , i.e. , equations possessing an infinite number of conserved quantities , like the ablowitz - ladik ( al ) model which is a discretization of the nonlinear schrdinger equation retaining this property @xcite . these models have analytic exponentially localized travelling solutions and completely lack pn - barrier , which accounts for the excellent mobility . unfortunately , in contrast to their many interesting mathematical properties , they seem to have very limited direct applications for real systems . thus other routes to mobility in discrete systems are sought , although these models are conveniently used as the starting point of perturbative approaches . in @xcite it is numerically shown how a mobile excitation develops a resonant tail when the governing equation deviates from the 1d integrable al - model . moreover , the amplitude of the tail grows when the equation is further away from the integrable limit , or alternatively , as the pn - barrier increases from zero . it is inferred that the effect of pinning is overcome by a periodic exchange of energy between the localized core and the oscillating tail as the excitation moves through the lattice and the periodic pn - potential . this result can also be extended to a 2d setting @xcite . however , there is a limit as to how far the mobile solutions can be continued away from the integrable limit , with broad solutions being more persistent as can be expected @xcite . hence , we can expect that mobility of localized modes is possible when the pn - barrier is small , and the ability to tailor the size of the pn - barrier may allow for a control of the mobility of narrow solutions in a lattice . recent research has focused on ways to minimize the pn - barrier , which in optics applications means moving beyond models based on the cubic dnls equation where the pn - barrier is an increasing function of the power of the stationary modes @xcite . similar to the problem of moving kinks in klein - gordon lattices @xcite the introduction of competing nonlinearities has proven fruitful . in dnls equations with saturable nonlinearity the vanishing of the pn - barrier at certain points and a subsequent enhanced mobility has been demonstrated in both 1d @xcite and 2d @xcite , and the same apply for a 1d dnls model with nonlinear interactions @xcite . as these models are derived from real physical settings they are interesting from an application point of view . note also that there is a large class of dnls models @xcite with an absence of pn - barrier ( not only zero at isolated points ) , different from the more physical models , as the former are special discretizations of the nonlinear schrdinger model that in some sense preserve a translational invariant @xcite . amongst them is the al - lattice and other integrable equations . a problem related to the vanishing of the pn - barrier at specific points is that of stability inversion between stationary solutions centred on a lattice point and between lattice points , respectively , which can be understood from the energy minimum principle . the solution with lowest energy at given power will be stable , and the change of sign of the energy difference is also related to an exchange of stability of the lowest energy modes . this can be observed for the above models @xcite , as well as for the dnls model with competing cubic - quintic nonlinearity @xcite and more general lattice models @xcite . in connection with the stability inversion there also exists an intermediate asymmetric solution interpolating between the stationary solutions centred on and in - between sites and bifurcating with these at the points where they change their stability properties @xcite . for the 1d saturable dnls model this solution can even be found analytically near the first zero of the pn - barrier @xcite . from this viewpoint , the mobility of a highly localized mode may be thought of as a transformation between stationary on- and off - site solutions , via the intermediate solution . hence a more accurate definition of the pn - barrier would need to include also the intermediate solution as this may be the mode with the highest energy . therefore , we will distinguish between the _ energy difference _ between two fundamental modes and the _ pn - barrier _ relating to the energy of all involved stationary modes . also , in 2d , one would in principle need to consider modes corresponding to different directions of possible propagation @xcite . though localized modes and their mobility are interesting for beam steering in waveguide arrays , transport of energy in the transversal direction ( from waveguide to waveguide ) can be achieved also by a collective excitation of the waveguides . with a plane wave propagating in the array , the same power will be transported along all waveguides , but there will also be a transfer of power between the waveguides depending on the wave number ( phase gradient ) of the plane wave . in @xcite it is demonstrated that the direction and magnitude of the transversal flow of power in a 1d array with nonlinear coupling are quite versatile under the variation of the amplitude and wave number of the plane waves . here the discussion is extended to 2d and will show an even greater flexibility . the outline of the paper is as follows : in sec . [ sec : model ] we present a model for optical waveguides coupled in two spatial dimensions and embedded in a nonlinear kerr material , discuss its relevance and introduce some of the important properties of the model . we should stress that our model does _ not _ attempt to describe configurations where the waveguides themselves are nonlinear , such as those presently experimentally realized in , e.g. , @xcite , but rather correspond to the original proposal of jensen @xcite with the nonlinearity residing solely in the _ embedding _ material . the stability of the fundamental localized modes are investigated in sec . [ sec : stationary ] and their relation to mobility of localized excitations is discussed in sec . [ sec : mobility ] . in sec . [ sec : compact ] some special solutions , like compact and complex modes , arising from a balance of the coupling terms in the model are presented . finally , in sec . [ sec : planewave ] the transport properties of plane waves are analyzed before we conclude in sec . [ sec : conclusion ] . we also include in the end of sec . [ sec : conclusion ] a brief discussion about some recent related works , which appeared in press after the original submission of this manuscript . the model we consider is a direct extension to two spatial dimensions of a one - dimensional model for the propagation of the electric field amplitudes in coupled optical waveguides embedded in a nonlinear material @xcite . the waveguides are identical and regularly spaced in the two spatial dimensions , thus forming a square lattice . coupling between the sites of the lattice is mediated by an evanescent field overlap of the modes ( single - mode propagation ) in the respective waveguides and it is assumed that the modes decay sufficiently fast outside the waveguides to motivate only nearest - neighbour coupling . further , the waveguides themselves are assumed to be constructed from a _ linear _ material and _ surrounded by a nonlinear kerr material_. this will tend to strengthen the effects of nonlinear coupling with respect to on - site nonlinear effects when compared to a system where also the waveguides are nonlinear . this motivates an extension of the generally employed cubic on - site dnls model to an equation that in two dimensions will read @xmath0 \\ & + 2q_{5}\big[2|\psi_{n , m}|^{2}\delta\psi_{n , m } + \psi_{n , m}^{2}\delta\psi_{n , m}^ { * } \\ & \phantom{+2q_{5}[0 } + \delta(\psi_{n , m}|\psi_{n , m}|^{2 } ) \big ] , \end{split}\ ] ] where @xmath1 is the complex amplitude of the electric field , @xmath2 measures the propagation along the waveguide array and the coupling parameters @xmath3@xmath4 are given by overlap integrals of the modes . the operator @xmath5 is defined by @xmath6 . ( note that the operator @xmath7 is the standard 2d discrete laplacian . ) the model ( [ eq : xdnls ] ) is also relevant in other contexts , e.g. , as a nonlinear tight - binding approximation , with @xmath8 argued to be negligible , for the time - evolution of bose - einstein condensates ( bec ) in a deep periodic optical potential @xcite , or for @xmath9 as a rotating - wave approximation to a fermi - pasta - ulam chain @xcite . in both these one - dimensional models the extension to two spatial dimensions is straightforward . one may also in a straightforward way consider extensions to other lattice geometries , such as e.g. triangular / hexagonal patterns ( which in some sense could be considered as the most natural packing of waveguides in two dimensions ) . in this work , we chose to consider only the square lattice which is the simplest 2d structure to analyze from a modelling point of view , and also for comparison with previously published work in the area which mainly dealt with square - lattice configurations . we expect that most of the qualitative results obtained here will persist also for different lattice geometries , although a more detailed discussion on these issues is beyond the scope of the present work . for general waveguide arrays @xmath10 and @xmath11 , @xmath12 , have the same sign as the kerr index of the nonlinear material . estimates of the relative strengths of the parameters show that the nonlinear coupling terms , @xmath8 and @xmath4 , can be up to the same order of magnitude as the on - site nonlinearity @xmath13 for an array of waveguides constructed from algaas with sizes in the tenth of micrometer regime and operated with a laser in the infrared ( @xmath14 ) . details of the derivation of the 1d model can be found in @xcite and a calculation of the coupling parameters for a slab waveguide array can be found in appendix a of @xcite ( see in particular fig . a2 in @xcite for an explicit illustration of realistic values for relative coupling strengths ) . when extending the model to two dimensions there is also the possibility of coupling in the diagonal direction between next - nearest neighbours . however , for experimentally relevant sizes , as given above , of an array of square waveguides the linear diagonal coupling can be estimated to be an order of magnitude smaller than the linear direct coupling ( see appendix a2 in @xcite ) . also , when operated in the power regime where the most interesting phenomena occur , the nonlinear coupling and on - site terms will be of the same order as the linear coupling , i.e. @xmath15 , @xmath12 . this can be shown to correspond to powers of a few kw @xcite , well within what is experimentally available . however , although these estimates of the parameter values are of relevance when considering experimental verifications and applications of the model they will not be taken as a restriction in the present paper , since we are also interested in the general effects of nonlinear coupling . further , the parameter @xmath3 can be made to vanish by the transformation @xmath16 and has no significant physical role . note also that the staggering transformation @xmath17 is equivalent to changing the sign of the parameters @xmath18 and @xmath4 , thus reducing the part of parameter space that needs to be investigated to get a complete picture of the equation . further , by rescaling the amplitudes and the time - like variable , two additional parameters can be fixed . here we will fix @xmath19 and @xmath20 , while restricting @xmath21 without loss of generality . from noether s theorem we know that any infinitesimal transformation leaving the action integral for an equation invariant leads to a quantity that is conserved under the evolution governed by that equation @xcite . essentially this means that any continuous symmetry of eq . ( [ eq : xdnls ] ) corresponds to a conserved quantity . in particular , the invariance under translations in @xmath2 will lead to conservation of the hamiltonian @xmath22 , with @xmath23 \\ & + 2q_5\big[\psi_{n , m}\psi_{n+1,m}(\psi_{n , m}^{*\,2}+\psi_{n+1,m}^{*\,2 } ) \\ & \phantom{+ 2q_50 } + \psi_{n , m}\psi_{n , m+1}(\psi_{n , m}^{*\,2}+\psi_{n , m+1}^{*\,2})\big ] \big\ } + c.c . \end{split}\ ] ] the evolution equation ( [ eq : xdnls ] ) , and its complex conjugate , can be obtained from the hamilton equations of motion using the canonical variables @xmath1 and @xmath24 , [ eq : heq ] @xmath25 further , the invariance of eq . ( [ eq : xdnls ] ) under global phase rotations , i.e. , @xmath26 , @xmath27 , corresponds to the conservation of the norm @xmath28 although the hamiltonian @xmath29 for all practical purposes constitutes an energy functional it has no direct connection to any physical energy for the array of waveguides . but from a mathematical viewpoint it can be treated as the energy of the system . in particular , for the cubic dnls model it holds that a bounded extremum ( a maximum in the present formulation since @xmath30 and hence @xmath29 should be regarded as negative energy ) of @xmath29 for given norm @xmath31 is lyapunov stable and may , with right , be called the ground state of the system @xcite . this ground state is conjectured to be unique up to a global phase factor , which is consistent with the absence of stability inversion and the non - vanishing pn - barrier in the cubic on - site dnls model . the norm @xmath31 , however , has a direct physical interpretation as the total power carried along the waveguides , provided that the modes are appropriately normalized . in the context of bec , the conservation of norm corresponds to boson number conservation . the conservation of hamiltonian and norm may also conveniently be expressed in terms of discrete continuity equations . for this purpose it is instructive to make a change to action - angle variables @xmath32 , where @xmath33 and @xmath34 are real canonical variables . the corresponding hamilton equations of motions are [ eq : heq_aa ] @xmath35 now , with @xmath36 from eq . ( [ eq : h ] ) and by introducing the phase difference between neighbouring sites in the two lattice directions as @xmath37 and @xmath38 eq . ( [ eq : heq_n ] ) for the norm density can be written as @xmath39 the norm current density in the @xmath40 direction is given by @xmath41 . \end{split}\ ] ] a similar expression holds for @xmath42 in the @xmath43 direction where the roles of @xmath44 and @xmath45 are interchanged . since the norm density @xmath33 is the power on site @xmath46 , the continuity equation expresses the obvious fact that a change of the power is related to a power transfer to neighbouring sites . hence the physical interpretation of @xmath47 is the flow of power from site @xmath46 to site @xmath48 . an equation on the form ( [ eq : neq ] ) can also be derived for the hamiltonian density @xmath36 , with a hamiltonian flux density @xmath49 and @xmath50 . for the present purposes it is sufficient to know that for a stationary monochromatic solution , i.e. , @xmath51 and @xmath52 , the relation to the norm current density is @xmath53 . our discussions on the flow of energy in the system ( sec . [ sec : planewave ] ) will thus be made using only the norm current density . initially we will turn our attention towards localized solutions or , more specifically , to time - periodic localized modes also known as discrete breathers ( dbs ) or intrinsic localized modes . the existence of dbs has been rigorously proven for hamiltonian lattice systems of arbitrary dimension subject to conditions of anharmonicity and non - resonance with linear phonons @xcite , and extended to more general systems in @xcite . in the context of coupled waveguides the variable @xmath2 will play the role of time . due to the global phase invariance of the dnls - type equations the time - periodicity leads to a significant simplification of the problem of finding localized solutions . for monochromatic solutions a simple gauge transformation to a rotating frame of reference will render these solutions stationary , i.e. , taking @xmath54 reduces eq . ( [ eq : xdnls ] ) to @xmath55 \\ & + 2q_{5}\big[2|\psi_{n , m}|^{2}\delta\psi_{n , m } + \psi_{n , m}^{2}\delta\psi_{n , m}^ { * } \\ & + \delta(\psi_{n , m}|\psi_{n , m}|^{2 } ) \big ] = 0 . \end{split}\ ] ] the idea of the existence proofs , and also the foundation for efficient numerical techniques to calculate dbs @xcite , is continuation from the anti - continuous limit , @xmath56 , @xmath57 . solutions are trivial in this uncoupled limit as the amplitude of the oscillators at each site can be chosen independently from the set @xmath58 , where we have made a restriction to real single - frequency solutions . from the configuration chosen in the anti - continuous limit it is possible to make a classification of the most fundamental localized solutions of eq . ( [ eq : xdnlsstat ] ) . the set of solutions obtained from continuation to non - zero coupling of the configuration with one site excited and all other at rest will be called @xmath59-site dbs . similarly , if two neighbouring sites are excited at the anti - continuous limit we will get symmetric or anti - symmetric @xmath60-site dbs , depending on the relative phases of the sites . since we ultimately are interested in the mobility of localized modes , we would like to compare the hamiltonian of different stationary solutions . as has been described in sec . [ sec : introduction ] ( especially for the corresponding 1d model @xcite ) , the enhanced mobility of localized modes is connected to the vanishing of a pn - barrier measuring the difference in hamiltonian between fundamental stationary modes . the idea is that for a small or vanishing barrier , mobility of a highly localized mode may be thought of as a continuous transformation between , e.g. , stationary @xmath59-site and @xmath60-site solutions , put into motion by adding energy to overcome the barrier ( or pinning ) . therefore , numerical continuation to non - zero coupling will be performed under the constraint of constant norm . as the parameters @xmath18 and @xmath13 are fixed from rescalings , the parameters @xmath8 and @xmath4 are varied , leaving the frequency @xmath61 as a free parameter to be determined by the stationary solution . the choice to keep @xmath31 fixed is motivated by the fact that it is a dynamical parameter that is conserved under the evolution of eq . ( [ eq : xdnls ] ) . hence , when considering a range of coupling parameters ( as will shortly be done ) , this will facilitate a meaningful comparison of the hamiltonians over that entire range . this would not be the case had the continuation been performed for constant frequency . however , for a fixed frequency and varying norm the relevant energy quantity would instead be the grand canonical free - energy @xmath62 , with @xmath61 acting as a chemical potential , as used in @xcite . the linear stability of stationary modes is most simply investigated by studying the evolution of a small perturbation in the rotating frame of reference . hence , for a given solution we make the ansatz @xmath63 { \mathrm{e}}^{-{\mathrm{i}}(\lambda+q_{1})z}$ ] in eq . ( [ eq : xdnls ] ) and arrive at a linear matrix eigenvalue problem for the growth rates @xmath64 of the real infinitesimal perturbations @xmath65 and @xmath66 @xcite . since the problem is hamiltonian this matrix is infinitesimally symplectic and will have the simultaneous eigenvalues @xmath67 , @xmath68 @xcite . due to the global phase invariance , @xmath69 is always a double eigenvalue . a stationary solution is linearly stable if all eigenvalues are located on the imaginary axis , i.e. , we can actually at most make a statement about marginal stability of the solution ( meaning that a perturbation at most will grow with a rate that is polynomial in @xmath2 ) . extensive calculations , covering a large portion of parameter space , have been done for the most fundamental stationary solutions . in fig . [ fig : stabbound ] the regions of stability ( shaded ) for the @xmath59-site and the two @xmath60-site solutions are shown for fixed norm @xmath70 . for the @xmath59-site case there actually exist three different solutions in part of the region @xmath71 ( the edge of the instability region ) , each accessible by continuation but via different paths from the anti - continuous limit . stability is indicated if one of the solutions is stable ( there is no multistability of these three 1-site solutions and some details can also be found in figs . [ fig : bifurc1 ] and [ fig : bifurc2 ] ) . the same behaviour appears also in 1d @xcite , and is presumably an effect of the nonlinear coupling terms that can lead to a vanishing effective coupling between lattice sites . see further the discussion in sec . [ sec : compact ] . from fig . [ fig : stabbound ] we see that both stable and unstable solutions of all kinds exist and we note that the stability regions for the @xmath60-site solutions together cover the region of instability of the @xmath59-site solutions . for a quite large range of parameter values we thus have stable propagation of both types of solutions along the array of waveguides . the stability boundaries of the different solutions do not coincide . in the 1d model @xcite , the stability boundaries very nearly coincide and in fact create a very narrow region of parameter space where the @xmath59-site solution and one of the @xmath60-site solutions have either simultaneous stability or instability . moreover , the exchange of stability across this narrow range is connected to the existence of an intermediate asymmetric solution in this region , interpolating between the two solutions . when crossing the region of stability inversion , the intermediate solution appears from a pitchfork bifurcation as one of the solutions ( say the @xmath59-site solution ) changes its stability and then disappears in a reversed pitchfork bifurcation with the other ( @xmath60-site ) solution @xcite . here , in 2d , although the stability boundaries lie rather far apart in parameter space , unstable asymmetric intermediate solutions still exist between the stability boundaries of the @xmath59-site solutions and both @xmath60-site solutions ( fig . [ fig : stabbound]d ) and appear as the result of similar bifurcations . examples of the solutions can be seen in fig . [ fig : sol ] . but for @xmath72 . note here that @xmath29 for the @xmath59-site solution is always larger than for the @xmath60-site solutions . ] moreover , comparing the hamiltonians of the obtained solutions reveals the existence of zeros of the energy difference along some boundaries in parameter space . these have been indicated in fig . [ fig : stabbound ] . to better clarify the relations among the different solutions , bifurcation diagrams for @xmath73 and @xmath72 are shown in figs . [ fig : bifurc1 ] and [ fig : bifurc2 ] , with hamiltonian @xmath29 for the solutions plotted against the parameter @xmath4 . note that the solution that maximizes @xmath29 always is stable . from fig . [ fig : bifurc1 ] it is clear that the intermediate solution appears in connection with the vanishing energy difference , i.e. , the difference in hamiltonian , of the @xmath59-site and @xmath60-site solutions @xmath74 , and through bifurcations provide the mechanism of their exchange of stability . the intermediate solution always lies very close but has a slightly lower value of @xmath29 and is therefore unstable , in contrast to the solutions in 1d that may be stable @xcite . from a mobility point of view , it is the energy of the intermediate solution that must be overcome to achieve mobility . this additional energy ( the pn - barrier ) is , with reference to the stable modes , small compared to the total energy ( @xmath75 ) . note also in fig . [ fig : bifurc1 ] that for each @xmath60-site solution there are two intersections with @xmath59-site solutions and hence two zeros of the energy difference . one is , as expected from the energy minimum principle , present in the region of existence of the intermediate solutions , where the @xmath59-site solution and the respective @xmath60-site solution have simultaneous stability and are the stationary modes with highest value of @xmath29 ( see the zoomed areas ) . the energy difference is also zero in a region where the involved solutions are not the ones with highest @xmath29 , and in particular the @xmath59-site solution is unstable in this region ( see intersections at @xmath76 ) . this intersection , as can be seen from the solution insets in fig . [ fig : bifurc1 ] , is with a solution possessing the ` wrong ' symmetry , i.e. , the nearest neighbours of the centre site have the opposite sign compared to the symmetry or anti - symmetry of the respective @xmath60-site solution . therefore exchange of stability , or mobility , can not be expected at this vanishing of the energy difference . in fig . [ fig : bifurc1 ] we also see that the branches of the @xmath59-site solutions terminate . two of the branches will merge at @xmath77 , thus the end of the branches are due to bifurcations . the behaviour of the three branches is not symmetric and the two branches lying close in the left part of the figure will not coincide . instead , the upper branch of these two is terminated at @xmath78 in a bifurcation with a symmetric @xmath79-site solution that at the anti - continuous limit has @xmath79 in - phase neighbouring sites excited in a cross configuration . the lower branch can be continued to lower values of @xmath4 where it will approach a plane wave solution . a bit surprisingly , when considering the results of the stability analysis of stationary solutions , the mobility of localized modes in the model ( [ eq : xdnls ] ) is very poor and in stark contrast to the very excellent mobility in the 1d model @xcite and the saturable model in 1d @xcite and 2d @xcite near points of minimal pn - barrier . the method used to try to induce mobility is to apply a phase gradient @xmath80 across one of the stationary real solutions . in applications this can be achieved by launching the laser beam that excites the waveguides at an angle to the face of the array @xcite . this perturbation will kick the solution in the direction of the phase gradient @xcite and will have an effect similar to the marginal mode perturbation at an instability boundary described in @xcite , which in a complex formulation will correspond to a pure imaginary ( velocity ) perturbation to the real ( position ) solution . the phase gradient perturbation also has the nice feature of preserving the norm . here we are strictly interested in axial propagation as this is the direction of possible propagation in the lattice suggested by the minimal pn - barrier of the stationary modes . therefore @xmath81 has consistently been used . mobility in the diagonal direction would for the nearest - neighbour coupled lattice be connected to a vanishing energy difference for the @xmath59-site solution and a @xmath82-site solution ( quadrupole ) . this does not appear in the present model and the @xmath82-site solutions generally has a much lower value of @xmath29 than the @xmath59-site solution . the reason that it is possible in the model in @xcite is the saturable nonlinearity that will limit the amplitude of the excitations , bringing the singly excited and multiply excited solutions closer in energy for high enough power . the same effect is also the reason for the multiple zeros of the energy difference , @xmath83 , in saturable media , as more and more excited sites reach the saturation limit . for the model ( [ eq : xdnls ] ) we should expect at most one zero of the pn - barrier with fixed coupling parameters and varying norm @xmath31 . -site solution for @xmath19 , @xmath20 , @xmath73 , @xmath84 and @xmath70 ( solution shown in fig . [ fig : sol]a ) with a phase gradient @xmath85 . in all cases @xmath29 is sufficiently lowered to be below the value of the lowest stationary ( the intermediate ) solution with @xmath86 . the solid line is for @xmath87 ( @xmath88 ) , the dotted line for @xmath89 ( @xmath90 ) and the dash - dotted line for @xmath91 ( @xmath92 ) . ] trying to kick any of the stationary solutions in the region of existence of the intermediate solution , where the pn - barrier is minimal , will result in an oscillatory behaviour near the initial position . the centre of energy for the axial direction @xmath40 , defined as @xmath93 will at most be displaced about one lattice site . the generic behavour is examplified in fig . [ fig : masscentre1 ] where the @xmath59-site solution in fig . [ fig : sol]a is perturbed with different phase gradients . although the perturbation is large enough to overcome the pn - barrier , set by the unstable intermediate solution , no mobility , except small displacements to neighbouring sites , is observed , indicating that the effect of pinning is stronger than suggested by the energy difference considerations . the final position of @xmath94 is not even always in the direction of the phase gradient kick , and is located either on or between sites corresponding to one of the stable stationary solutions . since the initial velocity of the excitation given by the perturbation can be large , the absence of mobility here is not the effect of resonance with the multi - spectral bands appearing at low velocity for travelling waves near a zero of the pn - barrier @xcite . -site solution for @xmath19 , @xmath20 , @xmath73 , @xmath95 and @xmath70 with a phase gradient @xmath89 . the motion depends sensitively on initial conditions and boundary conditions . here periodic boundary conditions on a @xmath96 lattice is used . ] if we instead start with solutions away from the region of existence of the intermediate solutions only small oscillations around the stable stationary solution in the present region is observed . displacement of the centre of energy a few sites is however possible . in fig . [ fig : masscentre2 ] an unstable symmetric @xmath60-site solution is perturbed and the centre of energy moves up to two sites . but the irregular oscillations are due to a reconfiguration of the excitation from an initial more symmetric profile to , at the final position , a more anti - symmetric profile being closer to the stable stationary mode at the given values of the parameters . the motion is highly unpredictable and depends sensitively on the initial conditions and boundary conditions . -site solution for @xmath19 , @xmath20 , @xmath73 , @xmath97 and @xmath70 . by symmetry , the eigenmode is degenerate with a corresponding eigenmode oriented along the @xmath43 direction . the related eigenvalue is @xmath98.,width=188,height=188 ] , relative to the value of the @xmath59-site solution ( @xmath99 ) , versus norm @xmath31 for @xmath19 , @xmath20 , @xmath73 and @xmath84 . the intermediate solution ( @xmath100 ) emerges from a pitchfork bifurcation with the symmetric @xmath60-site solution ( @xmath101 ) at @xmath102 . it is always the solution with the slightly lower value of @xmath29 and is unstable . as @xmath31 increases , although very close in @xmath29 , there will be no bifurcation with the @xmath59-site solution , implying that the intermediate solution will not connect the @xmath59-site and the @xmath60-site solutions when considered as a function of the dynamical parameter @xmath31 . the inset shows @xmath29 versus @xmath31 for the @xmath59-site solution , as a reference . ] , relative to the value of the @xmath59-site solution ( @xmath99 ) , versus norm @xmath31 for @xmath19 , @xmath20 , @xmath73 and @xmath103 . the intermediate solution ( @xmath100 ) appears at @xmath104 and disappears at @xmath105 in a bifurcation with the anti - symmetric @xmath60-site solution ( @xmath106 ) . the inset shows @xmath29 versus @xmath31 for the @xmath59-site solution , as a reference . ] , relative to the value of the @xmath59-site solution ( @xmath99 ) , versus norm @xmath31 for @xmath19 , @xmath20 , @xmath73 and @xmath107 . in this case the intermediate solution ( @xmath100 ) emerges from a pitchfork bifurcation with the @xmath59-site solution at @xmath108 and there is actually no zero of the energy difference . the anti - symmetric @xmath60-site solution ( @xmath106 ) always has a higher value of @xmath29 and is stable . the inset shows @xmath29 versus @xmath31 for the @xmath59-site solution , as a reference . ] at the edge of the stability boundary , where the bifurcation with the intermediate solution occurs for the fundamental stationary modes , the eigenmodes related to the real eigenvalues that are responsible for the instability has a form that may be indicative of mobility . the real part of this growing mode is shown in fig . [ fig : eigmode1 ] for the @xmath59-site solution . exactly at the stability boundary the imaginary part of the eigenmode is zero . from the form of the eigenmode we conclude that perturbing along this mode ( ` depinning mode ' in @xcite ) will promote a change of shape of the @xmath59-site solution towards a profile that resembles the symmetric @xmath60-site solution ( or the intermediate solution ) . the opposite relation holds at the stability boundary of the @xmath60-site solution . hence the form of the growing modes supports the heuristic view of mobility of narrow excitations as a transformation between stationary solutions ( see also fig . 3 in @xcite ) . but in the present model the stability boundaries are far apart in parameter space and such a transformation may not be possible . to be conclusive in this matter we will need to fix the coupling parameters and investigate the exchange of stability as a function of the dynamical parameters @xmath29 and @xmath31 . if @xmath31 is varied for a set of fixed coupling parameters inside the right region of existence of the intermediate solution in fig . [ fig : bifurc1 ] we can find a situation when the intermediate solution will only bifurcate with the symmetric @xmath60-site solution , and not with the @xmath59-site solution for reasonable values of the norm ( @xmath109 ) . this is illustrated in fig . [ fig : bifurc3 ] . in the event that a bifurcation occurs with both the fundamental solutions involved ( fig . [ fig : bifurc5 ] ) , the bifurcation points are far apart measured also in terms of the dynamical parameters . measuring the distance between the bifurcation points in norm ( @xmath110 ) we find that it is always comparable in size to the value of the norm at the bifurcation point with the higher value , i.e. , @xmath111 . the pn - barrier ( @xmath112 ) is small compared to @xmath29 for all values of the norm . since the bifurcation points are either far apart or the intermediate solution does not connect the @xmath59-site and @xmath60-site solutions at all , this indicates that there is no way to transform between the two . put in another way , this suggests that there is no trajectory in phase space that passes close to the stationary configurations and consequently the mobility is very poor . our numerical observations lead us to the conclusion that mobility is resisted since the site - centred and bond - centred configurations are too far apart for the adjustment of @xmath31 and @xmath29 to be possible by exchange with an accompanying oscillating background , or in this case mainly an adjustment of @xmath31 since the difference in energy is always relatively small . a tractable analysis could possibly be made along the lines of the method of an effective hamiltonian where the dynamics is averaged over periodic trajectories in phase space @xcite . the more explicit interpretation of the pn - barrier as referred to a trajectory in phase space would definitely be more suitable in the present context . however , some extensions are necessary as an evaluation of the method shows that it gives an accurate description of mobility only close to limits where the travelling excitations are exact @xcite . but this does not exclude that some insight might be given to the effect of pinning observed here . further , as shown in fig . [ fig : bifurc4 ] , there are cases when the pn - barrier does not vanish at all , but the intermediate solution still exists . an interesting effect of the nonlinear coupling terms in eq . ( [ eq : xdnls ] ) is that they may cancel the linear coupling and effectively give rise to a total coupling that is zero @xcite . this property is also the criterion for the existence of exact compact solutions having strictly zero amplitude outside an interval @xcite . the criterion will take the form of a parameter constraint and this implies that the compact solutions , as exact mathematical entities , are not robust with respect to parameter variation . they are interesting anyway from a physical perspective , since a near perfect localization prevails in the neigbourhood of the exact parameter values , and this family of strongly localized solutions may very well be dynamically stable . the solution will develop an exponentially decaying tail , but the contrast , i.e. , the ratio of the amplitude of neighbouring sites to the amplitude of the excited sites , is still large which can be desirable for applications @xcite . further , although we speak of exact compact solutions in the context of the lattice equation ( [ eq : xdnls ] ) , this does not imply compactness in the real system . the equation describes the amplitude of the modes in the respective waveguides , but the modes are themselves spatially exponentially decaying . the extreme localization represented by compact solutions will nonetheless be preferable if several excitations propagate in the system at the same time as these will constitute excitations with least mutual interaction . the existence of compact solutions has an impact on the properties of the system under study . the idea of the compactification is that for a particular amplitude the total coupling can vanish when the neighbouring sites have zero amplitude , leading to a decoupling of the lattice between these sites . take , e.g. , the @xmath59-site excitation with only the site @xmath46 excited and all others at rest and put this ansatz into eq . ( [ eq : xdnlsstat ] ) . from the equation on one of the neighbouring sites , e.g. , site @xmath113 , we get @xmath114 . for a compact solution with @xmath115 excited sites , this constraint generalises to @xmath116 if all sites have the same amplitude . note that @xmath117 is required for the existence of compact solutions ( as @xmath10 ) . this criterion corresponds to having a negative kerr index of the nonlinear material surrounding the waveguides , or for bec in an optical lattice an effective interatomic attraction in the dilute gas ( as for @xmath118li ) . solving for the @xmath59-site and @xmath60-site compact solutions the results are the same as in 1d @xcite . from eq . ( [ eq : qeff ] ) the constraint for the @xmath59-site solution will correspond to the line @xmath119 in fig . [ fig : stabbound ] , that precisely cuts through the edge of the instability region of the @xmath59-site solution and hence also seperates the two different regions of existence of the intermediate solutions . the loose interpretation of this constraint as an effective coupling is further strenghtened when considering that the two @xmath60-site solutions will have the same value of @xmath29 along the plane of existence of the @xmath59-site solution . note , e.g. , that the lines of @xmath29 intersect at @xmath119 in the bifurcation diagrams in figs . [ fig : bifurc1 ] and [ fig : bifurc2 ] . compact solutions with more than two sites excited can also be found . there are two different classes of real @xmath120-site solutions , with the center site being either in - phase or out - of - phase with the two outermost sites . having , in 2d , the freedom to choose the orientation of the three sites in the lattice as being all along one lattice direction or in an ` l'-configuration , the presence of coupling in two different lattice directions will also mean less freedom when compared to the 1d system . the extra restrictions imposed for the @xmath120-site solutions implies that all sites must have the same amplitude , which is not required in 1d . explicit calculations yield the same frequency @xmath121 for all solutions , and apart from the parameter constraint ( [ eq : qeff ] ) also the constraint @xmath122 , where the upper sign is for the in - phase solution and the lower sign for the out - of - phase solution . a check of the stability shows that only the out - of - phase solution will be stable for @xmath123 , regardless of the orientation in the lattice . it is also possible to solve analytically for compact @xmath82-site and @xmath79-site solutions , some of which are stable for large enough values of the norm ( @xmath124 ) . an interesting extension when going from one to two spatial dimensions is the possiblity to support localized excitations having a topological charge , which is proportional to the angular momentum carried by the excitation . we define the topological charge @xmath125 as the number of complete @xmath126 twists of the phase when moving on a contour around the excitation . such discrete vortex solitons have been observed in experiments @xcite and described in the context of dnls - type equations @xcite . in the model ( [ eq : xdnls ] ) _ compact vortices _ can be supported . the constraint ( [ eq : qeff ] ) must still be fulfilled , which means , as we have fixed @xmath18 and @xmath13 by rescalings , that the parameter @xmath4 is fixed if the norm @xmath31 and @xmath8 are used as free parameters . the most simple compact vortex is given by four excited sites in a square configuration with a phase difference of @xmath127 between neighbouring sites giving a topological charge @xmath128 . this solution will have the frequency @xmath129 and have a very small region of stability near @xmath130 ( e.g. , for @xmath131 it is stable for @xmath132 ) . for the square @xmath82-site configuration placed diagonally in the lattice , with a zero - amplitude mediating site in the middle , the frequency is instead @xmath133 and the region of stability is dramatically extended so that for @xmath131 the configuration is stable for @xmath134 . the excited sites are not nearest neighbours in this case and will only weakly interact . in fact for @xmath135 the instability will set in at @xmath136 , which exactly corresponds to the instability of the compact @xmath59-site solution for @xmath70 . with more excited sites it is possible to find larger @xmath128 vortices that are stable for certain parameter values . [ fig : vortex ] an example with eight sites is shown . also higher order compact vortices exist . the configuration with eight sites on a square contour with phase difference @xmath127 between the sites will have charge @xmath137 and frequency @xmath138 . it will also be stable for large norm ( @xmath124 ) and small @xmath8 . ( note that , by contrast , this @xmath137 configuration is always _ unstable _ in the standard on - site dnls model @xcite . ) and ( b ) the phase @xmath139 for a stable stationary compact vortex of topological charge @xmath128 for @xmath135 , @xmath19 , @xmath20 , @xmath140 and @xmath141 . the frequency is given by @xmath142 . ] the vortex solutions are necessarily complex solutions as they have a phase difference between neighbouring sites other than @xmath143 or @xmath144 , i.e. , they can not be transformed to real form using the global phase invariance of the dnls - type equations . this implies that the norm current , eq . ( [ eq : jn ] ) , is non - zero and that there is a flow of norm around the contour of the vortex balanced according to the discrete continuity equation ( [ eq : neq ] ) with constant norm density @xmath33 . as described for the 1d model in @xcite the presence of the nonlinear coupling will introduce a non - trivial phase dependence in the norm current density . in particular , the norm current @xmath47 in eq . ( [ eq : jn ] ) can become zero for specific values of the phase difference given by @xmath145 thus under the constraint ( [ eq : cosphi ] ) stationary complex solutions can exist without having to form a closed loop for the norm current flow , which is the case for a vortex , simply because there is no current . the compact complex @xmath60-site solution present in the 1d model @xcite exists also here , with @xmath146 and @xmath147 . it is also stable for a range of phase twists , e.g. , for @xmath131 stability is found for @xmath148 , equivalent to @xmath149 . also compact complex solutions with more sites excited can be constructed , e.g. , @xmath82-site non - vortex solutions that generally seem to be unstable . neither these solutions nor the vortex solutions above need to be compact , but can be continued to non - compact solutions by lifting the constraint set by eq . ( [ eq : qeff ] ) . although this may be an interesting further development the given examples serve to give an idea of the phenomena at hand . leaving the localized solutions , we will in this section focus on the transport properties of plane waves . the excitation of the lattice with a plane wave correponds to an even distribution of the energy over all sites in the lattice . a travelling wave on the form @xmath150 will have a norm density ( power per site ) @xmath151 and be a solution of eq . ( [ eq : xdnls ] ) for the frequency @xmath152 note that the dispersion relation ( [ eq : pwfreq ] ) can be derived from eq . ( [ eq : heq_theta ] ) in the form @xmath153 . though being a stationary solution , the plane wave ( [ eq : pw ] ) is associated with a balanced transfer of norm ( power ) between the sites due to the induced phase difference ( wave number ) . this balance is described by the discrete continuity equation ( [ eq : neq ] ) with constant norm density . from the interpretation of @xmath154 and @xmath155 as the norm transferred from a given site to its neighbours in the positive @xmath40 and @xmath43 directions we can , from eq . ( [ eq : jn ] ) , obtain the norm current flowing in the lattice for a plane wave solution , @xmath156\hat{\mathbf{n } } \\ & + 2\rho\sin\phi \big[q_{2}+4q_{4}\rho\cos\phi+4q_{5}\rho\big]\hat{\mathbf{m}}. \end{split}\ ] ] further , it is convenient to introduce a polar representation of the current defined by the relation @xmath157 , so that @xmath158 is the magnitude of the current and @xmath159-\pi,\pi[\ ] ] its direction measured counter clockwise relative to the @xmath40 direction . note that the power transported along the waveguides is proportional to @xmath160 and that the quantities ( [ eq : jvec ] ) , ( [ eq : jmag ] ) and ( [ eq : jdir ] ) describe the transport across the array . for a finite array such transport can only be sustained if there is a supply of power at one boundary and a drain at the opposite boundary . to ensure that the plane wave ( [ eq : pw ] ) has stable propagation its behaviour under a modulation is investigated . adding a modulating perturbation with wave numbers @xmath161 and @xmath162 we make the ansatz @xmath163 \\ & \times\exp[-{\mathrm{i}}(\varphi n+\phi m + \lambda z ) ] \end{split}\ ] ] in eq . ( [ eq : xdnls ] ) @xcite . keeping only terms linear in @xmath164 and @xmath165 , the modulational stability can be determined from the equations @xmath166 with @xmath167 b = & \phantom{0 } 2q_{3}\rho - 2q_{2}(\cos\varphi+\cos\phi ) \nonumber \\ & + 4q_{4}\rho[2\cos q(1+\cos 2\varphi)-\cos 2\varphi ] \nonumber \\ & + 4q_{4}\rho[2\cos p(1+\cos 2\phi)-\cos 2\phi ] \nonumber \\ & + 2(q_{2}+8q_{5}\rho)(\cos\varphi\cos q + \cos\phi\cos p ) \ , , \\[2 mm ] c = & \phantom{0 } 2q_{3}\rho + 4q_{4}\rho(\cos 2\varphi+2\cos q + \cos 2\phi+2\cos p ) \nonumber \\ & + 8q_{5}\rho[\cos\varphi(1+\cos q ) + \cos\phi(1+\cos p ) ] \,.\end{aligned}\ ] ] the plane wave will be stable if the eigenfrequencies @xmath168 are real for all @xmath169-\pi,\pi]$ ] . by explicit calculation we find that the eigenfrequencies can be written on the form @xmath170^{1/2}\!\!,\end{aligned}\ ] ] where we have introduced the ( energetic ) effective mass @xmath171 with a similar expression for @xmath172 , and the function @xmath173,\ ] ] entirely in correspondence with the result in 1d @xcite . from the first factor in eq . ( [ eq : weig ] ) we see that if the effective masses have opposite signs there will be at least some values of @xmath161 and @xmath162 that will yield instability . an important observation that follows is that if @xmath174 or @xmath175 change sign due to tuning of some available parameters , the plane wave can only change its stability or remain unstable ( unless the second factor in eq . ( [ eq : weig ] ) changes sign at the same time ) . generally , the direction of power flow in a lattice will be completely determined by the induced phase gradient of the plane wave . for the cubic on - site dnls equation ( @xmath176 ) , e.g. , the direction of the norm current is fixed by the relative phases between the sites , @xmath177 , while the magnitude is linear in the amplitude , @xmath178 . thus any control of the norm current by tunable parameters is restricted to the wave numbers of the plane wave . with current experimental techniques , a phase gradient is induced by having the exciting laser beam directed at an angle to the array of waveguides @xcite , so the wave numbers are easily adjustable but not rapidly tuned . also , for a finite array the wave numbers will be quantized in order to satisfy the imposed boundary conditions . as discussed in @xcite , with the introduction of nonlinear coupling terms in the model ( [ eq : xdnls ] ) , the norm current will have a nonlinear dependence on the amplitude . this will make @xmath160 more suitable as a tunable parameter , also from an application point of view . including the @xmath4 term , but keeping @xmath130 ( which would be relevant for application to bec in an optical potential @xcite ) , will introduce a nonlinear dependence in the magnitude @xmath179 through the factor @xmath180 . the sign of this factor will determine in which of the two possible directions ( @xmath181 ) , differing by an angle @xmath144 and fixed by the phase gradients @xmath182 and @xmath183 , the current will flow . for @xmath117 it is hence possible to make the current zero for a finite value of the amplitude @xmath160 and at the same time reverse its direction , just as in 1d @xcite . however , as the factor @xmath180 appears also in the effective mass ( [ eq : mass ] ) , the change of sign will affect the stability of the travelling plane wave . in particular , as noted in sec . [ sec : mi ] , the plane wave can not be modulationally stable for both directions of the current near the point of inversion . and ( b ) the direction @xmath181 of the norm current for varying amplitude @xmath160 with the parameters @xmath19 , @xmath20 , @xmath72 and @xmath107 . the solid line is for @xmath184 , and the plane wave is stable for all @xmath160 . the dashed line is for @xmath185 and @xmath186 , and the plane wave is stable for @xmath187 . ] inclusion also of the @xmath8 term will change this situation , as the zeros of @xmath188 and @xmath189 in general no longer will coincide . however , numerical investigations reveal that stable inversion is not possible in the axial direction , i.e. , with @xmath190 , for any values of @xmath182 and @xmath191 , but it is possible in other directions . taking @xmath192 will give a diagonal direction of the current , with @xmath193 depending on the relative signs of @xmath188 and @xmath194 , and always stability near the point of inversion . an example of such a flip of direction as the amplitude of the plane wave is varied is shown in fig . [ fig : current ] . an even greater flexibility is presented if different phase gradients are chosen for the @xmath40 and @xmath43 directions , @xmath195 . as illustrated in fig . [ fig : current ] we may have the remarkable situation that as the amplitude @xmath160 is tuned the magnitude of the current is near constant while _ the direction is continuously changed _ an angle @xmath127 over a stable range . such an exact control of the power flow by simply changing the amplitude of the plane wave may find useful applications . one possible application could be to use the array of waveguides , excited with a plane wave , as the link in a power coupling device . the advantage over a direct coupling would be a greater control of the flow facilitated by the control of the excitation in the waveguides . the effect of current inversion could form the basis for an amplitude - operated power - switch . however , any discussions in this direction other than mere speculative are beyond the scope of this paper . in conclusion , we have studied the extension to two spatial dimensions of a one - dimensional model describing coupled optical waveguides going beyond the lowest - order approximation and including nonlinear coupling , which , e.g. , is relevant for arrays of linear waveguides embedded in nonlinear media . the phenomenology of the 1d model presented in a series of papers @xcite has been revisited in the 2d setting . especially , we have shown that a vanishing energy difference between @xmath59-site and @xmath60-site stationary solutions and a subsequent small peierls - nabarro energy barrier , taking into account also the existence of asymmetric intermediate solutions , does _ not _ result in good mobility of highly localized modes . the main reason , and the difference to models that show good mobility @xcite , is that , despite small energy differences , the stationary solutions are still in some sense far apart . rather , the bifurcation points where the fundamental modes change their stability and where they exhibit a depinning mode promoting mobility are , although close in hamiltonian ( @xmath29 ) , far apart in norm ( @xmath31 ) ( figs . [ fig : bifurc3 ] and [ fig : bifurc5 ] ) . from this we conclude that for the stationary solutions that have near equal values of both hamiltonian and norm there is likely no trajectory in phase space passing close to them both . thus , in order to have good mobility of narrow modes in discrete systems , we need _ not only _ a small pn - barrier but also an exchange of stability between @xmath59-site and @xmath60-site solutions taking place _ in the vicinity_. we therefore also conjecture that the size of the oscillating background accompanying travelling modes @xcite is not only related to the size of the pn - barrier but also to the difference in other conserved quantities ( like @xmath31 ) of the involved stationary modes measured at the points where their stability is changed . the lack of mobility near a zero of the energy difference between fundamental modes was observed also in @xcite , but no further analysis was carried out . effects of a more delicate balance between linear and nonlinear coupling terms in eq . ( [ eq : xdnls ] ) have also been discussed . these include the existence and stability of compact solutions , both discrete breathers and discrete vortex solitons . the latter have to our knowledge not been previously reported . mathematically , these solutions are not robust with respect to parameter variation and require a balance of the coupling parameters in the equation for exact compactness , but near perfect localization persists in the neighborhood of the exact parameter values . an interesting question is if they can prevail also beyond the tight - binding approximation of eq . ( [ eq : xdnls ] ) . the answer is likely that they are only exact solutions for the present model , but that they will correspond to modes of a higher degree of localization in more accurate models , i.e. , they give an indication in which parameter regimes high localization can be achieved . as discussed in sec . [ sec : compact ] the compact solutions do not represent a true compactification in the real system . additional interesting effects present also in 1d arise from a non - trivial dependence of the power flow in the lattice on the amplitude and phase difference of an excitation . particularly , the current can become zero for a non - trivial phase twist which also in 2d leads to the existence of complex localized stationary modes with an open geometry , that may even be stable . further , we have shown how the transversal flow of power in the array of waveguides can be controlled with great flexibility by excitation with plane waves , and in particular explicitly demonstrated how the transport _ direction _ may be continuously tuned by _ amplitude _ variation . this may have applications for power - coupling devices . however , the use of a 2d array of the type studied here for multiport switching is discouraged due to the poor mobility of localized modes . for this purpose a 1d array @xcite or an array exhibiting a saturable @xcite or quadratic @xcite ( see discussion below ) nonlinearity instead of the kerr nonlinearity is better suited . in @xcite , abdullaev et.al extended the wannier - function approach of @xcite for the continuous 1d nls equation with periodic linear potential to the case with periodic variation also for the nonlinearity coefficient ( which is the case also considered in our work ) . within the tight - binding approximation , they derived under quite general conditions a lattice equation , which in a special case ( @xmath196 in @xcite ) is equivalent to the 1d version of our eq . ( [ eq : xdnls ] ) as derived in @xcite . an important conclusion of @xcite is the crucial importance to include also the nonlinear coupling terms ( corresponding to our @xmath197 and @xmath198 ) , as they for specific choices of nonlinear interactions were shown to be comparable with , or even stronger than , the on - site nonlinearity . thus , this supports the soundness of our approach to consider variation of the parameters @xmath197 and @xmath198 over a rather large range of values . several works discussing properties of moving solitons and ( vanishing ) pn - barriers in 1d generalized dnls models have appeared . dmitriev , khare et al . @xcite found analytically exact stationary and moving solutions for some of the exceptional , translationally invariant discretizations of @xcite . whether these are of relevance for any physically realizable system is , to our knowledge , still unclear . pelinovsky et al . @xcite developed a mathematical technique for analysis of persistence of traveling single - humped localized solutions from a certain limit , and found specifically that while travelling solutions terminated when continued from the integrable al - limit of the salerno model ( corresponding to the development of a resonant tail @xcite ) , they generally persisted in the translationally invariant models . oxtoby and barashenkov @xcite used the method of asymptotics beyond all orders to evaluate the amplitude of radiation from a moving small - amplitude soliton in the saturable dnls equation , and found it to be completely suppressed at certain ` sliding velocities ' , where , similarly as in @xcite , it was interpreted as an embedded soliton . the properties of the travelling solitary waves in the saturable model were also analyzed numerically in more detail by melvin et al . in @xcite . for the two - dimensional case , susanto et al . @xcite studied the mobility of discrete solitons in a square lattice with _ quadratic _ ( second - harmonic - generating ) nonlinearity . in this case , due to the absence of collapse instability in the continuum limit , good mobility of stable solutions was observed as long as the coupling constants were not too small ( ` quasicontinuum regime ' ) . in this regime , mobility in arbitrary directions ( not necessarily axial or diagonal ) was observed , which is not surprising since discreteness effects are smoothened out for broad , continuum - like solitons . however , no non - trivial zeros of the pn barrier were found for this model , and consequently strongly localized solutions were reported to be immobile . thus , in many aspects , the mobility properties of the 2d model with quadratic nonlinearity is similar to those of the 1d cubic on - site dnls model . finally , a very recent preprint by chong et al . @xcite extended the analysis of the on - site cubic - quintic dnls model of @xcite to higher dimensions . as concerns the two - dimensional mobility , the results were shown to be very similar to that of the saturable model @xcite : enhanced mobility was found in regimes close to stability inversion and associated with appearance of asymmetric intermediate solutions and low pn barrier . thus , as far as we are aware , the present work still provides the only known explicit example of a model where these properties do _ not _ lead to an enhanced mobility . m would like to thank the mathematics department at heriot - watt university , edinburgh , for its hospitality and especially j.c . eilbeck for guidance , support and very useful discussions . this work was partly carried out under the hpc - europa project ( rii3-ct-2003 - 506079 ) , with the support of the european community - research infrastructure action under the fp6 `` structuring the european research area '' programme . partial support from the swedish research council is acknowledged . pelinovsky , nonlinearity * 19 * , 2695 ( 2006 ) . kevrekidis , s.v . dmitriev and a.a . sukhorukov , mathematics and computers in simulation * 74 * , 343 ( 2007 ) . a. maluckov , m. stepi , d. kip and lj . hadievski , eur . j. b * 45 * , 539 ( 2005 ) . m. ster , _ stability and mobility of localized and extended excitations in nonlinear schrdinger models _ , doctoral thesis , linkping studies in science and technology . dissertations , no . 1072 , isbn 978 - 91 - 85715 - 83 - 1 , linkping university ( 2007 ) ; published electronically at http://urn.kb.se/resolve?urn=urn:nbn:se:liu:diva-8091 . m. ster and m. johansson , in _ proceedings of the conference on localization and energy transfer in nonlinear systems , san lorenzo de escorial , madrid , spain , 2002 _ , edited by l. vzquez , m.p . zorzano and r.s . mackay ( world scientific , singapore , 2003 ) .

we propose a scenario where blazars are classified as flat - spectrum radio quasars ( fsrqs ) , bl lacs , low synchrotron , or high synchrotron peaked objects according to a varying mix of the doppler boosted radiation from the jet , the emission from the accretion disk , the broad line region , and the light from the host galaxy . in this framework the peak energy of the synchrotron power ( ) in blazars is independent of source type and of radio luminosity . we test this new approach , which builds upon unified schemes , using extensive monte carlo simulations and show that it can provide simple answers to a number of long - standing issues including , amongst others , the different cosmological evolution of bl lacs selected in the radio and x - ray bands , the larger values observed in bl lacs , the fact that high synchrotron peaked blazars are always of the bl lac type , and the existence of fsrq / bl lac transition objects . objects so far classified as bl lacs on the basis of their _ observed _ weak , or undetectable , emission lines are of two physically different classes : intrinsically weak lined objects , more common in x - ray selected samples , and heavily diluted broad lined sources , more frequent in radio selected samples , which explains some of the confusion in the literature . we also show that strong selection effects are the main cause of the diversity observed in radio and x - ray samples , and that the correlation between luminosity and , that led to the proposal of the `` blazar sequence '' , is also a selection effect arising from the comparison of shallow radio and x - ray surveys , and to the fact that high - high radio power objects have never been considered because their redshift is not measurable . [ firstpage ] bl lacertae objects : general quasars : emission lines radiation mechanisms : non - thermal radio continuum : galaxies x - rays : galaxies

1110.4706

once considered rare sources , blazars , a type of radio loud active galactic nuclei ( agn ) pointing their jets in the direction of the observer ( see e.g. * ? ? ? * ; * ? ? ? * ) , are now being detected in increasingly larger numbers . recent results from the _ wilkinson microwave anisotropy probe _ ( wmap ) , the _ planck _ and _ fermi _ satellites have established that blazars are the most common type of extragalactic sources found at microwave and energies @xcite . so far about 3,000 blazars are known @xcite , but their number is steadily growing thanks to the _ fermi _ @xcite , the optical sloan digital sky survey ( sdss : * ? ? ? * ) , and the _ planck _ @xcite surveys . some faint blazars are also being detected as serendipitous sources in _ swift_-xrt images @xcite . while all blazars share the same property of emitting variable , non - thermal radiation across the entire electromagnetic spectrum , they also display diversity . namely , they come in two main subclasses , whose major difference is in their optical properties : 1 ) flat spectrum radio quasars ( fsrqs ) , which show strong , broad emission lines in their optical spectrum , just like radio quiet qsos ; and 2 ) bl lacs , which are instead characterized by an optical spectrum , which at most shows weak emission lines , sometimes displays absorption features , and in some cases can be completely featureless . historically , the separation between bl lacs and fsrqs has been made at the ( rather arbitrary ) rest - frame equivalent width ( ew ) of 5 ( e.g. * ? ? ? * ; * ? ? ? however , no evidence for a bimodal distribution in the ew of the broad lines of radio quasars has ever been found and , on the contrary , @xcite pointed out that radio - selected bl lacs were , from the point of view of the emission line properties , very similar to fsrqs but with a stronger continuum . most bl lacs selected in the x - ray band , on the other hand , had very weak , if any , emission lines , and @xcite , when studying the properties of the x - ray selected _ einstein _ medium sensitivity survey ( emss ) sample , had to introduce another criterion to identify bl lacs , this time to separate them from galaxies . this was based on the ca h&k break , a stellar absorption feature typically found in the spectra of elliptical galaxies . given that its value in non - active ellipticals is @xmath0 , @xcite chose a maximum value of @xmath1 to ensure the presence of a substantial non - thermal continuum superposed to the host galaxy spectrum . this was later revised to @xmath2 @xcite . blazar classification depends then on the details of their appearance in the optical band where they emit a mix of three types of radiation : 1 ) a non - thermal , jet - related , component ; 2 ) thermal radiation coming from the accretion onto the supermassive black hole and from the broad line region ( at least in most radio - selected sources ) ; 3 ) light from the host ( giant elliptical ) galaxy . figure [ fig : sed ] represents these three components as red , blue and orange lines , overlaid to the spectral energy distribution ( sed ) of four well - known blazars ( from * ? ? ? * ) . the strong non - thermal radiation , the only one that spans the entire electromagnetic spectrum , is composed of two basic parts forming two broad humps , the low - energy one attributed to synchrotron radiation , and the high - energy one , usually thought to be due to inverse compton radiation ( see e.g. * ? the peak of the synchrotron hump ( ) can occur at different frequencies , ranging from about @xmath3 hz to over @xmath4 hz ( see e.g. the cases of 3c 273 or 3c 279 and mkn 501 in fig . [ fig : sed ] ) reflecting the maximum energy at which particles can be accelerated ( e.g. * ? ? ? blazars where is lower than @xmath5 hz in their rest frame are called low synchrotron peaked ( lsp ) sources , while those where @xmath5 hz @xmath6 @xmath7 hz , and @xmath8 hz are called intermediate and high synchrotron peaked ( isp and hsp ) sources respectively @xcite . this definition extends the original division of bl lacs into lbl and hbl sources first introduced by @xcite . the large disparity between lsp and hsp blazars ( up to five orders of magnitude ) is the cause of very large differences between the intensity of the radiation emitted in different energy bands . for instance , for the same radio flux an hsp source can be a factor of a 100 brighter in the optical band , or even a factor of a 1,000 brighter in the x - ray band , than an lsp blazar . this induces very strong selection effects in blazar samples discovered in different bands and led to some confusion when comparing , for example , the first radio and x - ray - selected bl lac samples in the early 1990 s . + we feel that all the above factors , which play an important role in blazar classification , have not been properly taken into account so far . the purpose of this paper is then to propose a new hypothesis to explain the existence of apparently different properties of the blazar subclasses , which solves at once many of the open issues of blazar research . throughout this paper we use a @xmath9cdm cosmology with @xmath10 km s@xmath11 mpc@xmath11 , @xmath12 and @xmath13 @xcite . the two main blazar subclasses have many differences , which include : 1 . _ different optical spectra _ ( by definition ) . there are , however , a number of bl lac - fsrq transition objects , which include even bl lacertae itself , the prototype of the class , which displays at times moderately strong , broad lines ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) and 3c 279 , a well - studied fsrq , which can appear nearly featureless in a bright state @xcite ; 2 . _ different extended radio powers_. most bl lacs have extended radio powers and morphologies consistent with those of fanaroff - riley ( fr ) type i , while basically all fsrqs are fr ii - like ( * ? ? ? * and references therein ) . however , despite the difficulty of classifying the radio morphology of sources with their jets forming a small angle with respect to the line of sight , some radio - selected bl lacs are known to posses an fr ii - like structure ( e.g. * ? ? ? * ; * ? ? ? * ) ; 3 . _ very different redshift distributions . _ fsrqs , similarly to radio quiet qsos , are typically found at redshifts @xmath14 , and up to @xmath15 , while bl lacs are usually much closer with very few cases at @xmath16 . this is graphically shown in fig . [ fig : histoz ] , which plots the redshift distribution of all the fsrqs and bl lacs included in the third edition of bzcat @xcite , the largest compilation of blazars currently available . it must be stressed , however , that a large fraction of bl lacs ( @xmath17% in bzcat and @xmath18% of the bl lacs in the fermi 1 and 2 year agn catalogs : * ? ? ? * ; * ? ? ? * ) have no measured redshift , due to the lack of any detectable feature in their optical spectrum , despite the use of 10-m class optical telescopes for the spectroscopy identification campaign @xcite ; 4 . _ different cosmological evolutions . _ detailed analyses of both radio selected and x - ray selected samples have led to the conclusion that the cosmological evolution of the two blazar subclasses are very different , with fsrqs evolving strongly ( again similarly to radio quiet qsos ) and bl lacs evolving at a similar , or perhaps lower rate , in the radio band , or even showing no or negative evolution in the x - ray band ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * and references therein ) ; 5 . _ widely different mix of fsrqs and bl lacs in radio and x - ray selected samples _ , with the latter typically including a much larger fraction of bl lacs than the former . in fact , while only @xmath19 of wmap5 blazars are bl lacs ( section [ ingredients ] ) , this fraction is instead @xmath20 in the emss complete sample , which includes 41 bl lacs and 15 fsrqs @xcite ; 6 . _ widely different distributions of the synchrotron peak energy . _ the rest - frame distribution of fsrqs is strongly peaked at low energies ( @xmath21 @xmath22 hz ) and never reaches very high values ( @xmath23 hz ) independently of the selection method @xcite , while the distribution of bl lacs is shifted to higher values by at least one order of magnitude . it can also reach values as high as @xmath24 hz and its shape varies strongly depending on the selection band ( that is radio , x - ray or @xmath25-ray : * ? ? ? * ; * ? ? ? some of these differences have been explained by so - called unified schemes , which posit that bl lacs and fsrqs are simply fr i and fr ii radio galaxies with their jets forming a small angle with respect to the line of sight @xcite . ( radio galaxies would then be the `` parent '' population of blazars ) . due to relativistic beaming this has enormous effects on their apparent emitted power and luminosity functions ( lfs ) and can explain their different extended radio powers and , partly , their cosmological evolutions . however , unified schemes per se can not account for transition objects , and the different evolution of radio and x - ray selected bl lacs and distributions . our new hypothesis builds upon unified schemes by adding dilution and strong selection effects as new , vital components . the existence of bl lac - fsrq transition objects and the recent finding that unresolved radio quasars ( i.e. fsrqs ) appear to be redder than radio quiet agn @xcite suggest that the strong non - thermal jet emission in blazars may have a significant impact on the shape of their blue bump and on the ew of their broad lines . in particular , given this extra continuum component , the ews of broad - lined blazars ( fsrqs ) should be systematically lower than those of radio - quiet agn . to test this , we extracted ew data for samples of radio - quiet qsos and blazars from the sdss dr-7 spectral database . in order to have a sample that is not excessively large but still representative of the entire data set , we considered only radio - quiet qsos with @xmath26 , thus limiting the sample size to about 650 sources . to obtain the ew of a sizable sample of blazars we considered all fsrqs in bzcat with 1.4 ghz flux @xmath27 mjy ( to simulate a radio flux - limited sample ) and in the wmap5 catalogue . for each object in the three samples we retrieved , from the sdss on - line system , the line measurements parameters ( e.g. ew , @xmath28 of the fit etc . ) of the strongest emission lines in quasar spectra , that is ly@xmath29 , c iv , c iii , mg ii , h@xmath30 , h@xmath29 . details on the fitting algorithm can be found at the sdss website . to avoid problematic cases , for our analysis we considered only sdss fits to emission lines with @xmath31 . figure [ fig : ewrlrq ] shows the distribution of the ew of the mgii emission line , for which we have better statistics , for the sample of radio quiet sdss qsos ( 663 objects ) , bzcat ( 126 objects ) and wmap5 ( 35 objects ) fsrqs . the ew of both blazar samples is smaller than that of radio quiet qsos , with median values respectively equal to 16.4 ( bzcat ) , 14 ( wmap5 ) , and 18.4 ( radio - quiet qsos ) . a kolmogorov - smirnov ( ks ) test gives a probability @xmath32 that both fsrq samples have the same distribution as that of radio - quiet agn . moreover , we found that this dilution increases further ( i.e. the median ew gets smaller ) if we make a cut at higher radio fluxes in the bzcat fsrqs . similar results were obtained for the other emission lines . this demonstrates that the broad lines of blazars are significantly diluted compared to those of radio quiet qsos . the mean absolute magnitudes of the blazar and radio - quiet samples agree within @xmath33 magnitudes , which means that the difference can not be due to the well - known anti - correlation between ew and absolute magnitude , the so - called `` baldwin effect '' @xcite . we note that a number of papers have stated that the optical spectra of radio - quiet and radio - loud qsos are not significantly different ( e.g. * ? ? ? * and references therein ) . however , the radio - loud samples on which this conclusion was based were still optically selected and required the presence of a uv excess and/or strong emission lines . therefore , these radio - loud samples were biased towards objects with strong blue bumps . the considerations made in the previous sections lead us to propose a new , very simple scenario where the observed blazar optical spectrum is the result of a combination of an intrinsic ew distribution and the effects of three components : a non - thermal , jet related one , a thermal one due to the accretion disk , and emission from the host galaxy . different mixes of these components determine the appearance of the optical spectrum and therefore the classification of sources in fsrqs ( dominated by strong lines ) , bl lacs ( with diluted , weak lines , if a standard accretion disk is present ) , and radio - galaxies ( where the host galaxy swamps both the thermal and non - thermal nuclear emission present in blazars ) . the other novel component is a single lf whose evolution depends on radio power . this idea was tested through extensive monte carlo simulations , as described below . our simulations include the following ingredients , which we kept as simple as possible and tied as much as possible to observational data : 1 . * luminosity function * we derive the lf and evolution of blazars at 41 ghz from the wilkinson microwave anisotropy probe ( wmap5 ) sample @xcite . given that the mean blazar radio - sub - mm spectral index is @xmath34 @xcite this is practically equivalent to a radio - selected sample . we extend on the work of @xcite , which was based on wmap3 , and define a flux - limited sample of high galactic latitude sources ( @xmath35 jy , @xmath36 ) including 161 fsrqs , 29 bl lacs , and 10 blazars of unknown type . by applying a maximum likelihood technique to the wmap5 blazars ( see , e.g. * ? ? ? * for details ) we obtain , together with the evolution discussed below , a best - fit local lf @xmath37 ( in units of gpc@xmath38 p@xmath11 ) between @xmath39 and @xmath40 w / hz , which we assume in our radio simulations . * cosmological evolution * powerful ( @xmath41 w / hz ) radio sources are known to display a strong , positive evolution at moderately low redshifts followed by a decline at higher redshifts ( e.g. * ? ? ? * and references therein ) . we parametrize this behaviour with a simple model of the type @xmath42 , which allows for a maximum in the luminosity evolution followed by a decline . this was first suggested by @xcite and applied by @xcite to a sample of _ swift_/bat blazars . a maximum likelihood technique applied to the wmap5 blazar sample allows us to derive @xmath43 and @xmath44 in the @xmath45 redshift range ( which implies a peak at @xmath46 ) , which we assume in our simulations . the case of pure luminosity evolution ( @xmath47 ) is excluded with very high significance ( @xmath48 ) . lower luminosity ( @xmath49 w / hz ) , mostly fr i radio sources are known to display a much weaker cosmological evolution , which reaches @xmath50 zero at @xmath51 w / hz ( e.g. * ? ? ? * and references therein ) . we took this into account by using the radio lfs of bl lacs and fsrqs derived from those of fr is and fr iis and based on the beaming model of @xcite , which agree well with those of recent blazar samples @xcite . we then used the fraction of beamed fr i blazars in bins of radio power to simulate the fraction of non - evolving radio sources as a function of power . this fraction is equal to 1 for @xmath52 w / hz , decreases monotonically with power , and reaches 0 for @xmath53 w / hz . 3 . * non - thermal component * to represent the non - thermal / jet component , we assume a simple homogeneous synchrotron self - compton model ( ssc , see , e.g. * ? ? ? * and references therein ) with relativistic electrons distributed as a power law at low energies and as a log - parabola at high energies @xcite . this model represents well the synchrotron part of the observed seds , which always extends at least to the optical band where the classification of blazars as fsrqs or bl lacs occurs . as for the inverse compton emission , which can be important in the soft x - ray band , we set the compton dominance so as to reproduce the observed @xmath54 in fsrqs . this is sufficient for our purposes , since , complicating the emission with additional components , like thermal emission from accretion or inverse compton on an external field of photons , would only modify somewhat the amount of observed soft x - rays in lsp blazars but would not change the composition of the samples , nor alter any of our conclusions . + the lorentz factors of the electrons radiating at the peak of the synchrotron sed component ( @xmath55 ) are distributed as shown in fig . [ fig : gammae ] . the range of @xmath55 ( @xmath56 ) is that expected for typical parameters of the ssc model as shown in fig . 36 of @xcite . the particular shape of the distribution was chosen so as to empirically reproduce the observed distributions in radio and x - ray selected samples of blazars . the skewness to lower values is similar to that of the @xmath54 ( a proxy for ) distribution adopted by @xcite to unify x - ray selected and radio - selected bl lacs . as regards the doppler factor , we assumed a mean value of 15 , which was chosen to be consistent with the mean superluminal speed @xmath57 obtained by @xcite , and with the typical lorentz factor @xmath58 derived by @xcite ( since for the angle that maximizes the apparent velocity @xmath59 ) . * accretion disk and broad emission lines * + we use the quasar spectral template of @xcite , which has been built using an homogeneous dataset of over 2,200 sdss spectra ( see fig . 4 of * ? ? ? a standard accretion disk is likely to be present only in so - called `` high excitation '' radio galaxies ( hergs ) , while it appears not to be there , or be less efficient , in low - excitation ones ( lergs ) . almost all fr is are lergs , while most fr iis are hergs , although there is a population of fr ii lergs as well . @xcite and @xcite have in fact suggested that the obscuring torus , which is required by agn unification schemes , is absent in fr is , based on the high optical core detection rate and low x - ray intrinsic absorption , respectively . this would mean no accretion disk as well , otherwise one would see broad lines in their optical spectra , while , apart from a handful of objects ( e.g. 3c 120 ) , basically no fr i displays broad optical lines . @xcite have shown that the accretion flow luminosities of fr iis are typically several orders of magnitude higher than those of fr is . it then looks like lergs , which means all fr is and some fr iis , either do not possess an accretion disk , or if the disk is present is much less efficient than in fr iis ( i.e. of the advection dominated accretion flow [ adaf ] type ) . we have then associated the presence of a standard accretion disk only with beamed fr ii sources and assumed that all those with an fr i parent ( the non - evolving sources ) have no disk ( but see section [ discussion ] for an alternative scenario ) . * equivalent width distributions * the intrinsic ( before dilution ) distributions of the ew of the broad lines ( ly@xmath29 , c iv , c iii , mg ii , h@xmath30 , h@xmath29 ) have been assumed to be those of the radio quiet qsos included in the sdss database described in section [ sec : ew ] . we assumed gaussian distributions characterized by the measured means ( @xmath60ew@xmath61@xmath62= 200 , @xmath60ew@xmath63@xmath62= 23 , @xmath60ew@xmath64@xmath62= 18 , @xmath60ew@xmath65@xmath62= 16 , @xmath60ew@xmath66@xmath62= 20 , @xmath60ew@xmath67@xmath62= 47 ) and dispersions for the various lines . the disk to jet power ratio * the disk and jet components in blazars are known to be correlated and possibly of the same order of magnitude , although there are uncertainties associated with estimating them ( see , e.g. * ? ? ? * ; * ? ? ? we are interested in the somewhat simpler question of determining how the luminosity of the accretion disk ( blue bump intensity at 5000 ) scales with radio power ( at 5 ghz ) . the relevant data were derived by using the very large amount of multi - frequency information included in public databases and the tools that are now available to analyze seds @xcite . figure [ fig : sed ] gives some examples of representative objects . the amount of thermal flux in each fsrq was estimated by matching the composite optical qso spectrum of @xcite to the sed data in the blazar rest - frame . we have done that through a careful visual inspection of each sed and by adjusting the intensity of the composite qso spectrum until it overlapped well to the observed emission . the use of this manual approach was necessary as the heterogeneity of the available data and flux variability does not allow the implementation of a robust automatic procedure . [ fig : sed ] gives examples of the matching of the composite qso spectrum ( blue line ) to the data for the case of 3c273 or 3c279 . in those objects where the available optical / uv data are limited to a magnitude in one or two colors , we matched the composite qso spectrum to the flux level corresponding to the available magnitude(s ) , taking into account of redshift . + an upper limit was instead estimated for bl lac objects by placing the composite qso spectrum in the sed at an intensity such that the optical lines would not be detectable in the optical spectrum ( typically a factor ten below the observed flux . fig . [ fig : sed ] illustrates the case of mkn501 ) . this was done for a large number of blazars selected in four surveys , two radio - selected ( deep x - ray radio blazar sample [ dxrbs ] and wmap5 ) and two x - ray selected ( emss and _ swift_-bat ) . the results are plotted in fig . [ fig : arovsrlum ] where we can see that @xmath68 ( the slope between the 5 ghz luminosity and the blue bump intensity at 5000 and defined by @xmath69 ) correlates with radio luminosity , although with a large scatter . we interpret this as the result of relativistic beaming , in the sense that the larger the radio luminosity ( and therefore the beaming amplification ) the larger the ratio between non - thermal and thermal radiation . + we adopted a simple linear relationship between the two variables ( @xmath68 = 0.04*log(l@xmath70 , see fig . [ fig : arovsrlum ] ) and assumed a gaussian distribution around it with a dispersion of 0.1 . this has been derived by fitting the wmap5 and emss points in the figure ( see below ) . we did not take into account the lower limits on @xmath68 since the large majority of them are at low radio power ( @xmath71 w / hz ) and in our scenario these sources do not have a standard accretion disk . 7 . * host galaxy * + following @xcite and @xcite we have assumed that the host galaxy of blazars is a giant elliptical with fixed absolute magnitude of @xmath72 as estimated by @xcite using the same @xmath9cdm cosmology adopted in this paper . these authors have shown that the dispersion around this value is less than 1 magnitude and therefore such a giant elliptical galaxy can be considered a standard candle . for the spectral shape we used the galaxy template of @xcite , who derived it combining the data from 28 local elliptical galaxies observed in the wavelength range @xmath73 m . + figure [ fig : sed ] shows this template superposed to the sed of four well - known blazars . our monte carlo simulations consist of the sequential execution of the following steps : 1 . draw a value for the radio luminosity and redshift of a simulated blazar based on the luminosity function and evolution described above ; 2 . draw a value of the lorentz factor of the electron radiating at the peak of the synchrotron power ( @xmath55 ) from the distribution shown in fig . [ fig : gammae ] , which is assumed to be independent of blazar luminosity ; 3 . calculate the peak of the synchrotron power in the source rest frame . in our simulations we assume a simple ssc model , and therefore @xmath74 . the value of @xmath55 is derived in step ii ) , the magnetic field is fixed to b = 0.15 gauss , and the doppler factor @xmath75 is randomly drawn from a gaussian distribution with @xmath76 and @xmath77= 2 . calculate the observed radio flux density ( from the radio luminosity and redshift from step i ) ) and the non - thermal emission in the optical and x - ray bands under the assumption that the spectral shape of the observed emission is a log parabola around and that the low energy part of the sed ( cm and mm wavelengths ) is a power law , as is typically seen in blazars ( see e.g. fig . [ fig : sed ] for some representative examples , or @xcite for a much larger sample of blazar seds ) ; 5 . accept the source if its flux in the band under consideration ( radio or x - ray ) is above the flux limit chosen for the simulated survey ; 6 . add an accretion ( blue bump ) component as described above ( only for beamed fr ii sources ) , re - scaling the sdss quasar template to this value ; 7 . draw a value of the equivalent width of ly@xmath29 , c iv , c iii , mg ii , h@xmath30 , h@xmath29 starting from the ew distribution observed in the sdss radio quiet qsos ; 8 . add the optical light of the host galaxy assuming a standard giant elliptical observed in blazars ; 9 . calculate the total optical light and the observed equivalent width of all the broad lines considered by taking into account the dilution due to the non - thermal and host galaxy optical light ; 10 . classify the source as an fsrq if the rest - frame ew of at least one of the broad lines that enter the optical band in the observer frame ( which we assume to cover the @xmath78 range ) is @xmath79 . otherwise , the object is classified as a bl lac , unless the host galaxy dominates the optical light causing the ca h&k break to be larger than 0.4 @xcite , in which case the source is classified as a radio galaxy . a bl lac whose maximum ew is @xmath80 , or the non - thermal light is at least a factor 10 larger then that of the host galaxy @xcite , is deemed to have a redshift which can not be typically measured . it is important to stress that the scope of our simulations is _ not _ to reproduce _ all _ the observational details . while that could be possible in theory , in practice it would require a large number of parameters and some speculations . our approach is instead to keep the number of assumptions to a minimum , with the aim of allowing us to obtain robust , almost model - independent conclusions . our main results are in fact independent of the simulation details ( see section [ sec : stability ] ) . we simulated a radio flux density limited survey with @xmath81 jy , to match the wmap5 sample , and an x - ray flux limited survey down to @xmath82 in the @xmath83 kev band , in order to be able to compare it with the emss . to ensure good statistics each simulation run included 10,000 sources . in the x - ray case , since radio powers reach lower values than in the radio case , we extrapolated the radio lf down to @xmath84 w / hz assuming the same slope . in this section we make a detailed comparison between the results of our simulations and the observational data in terms of fractions of bl lacs and fsrqs , redshift and distributions , and cosmological evolution using the @xmath85 test @xcite , where @xmath86 is the volume out to the source and @xmath87 is the the volume at the distance where the object would be at the flux limit . we also show where our simulated sources end up on the plot used by @xcite when they first described the `` blazar sequence '' . table [ tab : radiosim ] summarizes our main results by giving the number of sources per class , their mean redshift , and . the number in parenthesis refers to the bl lacs with measurable redshift , to which the mean redshift and pertain . about 3/4 of our sources are classified as fsrqs , with the fraction of bl lacs being @xmath88 of blazars , which is consistent with the value of @xmath89 in the wmap5 sample ( where the @xmath90 errors are based on binomial statistics : * ? ? ? * ) . a small fraction ( @xmath91 ) of the simulated blazars are classified as radio galaxies . these are bona - fide blazars misclassified because their non - thermal radiation is not strong enough to dilute the host galaxy component . the mean redshift for our simulated fsrqs is in good agreement with the wmap5 value of @xmath92 , while for bl lacs this is slightly larger than the wmap5 mean ( @xmath93 ) . fig.[fig : wmap_red ] shows the overall good agreement between our simulated redshift distributions ( where we have only included sources with a measurable redshift ) and the observed ones . @xmath94 of our bl lacs have a redshift determination , in excellent agreement with the wmap5 value of @xmath95 . @xmath96 of the bl lacs ( @xmath97 of those with redshift ) have a standard accretion disk and are therefore broad - lined but are classified as bl lacs only because their observable emission lines are swamped by the non - thermal continuum . .results from a simulation of a radio flux density limited survey ( 0.9 jy ) [ cols="<,<,^,^ " , ] [ tab : xsim ] as regards @xmath85 , our simulated mean values agree reasonably well with the emss ones of @xmath98 and @xmath99 for fsrqs and bl lacs respectively , derived using the samples described in @xcite . [ fig : x_nupeak ] compares the distributions of of fsrqs and bl lacs in our simulation with those of blazars belonging to the soft x - ray sample of @xcite , which includes _ planck _ , _ swift _ and _ fermi _ observed blazars and it is therefore probably the sample with the best determination of values currently available . the agreement is clearly good reproducing well the fact that bl lacs have much higher values than fsrqs . the existence of a strong anti - correlation between bolometric luminosity and , known as the blazar sequence " has been the subject of intense discussions since its first proposal by @xcite and @xcite ( e.g. , * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? in this section we use our simulations to comment on the existence of such a sequence . figure [ fig : sequence ] shows our radio and x - ray selected simulated blazars in the log ( ) log(@xmath100l@xmath101(5 ghz ) ) plane . this reproduces the famous plot used by @xcite to propose the existence of the blazar sequence based on the correlation shown in this plane by fsrqs and bl lacs discovered in shallow radio surveys ( 2 and 1 jy samples ) and bl lacs found in the x - ray flux limited _ einstein _ slew survey ( @xmath102 ) . indeed , considered together the simulated radio and x - ray selected blazars display a broad correlation with radio selected fsrqs and bl lacs ( red open squares and open circles ) mostly filling the top left and central part of the diagram and x - ray selected bl lacs ( green open circles ) mostly confined to the lower right corner of the plot . this particular positioning of the points ( bright fsrqs of the lsp type vs. fainter hsp bl lacs ) is not due to any intrinsic correlation between luminosity and but is instead the result of a selection effect resulting from the fact that bright radio sources are mostly drawn from the high end of the blazar luminosity function , while bl lacs in x - ray flux limited samples are mostly high sources ( intrinsically rare ) drawn from the low end of the luminosity function where the source density is largest . the most important difference between fig . [ fig : sequence ] and the diagram of @xcite is in the high - luminosity - high part , where most of the radio and x - ray selected objects with no redshift ( red and green filled points ) are located . these sources could not be plotted by @xcite since the luminosity of blazars without redshift can not be estimated . this left the top right part of the @xcite diagram empty , thus contributing to make the displayed data look like a power sequence . to assess the dependence of our results on the adopted lf and evolution , we have made the following two checks : first , we varied their input values by @xmath103 adopting @xmath104 , @xmath105 , and @xmath106 ( luminosity peak at @xmath107 ) and @xmath108 , @xmath109 , and @xmath110 ( luminosity peak at @xmath111 ) ; second , we used as an alternative lf the sum of the bl lac and fsrq lfs based on the beaming model of @xcite ( converted to @xmath10 km s@xmath11 mpc@xmath11 ) assuming , for consistency with the way they were derived , a pure luminosity evolution of the type @xmath112 $ ] , where @xmath113 is the look - back time . a value of @xmath114 , which is consistent with the evolution of dxrbs fsrqs and bl lacs combined ( based on the samples in * ? ? ? * ) , was also assumed . we have also checked the dependence of our results on the assumed value of the doppler factor , running the simulation with values of @xmath115 ranging from 5 to 20 , and we have also considered the case of a dependence of @xmath75 on radio power as suggested by some authors ( e.g. * ? ? ? * ) , although @xcite , who studied a complete sample , have not been able to confirm such a correlation . in all cases our main results , that is the prevalence of fsrqs and bl lacs in radio and x - ray selected samples respectively , the higher redshifts and evolution of fsrqs as compared to bl lacs , and the no evolution of x - ray selected bl lacs , were confirmed , which shows that they are independent of the details of the lf and evolution . we also modified our assumption that standard accretion disks are associated only with sources having an fr ii parent but the simulations we obtained were inconsistent with observations . in fact , even if the relatively small fraction of @xmath116 of objects with fr i parents are randomly assigned an accretion disk and allowed to evolve , x - ray selected bl lacs reach too high values of @xmath117 and @xmath118 , 0.58 and 0.91 respectively , which are incompatible with observations . so far , blazars have been largely discovered in the radio , @xmath119-wave , x - ray or bands . we have shown that the wide diversity of observed in blazars results in strong differences in the observed flux at radio or x - ray frequencies ( up to three orders of magnitudes ) causing severe selection effects in surveys performed in these energy bands . this , combined with the fact that blazars are always identified on the basis of their optical spectrum , which reflects a mix of non - thermal , thermal , and host galaxy emission , has been the origin of a number of enduring open issues about the nature of blazars . a long - standing blazar puzzle is the difference in redshift distribution and cosmological evolution between fsrqs and bl lacs , selected both in the radio and in the x - ray band . bl lacs are mostly located at low redshifts and exhibit moderate , or even negative , evolution , while fsrqs evolve strongly just like radio quiet qsos and show a redshift distribution that peaks at @xmath120 @xcite . our simulations reproduce quite well both of these findings ( see fig . [ fig : wmap_red ] and tables [ tab : radiosim ] and [ tab : xsim ] ) implying that they are due to heavy selection effects . most of the simulated bl lacs found in radio surveys are luminous objects with broad lines that are diluted by non - thermal radiation beyond the 5 ew limit ( many of them just below , thus allowing a measurement of their redshift ) , while the bl lacs found in simulated x - ray surveys typically show high values ( and therefore are x - ray bright ) and are drawn from the low - power end of the radio luminosity function where non - evolving fr is are preferentially found ( section [ ingredients ] ) . we note that the of radio selected bl lacs is not too different from that of fsrqs , while the of x - ray selected bl lacs is significantly lower , as found in real surveys ( section [ radio_survey ] ) . another interesting outcome of our simulations is the fact that x - ray selected bl lacs with progressively larger values of are characterized by lower and lower values of ( see tab . [ tab : xsim ] ) . this is in full agreement with the puzzling , and so far unexplained , results of @xcite and of @xcite who reported that the of bl lacs is a function of their x - ray - to - radio flux ratio . recent results , based on radio and @xmath25-ray surveys , have revealed that bl lacs , on average , display a distribution of energies , which is shifted to values higher than those of fsrqs @xcite , expanding on the well - known fact that high objects ( hsps ) are always bl lacs . this experimental difference is well reproduced in our simulations ( see fig . [ fig : radio_nupeak ] ) which give @xmath121log()@xmath62 = 12.9 for fsrqs and @xmath122log()@xmath62 = 14.1 for bl lacs for the case of a radio survey . this distinction is due to the fact that blazars with higher values produce more non - thermal optical light than low sources , diluting more easily the broad line component , and are therefore classified more frequently as bl lac objects . we note that this has been interpreted in the literature as an intrinsic physical difference between lsp ( detected mostly in the radio band ) and hsp ( detected mostly in the x - ray and band ) blazars due to the fact that hsps are observationally characterized by a low intrinsic power and external radiation field , given their very weak or absent emission lines . as a consequence , cooling was thought to be less dramatic in hsp allowing particles to reach energies high enough to produce synchrotron emission well into the x - ray band @xcite . in our scenario , instead , all sources have exactly the same chance of being hsp or lsp ( that is , the value of @xmath55 is drawn from the distribution shown in fig . [ fig : gammae ] _ independently _ of luminosity ) and the very different distributions observed in radio and x - ray surveys arise from the strong selection effect discussed in section [ sec : seq ] and the emission line dilution mentioned above . our simulations predict the existence of a significant number of bl lacs with redshift that can not be measured , which occurs when both and radio power are so large that dilution becomes extreme . for example , @xmath123 of our simulated sources with @xmath8 hz and @xmath124 w / hz have no redshift . this is consistent with the fact that most bl lacs in current selected samples have no measured redshift , as _ fermi _ is known to preferentially select high bl lacs @xcite . this effect is also clearly shown in fig . [ fig : sequence ] where most of the simulated blazars with no measurable redshift ( filled circles with light colors ) occupy the top right part of the diagram . fig . [ fig : zdistofnoredshiftsources ] shows the intrinsic redshift distribution of these featureless bl lacs for the case of our simulation of a radio flux density limited survey . @xcite had originally suggested that the absence of broad lines in bl lacs was due to a very bright , doppler - boosted synchrotron continuum . in the years following that paper observations of various bl lacs , mostly selected in the x - ray band , showed that in many cases their optical spectrum was not swamped by a non - thermal component , as host galaxy features were very visible , and it was thought that most bl lacs had intrinsically weak lines . we have shown here that these two possibilities are not mutually exclusive and indeed are both viable , depending on radio power and , therefore , on the band of selection . one important consequence of our scenario is that objects so far classified as bl lacs on the basis of their _ observed _ weak , or undetectable , emission lines belong to two physically different classes : intrinsically weak lined objects , whose parents are lergs / fr is ( more common in x - ray selected samples , since they reach lower radio powers ) and heavily diluted broad - lined sources , which are beamed hergs / fr iis ( more frequent in radio selected samples ) . therefore , while the non - thermal engine is probably the same , the thermal one is obviously different . this solves at once the issue of the fsrq / bl lac transition objects and of the many differences between bl lacs selected in the radio and x - ray bands , which include line strength , extended radio emission and morphology , and evolution ( e.g. * ? ? ? * and references therein ) . this hypothesis , which explains also many other open issues of blazar research , is directly testable . our simulations , in fact , imply that the majority of sources in high - flux density radio - selected samples are identified as bl lacs only because all lines have ew @xmath125 _ in the optical observing window_. these sources should show a strong ( ew @xmath126 ) h@xmath29 line if observed in the near or mid - infrared , since this is the strongest emission line , and would then be considered fsrqs , which reinforces our point on the very strong effect that line dilution has on blazar classification . herg / fr ii bl lacs , then , should be easily recognizable through infrared spectroscopy . these sources , as expected , are also more dominant at higher radio powers : for example , according to our radio simulation the fraction of bl lacs with standard accretion disks is only @xmath127 for @xmath128 w / hz but becomes @xmath129 above this value . this also means that fainter radio - selected samples of bl lacs should be more and more similar to x - ray selected ones , apart from their values , since higher are easier to detect in the x - rays @xcite . the implications of our hypothesis for unified schemes is quite straightforward : the parent population of bl lacs need to include both lergs / fr is and hergs / fr iis , while that of fsrqs is made up of hergs / fr iis only . this should have only a small effect , for example , on the lf fitting done by @xcite , as herg / fr ii bl lacs make up the high - power end of the radio lf while most of the number density comes from lerg / fr i bl lacs . our new scenario has strong implications on blazar classification as well . if the relevant physical distinction for radio sources is between lergs ( mostly fr is ) and hergs ( fr iis ) , for most purposes then ( lfs , evolution , etc . ) herg / fr ii bl lacs should be simply grouped with fsrqs . how does one distinguish in practice herg / fr ii bl lacs from lerg / fr i bl lacs ? this is simple in the presence of _ any _ ( even weak ) broad lines or for transition objects . in other cases ( e.g. completely featureless spectrum or in presence of absorption features ) there is no easy way to distinguish between the two subclasses although , for example , radio power and/or morphology could help . given the paucity of known lerg / fr is at relative high redshifts , x - ray selected ( and also fainter radio - selected ) bl lac samples are also useful in selecting such sources , which are very relevant also for the study of the so - called `` agn feedback '' and the role that agn radio emission plays in galaxy evolution through the so - called `` radio - mode '' accretion @xcite . it should also be clear that bl lacs can be used to study the broader issue of the relationship between lergs and hergs , including their evolution . a by - product of our simulations has also been the realization that some sources classified as radio - galaxies do _ not _ have their jets oriented at large angles with respect to the line of sight , as expected , but are instead moderately beamed blazars with their non - thermal emission swamped by the galaxy ( note that none of these objects has a standard accretion disk ) . these sources , which are all local ( @xmath130 at @xmath131 ) should be recognizable by their blazar - like seds and indeed some of them have already been identified by @xcite , @xcite and @xcite . recently @xcite have proposed a new classification scheme to divide bl lacs from fsrqs , which is based on the broad line region ( blr ) luminosity in eddington units and set at a dividing value of @xmath132 . this turns out to be also the value , which separates radiatively efficient ( i.e. , standard accretion disks ) from radiatively inefficient ( i.e. , adafs ) regimes , and therefore coincides with our herg / fr ii lerg / fr i division . therefore , @xcite are also suggesting that herg / fr ii bl lacs belong with the fsrqs . we have tackled the open issue of the relationship between the various blazar subclasses , and of the consequences of selection effects , through extensive monte carlo simulations . our approach is based on robust observational input , and on various results obtained by many authors over the past 20 years or so , which had never before been put together in a comprehensive way . we kept the number of assumptions and parameters to a minimum so as to draw robust conclusions . our starting point are two populations of high - excitation , high radio - power , evolving and low - excitation , low - power , non - evolving radio sources . to these we add : an homogeneous synchrotron self - compton component with a distribution of electron lorentz factors peaked at relatively low values , an optical / uv quasar template , the ew distributions of the main broad lines observed in radio - quiet quasars , a distribution of disk / jet ratios , and finally an elliptical host galaxy . our main results can be summarized as follows : 1 . we explain the main properties of all blazar subclasses , namely fsrqs and bl lacs selected in the radio and x - ray band . our simulations reproduce well the prevalence of fsrqs and bl lacs in radio and x - ray selected samples respectively , the higher redshifts , evolution and lower of fsrqs as compared to bl lacs , the non evolution of x - ray selected bl lacs , and the main differences between radio and x - ray - selected bl lacs in a manner that does not dependent significantly on the adopted lf and evolutionary details ; 2 . objects classified until now as bl lacs on the basis of their _ observed _ weak , or undetectable , emission lines actually belong to two physically different classes : most bl lacs in high flux density , radio - selected samples are actually beamed radio quasars with their emission lines heavily diluted by the non - thermal continuum and for all purposes should then be grouped with fsrqs . infrared spectroscopy should reveal in many such sources the presence of strong ( ew @xmath126 ) h@xmath29 ; most bl lacs selected in the x - ray band are instead intrinsically weak - lined , low - excitation radio galaxies with a strong non - thermal , jet component ; 3 . there are only two main intrinsic blazar types : low - ionization ( mostly beamed fr is ) and high - ionization ( beamed fr iis ) ones . all other classifications and classes proposed so far are not physically relevant and are simply due to severe selection effects and different relative strengths of the non - thermal , thermal , and galaxy components , which make up blazars optical emission ; 4 . some sources classified as radio - galaxies on the basis of their optical properties are instead moderately beamed blazars ( that is , sources with their jets forming a small angle with respect to the line of sight ) with their non - thermal emission swamped by the host galaxy . some of these sources have already been recognized from their seds ; 5 . our simulations show that the purported correlation between luminosity and , the so - called `` blazar sequence '' , is likely a selection effect resulting from comparing shallow radio surveys with shallow x - ray surveys . we also show that blazars with featureless optical spectra , and therefore without a redshift determination , are mostly high luminosity high sources , a type of blazar that should not exist in the `` blazar sequence '' scenario . this is consistent with the lack of redshift in the majority of bl lacs in current selected samples , as _ fermi _ is known to preferentially select high bl lacs ; 6 . the topics addressed in this paper are not relevant only to blazars but are also related to the broader issues of low - ionization radio galaxies and radio - mode `` agn feedback '' . in this paper we limited our simulations to the radio and the x - ray bands where ssc is a fair approximation of the observed non - thermal emission . the properties of detected blazars are instead not consistent with simple ssc models ( e.g. * ? ? ? * ) , and almost half of the radio and x - ray selected lsp blazars are quiet ( e.g. * ? ? ? the present approach must therefore be integrated with additional information about the properties of the inverse compton emission before it can be used to simulate surveys . we are planning to extend our simulations to the band by taking into account the recent results of @xcite who determined the properties of blazar samples selected in different bands . we thank a. pollock for careful reading of the manuscript and useful suggestions . st would like to thank mark subbarao and rich plotkin for useful discussions on the sdss spectroscopic pipeline . we also thank matteo perri for providing part of the software used for our monte carlo simulations . we acknowledge the use of data and software facilities from the asi science data center ( asdc ) , managed by the italian space agency ( asi ) . part of this work is based on archival data and on bibliographic information obtained from the nasa / ipac extragalactic database ( ned ) and from the astrophysics data system ( ads ) . funding for the sdss and sdss - ii was 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 was managed by the astrophysical research consortium for the participating institutions . abdo a. a. , et al . , 2010a , apj , 715 , 429 abdo a. a. , et al . , 2010b , apjs , 188 , 405 abdo a. a. , et al . , 2010c , apj , 716 , 30 abdo a. a. , et al . , 2011 , submitted to apj , ( arxiv:1108.1435 ) ackermann m. , et al . , 2011 , submitted to apj , ( arxiv:1108.1420 ) ajello m. , et al . , 2009 , apj , 699 , 603 antn s. , browne i. w. a. , 2005 , mnras , 356 , 225 baldwin j. a. , 1977 , apj , 214 , 679 blandford r. d. , rees m. j. , 1978 , in pittsburg conference on bl lac objects , ed . a. m. wolfe , pittsburgh , university of pittsburgh press , p. 328 beckmann v. , engels d. , bade n. , wucknitz o , 2003 , a&a , 401 , 927 capetti a. , raiteri c. m. , buttiglione s. , 2010 , a&a , 516 , a59 caccianiga a. , march m. j. , mnras , 348 , 937 chiaberge m. , capetti a. , celotti a. , 1999 , a&a , 349 , 77 croton d. j. , et al . , 2006 , mnras , 365 , 11 dennett - thorpe j. , march m. j. , 2000 , a&a , 361 , 480 delia v. , padovani p. , landt h. , 2003 , mnras , 339 , 1081 donato d. , sambruna r. m. , gliozzi m. , 2004 , apj , 617 , 915 evans d. a. , worrall d. m. , hardcastle m. j. , kraft r. p. , birkinshaw m. , 2006 , apj , 642 , 96 fossati g. , maraschi l. , celotti a. , comastri a. , ghisellini g. , 1998 mnras , 299 , 433 francis p. j. , hooper e. j. , impey c. d. , 1993 , aj , 106 , 417 gehrels n. , 1986 , apj , 303 , 336 gendre m. a. , best p. n. , wall j. v. , 2010 , mnras , 450 , 1719 ghisellini g. , celotti a. , fossati g. , maraschi l. , comastri a. , 1998 , mnras , 301 , 451 ghisellin g. , & tavecchio f. , 2008 , mnras , 387 , 1669 ghisellini g. , tavecchio f. , foschini l. , ghirlanda g. , 2011 , mnras , 414 , 2674 giommi p. , menna m.t . , padovani , p. , 1999 , mnras , 310 , 465 giommi p. , piranomonte , s. , perri , m. , padovani , p. , 2005 , aa&a , 434 , 385 giommi p. , perri m. , piranomonte s. , padovani p. , 2002 , blazar astrophysics with bepposax and other observatories , eds . p. giommi , e. massaro , g. palumbo , esa - esrin , 2002 , p. 123 giommi p. , capalbi m. , cavazzuti e. , et al . , 2007 , a&a , 468 , 571 giommi p. , colafrancesco s. , padovani p. , gasparrini d. , cavazzuti e. , cutini s. , 2009 , a&a , 508 , 107 giommi p. , et al . , 2011 , a&a , submitted , arxiv:1108.1114 hovatta t. , valtaoja e. , tornikoski m. , lhteenmki a. , 2009 , a&a , 494 , 527 kimball a. e. , ivezi . , wiita p. j. , schneider d. p. , 2011 , aj , 141 , 182 kollgaard r. i. , wardle j. f. c. , roberts d. h. , gabuzda d. c. , 1992 , aj , 104 , 1687 komatsu e. , et al . , 2011 , apjs , 192 , 18 landt h. , padovani p. , giommi p. , 2002 , mnras , 336 , 945 lister m. l. , et al . , 2009 , aj , 138 , 1874 mannucci f. , basile f. , poggianti b. m. , cimatti a. , daddi e. , pozzetti l. , vanzi l. , 2001 , mnras , 326 , 745 march m. j. m. , browne i. w. a. , impey c. d. , smith p. s. , 1996 , mnras , 281 , 425 massaro e. , perri , m. , giommi , p. , nesci , r. , 2004 , a&a , 413 , 489 massaro e. , tramacere , a. , perri , m. , giommi , p. , tosti , g .. , 2006 , a&a , 448 , 861 massaro e. , giommi p. , leto c. , marchegiani p. , maselli a. , perri m. , piranomonte s. , sclavi s. , 2009 , a&a , 495 , 691 massaro e. , giommi p. , leto c. , marchegiani p. , maselli a. , perri m. , piranomonte s. , 2011 , in preparation nieppola e. , tornikoski m. , & valatoja e. , 2006 , a&a , 445 , 441 nieppola e. , valatoja e. , tornikoski m. , hovatta t. , kotiranta m. , 2008 , a&a , 488 , 867 padovani , perlman e. s. , landt h. , giommi p. , perri m. , 2003 , apj , 588 , 128 padovani p. , giommi p. , 1995 , apj , 444 , 567 padovani p. , giommi p. , landt h. , perlman e. s. , 2007 , apj , 662 , 198 padovani p. , miller n. , kellermann k. i. , mainieri v. , rosati p. , tozzi p. , 2011 , apj , in press padovani p. , perlman e. s. , landt h. , giommi p. , perri m. , 2003 , apj , 588 , 128 padovani p. , 2007 , ap&ss , 309 , 63 pian e. , et al . , 1999 , apj , 521 , 112 piranomonte s. , perri m. , giommi p. , landt h. , padovani p. , 2007 , a&a , 470 , 787 planck collaboration , 2011 , a&a , in press ( arxiv:1101.2041 ) plotkin r. m. , et al . , 2010 , aj , 139 , 390 rector t. a. , stocke j. t. , perlman e. s. , morris s. l. , gioia i. m. , 2000 , aj , 120 , 1626 rector t. a. , stocke j. t. , 2001 , aj , 122 , 565 sbarufatti b. , treves a. , falomo r. , 2005 , apj , 635 , 173 scarpa r. , falomo r. , 1997 , a&a , 325 , 109 scarpa r. , urry c. m. , falomo r. , pesce j. e. , treves a. , 2000 , apj , 532 , 740 schmidt m. , 1968 , apj , 151 , 393 shaw m. s. , romani r. w. , healey s. e. , cotter g. , michelson p. f. , readhead a. c. s. , 2009 , apj , 704 , 477 shaw m. s. , romani r. w. , cotter g. , readhead a. c. s. , healey s. e. , michelson p. f. , fermi lat collaboration , 2010 , aas , 42 , # 434.12 stickel m. , fried j. w. , khr h. , 1993 , a&as , 98 , 393 stickel m. , padovani p. , urry c. m. , fried j. w. , khr h. , 1991 , apj , 374 , 431 stocke j. t. , liebert j. , stockman h. , danziger j. , lub j. , maccacaro t. , griffiths r. , giommi p. , 1982 , mnras , 200 , 27 stocke j. t. , morris s. l. , gioia i. m. , maccacaro t. , schild r. , wolter a. , fleming t. a. , henry j. p. , 1991 , apjs , 76 , 813 stratta g. , capalbi m. , giommi p. , primavera r. , cutini s. , gasparrini d. , 2011 , arxiv:1103.0749 tramacere a. , giommi p. , perri m. , verrecchia f. , tosti g. , 2009 , a&a , 501 , 879 turriziani s. , 2010 , ph.d . thesis turriziani s. , 2011 , aas , 43 , # 125.03 urry c. m. , padovani p. , 1995 , pasp , 107 , 803 urry c. m. , scarpa r. , odowd m. , falomo r. , pesce j. e. , treves a. , 2000 , apj , 532 , 816 vanden berk , d. e. , et al . , 2001 , aj , 122 , 549 vermeulen r. c. , taylor , g. b. , readhead a. c. s. , browne , i. w. a. , aj 111 , 1013 wall j. v. , jackson c. a. , shaver p. a. , hook i. m. , kellermann k. i. , 2005 , a&a , 434 , 133 wall j. v. , pope a. , scott d. , 2008 , mnras , 383 , 435 wright e. l. , et al . , 2009 , apjs , 180 , 283

the control landscape for various canonical quantum control problems is considered . for the class of pure - state transfer problems , analysis of the fidelity as a functional over the unitary group reveals no suboptimal attractive critical points ( traps ) . for the actual optimization problem over controls in , however , there are critical points for which the fidelity can assume any value in , critical points for which the second order analysis is inconclusive , and traps . for the class of unitary operator optimization problems analysis of the fidelity over the unitary group shows that while there are no traps over , traps already emerge when the domain is restricted to the special unitary group . the traps on the group can be eliminated by modifying the performance index , corresponding to optimization over the projective unitary group . however , again , the set of critical points for the actual optimization problem for controls in is larger and includes traps , some of which remain traps even when the target time is allowed to vary . ,

1004.3492

although quantum theory has been in existence for about a century , until recently , the main emphasis in the field was on constructing hamiltonian models and solving the schrodinger equation . although it was recognized that external fields and potentials could change the hamiltonian and thus the dynamics of a system , and such external fields were certainly used in many areas from nuclear magnetic resonance to atomic physics to effect changes to the system , it was only rather recently that the full potential of using such external fields was recognized . since then the subject of control of quantum systems has developed from a niche area into a subject of rapidly growing interest and importance , with an ever increasing number of applications ranging from quantum chemistry to quantum information processing ( see e.g. @xcite ) . quantum control has been applied , for instance , to influence the outcome of chemical reactions @xcite , to prepare entangled states @xcite , which are a resource in quantum metrology and information processing , and to realize quantum gates @xcite , which are fundamental building blocks for a quantum computer . the scope of the applications is vast . as the potential and importance of quantum control was realized , attempts were made to formulate quantum control problems abstractly , and establish solid mathematical foundations . some pioneering attempts were necessarily imperfect and the theory is still under development , but there have been a number of interesting theoretical results . one important area is the control landscape @xcite . while it is not too difficult in many cases to formulate a particular task in terms of a control optimization problem , i.e. , of finding an admissible control that maximizes a performance index , solving the resulting control optimization problems is challenging , especially for large or complicated systems , and in almost all cases they can only be solved computationally using numerical optimization strategies . since the evaluation of the performance index for quantum control problems generally requires the solution of an operator or partial differential equation , evaluation of the latter becomes computationally demanding and efficient algorithms are therefore crucial . such algorithms exist for finding local extrema , especially when gradients can be computed relatively cheaply compared to the computational overhead involved in evaluating the performance index , as can be achieved in our case . we are however primarily interested in global optima of that function , and this is where the `` landscape '' defined by the objective function becomes relevant . if the landscape is littered with many suboptimal , locally attractive critical points then optimization strategies designed to find local extrema have a very low probability of finding a global optimum from a generic starting point . in such a situation global search strategies are usually required , involving substantially greater computational overheads . if the landscape is nice , however , i.e. , if most local extrema are actually global extrema , then algorithms that find local extrema are likely to succeed in finding a global optimum and the computational overhead of global search algorithms can be avoided . the landscape for quantum control problems was initially expected to be complicated , but it was quickly realized that optimization algorithms designed to find local extrema were often very successful at finding controls for which the performance index to be optimized took values very close to its _ global _ optimum . this has led a number of researchers to study the issue and put forward certain arguments why this should be the case . the core argument was that , contrary to expectations , the landscape for quantum control was actually extremely simple and devoid of any traps , i.e. , suboptimal locally attractive critical points . the main objective of this paper is to revisit this topic , taking a closer look at the control landscape . previous studies @xcite have almost all yielded results that only apply to regular critical points , in particular showing that these can _ not _ be traps for typical quantum control problems . however , it is well - known in the broader field of optimal control that these are not the only critical points @xcite , and that the role of such non - regular critical points requires analysis @xcite . for quantum control problems the existence of non - regular critical points has only recently been acknowledged @xcite but their properties have not been studied . in particular , it is not known if such critical points can be traps . the existence of such traps would complicate the control landscape and invalidate some general results arrived at by considering only regular points . the main objective of this paper is to show that non - regular critical points do exist for a large class of quantum control systems and that at least some of these are _ suboptimal extrema _ for which the performance index can take many values . the particular examples we construct to rigorously prove this claim are quite artificial , but they prove that such bad points _ do _ exist . for problems of practical interest it is generally extremely difficult to rigorously prove that a critical point is a trap as these problems can only be solved numerically . however , we present numerical evidence that suggests trapping behavior does occur for practical quantum optimal control problems even if such behavior appears to be relatively rare , at least for the simple systems that have been studied extensively to date . this suggests that further analysis is required to fully characterize the control landscape , including the nature of potential traps and their practical relevance , e.g. , for the design of efficient algorithms . the paper is organized as follows . in sec . ii , we formally define the problem and give the mathematical prerequisites . in section iii , we take a closer look at the landscape for pure - state transfer problems . in sec . iv we consider the landscape for unitary operator optimization problems . finally , in sec . v , we present the results of a numerical investigation of the control landscape for a three - qubit gate optimization problem , before concluding with a brief discussion in sec . the focus of this paper is quantum systems subject to hamiltonian evolution governed by the schrodinger equation . although the latter generally applies to states evolving over an infinite - dimensional hilbert space , we restrict our attention to finite - dimensional systems . this is the usual context for work on control of quantum systems and motivated by the fact that in practice one is usually either dealing with a system that intrinsically has a finite number of degrees of freedom such as a collection of spins , or for which the hilbert space can be faithfully truncated , and the controlled dynamics is restricted to finite - dimensional subspace of interest . assuming this hilbert space has dimension @xmath3 , the evolution of the system can be described by a unitary operator @xmath4 satisfying the schrodinger equation @xmath5 where @xmath6 is the identity matrix in @xmath2 and @xmath7 is the hamiltonian of the system , which in our case will be control - dependent , as will be indicated by the subscript @xmath8 . for simplicity the control dependence will be assumed to be linear in the ( real - valued ) controls @xmath9 , @xmath10 where @xmath11 is the system s intrinsic hamiltonian and @xmath12 are control interaction hamiltonians . in the following we set @xmath13 unless otherwise stated , choose units such that @xmath14 , and define the two - point propagator @xmath15 . most typical control problems for ( hamiltonian ) quantum systems fall in one of the following three categories : pure - state transfer , observable control and unitary operator implementation . the pure - state transfer problem consists of preparing the system in a desired state @xmath16 , usually at a given time @xmath17 , assuming it is initialized to some state @xmath18 at time @xmath19 . a natural measure of success in achieving this goal is the transfer probability @xmath20 where in our case @xmath21 and @xmath22 is a solution of ( [ eq : seu ] ) . for a given hamiltonian , initial and target states , and fixed target time @xmath17 , the transfer probability depends only on the choice of control @xmath8 , i.e. , if the controls vary over @xmath0 then @xmath23 . the observable control problem lies in taking a system characterized by an initial density operator @xmath24 at time @xmath19 to a state @xmath25 at time @xmath17 for which the selected observable @xmath26 attains its maximal possible value , i.e. , we aim to maximize the expectation value @xmath27,\ ] ] where @xmath28 and @xmath22 is again a solution of ( [ eq : seu ] ) . the gate synthesis control problem consists in choreographing the system dynamics to implement the unitary gate @xmath29 over the time interval @xmath30 $ ] . using the hilbert - schmidt ( frobenius ) norm to measure distance from the target gate @xmath29 , this leads to the gate fidelity @xmath31\ ] ] as a natural performance index , where @xmath22 is a solution of ( [ eq : seu ] ) as before . broadly speaking , the objective of quantum control is to find a control @xmath8 to maximize one of these performance indices , where @xmath8 is allowed to vary over a large function space , while @xmath17 is fixed unless stated otherwise . in practical implementations of optimization algorithms for these problems , @xmath8 will always be restricted to a finite - dimensional subspace , but this subspace may be arbitrarily large , and we will be chiefly interested in the ideal case where @xmath8 varies over @xmath0 in this paper , except in sec . so in general , we wish to find the ( global ) maximum of a ( non - linear ) functional @xmath32 , which in our case is either @xmath33 , @xmath34 , or @xmath35 . it will often be fruitful to think of @xmath36 as a composition of the solution functional @xmath37 taking a control field @xmath8 to the corresponding propagator at time @xmath17 , @xmath38 , via ( [ eq : seu ] ) , and the fidelity on the lie group @xmath39 , and we shall use @xmath40 , \quad \g_v(u)= \frac{1}{n}\re\tr[v^\dag u],\ ] ] respectively , to denote the equivalents of @xmath33 , @xmath34 and @xmath35 above . whether we allow the fields @xmath41 to range over the whole of @xmath0 or a smaller space such as the continuous or piecewise constant functions , we must consider whether the aforementioned control problems are feasible , i.e. , whether the maximum of the performance index is actually attainable , or the question of controllability . more specifically , we are interested in whether for a given system @xmath7 , and all endpoint conditions either @xmath42 , @xmath43 or @xmath44 , there exists a time @xmath17 and an ( admissible ) control @xmath45 that solves the corresponding control problem . in general , for bilinear control systems on a lie group , as in our case , this problem reduces to one about the lie algebra @xmath46 generated by the matrices @xmath47 . for pure - state control problems controllability is equivalent to @xmath46 being a representation of either @xmath48 , @xmath49 , @xmath50 or @xmath51 . for mixed - state ( density - operator ) control problems , we require @xmath50 or @xmath51 , while the system is fully unitary operator controllable only when @xmath52 @xcite . note that the distinction between @xmath50 and @xmath51 is artificial in the quantum case as the extra dimension contributes only a global phase , which is generally not physically observable . since it is convenient , and indeed common , to fix the final time @xmath17 when carrying out control optimization , we are also interested in the stronger notion of exact - time controllability , namely , whether a system has some critical time @xmath53 such that all endpoint conditions give solvable control problems for any @xmath54 . it turns out that , neglecting global phase , these notions of controllability are equivalent to the previous three by theorem 13 of @xcite . apart from these now standard results , we will need the following . let @xmath55 be a control system , @xmath52 or @xmath50 and @xmath56 or @xmath57 be associated lie group . * if the set of all commutator expressions in @xmath58 and @xmath59 joined with @xmath58 and @xmath59 span @xmath46 ( lie algebra rank condition ) , then there exists a maximal time @xmath60 and neighborhood @xmath61 of @xmath62 in @xmath63 such that for all systems @xmath64 and elements @xmath65 , there is a control taking @xmath66 to @xmath67 in time @xmath68 . * if the set of all commutator expressions in @xmath58 and @xmath59 joined with @xmath69 span @xmath46 ( exact - time lie algebra rank condition . ] ) , then there is a critical time @xmath53 and neighborhood @xmath61 of @xmath70 such that for all @xmath64 , @xmath65 and @xmath54 , there is a control taking @xmath66 to @xmath67 in time @xmath17 . assuming that our system specified by @xmath62 satisfies the lie algebra rank condition , by theorem 1 of @xcite there is a set of values @xmath71 such that the function @xmath72 has jacobian @xmath73 of full rank at some point @xmath74 . this implies that the extended function @xmath75 also has jacobian @xmath76 of full rank at the point @xmath77 . thus by the inverse function theorem , choosing neighborhoods @xmath61 of @xmath62 and @xmath29 of @xmath74 bounded , there is a neighborhood @xmath78 of @xmath79 such that @xmath80 includes @xmath81 for every @xmath82 . by compactness ( and connectedness ) , some power @xmath83 covers the entire group @xmath84 . letting @xmath17 be the supremum of @xmath85 over @xmath29 , we see that for all systems in @xmath61 the entire group is reachable in less than time @xmath86 . assuming @xmath62 satisfies the exact - time lie algebra rank condition , the same argument may be repeated with @xmath87 as constructed in the proof of theorem 3 in @xcite , where now @xmath88 has domain restricted to @xmath89 . this alteration shows that the critical time @xmath53 of exact - time controllability can indeed be chosen uniformly about @xmath62 , as claimed . finding the global optimum of a function is generally a very difficult task . on the other hand , there are many efficient algorithms to find local extrema , i.e. , attractive critical points of a function . for this reason we are particularly interested in the nature of the critical points of the various functionals @xmath36 , which depends only on certain local properties of the solution functional @xmath90 . to study the latter , we need the identity @xmath91u_{f+{\delta}f}(\tau)\,\dd\tau\ ] ] which can be verified by differentiating both sides , and used to derive @xmath92 where @xmath93 is just @xmath94 in the @xmath13 case . for @xmath95 we define the linear map @xmath96 corresponding to the gradient of @xmath22 at @xmath8 , and the bilinear map @xmath97\dd\sigma\dd\tau\ ] ] corresponding to its hessian . indeed , the second order perturbative expansion ( [ eq : pert ] ) is the functional analog of the taylor expansion of a function , with the gradient and hessian functionals corresponding to the first and second order terms , respectively . since the hessian is naturally a linear operator mapping @xmath0 to itself , it is convenient to have notation for referring directly to some such operators , along with the elements of @xmath98 themselves . we shall use @xmath99 to denote the unnormalized projection onto its argument , and @xmath100 as a placeholder for the argument of the anonymous function it is a constituent of . for example , @xmath101 $ ] refers to the projector onto the function @xmath102 , that is , the operator that maps each vector @xmath103 to the vector @xmath104 in @xmath0 . the importance of the gradient and the hessian functionals is that the former determines the critical points of the functional @xmath36 , and the latter their nature . in particular , if the hessian at a critical point @xmath8 is negative definite , then the critical point is attractive and corresponds to a local maximum over any finite - dimensional subspace of @xmath0 ; similarly , if it is positive definite , the critical point is repulsive and corresponds to a local minimum over any finite - dimensional subspace , and if the hessian at the critical point can take both positive and negative values , the critical point is a saddle . as mentioned in the introduction the problem of finding a control @xmath105 to maximize the pure - state fidelity functional ( [ eq : fp ] ) can be viewed a composition of finding a @xmath106 or @xmath57 that maximizes @xmath33 , and an optimization problem of finding a control in @xmath0 that realizes @xmath107 . defining @xmath108 and @xmath109 , we have @xmath110 $ ] , i.e. , we can rewrite the pure - state optimization problem as a general observable optimization problem . it is easy to show that a necessary and sufficient condition for @xmath111 or @xmath57 to be a critical point of @xmath112 $ ] is that @xmath26 and @xmath113 commute , i.e. , @xmath114=0 $ ] . in our case , as @xmath26 and @xmath24 ( and thus @xmath113 ) are projectors onto pure states , there are only two types of critical points : ( i ) @xmath26 and @xmath113 are projectors onto orthogonal subspaces of @xmath115 , in which case we have @xmath116=0 $ ] , and ( ii ) @xmath26 and @xmath113 are projectors onto the same 1d subspace of @xmath115 , in which case we have @xmath116=1 $ ] . thus , the landscape for pure - state optimization over @xmath2 or @xmath57 is very simple : there are only two types of critical points , corresponding to extremal values of the fidelity , i.e. , global extrema , and no saddles or other critical points for which the fidelity assumes values in @xmath1 . the critical points over @xmath57 or @xmath2 are sometimes referred to as _ kinematic _ critical points @xcite . since then several papers have attempted to show that this result extends to the actual optimization problem over @xmath0 , i.e. , that the fidelity @xmath33 as a functional over @xmath0 only has critical points for which it achieves either its global minimum @xmath19 or maximum @xmath117 . however , the arguments put forward in @xcite and @xcite rely , without rigourous justification , on the property that the solution functional @xmath118 is of full rank _ everywhere_. @xcite gives another proof but relies on the same problematic assumption . @xcite argues that a certain sequence of expressions starting with equation ( 14 ) and ( 15 ) therein generate the full lie algebra generated by @xmath119 and @xmath69 , and hence , assuming controllability , span the entire lie algebra @xmath50 . but this procedure only allows one to generate specific linear combinations of commutators , and , although it may seem plausible , these are not guaranteed to span @xmath50 . for the more general density matrix control problem , @xcite similarly assumes that the map @xmath118 is of full rank everywhere to derive that the final state @xmath120 must commute with the observable @xmath26 at critical points , and hence that @xmath19 and @xmath117 are the only critical values for pure states . @xcite makes use of the same problematic assumption , referring to @xcite as evidence of its validity , to re - derive this critical point characterization , along with the result that all but the global maxima and minima are saddle points . @xcite recognises the possibility of singular controls @xmath8 , here of @xmath121 not being full rank for a particular control @xmath8 , along with the issues this raises , yet defers analysis of such points to future work . @xcite finally gives a characterization of singular controls and acknowledges that these can be critical points of the performance index that do not correspond to kinematical critical points , but considers only one example involving eigenstates of a four - level system , to conclude based on numerical simulations that singular controls do not appear to be traps for this problem . therefore , the question of the existence of non - kinematic critical points for the pure - state fidelity ( [ eq : fp ] ) , i.e. , critical points @xmath122 for which @xmath33 does not assume extremal values , remains open . we now show that such critical points do exist for all control - linear hamiltonian systems , and while most of them can not be expected to be attractive , one can construct examples where calculation of the hessian at such a non - kinematic critical point shows it to be strictly negative definite . thus , not only do non - kinematic critical points exist but they can be attractive , i.e. , traps . hessian calculations also show that the hessian need not have finite rank at such critical points . the main implication of these examples is that landscape analysis based on considering regular points only is incomplete , and that in some cases at least , traps definitively do exist . they may be generally rare but even a single trapping point can be problematic if it has a large domain of attraction . it should be noted here that the mere existence of singular controls implies that the end - point map @xmath121 can not be full rank everywhere , and thus that the hypothesis on which most of the landscape results cited above are predicated , can not be satisfied in general , so that further analysis is necessary . but this still leaves open the question whether the results are true in the absence of this hypothesis , and the significance of exhibiting examples for which traps provably exist is to show that the landscape is not universally trap - free . let us start by considering , for any choice of final time @xmath17 , a pure state control problem for a two - level system of the general form : [ genpure ] @xmath123 e^{i t h_0 } & , \quad \ket{\psi_0 } = \frac{1}{\sqrt{2 } } [ e^{i \theta } , e^{- i \left ( \theta + \varphi \right)}]^{t } , \end{aligned}\ ] ] where we can exclude the degenerate case @xmath124 by choosing @xmath125 . for the field @xmath126 which is identically zero over @xmath30 $ ] , we can calculate @xmath127 , so that the fidelity @xmath33 is @xmath128 . moreover , @xmath126 is a critical point since the general gradient formula @xmath129 in this case evaluates to @xmath130 a value of @xmath131 can be chosen to achieve any fidelity in the interval @xmath1 , and this can be done avoiding the degenerate case by choosing the right sign for @xmath131 when necessary . so we already see there is a large family of control problems admitting critical points for which the fidelity assumes any possible value but this is simply a special case of the more general fact that follows . [ thm:1 ] given any bilinear control system with control hamiltonian @xmath132 , any target time @xmath133 , and any @xmath134 , there exist pairs of initial and target states @xmath135 and a control @xmath8 such that @xmath8 is a critical point of the fidelity @xmath33 with critical value @xmath136 . any bilinear control system @xmath137 with @xmath138 is equivalent to one with hamiltonians @xmath139 and control @xmath140 . case 1 : suppose there is a value of @xmath141 such that @xmath142 has degenerate eigenvalues . choose a two - dimensional subspace @xmath143 of this eigenspace . then , restricted to the subspace @xmath143 the hamiltonians take the form @xmath144 it easy to see that @xmath145 is a critical point with fidelity @xmath146 for any @xmath17 for the state transfer problem @xmath147 on @xmath143 with @xmath148 , \quad \ket{\psi_0 } = [ \cos ( \varphi ) , \sin ( \varphi ) e^{i \gamma } ] ^{t}\ ] ] if the hamiltonians restricted to @xmath143 take the form ( [ unconpure ] ) . since @xmath143 is an invariant space of @xmath149 , if the initial and final states are in @xmath143 , the formulas for the fidelity and gradient simplify to their analogues restricted to @xmath143 . therefore , @xmath150 is a critical point for the state transfer problem @xmath147 for the original system if we embed the initial and target state into the full hilbert space @xmath115 in the obvious manner . case 2 : if the eigenvalues of @xmath151 are always distinct then we can continuously parametrize the ( unique ) eigenvectors @xmath152 and @xmath153 corresponding to the maximal and minimal eigenvalues of @xmath154 , respectively . letting @xmath155 , @xmath156 be the maximal and minimal eigenvalues of @xmath157 , we have the identities @xmath158 hence by continuity we must have @xmath159 for some ( finite ) @xmath160 , unless @xmath161 , which would mean that @xmath157 is a multiple of the identity and so any value of @xmath160 would do . using this @xmath160 leads to an effective system @xmath162 which , when restricted to the subspace @xmath143 spanned by @xmath163 , and expressed in this basis , is precisely of the form ( [ genpure ] ) . again we can extend the initial and target states in ( [ genpure ] ) to the full hilbert space in a trivial way so that @xmath145 is a critical point with fidelity @xmath164 for the resulting state transfer problem with the hamiltonian @xmath165 . restriction to @xmath143 is again a well - behaved procedure since the span of each eigenvector is invariant under the time evolution generated by the corresponding hamiltonian @xmath142 . this theorem shows that for any control system , in particular any controllable one , there exist pairs of initial and target states for which we can achieve critical values of the fidelity in @xmath1 . however , the initial and target states which exhibit such critical points in the proof above will not in general be eigenstates of @xmath11 . one might wonder whether critical points for which the fidelity does not achieve extremal values might be impossible if the initial and target states are restricted to eigenstates of the system hamiltonian @xmath11 , as is the case in many applications , but even in this case we can find counterexamples . * non - extremal critical points for eigenstate transfer problem . * consider the system @xmath166 with @xmath167 and the initial and target states @xmath168^\t$ ] , @xmath169^\t$ ] . @xmath33 for this system has a critical point at @xmath170 as @xmath171\equiv0 $ ] with @xmath172 , where @xmath17 is the target time . indeed , we can even verify that this system is controllable as @xmath11 has distinct transition frequencies and @xmath157 is connected . we can furthermore show that the critical points above , for which the fidelity does not assume extremal values , are not necessarily saddle points either . we show this for pure state examples but the problems can be reformulated as observable optimization problems with @xmath108 for a density matrix that is a projector onto the pure state @xmath173 . hence the examples also show the existence of non - kinematic critical points for which the observable fidelity @xmath174 $ ] assumes non - extremal values and which are not saddle points . * critical saddle points with negative semi - definite hessian . * let @xmath175 with @xmath176 given any target time @xmath17 , choose the initial and target states @xmath177^\t , \quad \ket{\psi_g}=[e^{i2t}\cos\theta,0,0,e^{i9t}\sin\theta]^\t.\end{aligned}\ ] ] then @xmath178 is a critical point of the fidelity , since @xmath179 is diagonal so that @xmath180 , with critical value @xmath181 . the hessian of the fidelity at this point can be computed as @xmath182 \left [ \alpha ( \tau ) \beta ( \sigma ) + \beta ( \tau ) \alpha ( \sigma ) \right ] \dd \sigma \dd \tau\end{gathered}\ ] ] which just constitutes two instances of the @xmath183 operator from [ appendix ] and so can more clearly be written as @xmath184+\pi[\sin(2\bullet)]\big)\quad\\ + \sin\theta\sin\phi \big(\pi[\cos(4\bullet)]+\pi[\sin(4\bullet)]\big)\big].\end{aligned}\ ] ] this expression is clearly negative semi - definite whenever @xmath185 and @xmath186 are both positive , equivalently , when @xmath131 and @xmath187 lie in the interior of the same quadrant . in this circumstance we are exhibiting a point @xmath8 which to second order is not a saddle . for example , for @xmath188 and @xmath189 we obtain @xmath190 and hessian taking the form @xmath191+\pi[\sin(2\bullet ) ] + \pi[\cos(4\bullet ) ] + \pi[\sin(4\bullet)]\big]$ ] . + by extending ( [ eq : pert ] ) , we can compute the third order derivative of @xmath33 in this case to be @xmath192 \delta f ( \rho ) \delta f ( \sigma ) \delta f ( \tau ) \dd \rho \dd \sigma \dd \tau\end{gathered}\ ] ] in direction @xmath193 . when @xmath194 , the first and second order derivatives of @xmath33 in direction @xmath195 vanish , with the third order one being @xmath196 . the condition already imposed on @xmath131 and @xmath187 implies that @xmath197 , hence , unless @xmath198 thus @xmath199 , one of @xmath200 is a direction of increase , showing that @xmath8 is nonetheless a saddle point . this example shows that we can not rule out the existence of traps by just looking at the gradient and hessian . in fact , we can construct examples for which we have attractive critical points with fidelity in @xmath1 , i.e. , traps . * non - extremal critical points with negative definite hessian ( traps ) . * consider the system @xmath175 with @xmath201 taking the target time to be @xmath202 , choose the initial and target states @xmath203^\t , \quad \ket{\psi_g } = \frac{1}{\sqrt{2}}[e^{-it(1+\eps)},0,0,e^{-i2t}]^\t.\end{aligned}\ ] ] for @xmath178 it is easy to verify that the gradient vanishes identically as before , so this is a critical point , with fidelity @xmath204 . the hessian is @xmath205- \!\!\!\!\!\!\underset{0<\sigma<\tau<\frac{\pi}{\eps}}{\int\!\!\int}\!\!\!\!\!\ ! \cos\theta [ \cos\theta \cos(\eps(\tau-\sigma ) ) -\sin\theta\sin(\eps(\tau-\sigma ) ) ] [ \alpha(\sigma)\beta(\tau)+\beta(\sigma)\alpha(\tau ) ] \dd \sigma \dd \tau\end{aligned}\ ] ] with @xmath206 , which , using [ appendix ] , can be rewritten as @xmath207 + \sum_{k=1}^\infty a_k \big ( \pi[\sin(2k\eps\bullet)]+\pi[\cos(2k\eps\bullet ) ] \big ) -\cos^2\theta \big ( \pi[\sin(\eps\bullet)]+\pi[\cos(\eps\bullet ) ] \big ) \end{aligned}\ ] ] with the coefficients given by @xmath208 and @xmath209 for @xmath210 . as @xmath211 is always negative , the hessian will be strictly negative definite if @xmath212 for all @xmath213 , which is equivalent to @xmath214 . with @xmath215 , for instance , this double inequality is satisfied for all @xmath216 $ ] , and this range of @xmath131 yields fidelities @xmath217 . therefore , for all of these values of @xmath131 , the critical point @xmath178 corresponds to a local maximum , at least over any finite dimensional subspace , with @xmath218 . note that this example also shows that the hessian at a ( non - kinematical ) critical point need not have finite rank . by computing the rank of the relevant lie algebra , we can verify that @xmath219 specifies an ( exact - time ) controllable system at @xmath220 , so that our uniform controllability result applies . time-@xmath17 controllability can therefore be ensured by choosing a sufficiently small @xmath221 , which both makes @xmath219 lie in the neighborhood of uniform controllability and @xmath202 be larger than the minimal time required for controllability . theorem [ thm:1 ] and the previous examples relied on constant controls to show that critical points with non - extremal values of the fidelity exist for all systems with @xmath175 , i.e. , any choice of @xmath11 and @xmath157 , and that such critical points need not be saddle points but can in fact be traps . it is also interesting to consider how the critical point condition @xmath222 could be satisfied for a given system for non - constant controls @xmath223 . let @xmath224 , then the critical point condition reads @xmath225 where @xmath226 must satisfy the schrodinger equation ( [ eq : seu ] ) , i.e. , we have a differential - algebraic equation system . note that since @xmath22 is of course not known in advance , we set @xmath227 for some @xmath228 and then @xmath229 , a posteriori . while the existence and uniqueness of solutions to such systems is in general not trivial to ascertain , differentiating the constraint and using the schrodinger equation leads to a more explicit form @xmath230\ket{\psi_0 } \\ 0 & \equiv \im \bra{\psi_b } u_f(t)^\dag [ h_0+f(t)h_1,[h_0,h_1 ] ] u_f(t ) \ket{\psi_0}.\end{aligned}\ ] ] assuming the first two equations hold , which can be thought of as a two ( real ) dimensional constraint on @xmath231 , the system can be solved , at least locally about @xmath232 , by adjoining the constraint @xmath233 u_f(t ) \ket{\psi_0 } } { \im \bra{\psi_b } u_f(t)^\dag [ h_1,[h_0,h_1 ] ] u_f(t ) \ket{\psi_0}}\ ] ] to the schrodinger equation , under the generically true condition that the denominator does not vanish at @xmath232 . the additional , also generic , property that @xmath234 guarantees that the critical point @xmath8 in question is not a constant function . as the fidelity @xmath33 for the control @xmath8 is @xmath235 we can construct critical points for which the fidelity does not take extremal values . ( [ eq : nonkin ] ) can be solved numerically to find non - constant non - kinematic critical points for which the fidelity assumes any desired value , as has been done for a four - level system in @xcite . while some of these critical points may be non - attractive , as appeared to be the case for the example studied in @xcite there is no particular reason why _ all _ such critical points should be non - attractive in general . unfortunately , it is difficult to prove this , however , due to the difficulty involved in calculating the hessian and proving it to be strictly negative definite given only a numerical solution of ( [ eq : nonkin ] ) . the set of critical points of the fidelity as a map from the unitary group @xmath2 to @xmath236 given by @xmath237 $ ] is equal to all @xmath238 for which @xmath239 $ ] vanishes for every anti - hermitian matrix @xmath240 . since the bilinear map @xmath241 is an inner product over complex matrices for which the hermitian and anti - hermitian matrices constitute orthogonal subspaces , the latter condition is equivalent to @xmath242 being hermitian . as @xmath78 is also unitary it must be of the form @xmath243 , where @xmath244 is a projector onto a subspace @xmath245 of @xmath246 , and @xmath247 the projector onto the orthogonal complement of @xmath245 . from this we see immediately that @xmath248 is equal to @xmath249 , where @xmath250 is the dimension of the subspace @xmath245 , and thus there are @xmath251 critical manifolds corresponding to critical values @xmath252 of @xmath253 for @xmath254 . moreover , the second derivative of @xmath241 is @xmath255 $ ] , where @xmath256 can be any negative semi - definite matrix . thus @xmath257 , corresponding to @xmath258 are global extrema , and all other possible @xmath78 have both positive and negative eigenvalues , so that @xmath259 $ ] can take any value , showing that all other critical points @xmath238 are saddle points . this characterization of the critical points on @xmath2 , which can be traced back to at least @xcite , provides the motivation for the claim in @xcite that the landscape for unitary operator optimization has critical points only at these values of the fidelity , and that all critical points @xmath238 , except @xmath257 , which correspond to the global maximum and minimum of @xmath253 , respectively , are saddle points . as in the pure - state transfer case , to conclude this for the optimization problem over @xmath0 from the observations about the critical points of @xmath241 as a functional on @xmath2 , it is implicitly assumed that the solution functional @xmath22 is regular , in the differential geometric sense of having a jacobian of full rank , everywhere over @xmath0 . @xcite includes the condition that variations of the controls can be used to generate any local variation of @xmath38 as part of the definition of controllability . as before , the full - rank - everywhere assumption is problematic as it can not be satisfied in general . indeed , for any constant control @xmath260 , the trajectory @xmath261 will be a linear combination of expressions @xmath262 , where @xmath263 are the eigenvalues of @xmath264 . the gradient of the solution operator at this point @xmath265 is therefore a linear combination of functions @xmath266 for @xmath267 , or equivalently of the real functions @xmath268 for @xmath269 , of which there are only @xmath270 . more generally the rank of the jacobian where all controls are constants @xmath271 is at most @xmath272 because , in the eigenbasis of @xmath273 , the diagonal elements of each @xmath274 are constant functions . the observation that constant controls are singular in the sense of not being full - rank has also been made recently in @xcite but without fully considering the implications for the validity of the landscape results . also , the application of results in @xcite as suggested in @xcite is problematic , as the former work , originating in sub - riemannian geometry requires a ( strictly ) positive definite running costs on the controls , which serves as a regularizing term and is not present in our case , and highly undesirable as it would prevent us from ever reaching the global maximum of the actual objective function . also properties generic over the infinite dimensional space of vector fields , such as the main results of @xcite or @xcite , need not hold for any instance within the finite dimensional subset of right - invariant vector fields which we restrict attention to . thus , it is still not clear whether the fidelity @xmath35 over @xmath0 has critical points that do not satisfy the criteria above , and whether these can be attractive . before we study the fidelity over @xmath0 , it is useful to briefly consider the critical points of @xmath275 on the lie group in more detail , as there are some subtleties one should be aware of ; in particular when the system evolution is restricted to a subgroup of @xmath2 such as @xmath57 , attractive critical points may emerge even on the group and it may desirable to eliminate these by modifying the performance index . many problems in quantum control involve control hamiltonians @xmath157 that have zero trace , i.e. @xmath276 , in which case exact - time @xmath277 controllability can not hold . indeed , we always have @xmath278 = e^{- it / n \tr [ h_0 ] } = e^{i \phi}$ ] , regardless of the control @xmath279 , so the reachable set at time @xmath17 is restricted to matrices in @xmath280 times the fixed phase @xmath281 . hence , even assuming that the system is @xmath277 controllable and @xmath282 are chosen such that @xmath283 to make @xmath29 reachable , when considering the critical points of the fidelity on the group , we should really consider the critical points of @xmath284 $ ] as a functional over @xmath285 . the critical point condition is now that @xmath286 = 0 $ ] for all traceless @xmath287 , and thus that @xmath78 be equal to a hermitian matrix @xmath288 plus a multiple @xmath289 of the identity with @xmath290 real . both this and the unitary conditions are properties of the spectrum alone , so we have in general that , @xmath78 is a critical point of @xmath241 whenever its eigenvalues are @xmath291 and @xmath292 with multiplicities @xmath250 and @xmath293 , for some @xmath294 . finally , the unit determinant condition on @xmath78 forces @xmath295 to equal @xmath296 times any @xmath297 root of unity . note that when @xmath298 , this set of points is larger than the set of critical points over @xmath277 which also happen to lie in @xmath280 . the second derivative of @xmath299 is again @xmath300 $ ] , but now with @xmath301 , and only the hermitian part @xmath288 of @xmath78 contributes in this expression , since @xmath256 is negative semi - definite . it is clear that if @xmath288 has eigenvalues of both signs , implying @xmath302 , then the corresponding critical point @xmath78 is a second order saddle . otherwise @xmath303 , and the critical point is a local maximum or minimum if @xmath288 is a positive or negative multiple of the identity , while in case @xmath304 , we can find curves @xmath305 for which @xmath241 does not vanish to third order in @xmath306 , so @xmath78 is a saddle point . this characterizes the attractive critical points of @xmath307 as those @xmath238 of the form @xmath308 for some @xmath309 root of unity @xmath310 having positive real part , and continuity of the solution operator @xmath311 then implies : for any controllable system @xmath55 with @xmath312 , and any @xmath313 with @xmath314 , there exists an open subset @xmath315 of @xmath0 such that any monotonically increasing local optimization algorithm started with @xmath316 will never exceed a fidelity of @xmath88 . these results show that that for optimization of @xmath57 , there are traps , i.e. , attractive critical points of @xmath253 with @xmath318 , and this is relevant for optimization problems over @xmath0 with @xmath319=0 $ ] . as the attractive critical points differ only by a global phase factor @xmath320 from the target gate @xmath29 , however , and we usually do not care about the global phase of a gate , these traps are not really a problem in practice , provided the convergence condition of the algorithm is more sophisticated than checking if @xmath321 is less than a certain tolerance . however , if we do not care about the global phase of the gate @xmath29 , a better choice of the performance index to be optimized would be @xmath322 which is equivalent to optimization over the projective unitary group @xmath317 . to see why it is a better choice , note that the critical point condition for @xmath323 is @xmath324=0 $ ] for every @xmath325 with @xmath242 as before , and @xmath326 $ ] . this is satisfied only when @xmath327 for some real @xmath290 and hermitian @xmath288 with all eigenvalues of the same magnitude . the second derivative at these points is @xmath328 + 2|\tr[w a]|^2 $ ] , the first term can only fail to ever be positive for any @xmath325 if @xmath288 is a non - negative multiple of the identity , or @xmath329 . in the first case , either @xmath330 or @xmath331 , which corresponds to @xmath323 attaining its minimal or maximal values respectively otherwise , @xmath288 has eigenvalues of opposite sign , so the second derivative s first term must vanish for every @xmath325 , and if its second term always vanishes then @xmath238 must have equal eigenvalues , thus maximizes @xmath323 . the previous section shows that suboptimal attractive critical points do arise for optimization of @xmath57 , and that this case is relevant for quantum control problems , but these can easily be eliminated by changing the performance index . we now turn our attention to the existence of sub - optimal attractive critical points for the actual optimization problem of interest , i.e. , the problem of maximizing the fidelity @xmath35 over @xmath0 . in particular we are interested in whether there are suboptimal attractive critical points , i.e. , points @xmath22 for which the gradient of @xmath35 vanishes , the hessian is negative definite , and @xmath332 for systems that are time-@xmath17 controllable . * traps for unitary operation optimzation problem . * consider the control problem specified by : @xmath333 with @xmath202 , whose gradient at @xmath126 can be calculated as @xmath334 and whose hessian at the same @xmath126 expands to @xmath335+\frac{1}{3 } x \sum_{k=1}^{\infty } \frac{4}{\pi(4k^2 - 1)}\big(\pi[\sin(2k\eps\bullet)]+\pi[\cos(2k\eps\bullet)]\big)\\ + \frac{1}{3 } y\big(\pi[\cos(\bullet)]+\pi[\sin(\bullet)]\big)- \frac{1}{3 } z\big(\pi[\cos(\eps\bullet)]+\pi[\sin(\eps\bullet)]\big)\end{gathered}\ ] ] where the constants @xmath336 are defined in ( [ egconsts ] ) below . this formula for the hessian again comes from applying the results of [ appendix ] to the original operator expression @xmath337 [ \alpha(\sigma)\beta(\tau)+\alpha(\tau)\beta(\sigma ) ] \dd \sigma \dd \tau\end{gathered}\ ] ] therefore , a sufficient set of conditions for this operator to be negative definite , thus the fidelity @xmath338 to have a local maximum over all finite dimensional subspaces at @xmath126 , are the constraints : [ egconsts ] @xmath339 we are interested in whether these conditions , which are independent of @xmath221 , can be satisfied for some choice of the other parameters , while having @xmath219 satisfy the ( exact - time ) controllability lie algebra rank condition for @xmath220 , so that by our uniform controllability result a sufficiently small choice of @xmath221 would make this system exact - time controllable at @xmath17 . indeed , several such choices of parameters exist , for example : in the @xmath340 case , @xmath341 , @xmath342 , @xmath343 , @xmath344 , @xmath345 with fidelity @xmath346 , or in the @xmath347 case , @xmath348 , @xmath349 , @xmath350 , @xmath351 , @xmath352 with a fidelity of @xmath353 . in typical quantum control problems , the target time @xmath17 is fixed , but @xmath17 can be allowed to vary . if the target time @xmath17 of control problems is not fixed in advance , and is consequently allowed to vary in the corresponding optimization formulation , there is some ambiguity as to how proximity between control fields is to be measured when these lie in different spaces @xmath0 . we choose to resolve this difficulty by optimizing over hamiltonians of the form @xmath354 with @xmath17 fixed , since propagating @xmath355 up to time @xmath17 is equivalent to propagating @xmath356 up to time @xmath357 . the expressions for the gradient ( [ eq : grd ] ) and hessian ( [ eq : hess ] ) derived earlier from the perturbative expansion can be adapted to this variable time optimization framework simply by replacing all instances of @xmath358 by @xmath359 . in particular , at a point where @xmath126 , we find the solution functional has gradient simplifying to @xmath360 while its hessian is the bilinear map on @xmath361 given by @xmath362 \dd\sigma\dd\tau \\ - u(t)\int_0^t \left([u_f(\tau)^{\dag}h_1u_f(\tau),h_0]\tau + h_0u(\tau)^{\dag}h_1u_f(\tau ) t \right ) [ \alpha(\tau)\ell_b+\beta(\tau)\ell_a]\dd\tau \\ - u_f(t)t^2 h_0 ^ 2 \ell_a \ell_b.\end{gathered}\ ] ] in this framework we consider the unitary control problem specified by : @xmath363 @xmath364 with @xmath202 , where a fortiori @xmath365 , and we shall set @xmath366 , @xmath367 , @xmath368 , @xmath369 , @xmath370 , @xmath351 , @xmath371 with @xmath372 arbitrary . with these parameters , the gradient at @xmath126 , @xmath373 always vanishes , and the hessian evaluates to @xmath374 & - 1.60\ \big ( \pi[\cos]+\pi[\sin ] \big ) - 0.28\ \big ( \pi[\cos(\eps\bullet)]+\pi[\sin(\eps\bullet ) ] \big)\\ & -\sum_{k=1}^{\infty } \frac{4}{\pi(4k^2 - 1 ) } \big(\pi[\sin(2k\eps\bullet)]+\pi[\cos(2k\eps\bullet ) ] \big),\end{aligned}\ ] ] which is always positive definite , since @xmath183 , although dependent on @xmath221 , must be at least @xmath375 , and the fidelity @xmath338 here is @xmath376 . having demonstrated the existence of suboptimal attractive critical points for quantum control optimization problems , the question arises how significant such points are in practical applications . the examples in the previous sections for which we can rigorously prove the existence of such bad points are clearly quite artificial , and mainly serve to prove a theoretical point . unfortunately , it does not appear that any theoretical machinery is available to provide a general answer to this question for far more complex problems that are of immediately practical interest , so we will use numerical simulations to give a glimpse at the range of possibilities . for this purpose , we consider the problem of implementing a quantum fourier transform gate for a three - qubit ising - coupled spin chain with full local control . explicitly , the system has the time varying hamiltonian @xmath377 with six control fields @xmath378 , where @xmath379 are the unnormalized pauli matrices and @xmath380 is a shorthand for the tensor product @xmath381 , etc . the quantum fourier transform objective @xmath29 is specified by @xmath382{i}}{\sqrt{8 } } \omega^{jk}\ ] ] where @xmath383 is the canonical choice @xmath384 , and the prefactor @xmath385{i}=e^{2\pi i(m+\frac{1}{4})/8}$ ] for some integer @xmath386 is needed to ensure @xmath387 as our hamiltonians are in @xmath388 . we choose @xmath389 . to solve the optimization problem numerically we use the standard parameterization of the controls @xmath390 in terms of piecewise constant functions over @xmath245 intervals forming a uniform partition of @xmath30 $ ] , i.e. , effectively we define @xmath391 whenever @xmath392 , and optimize over all @xmath393 , @xmath394 variables . this system satisfies the condition of exact - time controllability over @xmath395 , the case we are usually interested in , although , as is generally the case , we do not know the critical time @xmath53 required to render the whole group accessible . neither do we know how to ensure having a suitable number of time slices @xmath245 , except that we must satisfy the dimensionality constraint @xmath396 , otherwise the optimization will either get trapped or converge to unbounded fields , for almost all objectives @xmath29 . however , the fact that we were always able to achieve errors @xmath397 below @xmath398 for 50 randomly selected target gates suggests that our choice of @xmath399 is sufficiently large , and of @xmath400 appropriate . ideally , we would like control optimization to be effective even under the weaker condition that the target gate @xmath29 be accessible at the given time @xmath17 , using @xmath245 time slices . this is certainly true , at the @xmath398 accuracy level , for our @xmath29 . ) show that the good runs terminate with errors @xmath401 , with many reaching smaller errors . the histogram of the errors for the 42 bad runs ( inset ) show that all but one of these terminate with errors @xmath402 , orders of magnitude larger than the terminal errors for the good runs.,scaledwidth=60.0% ] for the optimization we primarily used a quasi - newton method @xcite , relying on the bfgs formula to construct an approximate hessian , and an exact analytic formula for the gradients . this choice was motivated by related work in which this procedure performed well relative to various others , in particular at high fidelities . each run of the algorithm was initialized by setting the @xmath393 variables independently at random from a standard normal distribution , a choice which was deemed suitable considering the typical norm of high fidelity solutions . we carried out 1,000 optimization runs and found 958 of them successfully passed our pre - selected error threshold of @xmath403 , while of the remaining 42 all but one failed to reach even the @xmath404 level . a histogram of the terminal errors for 100 selected good runs versus the 42 bad runs ( fig . [ fig : hist ] ) shows that the final fidelites achieved by the good runs are many orders of magnitude higher than the terminal fidelities for the bad runs . [ fig : conv ] further shows that the convergence behavior for the successful runs is distinctly different from that of the failed ones . considering that the successful runs reached fidelities @xmath405% on average in a few dozen iterations taking mere seconds , the bad runs constitute effective traps in that the otherwise highly efficient quasi - newton optimization routine gets stuck at these points and the runs are extremely time - consuming . we further explored the behavior of the fidelity in neighborhoods of these bad points by evaluating it for @xmath406 small random field perturbations around each of the presumed traps @xmath407 . the results show that the fidelities for _ all _ of these fields are smaller than the fidelity @xmath408 at the presumed trap @xmath407 , suggesting that the @xmath407 are indeed in the trapping neighbourhood of a local maximum for all 42 runs that terminated at fidelities @xmath409 . an example of a scatter - plot of the fidelities of the perturbed fields @xmath45 as a function @xmath410 in a neighborhood of trap 1 is shown in fig . [ fig : trap1 ] . we know that this is not a result of the target time @xmath17 being too short or the dimension of the subspace of @xmath98 being too small as there are solutions for the fields for @xmath399 and @xmath400 with errors @xmath411 on the order of @xmath412 , close to machine precision . to ascertain whether these bad fields @xmath407 indeed attracted trajectories , we also restarted the optimization with perturbed fields @xmath413 . even when the size of the perturbations was increased to @xmath414 on the order of @xmath415 the runs tended to remain stuck in the trapping region as illustrated in fig . [ fig : pert ] . finally , we also tested other implementations , including several variants of sequential update algorithms , similar to the krotov method _ without _ penalty , but starting with the bad points @xmath407 none of the runs managed to escape the trapping regions . it is difficult to ascertain if there is a single trap or possibly multiple traps in each region , and there are clearly limitations of this computational investigation , but taken together , the simulation results strongly suggest that there are suboptimal extrema and effective trapping regions , at least for the @xmath416-dimensional search space we chose to optimize over . is negative , bounded away from @xmath19 , for all sample points , showing that the fidelity at all of these random samples is less than the fidelity at the presumed local extremum @xmath407 , suggesting that @xmath407 is indeed a local extremum.,scaledwidth=60.0% ] . for each @xmath407 , for clarity , the _ difference _ of the fidelity as a function of the iteration number , @xmath417 , and the fidelity of the nearest `` trap '' @xmath418 is shown for 10 runs starting with @xmath419 . for each run the difference in the fidelites vanishes , i.e. , the fidelities seem to converge to the respective fidelity value for the closest trap . inset shows close - up of a small portion to better resolve the trajectories.,scaledwidth=60.0% ] it is difficult to ascertain whether these bad points would constitute traps if each of our six control fields @xmath41 were not restricted to the 140-dimensional subspace of @xmath420 consisting of piecewise constant functions . to get a glimpse of how the convergence behavior changes when we enlarge the subspace of @xmath98 we project onto , we subdivided the time grid three - fold to @xmath421 , and restarted the optimization , starting with the field @xmath422 for which the optimization got trapped on the original subspace . as one would expect , tripling the dimension of the parameter space we optimize over from @xmath416 to @xmath423 does increase the attainable fidelities slightly , but in all the cases we studied the improvements were small , and the optimization over the larger subspace exhibited trapping behavior similar to that on the original control space , terminating with substantially lower fidelities after many ( time - consuming ) iterations than typical successful runs ( see fig . [ fig : refinement ] ) . that refining the mesh would lead to higher fidelities is rather inevitable , since even if there is a local maximum in infinite dimensions , any finite dimensional subspace will almost surely not contain this point , but the distance between increasingly larger subspaces and the point must decrease , tending to zero , allowing the fidelity to increase towards an asymptotic value . this phenomenon can conveniently be visualised by considering a non - degenerate quadratic potential on the plane and a generic line through the origin : the maximum on the line is lower than the maximum over the plane . the fact that fidelity increases so little ( relative to the error ) after we refine the mesh is , on the other hand , an indication that said asymptotic value is likely to strictly below the global maximum of @xmath117 , and the data certainly suggests that the neighborhoods of the @xmath407 are really bad neighborhoods for the optimization . error threshold or exceeded a thousand bfgs iterations ( blue , largest ) , ( 2 ) after increase in fidelity dropped below @xmath424 in additional newton iteration steps ( red , intermediate ) , ( 3 ) after increase in fidelity obtained by a newton iteration remains below @xmath425 even after sub - division of time intervals ( yellow , smallest ) . the inset focuses on the failed runs . since the rank order of the final errors between runs did not change much across the three stages , one can see that the gains made by going beyond a thousand iterations or refining the mesh are marginal . in fact , the largest reduction in the error obtained by refining the mesh was only @xmath426% , orders of magnitude less than what would have been achieved by a successful run with less computational effort.,scaledwidth=60.0% ] we have revisited the control landscape for several classes of canonical quantum control problems , in particular pure - state transfer and unitary gate optimization problems . although the pure - state transfer fidelity as a function over the unitary group only has two types of critical points , corresponding to either the global minimum @xmath19 or the global maximum @xmath117 , detailed analysis shows that the class of critical points for the actual optimization problem over functions in @xmath0 is strictly larger than the set of kinematic critical points . in particular , for any bilinear control system @xmath175 and any fixed target time @xmath17 , there exist pairs of initial and target states , such that some @xmath427 is a critical point of the system , and we can achieve any value of the fidelity between @xmath19 and @xmath117 for such critical points . moreover , while these critical points are not expected to be attractive in most cases , we have presented examples of systems with attractive suboptimal critical points , showing that traps _ do _ exist for such problems , and the fidelity at these traps can take many values , unlike the fidelity for kinematic critical points , which is limited to 0 or 1 for pure - state optimization problems . for the problem of unitary operator optimization we demonstrated that there are indeed no traps when we consider the fidelity as a functional over the unitary group . although there are critical manifolds on which the fidelity takes values between @xmath428 ( global minimum ) and @xmath429 ( global maximum ) , all of these critical points are indeed saddle points , but the situation changes when the analysis is restricted to the special unitary group @xmath57 . this case may appear artificial but it is actually highly relevant for quantum control , as many quantum control problems involve control hamiltonians that have zero - trace , and hence we have no global phase control . although the system may be @xmath2 controllable when @xmath319=0 $ ] , if the target time is fixed , so is the global phase . perhaps surprisingly , there are more critical points in this case than for @xmath2 , and some of these are attractive critical points at which the fidelity assumes values @xmath430 , i.e. , traps . more careful analysis shows that these traps correspond to solutions @xmath431 , where @xmath320 is a root of unity , which immediately shows that these solutions have fidelity @xmath430 , according to the standard definition of the fidelity . nonetheless , these solutions are perfectly adequate for most practical purposes as we usually do not care about the global phase of an operator , and this problem can be avoided entirely simply by modifying the performance index to reflect the fact that we do not care about the global phase . optimizating over the projective unitary group @xmath317 there are indeed no traps , at least on the group . again , the situation is more complicated for actual optimization problem over controls in @xmath0 . not only are there critical points @xmath8 for which the fidelity assumes critical values other than those permitted over @xmath2 or @xmath57 , but examples again show that the hessian at these critical points can be negative definite , implying that they are locally attractive on any finite - dimensional subspace and therefore traps . the results can even be extended to the case where the target time @xmath17 is variable , again proving the existence of traps . the specific examples of traps constructed prove a theoretical point about the existence of traps , but they are not of particular interest in their own right . unfortunately , due to the difficulty of exactly calculating the hessian for complex systems with time - varying controls , it is difficult to _ prove _ the existence of traps for more interesting cases . therefore we turned to numerical simulations to investigate any evidence for the existence of trapping regions in practice . for a three - qubit gate optimization problem we studied the simulation results suggests that there are genuine traps , i.e. , attractive critical points for which the fidelity is below a certain threshold . although such traps do not appear to be very common in the example considered we observed trapping in about 42 out of 1000 cases , they do complicate the control landscape and obstruct convergence to a global extremum , and perhaps even more importantly , trapped runs can be extremely time - consuming , which suggests that it is important to consider the possibility of trapping in the algorithm design , to ensure that trapped runs are quickly identified and terminated . furthermore , the example considered was a rather simple system of dimension @xmath432 with six independently variable controls . it is quite plausible that the number of traps and the likelihood of a local optimization algorithm converging to these increases with the system dimension or complexity . this is an important issue of practical concern that requires a systematic further analysis . we would like to thank thomas schulte - herbrggen , robert zeier and john v. leahy for helpful discussions and suggestions . sgs acknowledges funding from epsrc arf grant ep / d07192x/1 , the epsrc qip interdisciplinary research collaboration ( irc ) , hitachi and nsf grant phy05 - 51164 . pdf acknowledges funding from epsrc and hitachi ( case / cna/07/47 ) . we define the operators @xmath183 and @xmath245 via their action on functions @xmath95 [ eq : sc ] @xmath433\dd\sigma\dd\tau , \\ \ip{\beta}{s\alpha}_{l^2 } & = \!\!\!\underset{0<\sigma<\tau < t}{\int\int}\!\!\ ! \sin(\omega(\sigma-\tau ) ) [ \alpha(\tau)\beta(\sigma)+\beta(\tau)\alpha(\sigma)]\dd\sigma\dd\tau \end{aligned}\ ] ] for @xmath183 , using the trigonometric identity @xmath434 , and symmetrization the double integral , we find @xmath435 + \pi[\cos(\omega\bullet)]\ ] ] with @xmath383 arbitrary . when @xmath436 , @xmath245 can be expressed in terms of its eigenfunction decomposition , constituted of the real fourier basis @xmath437 + \sum_{k=1}^\infty \frac{4}{\pi(4k^2 - 1 ) } \big ( \pi[\sin(2k\omega \bullet ) ] + \pi[\cos(2k\omega\bullet)]\big)\ ] ] recall that here @xmath99 denotes the unnormalized projection onto its argument and @xmath100 is a placeholder for the argument to an anonymous function . for example , @xmath438 $ ] is the projector onto the function @xmath439 , i.e. the operator sending each element @xmath440 to the element @xmath441 in @xmath0 . 10 url # 1#1urlprefix[2][]#2 rabitz h 2009 _ new journal of physics _ * 11 * 105030 issn 1367 - 2630

continuous - time random walks combining diffusive scattering and ballistic propagation on lattices model a class of lvy walks . the assumption that transitions in the scattering phase occur with exponentially - distributed waiting times leads to a description of the process in terms of multiple states , whose distributions evolve according to a set of delay differential equations , amenable to analytic treatment . we obtain an exact expression of the mean squared displacement associated with such processes and discuss the emergence of asymptotic scaling laws in regimes of diffusive and superdiffusive ( subballistic ) transport , emphasizing , in the latter case , the effect of initial conditions on the transport coefficients . of particular interest is the case of rare ballistic propagation , in which case a regime of superdiffusion may lurk underneath one of normal diffusion .

1501.05216

stochastic processes in which independent particles scatter randomly at finite speed and may occasionally propagate over a long distance in single bouts are known as lvy walks . use of these models has become ubiquitous in the study of complex diffusive processes . they are particularly relevant to situations such that the probability of a long jump decays slowly with its length @xcite . the scale - free superdiffusive motion of lvy walkers @xcite has been identified as an efficient foraging strategy @xcite , spurring a large interest in these models , including with regards to human mobility patterns @xcite . a central quantity in this formalism is the distribution of free path lengths , which gives the probability that a particle propagates over a distance @xmath0 between two successive scattering events . its asymptotic scaling determines how the moments of the displacement asymptotically scale with time . assuming the probability of a free path of length @xmath0 scales asymptotically as @xmath1 , one finds , in the so - called velocity picture of lvy walks , the following scaling laws for the mean squared displacement after time @xmath2 @xcite : @xmath3 see below for a precise definition of this quantity . a scaling parameter value @xmath4 is such that the variance of the distribution of free paths diverges , in which case the process is often called scale - free @xcite . correspondingly , the asymptotic divergence of the mean squared displacement gives rise to anomalous transport in the form of superdiffusion . in a recent paper @xcite , we considered lvy walks on lattices and generalized the standard description of the velocity picture of lvy walks , according to which a new jump event takes place as soon as the previous one is completed , to include an exponentially - distributed waiting time separating successive jumps . as emphasised in reference @xcite , this additional phase induces a differentiation between the states of particles which are in the process of completing a jump and those that are waiting to start a new one . we call the former states propagating and the latter scattering . the distinction between these states leads to a new theoretical framework of lvy walks in terms of multistate processes @xcite , whereby the generalized master equation approach to continuous time random walks @xcite translates into a set of delay differential equations for the corresponding distributions . it is the purpose of this paper to show that a complete characterization of the solutions of such multistate processes can be obtained , which yields exact time - dependent analytic expressions of their mean squared displacement . these expressions can , on the one hand , be compared with the asymptotic solutions already reported in @xcite , and , on the other , also prove useful when the asymptotic regime is not reached , which is often the case with studies dealing with observational data . such a situation arises when the probability of a transition from scattering to propagating states is small . a particle will then spend most of its time undergoing transitions among scattering states , performing a seemingly standard continuous - time random walk , only seldom undergoing a transition to a propagating state , during which it moves ballistically over a distance distributed according to the scaling parameter @xmath5 . if @xmath5 is small enough ( @xmath6 ) , these occurrences , although rare , have a dramatic effect on the asymptotic scaling properties of the mean squared displacement , such that a crossover from normal to anomalous diffusive transport is observed . this crossover time may in some situations , however , be much larger than the times accessible to measurements . here we provide analytic results which give a precise characterisation of the emergence of an anomalous contribution out of a normally scaling one , showing how the asymptotic scaling comes about . this is of particular relevance to the regime @xmath7 , for which the logarithmic divergence in time of the mean squared displacement is indeed very slow . our results are tested and confirmed by numerical simulations of the processes under consideration . the paper is organized as follows . in , we introduce the multi - state description of lvy walks with respect to scattering and propagating states , and define the transition probabilities between them in terms of two parameters , one relating to the probability of a transition from a scattering to a propagating state and the other characterizing the asymptotic scaling of the free path distribution . the fraction of particles in a scattering state evolves according to a delay differential equation which is derived and solved in . these results are exploited in to obtain the mean squared displacement of the processes . a comparison with numerical simulations is provided in . conclusions are drawn in . we consider a continuous - time random walk on a square lattice which generalizes the standard model of montroll & weiss @xcite in that it assumes displacements to non - neighbouring sites occur in a time span given by the ratio of the distance traveled to the walker s speed @xmath8 , which itself remains fixed throughout . in this sense , the model is similar to the so - called velocity picture of lvy walks @xcite , the difference being that successive propagations , irrespective of their lengths , are separated by random waiting times , which we assume have exponential distributions . the model itself is not new and , in some sense , is a simplification of other models of lvy walks interrupted by rests ; see reference @xcite . as we show below , however , the combination of a discretized spatial structure and exponentially - distributed waiting times yields a novel description of the process in terms of delay differential equations amenable to analytic treatment . a natural distinction arises between the states of walkers which are moving across the lattice structure and those at rest . we call _ propagating _ the state of a particle which is undergoing a displacement phase and _ scattering _ the state of a particle at rest , waiting to start a new displacement . in the framework of intermittent random walks @xcite , the former state is usually referred to as ballistic and the latter as diffusive or reactive , depending on context . a scattering state may therefore be thought of as one associated with a local diffusive process bound to the scale @xmath9 of the distance between neighbouring lattice sites . the interplay between the two states is as follows . a particle in a propagating state switches to a scattering state upon completing a displacement . a particle in a scattering state , on the other hand , can make transitions to both scattering and propagating states ; as soon as its randomly - distributed waiting time has elapsed , it moves on to a neighbouring site and , in doing so , may switch to a propagating state and continue its motion to the next site , or start anew in a scattering state . consider a @xmath10-dimensional cubic lattice of individual cells @xmath11 . the state of a walker at position @xmath12 and time @xmath2 can take on a countable number of different values , @xmath13 , specified by an integer @xmath14 , and direction @xmath15 , where @xmath16 denotes the coordination number of the lattice , that is the number of different lattice directions . states @xmath17 are associated with a scattering state , irrespective of direction @xmath18 , while states @xmath13 with @xmath19 refer to propagating states , with @xmath20 being the remaining number of lattice sites the walker will travel in direction @xmath18 to complete its displacement . time evolution proceeds in steps , characterized by a waiting - time density function and a transition probability . a particle at position @xmath21 in a scattering state will wait for a random time @xmath2 , exponentially distributed with mean stands for residence as in `` residence time . '' in contrast , b in @xmath22 stands for ballistic as in `` ballistic propagation time . '' ] @xmath23 , before updating its state to @xmath24 with probability @xmath25 , simultaneously changing its location to @xmath26 , where @xmath27 denotes the lattice vector ( of length @xmath9 ) associated with direction @xmath18 . in contrast , a particle in a propagating state @xmath13 , @xmath19 , will change its state to @xmath28 after a time @xmath29 , simultaneously moving from site @xmath21 to @xmath30 . the waiting times associated with scattering states are drawn from a standard poisson process , while the renewal process generated by the combination of scattering and propagating states has arbitrary holding times , whose distribution is determined by the transition probabilities @xmath31 . the waiting - time density of the process is thus the function @xmath32 where @xmath33 denotes the dirac delta function . when a step takes place , the transition probability to go from state @xmath24 to state @xmath34 is @xmath35 where @xmath36 is the kronecker symbol . a particle which makes a transition from the scattering state to a propagating state @xmath24 will therefore travel a distance @xmath37 away in direction @xmath18 , until it eventually comes back to the scattering state and may change directions at the next transition . we now introduce a characterization of the transition probabilities @xmath31 in terms of two parameters . the first , which we refer to as the _ scattering parameter _ , is denoted by @xmath38 , @xmath39 , and gives the total probability of a transition from the scattering state , @xmath40 , into a propagating state , @xmath41 : @xmath42 the remaining transition probability , @xmath43 is that of a transition from a scattering state into another scattering state . the value @xmath44 thus corresponds to the absence of propagating states : the process is then a simple continuous - time random walk with transitions to nearest neighbouring sites only and exhibits normal diffusion with coefficient @xmath45 . the opposite extreme , @xmath46 , assigns zero probability to transitions from scattering to scattering states , which means that every transition involves a displacement over a distance of at least two sites . scattering states remain populated , however , due to the decay of propagating states . the second parameter , @xmath47 , is the _ scaling parameter _ of the transition probabilities @xmath31 , which controls their asymptotic behaviour , @xmath48 and determines the scaling law of the mean squared displacement . the specific form of @xmath31 has no effect on this scaling law , but is does affect the time - dependent properties of the mean squared displacement . to be specific , we consider in this paper the following double - telescopic form for the transition probabilities @xmath31 : @xmath49 \qquad ( k\geq1 ) ; \label{eq : rhok2}\ ] ] its structure is particularly helpful for some of the computations presented below and is motivated by our study of anomalous transport in the infinite - horizon periodic lorentz gas @xcite . in this respect , the model is slightly different from that presented in reference @xcite , where a simple telescopic structure of the transition probabilities was used . for future reference , we define @xmath50 to be the probability of a transition to a state larger than or equal to @xmath20 , such that , in particular , @xmath51 and @xmath52 , and , for the choice of transition probabilities , @xmath53.\ ] ] we also note the following two identities : @xmath54 \big \ } , \\ \sum_{j = 1}^k j \rho_j & = \epsilon ( 1 - 2^{1-\alpha})^{-1 } [ 1 - ( k + 1)^{2-\alpha } + ( k + 2)^{2-\alpha } - 2 ( k+2)^{1 - \alpha } ] . \end{split}\ ] ] a quantity which plays a central role in the analysis of the process generated by the waiting - time density and transition probabilities is the average return time to the scattering state , @xmath55 the average return time is finite only when must decay asymptotically faster than @xmath56 for the average return time to be finite . ] the process is then said to be _ positive recurrent_. in the remainder , we restrict our attention to this range of parameter values . the null - recurrent case , when @xmath58 , for which the average return time to the scattering state diverges , is considered in reference @xcite . the occupation probability of particles at site @xmath21 and time @xmath2 , @xmath59 , is a sum of the probabilities over the different states , @xmath60 : @xmath61 in reference @xcite , we obtained the following set of delay differential equations for the time - evolution of these occupation probabilities : @xmath62 + \sum_{k ' = 0}^\infty \big [ \sigma_{k+k'\ ! , j}(\mathbf{n } - k ' \mathbf{e}_{j } , t - k ' { \tau_\mathrm{b } } ) \cr & \quad - \sigma_{k+k'\ ! , j}(\mathbf{n } - k ' \mathbf{e}_{j } , t - ( k ' + 1 ) { \tau_\mathrm{b } } ) \big ] \ , , \label{eq : dtpkj}\end{aligned}\ ] ] where the inclusion of terms @xmath63 accounts for the possibility of external sources , such that @xmath64 is the rate of injection of particles at site @xmath21 and time @xmath2 in the state @xmath24 . to simplify matters , we assume that these source terms are independent of the lattice direction and write @xmath65 . furthermore , we will usually let @xmath66 for @xmath67 , which amounts to assuming that particles are injected only in a scattering state . the injection of particles in a propagating state is easily treatable as well , and will be considered explicitly in order to initiate the process in a stationary state of equations . we will , however , limit such considerations to this specific choice and avoid the possibility that source terms interfere with the asymptotic scaling of the process generated by particles initially injected in a scattering state , as might arise from alternative choices of injection rates of propagating states . of particular interest is the fraction of particles in the scattering state , @xmath68 whose time - evolution is described by the following delay differential equation : @xmath69 here , @xmath70 denotes a source term for the rate of injection of particles in state @xmath20 , irrespective of their positions , that is , @xmath71 . the first two terms on the right - hand side of this equation have the typical gain and loss structure of jump processes . indeed , the second term corresponds to particles lost by scattering states due to transitions to propagating states , which occur at rate @xmath72 . each such transition is gained back after a delay given by the length of the ballistic segment in a propagating state . the sum of those terms yields the first term on the right - hand side of equation . before turning to the integration of this differential equation , we note that the asymptotic value of @xmath73 , @xmath74 , has a simple expression . by ergodicity of the process , the equilibrium ratio of particles in the scattering state is given by the ratio between the average time spent in the scattering state , @xmath23 , and the average return time to it , @xmath75 , equation , @xmath76 assuming a positive recurrent process , such as when the transition probabilities @xmath31 are specified by equation with @xmath77 , the average return time @xmath78 is finite , so that the asymptotic fraction of particles in a scattering state is strictly positive . for the particular choice of the parameters @xmath31 , equation , the return time is given by equation and equation becomes @xmath79 note , however , that the convergence to this asymptotic value follows a power law whose exponent goes to zero as @xmath80 ; see . furthermore , @xmath73 converges to @xmath81 when the scaling parameter falls into the null recurrent regime , @xmath82 . correspondingly , the fraction of particles in the propagating states , @xmath83 tends to @xmath84 which follows from the observation that a particle that makes a transition to state @xmath20 spends an equal amount of time in all states @xmath18 such that @xmath85 . the integration of equation relies on the specification of initial conditions . for instance , the choice @xmath86 corresponds , when all particles are injected at the lattice origin , to the injection rate @xmath87 and will be henceforth referred to as the _ all - scattering initial condition_. the time - dependent fraction of particles in the scattering state , @xmath73 , @xmath88 , may then be obtained by the method of steps @xcite , which consists of integrating the differential equation successively over the intervals @xmath89 , @xmath90 , matching the solutions at the upper and lower endpoints of successive intervals . for the all - scattering initial condition , the fraction of particles in the scattering state is , for times @xmath91 , @xmath92 where @xmath93 ( resp . @xmath94 , used below ) denotes the largest ( resp . smallest ) integer smaller ( resp . larger ) than or equal to @xmath95 , and each coefficient @xmath96 is the sum of all possible combinations of products @xmath97 , where the sequences @xmath98 are such that @xmath99 , @xmath100 the contributions to @xmath96 consist of all distinct ways of travelling a distance @xmath20 in @xmath101 steps , divided by their number of permutations ; the derivation of equation is provided in [ app:1 ] . a comparison between the time - dependent fraction of particles in the scattering state and the asymptotic value is illustrated in , where the difference between the two expressions is plotted vs. time for different values of the scaling parameter @xmath5 and a specific choice of the scattering parameter @xmath38 , with the transition probabilities @xmath31 specified by equation . 0.49 , computed from equation , to the asymptotic value . the transition probabilities are taken according to equation , with @xmath102 and @xmath103 . the value of the scattering parameter is chosen such that @xmath104 ( left panel ) and @xmath105 ( right panel ) . the left panel shows the difference between @xmath73 and its asymptotic value . in both panels , the scaling parameter takes the values @xmath106 ( green curve ) , @xmath107 ( cyan curve ) , and @xmath108 ( red curve ) . the right panel corresponds to a perturbative regime , such that @xmath73 can be approximated by a first - order polynomial in @xmath38 ( darker curves ) whose first - order coefficient varies with time ; see equation . the agreement between the exact and the approximate solutions improves as the value of @xmath38 decreases . , title="fig : " ] 0.49 , computed from equation , to the asymptotic value . the transition probabilities are taken according to equation , with @xmath102 and @xmath103 . the value of the scattering parameter is chosen such that @xmath104 ( left panel ) and @xmath105 ( right panel ) . the left panel shows the difference between @xmath73 and its asymptotic value . in both panels , the scaling parameter takes the values @xmath106 ( green curve ) , @xmath107 ( cyan curve ) , and @xmath108 ( red curve ) . the right panel corresponds to a perturbative regime , such that @xmath73 can be approximated by a first - order polynomial in @xmath38 ( darker curves ) whose first - order coefficient varies with time ; see equation . the agreement between the exact and the approximate solutions improves as the value of @xmath38 decreases . , title="fig : " ] suppose that the scattering parameter is small , @xmath109 , so that the overwhelming majority of transitions occur between scattering states , and only rarely does a particle make an excursion into a propagating state , which may , however , last long , depending on the value of the scaling parameter @xmath77 . the first - order expansion of equation yields the following approximation of @xmath110 : @xmath111 that is to say , for @xmath112 , @xmath73 is a sequence of straight line segments joining the values the function takes at integer multiples of the propagation time @xmath22 , @xmath113\ , . \label{eq : s0expinteger}\end{aligned}\ ] ] in this regime , @xmath73 converges asymptotically to the constant value , which , for the choice of parameters , is given by equation , i.e. , with all other parameters being fixed , @xmath114 for these parameters , we make use of identities to evaluate equation , @xmath115 \frac{{\tau_\mathrm{b}}}{{\tau_\mathrm{r}}},\ ] ] or , for general time values , @xmath116 \cr & \quad + \frac{{\tau_\mathrm{b}}}{{\tau_\mathrm{r } } } \big [ 1 - \big ( { { \left\lfloor\hspace{-.25 mm } { t}/{{\tau_\mathrm{b } } } \hspace{-.25 mm } \right\rfloor } } + 1\big)^{2 - \alpha } + \big ( { { \left\lfloor\hspace{-.25 mm } { t}/{{\tau_\mathrm{b } } } \hspace{-.25 mm } \right\rfloor } } + 2\big)^{2 - \alpha } \cr & \quad - 2\big ( { { \left\lfloor\hspace{-.25 mm } { t}/{{\tau_\mathrm{b } } } \hspace{-.25 mm } \right\rfloor } } + 2\big)^{1 - \alpha } \big ] \big\}. \label{eq : s0expepsi}\end{aligned}\ ] ] a comparison between this approximate value and the exact one , equation , is displayed in , where the value of @xmath38 was taken to be large enough that the curves remain distinguishable . assuming initial injection of particles at the origin , the mean squared displacement of particles as a function of time , given by the second moment of the displacement vector @xmath117 , @xmath118 , where @xmath119 and @xmath120 , evolves according to the differential equation @xmath121 where , in contrast to the expression derived in reference @xcite , we have here added possible contributions from source terms @xmath122 , @xmath123 , which we assume to be concentrated at site @xmath124 only . to conform with stationarity of the process , propagating states should be uniformly injected in the time interval @xmath125 so they decay uniformly in the time interval @xmath126 . we thus let ( @xmath19 ) @xmath127 where @xmath128 is a sequence of positive numbers such that @xmath129 and @xmath130 is the heaviside step function , such that the product @xmath131 guarantees the uniform distribution of propagating source terms in the desired time interval . equation transforms to @xmath132 integrating over time , we obtain the mean squared displacement : @xmath133 there are two contributions to this expression ; the first arises from the fraction of particles in a scattering state , and the second from particles initially distributed among propagating states . this term transforms to @xmath134 the _ stationary initial condition _ , i.e. such that the initial fractions of particles in scattering and propagating states are stationary solutions of equations - , are realized if , in equation , we identify @xmath135 with the stationary fraction of particles in state @xmath20 , @xmath136 cf . equation , and let @xmath137 with this choice , the mean squared displacement yields the exact expression @xmath138 \bigg\ } \cr & \quad + \frac{\epsilon { \tau_\mathrm{b } } } { ( 1 - 2^{1-\alpha } ) { \tau_\mathrm{r}}+ \epsilon { \tau_\mathrm{b } } } \bigg\ { \sum_{k=1}^{{{\left\lfloor\hspace{-.25 mm } t/{\tau_\mathrm{b}}\hspace{-.25 mm } \right\rfloor } } } \frac{2k - 1 } { k^{\alpha - 1 } } + 2 ( t/{\tau_\mathrm{b}}- { { \left\lfloor\hspace{-.25 mm } t/{\tau_\mathrm{b}}\hspace{-.25 mm } \right\rfloor } } ) \frac{{{\left\lfloor\hspace{-.25 mm } t/{\tau_\mathrm{b}}\hspace{-.25 mm } \right\rfloor } } + 1/2 } { ( { { \left\lfloor\hspace{-.25 mm } t/{\tau_\mathrm{b}}\hspace{-.25 mm } \right\rfloor } } + 1)^{\alpha - 1 } } \bigg\ } \ , . \label{eq : n2ss}\end{aligned}\ ] ] treating @xmath139 as an integer and letting @xmath140 denote the generalized harmonic numbers , the two terms with sums over @xmath20 in the above expression evaluate respectively to @xmath141 \simeq 1 - 4 n^{-1 } h_n^{(\alpha -2 ) } + 2 h_n^{(\alpha -1)}\ , , \label{eq : n2asym_evalss}\end{aligned}\ ] ] for the term arising from the fraction of particles in a scattering state , and to @xmath142 for the term arising from particles initially distributed among the propagating states . depending on the value of the scaling parameter @xmath5 , we have the following three asymptotic regimes . for parameter values @xmath143 , the asymptotic properties of the harmonic numbers , @xmath144 are such that only the terms arising from the fraction of particles in a scattering asymptotically contribute to equation : @xmath145 } { ( 1 - 2^{1-\alpha}){\tau_\mathrm{r}}+ \epsilon { \tau_\mathrm{b } } } , \label{eq : msdasympt_agt2}\ ] ] which , up to a factor @xmath146 , is the diffusion coefficient of the process . for the marginal parameter value @xmath7 , the harmonic numbers in equation evaluate to @xmath147 where @xmath148 is euler s constant . we therefore have the asymptotic limit of equation , @xmath149 which is due to the fraction of particles in a scattering state alone . this result is , however , of limited use because the asymptotic regime only emerges provided @xmath150 ( where @xmath151 corresponds to the order of the sub - leading term ) , which may not be attainable , especially when the scattering parameter is small . for this reason , it is preferable to retain also the next - order term in equation , @xmath152 \big\}\ , , \label{eq : msdasymptsim_aeq2}\end{aligned}\ ] ] where a fraction @xmath153 is contributed by particles initially distributed among propagating states . in the range of parameters @xmath154 , we substitute the scaling properties of the harmonic numbers , @xmath155 and obtain from equation the limit @xmath156 \frac { 2\epsilon { \tau_\mathrm{b } } } { ( 1 - 2^{1-\alpha}){\tau_\mathrm{r}}+ \epsilon { \tau_\mathrm{b } } } \ , , \cr & = \frac{1}{(2 - \alpha)(3 - \alpha ) } \frac { 2\epsilon { \tau_\mathrm{b } } } { ( 1 - 2^{1-\alpha}){\tau_\mathrm{r}}+ \epsilon { \tau_\mathrm{b } } } \ , . \label{eq : msdasympt_alt2}\end{aligned}\ ] ] as emphasized by the first line of this equation , the leading - order coefficient is the sum of two distinct non - trivial contributions from the two terms on the right - hand side of equation : one from the fraction of particles in a scattering state and the other from particles initially injected in propagating states . as a consequence , the transport coefficient corresponding to the all - scattering initial condition differs by a factor @xmath157 from that above , corresponding to the stationary initial condition . this observation is consistent with the results of reference @xcite , where non - recurrent regimes of the scaling parameter @xmath5 were also investigated . the following exact expression of the mean squared displacement is obtained after substitution of the fraction of particles in a scattering state , equation , and applies to the all - scattering initial condition : @xmath158 \nonumber\\ & \quad + \sum_{m=1}^{{{\left\lfloor\hspace{-.25 mm } t/{\tau_\mathrm{b}}\hspace{-.25 mm } \right\rfloor}}-k } \sum_{n=1}^m \epsilon^{-n } a_{(n|m ) } \ , e^{-\epsilon ( { { \left\lfloor\hspace{-.25 mm } t/{\tau_\mathrm{b}}\hspace{-.25 mm } \right\rfloor}}-k - m ) { \tau_\mathrm{b}}/{\tau_\mathrm{r } } } \sum_{i=0}^{n } \frac{n!}{i ! } \epsilon^{i } \left(\frac{{\tau_\mathrm{b}}}{{\tau_\mathrm{r}}}\right)^{i } \nonumber\\ & \qquad \times \big[({{\left\lfloor\hspace{-.25 mm } t/{\tau_\mathrm{b}}\hspace{-.25 mm } \right\rfloor}}-k - m)^{i } - ( t/{\tau_\mathrm{b}}-k - m)^{i } e^{-\epsilon(t-{{\left\lfloor\hspace{-.25 mm } t/{\tau_\mathrm{b}}\hspace{-.25 mm } \right\rfloor } } { \tau_\mathrm{b}})/{\tau_\mathrm{r}}}\big ] \bigg\ } , \label{eq : n2sol}\end{aligned}\ ] ] where , for consistency , one must interpret terms such as @xmath159 as unity , even when @xmath160 . 0.49 , computed from equation . the transition probabilities were set according to equations , with @xmath102 and @xmath161 . the value of the scattering parameter was chosen such that @xmath104 ( left panel ) and @xmath105 ( right panel ) . in both figures , the scaling parameter takes the values @xmath106 ( green curve ) , @xmath107 ( cyan curve ) , and @xmath108 ( red curve ) . for comparison , in the left panel , the asymptotic scaling values predicted by equations , and are displayed in dashed lines with matching colors . the right panel shows a comparison between the exact solution and an approximated solution , exact to first order in @xmath38 ( darker curves ) ; see equation . , title="fig : " ] 0.49 , computed from equation . the transition probabilities were set according to equations , with @xmath102 and @xmath161 . the value of the scattering parameter was chosen such that @xmath104 ( left panel ) and @xmath105 ( right panel ) . in both figures , the scaling parameter takes the values @xmath106 ( green curve ) , @xmath107 ( cyan curve ) , and @xmath108 ( red curve ) . for comparison , in the left panel , the asymptotic scaling values predicted by equations , and are displayed in dashed lines with matching colors . the right panel shows a comparison between the exact solution and an approximated solution , exact to first order in @xmath38 ( darker curves ) ; see equation . , title="fig : " ] figure [ fig : msd ] displays the results of explicit evaluations of equation for three different values of the scaling parameter , @xmath77 , with @xmath162 . according to the asymptotic results , and , the scaling parameter value @xmath163 gives rise to a logarithmic divergence of @xmath164 , which separates the superdiffusive regime , for @xmath165 , from that of normal diffusion , for @xmath166 . a comparison between the exact solution and the asymptotic ones is displayed in . on the time scale of the figure , convergence to the asymptotic regimes is convincingly observed only in the superdiffusive case . note that while the computation of equation up to @xmath167 takes less than one half hour of cpu time on a reasonably fast computer , doubling the time range would increase the required cpu time to over ten days . although equation gives the exact time - dependence of the mean squared displacement of the process for the initial condition , it does not give much insight into its time development . in particular , how to extract the large - time behaviour of the mean squared displacement and connect this result to the asymptotic scalings , and is not transparent . as reflected by , a numerical evaluation of equation is necessarily limited to moderately large times , due to the number of terms involved . a regime of specific interest which allows us to infer the emergence of the asymptotic scalings , and from the solution is , however , that of small values of the scattering parameter @xmath38 , i.e. , such that transitions from scattering to propagating states occur only rarely . this is discussed below . we will otherwise have to resort to numerical simulations of the underlying stochastic processes to observe this convergence . those results are presented in . for small values of the scattering parameter @xmath38 , when transitions from scattering to propagating states are much rarer than transitions between scattering states , an expansion of equation in powers of this parameter provides an approximate expression of the mean squared displacement from which the different asymptotic regimes discussed in can be inferred . in the anomalous regime of the scaling parameter , @xmath168 , terms constant in time , that carry a normal contribution to the mean squared displacement , may bring about contributions which , even for large times , may be much larger than that of terms diverging in time ; an anomalous contribution to the mean squared displacement may then be masked by a normal one . to proceed , we substitute equation into equation and let @xmath169 . having dropped all terms of order @xmath170 and higher , we obtain @xmath171 \cr & \quad - \frac{{\tau_\mathrm{b}}}{{\tau_\mathrm{r } } } ( x - { { \left\lfloor\hspace{-.25 mm } x \hspace{-.25 mm } \right\rfloor } } ) \sum_{i = { { \lceil\hspace{-.25 mm } x \hspace{-.25 mm } \rceil}}}^\infty \rho_i \big\ } + \frac{{\tau_\mathrm{b}}}{{\tau_\mathrm{r } } } \sum_{j = 1}^{k } ( 2j+1 ) ( k - j ) \sum_{i = j}^{\infty}\rho_{i } \cr & = \frac{k { \tau_\mathrm{b}}}{{\tau_\mathrm{r } } } \big\ { 1 - \frac{{\tau_\mathrm{b}}}{{\tau_\mathrm{r } } } \epsilon ( 1 - 2^{1-\alpha})^{-1 } \big [ 1 - k ^{-1 } ( h_k^{(\alpha - 1 ) } - 1/2 + 1/2(k+1)^{1 - \alpha } ) \big ] \cr & \quad + \epsilon ( 1 - 2^{1-\alpha})^{-1 } \big [ 1 - ( k + 1)^{1 - \alpha } + ( 2 - k^{-1 } ) ( h_{k+1}^{(\alpha - 1 ) } - ( k + 1)^{2 - \alpha } ) \cr & \quad - 2 k^{-1 } ( 2 h_{k+1}^{(\alpha - 2 ) } - h_{k+1}^{(\alpha - 1 ) } - ( k+1)^{3 - \alpha } ) \big ] \big\}\ , , \cr & \simeq \frac{k { \tau_\mathrm{b}}}{{\tau_\mathrm{r } } } \big\ { 1 + \epsilon ( 1 - 2^{1-\alpha})^{-1 } \big [ 1 + 2 h_{k}^{(\alpha - 1 ) } - 4 k^{-1}h_{k}^{(\alpha - 2 ) } - \frac{{\tau_\mathrm{b}}}{{\tau_\mathrm{r } } } \big ] \big\}\ , , \label{eq : n2epsi}\end{aligned}\ ] ] where , in the last line , we have omitted terms that decay to zero as @xmath20 grows large . plugging into this expression the asymptotic forms of the generalized harmonic numbers , and , @xmath172 we obtain approximations of the mean squared displacement , which can be compared with the asymptotic expressions , and . in the regime of normal diffusion , @xmath173 , equation reduces to the first order expansion in @xmath38 of equation . for @xmath174 , the mean squared displacement displays normal diffusion at short times and anomalous diffusion at large times . thus in the weak superdiffusive regime , @xmath7 , equation reduces to @xmath175 \big\ } , \label{eq : n2epsi_aeq2}\ ] ] which differs from the stationary expression by a factor @xmath176 , subleading with respect to the logarithmically diverging term , identical in both expression . this difference stems from the absence of particles populating propagating states in our choice of initial condition . in the superdiffusive regime , @xmath177 , the divergence of the harmonic numbers , equation , dominates the large time behaviour , @xmath178 \big\}. \label{eq : n2epsi_alt2}\ ] ] here again we note a difference between this expression and equation , obtained in the stationary regime . at variance with the case @xmath7 , however , the difference occurs in the coefficient of the leading term on the right - hand side of equation , whose numerator is @xmath179 instead of @xmath151 in the stationary regime . the transport coefficient on the right - hand side of equation therefore changes , depending on the choice of initial conditions . at large times , after the mean squared displacement crosses over from normal diffusion to superdiffusion , the above expressions are equivalent to @xmath180 expansions of the asymptotic values and . the value of the crossover time , @xmath181 , that separates the short time normal diffusion from the large time anomalous diffusion can be inferred from the above expressions : @xmath182^{\frac{1}{2 - \alpha } } , & 1 < \alpha < 2,\\ \exp \left ( \frac{1}{4\epsilon } \right ) , & \alpha = 2 , \end{cases}\ ] ] which can be large , in particular when @xmath7 . for short times , equation can be improved by removing the assumption that @xmath2 is an integer multiple of @xmath22 . one then finds @xmath183 \cr & \quad + \frac{{\tau_\mathrm{b}}}{{\tau_\mathrm{r } } } \big [ 2 ( { { \left\lfloor\hspace{-.25 mm } t/{\tau_\mathrm{b}}\hspace{-.25 mm } \right\rfloor } } + 1)^{3 - \alpha } + ( { { \left\lfloor\hspace{-.25 mm } t/{\tau_\mathrm{b}}\hspace{-.25 mm } \right\rfloor } } + 1)^{2 - \alpha } - 4 h_{{{\left\lfloor\hspace{-.25 mm } t/{\tau_\mathrm{b}}\hspace{-.25 mm } \right\rfloor } } + 1}^{(\alpha - 2 ) } + h_{{{\left\lfloor\hspace{-.25 mm } t/{\tau_\mathrm{b}}\hspace{-.25 mm } \right\rfloor } } + 1}^{(\alpha - 1 ) } \big ] \cr & \quad - \left ( \frac{{\tau_\mathrm{b}}}{{\tau_\mathrm{r } } } \right)^2 \big\ { { { \left\lfloor\hspace{-.25 mm } t/{\tau_\mathrm{b}}\hspace{-.25 mm } \right\rfloor } } - h_{{{\left\lfloor\hspace{-.25 mm } t/{\tau_\mathrm{b}}\hspace{-.25 mm } \right\rfloor}}}^{(\alpha - 1 ) } + 1/2 - 1/2 ( { { \left\lfloor\hspace{-.25 mm } t/{\tau_\mathrm{b}}\hspace{-.25 mm } \right\rfloor } } + 1)^{1 - \alpha } \cr & \qquad + 1/2 \big [ ( t/{\tau_\mathrm{b}})^2 - { { \left\lfloor\hspace{-.25 mm } t/{\tau_\mathrm{b}}\hspace{-.25 mm } \right\rfloor}}^2 \big ] \big [ ( { { \left\lfloor\hspace{-.25 mm } t/{\tau_\mathrm{b}}\hspace{-.25 mm } \right\rfloor } } + 1)^{1 - \alpha } - ( { { \left\lfloor\hspace{-.25 mm } t/{\tau_\mathrm{b}}\hspace{-.25 mm } \right\rfloor } } + 2)^{2 - \alpha } \big ] \cr & \qquad + \big ( t/{\tau_\mathrm{b}}- { { \left\lfloor\hspace{-.25 mm } t/{\tau_\mathrm{b}}\hspace{-.25 mm } \right\rfloor } } \big ) \big [ 1 - ( { { \left\lfloor\hspace{-.25 mm } t/{\tau_\mathrm{b}}\hspace{-.25 mm } \right\rfloor } } + 1)^{2 - \alpha } + ( { { \left\lfloor\hspace{-.25 mm } t/{\tau_\mathrm{b}}\hspace{-.25 mm } \right\rfloor } } + 2)^{2 - \alpha } \cr & \qquad -2 ( { { \left\lfloor\hspace{-.25 mm } t/{\tau_\mathrm{b}}\hspace{-.25 mm } \right\rfloor } } + 2)^{1 - \alpha } \big ] \big\ } \bigg\}\ , . \label{eq : n2epsitime}\end{aligned}\ ] ] a comparison between this approximate solution and the exact one , equation , is shown in for different values of the scaling parameter , @xmath5 . numerical simulations of the process with rates and probabilities are based on a classic kinetic monte carlo algorithm @xcite , taking into account the possibility of ballistic motion of particles in a propagating phase . a collection of independent walkers are initialized at time @xmath184 at the origin of the two - dimensional square lattice @xmath185 , either in the scattering state , @xmath186 , for the all - scattering initial condition , or in state @xmath187 with relative weights @xmath188 specified by equations , for the stationary initial condition . for each walker , we generate a sequence @xmath189 of successive states @xmath190 , @xmath191 , @xmath192 , and corresponding times @xmath193 as follows . when in a state @xmath194 , the next transition is determined by drawing the following three random numbers . this is referred to as a _ scattering step_. * the first random number , @xmath195 $ ] , is drawn from a uniform distribution , and yields the state @xmath196 , such that @xmath197 . * the corresponding waiting time , i.e. , the time @xmath198 the particle waits in the state @xmath194 before the transition to state @xmath196 takes place , is obtained by drawing a second random number , whose distribution is exponential with mean @xmath23 . * a third random integer , with uniform distribution among the set of lattice directions @xmath199 , identifies the direction @xmath200 of the corresponding displacement by one lattice unit . depending on the new state @xmath196 , another scattering step is taken if @xmath201 or , if @xmath202 , a sequence of @xmath196 _ propagating steps _ @xmath203 takes place , such that : * the state decreases by one unit , @xmath204 , * the direction remains unchanged @xmath205 , and * the time step takes value @xmath206 , corresponding to the propagation time over a single lattice cell . the above loop repeats itself until @xmath207 , that is until @xmath208 , at which point another scattering step is taken . keeping track of the positions of all walkers as functions of time , which we measure at regular intervals on a logarithmic timescale , measurements of the mean squared displacement @xmath209 rescaled by @xmath210 are performed by taking averages of this quantity over the set of all walkers , which typically consists of @xmath211 trajectories . such measurements are reported below for the different scaling regimes in the positive recurrent range of the scaling parameter values , @xmath57 . throughout this section , we set the transition probabilities according to equation and , for convenience , change the scattering parameter @xmath38 to @xmath212 we further let @xmath213 such that the asymptotic fraction of particles in the scattering state becomes @xmath214 with the choice of parameters - , the mean squared displacement , for @xmath2 large and with the all - scattering initial condition , is expected to scale as @xmath215\ , , & 1 < \alpha < 2\ , , \end{cases } \quad \mbox{(all - scattering i.~c.)}\ ] ] where , in the two anomalous cases , we kept terms constant in time to reflect the possibility that , when @xmath216 is small , the normal term may not be negligible with respect to the anomalous one over a large range of times . this is particularly relevant for the marginal case , @xmath217 , where @xmath218 and @xmath219 remain of comparable sizes throughout the time range accessible to numerical computations . in comparison , for the stationary initial condition , the above expression remains unchanged in the normal diffusive case , but is modified in the two anomalous cases , @xmath220\ , , & 1 < \alpha < 2\ , . \end{cases } \quad \mbox{(stationary i.~c.)}\ ] ] the first order expansion in @xmath216 of the two anomalous regimes in equation ( in all - scattering initial condition ) yields results equivalent to equations and respectively . 0.5 0.5 0.5 a computation of the time evolution of the mean squared displacement in regimes of normal ( @xmath143 ) and anomalous ( @xmath221 ) diffusion is shown in , providing a comparison between the two initial conditions analyzed in , with the two corresponding sets of asymptotic regimes given by equations and . the scaling parameter values are set to @xmath222 ( ) , @xmath107 ( ) , and @xmath223 ( ) . the scattering parameter value is set to @xmath46 throughout , such that transitions between scattering states are forbidden . as expected , the numerically computed mean squared displacement obtained from the stationary initial condition ( magenta curves in figures [ fig : msdss_alphagt2]-[fig : msdss_alphalt2 ] ) follow precisely the analytic result in all three regimes . we further note that , in the regime of normal diffusion ( @xmath224 ) , both data sets in display consistent convergence to the same asymptotic regime , given by the first lines of equations and , @xmath225 . a similar result is observed in , where the convergence of the two data sets to the common leading value of equations and , @xmath226 , is apparent . the effect of the differing subleading terms , the one in negative , the other in positive ( the latter value is about one half the absolute value of the former for @xmath227 ) , is , however , manifest . finally , in the superdiffusive regime , @xmath228 , exhibits the two asymptotic regimes of the anomalously rescaled mean squared displacement given by equations and , @xmath229 ( all - scattering initial condition ) or @xmath230 ( stationary initial condition ) , whose values differ by a @xmath231 ratio for this value of the scaling parameter . 0.49 for the all - scattering initial condition . the scattering parameter takes values in or near the perturbative regime : ( a ) , ( c ) , ( e ) @xmath232 , and ( b ) , ( d ) , ( f ) @xmath233 . the data are compared with the asymptotic solution ( black dotted curves ) and the @xmath180 solution ( dashed curves ) . insets : time evolution of the fraction of particles in the scattering state with the stationary value ( dotted lines ) and the the @xmath180 solution ( dashed curves ) . , title="fig : " ] 0.49 for the all - scattering initial condition . the scattering parameter takes values in or near the perturbative regime : ( a ) , ( c ) , ( e ) @xmath232 , and ( b ) , ( d ) , ( f ) @xmath233 . the data are compared with the asymptotic solution ( black dotted curves ) and the @xmath180 solution ( dashed curves ) . insets : time evolution of the fraction of particles in the scattering state with the stationary value ( dotted lines ) and the the @xmath180 solution ( dashed curves ) . , title="fig : " ] 0.49 for the all - scattering initial condition . the scattering parameter takes values in or near the perturbative regime : ( a ) , ( c ) , ( e ) @xmath232 , and ( b ) , ( d ) , ( f ) @xmath233 . the data are compared with the asymptotic solution ( black dotted curves ) and the @xmath180 solution ( dashed curves ) . insets : time evolution of the fraction of particles in the scattering state with the stationary value ( dotted lines ) and the the @xmath180 solution ( dashed curves ) . , title="fig : " ] 0.49 for the all - scattering initial condition . the scattering parameter takes values in or near the perturbative regime : ( a ) , ( c ) , ( e ) @xmath232 , and ( b ) , ( d ) , ( f ) @xmath233 . the data are compared with the asymptotic solution ( black dotted curves ) and the @xmath180 solution ( dashed curves ) . insets : time evolution of the fraction of particles in the scattering state with the stationary value ( dotted lines ) and the the @xmath180 solution ( dashed curves ) . , title="fig : " ] 0.49 for the all - scattering initial condition . the scattering parameter takes values in or near the perturbative regime : ( a ) , ( c ) , ( e ) @xmath232 , and ( b ) , ( d ) , ( f ) @xmath233 . the data are compared with the asymptotic solution ( black dotted curves ) and the @xmath180 solution ( dashed curves ) . insets : time evolution of the fraction of particles in the scattering state with the stationary value ( dotted lines ) and the the @xmath180 solution ( dashed curves ) . , title="fig : " ] 0.49 for the all - scattering initial condition . the scattering parameter takes values in or near the perturbative regime : ( a ) , ( c ) , ( e ) @xmath232 , and ( b ) , ( d ) , ( f ) @xmath233 . the data are compared with the asymptotic solution ( black dotted curves ) and the @xmath180 solution ( dashed curves ) . insets : time evolution of the fraction of particles in the scattering state with the stationary value ( dotted lines ) and the the @xmath180 solution ( dashed curves ) . , title="fig : " ] the time evolution of the mean squared displacement in the perturbative regime of the scattering parameter , @xmath234 , is analyzed in for particles initially distributed in the scattering state at the origin , where the rescaled mean squared displacement is compared with the @xmath180 solution , equation . excellent agreement between the analytic and numerical results is observed throughout the time range when the parameter is small enough ( @xmath233 figures [ fig : msd_alphagt2_delta001 ] , [ fig : msd_alphaeq2_delta001 ] , [ fig : msd_alphalt2_delta001 ] ) . numerical computations are also consistent with the asymptotic regime for all cases . the scattering parameter value @xmath235 , shown in figures [ fig : msd_alphagt2_delta01 ] ( @xmath222 ) , [ fig : msd_alphaeq2_delta01 ] ( @xmath7 ) , and [ fig : msd_alphalt2_delta01 ] ( @xmath106 ) , is on the one hand small enough that , for short times , the computed mean squared displacement is barely distinguishable from the perturbative expansion . the effect of next order corrections is , on the other hand , apparent for larger times ( @xmath236 ) , beyond which a convergence to the asymptotic result is observed . we note that the crossover time is @xmath237 for @xmath232 ( ) and much larger yet for @xmath233 . for @xmath7 , subleading terms in equation therefore dominate the logarithmically divergent term throughout the time range of measurements . similarly , in the superdiffusive case , @xmath106 , the respective crossover times are @xmath238 ( ) and @xmath239 ( ) , such that subleading terms in equation remain important throughout the time range of measurements in both cases . we conclude this section by pointing out that the time range of numerical measurements such as presented in is limited by the finite precision of the random number generator used to draw the transition probabilities @xmath31 and sets an effective maximal scale of free flights , @xmath240 . this observation is analogous to the effect of machine - dependent limitations recently reported in reference @xcite and is particularly relevant to the regime of weak superdiffusion . the effective maximal scale @xmath240 induces a saturation of the logarithmic growth of the second moment for times larger than @xmath241 . the process would thus become diffusive for times sufficiently large . although this effect can be pushed upward to larger times by increasing the precision of the random number generator , it can not be eliminated altogether . the inclusion of exponentially - distributed waiting times separating the successive jump events of lvy walkers leads to a natural distinction between propagating and scattering states , whose respective concentrations evolve in time according to a set of generalized master equations . in particular , the fraction of walkers in a scattering state obeys a linear delay differential equation with a countable hierarchy of delays whose analytic solutions were obtained . as opposed to classic methods based on the fourier laplace transform of an integral kernel and the use of tauberian theorems @xcite , the approach presented in this paper is based solely on the solutions of such delay differential equations . our method thus yields a simple expression of the mean squared displacement of lvy walkers in terms of the distribution of free paths and the time integral of the fraction of particles in the scattering state . both exact and asymptotic expressions of the mean squared displacement of walkers were obtained in regimes ranging from normal to superdiffusive subballistic transport . in these regimes , the mechanism through which a walker can change directions between successive propagation phases plays an important role in determining the values of the transport coefficients , whether normal or anomalous . the transition through a scattering phase brings about a description in terms of two parameters , the first specifying the typical timescale of scattering events as opposed to the timescale of propagation across an elementary cell , the second weighting the probability of transitions between scattering and propagating states . relying on the stationary fractions of particles in scattering and propagating states , a detailed derivation of the transport coefficients , similar to that reported in @xcite , was given , exhibiting the precise effect of the two scattering parameters on these coefficients , as well as the influence of the initial distribution of walkers @xcite . our formalism also yields the exact time evolution of the mean squared displacement , which was investigated when particles are initially found in a scattering state . the comparison between the exact and asymptotic solutions is generally not immediate , but is fairly straightforward in perturbative regimes such that the likelihood of a transition from scattering to propagating states is small . a case in point though by far not the only application of the present results is given by the infinite - horizon lorentz gas @xcite , for which the scaling parameter of the distribution of free paths takes on the marginal value @xmath7 . as one might expect , measuring the logarithmic divergence in time of the rescaled mean squared displacement is typically hindered by dominant constant terms . this is particularly so in the regime of narrow corridors , which yields a stochastic description in terms of multistate lvy walks such that the scattering parameter is small , @xmath112 @xcite . the parameter @xmath38 induces a further separation between the time scales of the scattering and propagating phases , @xmath242 . to accurately model the scattering phase is therefore of paramount importance to understanding the dynamics of the infinite - horizon lorentz gas in terms of a lvy walk . the results obtained in this paper provide the framework to transposing such results to a range of applications exhibiting other regimes of transport , such as have been studied in the context of optimal intermittent search strategies @xcite . our theoretical framework brings about new perspectives to extend the study of intermittent walks to power - law distributed ballistic phases , which may be relevant to the increasing body of literature on optimal search strategies @xcite . this work was partially supported by firb - project no . rbfr08uh60 ( miur , italy ) , by sep - conacyt grant no . cb-101246 and dgapa - unam papiit grant no . in117214 ( mexico ) , and by frfc convention 2,4592.11 ( belgium ) . is financially supported by the ( belgian ) frs - fnrs . to derive equation , we note that @xmath243 and , for @xmath244 , @xmath245 by recursive application of this formula , we obtain @xmath246 which is equivalent to equation . taking the derivative of equation , we verify equation : @xmath247.\end{aligned}\ ] ] we have to show that the last term on the rhs vanishes , which amounts to proving the identity @xmath248 = 0 , \label{eq : id1}\end{aligned}\ ] ] or , after rearrangement , @xmath249 the second term in this expression , by successively changing the index @xmath18 to @xmath250 , then swapping the sums over @xmath20 and @xmath18 and exchanging the indices @xmath18 and @xmath20 , transforms to @xmath251 now swapping the sums over @xmath18 and @xmath101 in the last line and plugging this expression back into equation , we obtain , after factorization of the common factors , the condition @xmath252 { \tau_\mathrm{r}}^{-n } ( t - k { \tau_\mathrm{b}})^{n } = 0 , \label{eq : s0.id}\ ] ] which holds by identity , thus completing the proof that @xmath73 as specified by equation is the solution to equation for the initial condition .

we solve and generalize an open problem posted by james propp ( problem 16 in _ new perspectives in geometric combinatorics _ , cambridge university press , 1999 ) on the number of tilings of quasi - hexagonal regions on the square lattice with every third diagonal drawn in . we also obtain a generalization of douglas theorem on the number of tilings of a family of regions of the square lattice with every second diagonal drawn in .

1309.5376

the field of exact enumeration of tilings ( equivalently , perfect matchings ) dates back to the early 1900 s when macmahon proved his classical theorem on the number of plane partitions that fit in a given box ( see @xcite ) . this theorem is equivalent to the fact that the number of unit rhombus ( or lozenge ) tilings of a hexagon @xmath0 of side - lengths @xmath1 , @xmath2 , @xmath3 , @xmath1 , @xmath2 , @xmath3 ( in cyclic order ) drawn on the triangular lattice is equal to @xmath4 ( we use the operator @xmath5 to denote both the number of tilings of a lattice region and the number of perfect matchings of a graph , as the two objects can be identified by a well known bijection ) . another classical result , from the early 1960 s , is the enumeration of domino tilings of a rectangle on the square lattice , due independently to kasteleyn @xcite and temperley and fisher @xcite . this states that the number of domino tilings of a @xmath6 by @xmath7 rectangle on the square lattice equals @xmath8 in the early 1990 s , elkies , kuperberg , larsen and propp @xcite considered another family of simple regions on the square lattice called aztec diamonds ( see figure [ aztecdiamond ] for an example ) , and proved that the number of domino tilings of the aztec diamond of order @xmath9 is given by the simple formula @xmath10 a large body of related work followed ( see e.g. @xcite , @xcite , @xcite , @xcite , @xcite , @xcite , and the references in @xcite for a more extensive list ) , centered on families of lattice regions whose tilings are enumerated by simple product formulas . the state of affairs at the end of that decade is captured by propp s paper @xcite published in 1999 , which presented a list of 32 open problems in the field of enumeration of perfect matchings . most of those 32 problems have been solved in the meanwhile , but some are still open . in this paper we solve and generalize one of these open problems ( problem 16 on propp s list @xcite ) . our methods also provide a new proof and a generalization for a related result of douglas @xcite . problem 16 on propp s list @xcite concerns a family of quasi - hexagonal regions on the lattice obtained from the square lattice by drawing in every third southwest - to - northeast diagonal . the case when the side - lengths of the quasi - hexagon are @xmath11 , @xmath12 , @xmath12 , @xmath11 , @xmath12 , @xmath12 ( clockwise from top ) is illustrated in figure [ hexagonp ] . problem 16 of @xcite asks for a formula for the number of tilings of the quasi - hexagon of sides @xmath1 , @xmath2 , @xmath3 , @xmath1 , @xmath2 , @xmath3 ( the sides of length @xmath1 are the ones along _ _ diagonals _ _ of the square lattice ) , where the allowed tiles are unions of two fundamental regions of the resulting dissection of the square lattice sharing an edge . ( 0,0 ) # 1#2#3#4#5 @font ( 4289,2665)(222,-1852 ) ( 237,485)(0,0)[lb ] ( 2009,603)(0,0)[lb ] ( 2836,-224)(0,0)[lb ] ( 2481,-1523)(0,0)[lb ] ( 1182,-1759)(0,0)[lb ] ( 237,-933)(0,0)[lb ] as mentioned in @xcite , the case @xmath13 ( to which also the cases @xmath14 and , by symmetry , @xmath15 turn out to reduce ) has been solved by ben wieland ( in unpublished work ) , who showed that the number of tilings is given in these situations by powers of 2 . in this paper we prove a simple product formula in the case @xmath16 , which was previously open . the result turns out to fully justify propp s initial motivation , stated in @xcite : `` one reason for my special interest in problem 16 is that it seems to be a genuine hybrid of domino tilings of aztec diamonds and lozenge tilings of hexagons . '' indeed , we show that by a sequence of graph transformations which can also be used to prove the aztec diamond theorem ( [ diamond ] ) the dual graph can be transformed into a honeycomb graph whose number of perfect matchings is given by macmahon s formula ( [ mcmahon ] ) . we in fact enumerate the tilings of more general regions , which we describe next . rather than considering quasi - hexagonal regions on the square lattice on which diagonals are drawn in so that the distance . ] between any two successive ones is 3 , we allow the general situation when the distances between successive diagonals are _ arbitrary_. let @xmath17 be a fixed diagonal ( this will contain the middle two vertices of the quasi - hexagon ) , and assume that @xmath18 diagonals have been drawn in above it , with the distances between successive ones , starting from top , being @xmath19 ( @xmath20 is the distance between the bottommost of these @xmath18 diagonals and @xmath17 ) . assume also that @xmath21 diagonals have been drawn in below @xmath17 , with the distances between successive ones , starting from the bottom , being @xmath22 ( @xmath23 is the distance between the topmost of these @xmath21 diagonals and @xmath17 ) . given a positive integer @xmath1 , we define the quasi - hexagonal region @xmath24 as follows ( see figure [ hexagonregion ] for an example ) . its top and bottom boundaries are along the top and bottom diagonals that have been drawn in . its southwestern and northeastern boundaries are defined in the next two paragraphs . color the resulting dissection of the square lattice black and white so that any two fundamental regions that share an edge have opposite color , and assume that the triangles just below the top diagonal are white . let @xmath25 be a lattice point on the top diagonal . start from @xmath25 and take unit steps south or east so that for each step the color of the fundamental region on the left is black . we arrive @xmath17 at a lattice point @xmath26 . continue downward from @xmath26 in similar fashion , with the one difference that now we require the fundamental region on the left to be white for each step . let @xmath27 be the lattice point on the bottom diagonal that is reached by these steps . the described path from @xmath25 to @xmath27 is the southwestern boundary of our region . ( 0,0),title="fig : " ] # 1#2#3#4#5 @font ( 5276,5169)(406,-4616 ) ( 421,-1200)(0,0)[lb ] ( 1350,-2900)(0,0)[lb ] ( 3200,-4141)(0,0)[lb ] ( 4500,-2986)(0,0)[lb ] ( 1500,0)(0,0)[lb ] ( 3900,-1329)(0,0)[lb ] ( 1239,-586)(0,0)[lb ] ( 4300,-646)(0,0)[lb ] ( 3025,131)(0,0)[lb ] ( 3400,-283)(0,0)[lb ] ( 3500,-706)(0,0)[lb ] ( 2500,-4231)(0,0)[lb ] ( 1800,-3700)(0,0)[lb ] the northeastern boundary is defined analogously , starting from the lattice point @xmath28 on the top diagonal that is @xmath1 unit square diagonals to the northeast of @xmath25 ( i.e. @xmath29 ) . take unit steps south or east so that all fundamental regions on the left are white , until a point @xmath30 on @xmath17 is reached , and continue with similar unit steps so that the fundamental regions on the left are black , until a point @xmath31 on the bottom diagonal is reached . the described path from @xmath28 to @xmath31 completes the boundary of @xmath32 . we are only interested in connected regions , so we assume in addition that _ the northeastern and the southwestern boundaries do not intersect each other . _ our generalization of problem 16 of @xcite consists of providing an explicit formula for the number of tilings of such regions . we need some additional definitions and terminology as below . call the fundamental regions inside @xmath32 _ cells _ , and call them black or white according to the coloring described above . note that there are two kinds of cells , square and triangular . the latter in turn come in two orientations : they may point towards @xmath17 or away from @xmath17 . a cell is called _ regular _ if it is a square cell or a triangular cell pointing away from @xmath17 . a _ row of cells _ consists of all the triangular cells of a given color with bases resting on a fixed lattice diagonal , or consists of all the square cells ( of a given color ) passed through by a fixed lattice diagonal . define the _ upper height _ of our region to be the number of rows of regular black cells above @xmath17 . the _ lower height _ is the number of rows of white regular cells below @xmath17 ( for example , the quasi - hexagon in figure [ hexagonregion ] has both the upper and lower heights equal to 6 ) . we are now ready to state our main result . [ main ] let @xmath1 be a positive integer , and let @xmath33 be positive integers . @xmath36 unless the upper and lower heights are equal and the bottom row of cells is black . @xmath34b@xmath35 . suppose that the upper and lower heights of @xmath32 are both equal to @xmath37 , and that the bottom row of cells is black . let @xmath27 be the number of regular black cells above @xmath17 , and @xmath38 the number of regular white cells below @xmath17 . then for any positive integer @xmath1 we have @xmath39 where @xmath40 is the number of rows of black triangular cells above @xmath17 pointing away from @xmath17 , and @xmath9 is the number of rows of black triangular cells above @xmath17 pointing towards @xmath17 . by macmahon s formula ( [ mcmahon ] ) and equality ( [ maineq ] ) , the number of tilings of the quasi - hexagon is equal to @xmath41 when @xmath42 . note that the original problem corresponds to @xmath43 and @xmath44 . the resulting region is then precisely the quasi - hexagon of sides @xmath1 , @xmath18 , @xmath18 , @xmath1 , @xmath18 , @xmath18 . in particular , for @xmath45 and @xmath46 , ( [ maineq ] ) states that the number of tilings of the quasi - hexagon shown in figure [ hexagonp ] is equal to @xmath47 , thus explaining how this factorization ( mentioned explicitly in @xcite ) comes about . in the case of odd @xmath48 s and @xmath49 s , our quasi - hexagons have the following nice facts : the bottom row of cells is always black , all the black cells above @xmath17 and all the white cells below @xmath17 are regular , and the upper and lower heights are equal to @xmath50 and @xmath51 , respectively . therefore , we have a special form of theorem [ maineq ] as follows . [ odd - main ] let @xmath19 and @xmath22 be odd positive integers . assume that @xmath52 let @xmath27 be the number of black cells above @xmath17 , and @xmath38 be the number of white cells below @xmath17 . then for any positive integer @xmath1 we have @xmath53 note that when @xmath43 in the original quasi - hexagons , the balancing condition ( [ balancing ] ) is equivalent to @xmath44 . as mentioned in @xcite , we do not have a simple product formula for the number of tilings of the quasi - hexagon of problem 16 in @xcite when @xmath1 , @xmath2 , @xmath3 are all distinct . however , we show that our graph transformations can still be used to turn the dual graph of the quasi - hexagon into a conceptually simple graph consisting of a honeycomb part and an aztec rectangle part , and therefore the number of perfect matchings of the former is equal to a power of @xmath12 times the number of perfect matchings of the latter ( see theorems [ asymodd ] and [ asymmetric1 ] ) . ( 0,0 ) , @xmath54 and @xmath55.,title="fig : " ] # 1#2#3#4#5 @font ( 6283,2966)(402,-2292 ) ( 2363,-1641)(0,0)[lb ] ( 6260,-1878)(0,0)[lb ] ( 828,-696)(0,0)[lb ] we also extend a theorem of douglas @xcite concerning a particular family of regions on the lattice obtained from @xmath56 by drawing in every _ second _ diagonal ( see figure [ douglas ] ) , to the case of arbitrary distances between the diagonals . let @xmath57 be the portion of @xmath32 that is above the diagonal @xmath17 ( i.e. the top half of the generalized quasi - hexagon ) ; we call it a _ generalized douglas region_. the _ height _ of @xmath57 is defined to be equal to the upper height of @xmath32 . a triangle pointing towards @xmath17 in @xmath57 is called a _ down - pointing triangle _ , and a triangle pointing away from @xmath17 in this region is called a _ up - pointing triangle_. our extension of douglas result is the following . [ douggen ] let @xmath1 be a positive integer , and let @xmath19 be positive integers . let @xmath37 be the height of @xmath57 , @xmath40 ( resp . , @xmath9 ) be the number of rows of up - pointing ( resp . , down - pointing ) triangular black cells in @xmath57 , and let @xmath27 be the number of black cells that are not down - pointing triangles . @xmath34a@xmath35 . if the the bottom row of cells of @xmath57 is black , or if the equality @xmath58 fails , then @xmath59 . @xmath34b@xmath35 . suppose ( [ balancing2 ] ) holds and the bottom row of cells of @xmath57 is white . then we have @xmath60 douglas theorem corresponds to the special case when @xmath18 is odd and @xmath61 , @xmath62 , when the balancing condition ( [ balancing2 ] ) requires @xmath63 ( strictly speaking , in douglas original definition the top and bottom rows of cells consist of squares rather than triangles ; however , the dual graph is isomorphic to the dual graph of our region , so the two versions have the same number of tilings ) . thus , one can view theorem [ main ] as a common generalization of douglas theorem @xcite and propp s open problem on quasi - hexagons . a _ perfect matching _ of a graph @xmath64 is a collection of edges such that each vertex of @xmath64 is incident to precisely one edge in the collection . the number of perfect matchings of @xmath64 is denoted by @xmath65 . more generally , if the edges of @xmath64 have weights on them , @xmath65 denotes the sum of the weights of all perfect matchings of @xmath64 , where the weight of a perfect matching is the product of the weights on its constituent edges . given a lattice in the plane , a ( lattice ) _ region _ is a finite connected union of fundamental regions of that lattice . tile _ is the union of any two fundamental regions sharing an edge . a _ tiling _ of the region @xmath66 is a covering of @xmath66 by tiles with no gaps or overlaps . the tilings of a region @xmath66 can be naturally identified with the perfect matchings of its dual graph ( i.e. , the graph whose vertices are the fundamental regions of @xmath66 , and whose edges connect two fundamental regions precisely when they share an edge ) . in view of this , we denote the number of tilings of @xmath66 by @xmath67 . a _ forced edge _ of a graph is an edge contained in every perfect matching of @xmath64 . assume that @xmath64 is a weighted graph with weight function @xmath68 on its edges , and @xmath69 is obtained from @xmath64 by removing forced edges @xmath70 , and removing the vertices incident to those edges . then one clearly has @xmath71 _ hereafter , whenever we remove some forced edges , we remove also the vertices incident to them . _ the main results of this section are two graph transformation rules that change the number of perfect matchings in a simple predictable way ( see lemmas 3.4 and 3.5 ) . in proving them we will employ three basic preliminary results stated below . ( 0,0 ) # 1#2#3#4#5 @font ( 4188,1361)(593,-556 ) ( 1336,591)(0,0)[lb ] ( 3549,621)(0,0)[lb ] ( 3757 , 89)(0,0)[lb ] ( 3999,621)(0,0)[lb ] ( 820,-541)(0,0)[lb ] ( 1840,-535)(0,0)[lb ] ( 3031,-535)(0,0)[lb ] ( 4426,-484)(0,0)[lb ] [ vertex - splitting lemma][vs ] let @xmath64 be a graph , @xmath72 be a vertex of it , and denote the set of neighbors of @xmath72 by @xmath73 . for any disjoint union @xmath74 , let @xmath69 be the graph obtained from @xmath75 by including three new vertices @xmath76 , @xmath77 and @xmath78 so that @xmath79 , @xmath80 , and @xmath81 ( see figure [ vertexsplitting ] ) . then @xmath82 . [ star ] let @xmath64 be a weighted graph , and let @xmath72 be a vertex of @xmath64 . let @xmath69 be the graph obtained from @xmath64 by multiplying the weights of all edges incident to @xmath72 by @xmath83 . then @xmath84 . part ( a ) of the following result is a generalization ( due to propp ) of the urban renewal " trick first observed by kuperberg . parts ( b ) and ( c ) are due to ciucu ( see lemma 2.6 in [ 5 ] ) . ( 0,0 ) # 1#2#3#4#5 @font ( 4054,1735)(340,-948 ) ( 355,-156)(0,0)[lb ] ( 1064,-933)(0,0)[lb ] ( 1891,-106)(0,0)[lb ] ( 1182,603)(0,0)[lb ] ( 2717,-189)(0,0)[lb ] ( 3426,-933)(0,0)[lb ] ( 4253,-106)(0,0)[lb ] ( 3426,603)(0,0)[lb ] ( 904,-382)(0,0)[lb ] ( 1396,-388)(0,0)[lb ] ( 1418,130)(0,0)[lb ] ( 946,130)(0,0)[lb ] ( 2968,284)(0,0)[lb ] ( 3934,311)(0,0)[lb ] ( 3964,-544)(0,0)[lb ] ( 2965,-526)(0,0)[lb ] ( 2197,-817)(0,0)[lb ] [ spider lemma][spider ] ( a ) let @xmath64 be a weighted graph containing the subgraph @xmath85 shown on the left in figure [ spider1 ] ( the labels indicate weights , unlabeled edges have weight 1 ) . suppose in addition that the four inner black vertices in the subgraph @xmath85 , different from @xmath86 , have no neighbors outside @xmath85 . let @xmath69 be the graph obtained from @xmath64 by replacing @xmath85 by the graph @xmath87 shown on right in figure [ spider1 ] , where the dashed lines indicate new edges , weighted as shown . then @xmath88 . \(b ) consider the above local replacement operation when @xmath85 and @xmath87 are graphs shown in figure [ spider2](a ) with the indicated weights ( in particular , @xmath89 has a new vertex @xmath31 , that is incident only to @xmath25 and @xmath27 ) . then @xmath90 . \(c ) the statement of part ( b ) is also true when @xmath85 and @xmath87 are the graphs indicated in figure [ spider2](b ) ( in this case @xmath69 has two new vertices @xmath27 and @xmath31 , they are adjacent only to one another and to @xmath26 and @xmath25 , respectively ) . , ( b ) @xmath91 , ( c ) @xmath92 , and ( d ) @xmath93,width=340 ] the following four families of graphs will play a special role in our proofs . consider a @xmath94 rectangular chessboard @xmath26 and suppose the corners are black . aztec rectangle _ @xmath95 is the graph whose vertices are the white squares and whose edges connect precisely those pairs of white squares that are diagonally adjacent ( the case @xmath96 , @xmath97 is shown in figure [ arvariants](a ) ) . if one removes the bottom row of the board @xmath26 and then applies the same procedure , the resulting graph is denoted by @xmath98 , and called a _ baseless aztec rectangle _ ( see figure [ arvariants](c ) for an illustration ) the _ combed aztec diamond _ @xmath99 is the graph obtained from @xmath95 by appending a vertical edge to each of its bottom vertices ( an example is shown in figure [ arvariants](b ) ) . the _ combed baseless aztec rectangle _ @xmath100 is the graph obtained from @xmath98 by appending a vertical edge to each of its bottom vertices ( see figure [ arvariants](d ) for an example ) . the _ connected sum _ @xmath101 of two disjoint graphs @xmath64 and @xmath69 along the ordered sets of vertices @xmath102 and @xmath103 is the graph obtained from @xmath64 and @xmath69 by identifying vertices @xmath104 and @xmath105 , for @xmath106 . [ t1 ] let @xmath64 be a graph and let @xmath107 be an ordered subset of its vertices . then @xmath108 where the connected sum acts on @xmath64 along @xmath107 , and on @xmath109 and @xmath110 along their bottom @xmath111 vertices ( ordered from left to right ) . for @xmath112 , the above transformation is illustrated in figures [ lemmat1](a ) and ( e ) . let @xmath113 be the graph obtained from @xmath114 by applying the vertex - splitting lemma at all vertices of @xmath115 that are not @xmath116 and not adjacent to any of them ( figures [ lemmat1](a ) and ( b ) illustrate this for the case + @xmath117 and @xmath118 ) . apply the spider lemma around all @xmath119 shaded diamond in the graph @xmath113 , and remove the @xmath120 edges adjacent to a vertex of degree 1 , which are forced edges ( see figure [ lemmat1](c ) , the forced edges are the circled ones ) . the resulting graph is isomorphic to @xmath121 , where @xmath122 is the graph obtained from @xmath123 by changing all the weights of edges to @xmath124 ( see figure [ lemmat1](d ) ; dotted edges have weight @xmath124 ) . by lemmas [ vs ] and [ spider ] , we have @xmath125 applying the star lemma with factor @xmath126 to all @xmath127 shaded vertices of the graph @xmath122 ( see figure [ lemmat1](d ) ) , the graph @xmath121 is turned into @xmath128 . by ( [ t1eq2 ] ) and lemma [ star ] , we obtain @xmath129 which proves ( [ t1eq1 ] ) . [ t2 ] let @xmath64 be a graph and let @xmath130 be an ordered subset of its vertices . then @xmath131 where the connected sum acts on @xmath64 along @xmath130 , and on @xmath132 and @xmath133 along their bottom @xmath134 vertices ordered from left to right ( see figures [ lemmat2](a ) and ( e ) , for @xmath135 ) . one can prove lemma [ t2 ] similarly to lemma [ t1 ] by using vertex - splitting lemma , spider lemma and star lemma to transform the graph on the left hand side of ( [ t2eq1 ] ) into the graph on the right hand side . this transforming process is illustrated by figure [ lemmat2 ] , for @xmath135 . an _ induced subgraph _ of a graph @xmath64 is a graph whose vertex set is a subset @xmath136 of the vertex set of @xmath64 , and whose edge set consists of all edges of @xmath64 with endpoints in @xmath136 . for a graph @xmath64 , we denote by @xmath137 its vertex set . [ graphsplitting ] let @xmath64 be a bipartite graph , and let @xmath138 and @xmath139 be the two vertex classes . let @xmath140 be an induced subgraph of @xmath64 . \(a ) assume that @xmath140 satisfies the following two conditions . 1 . the _ separating condition _ : there are no edges of @xmath64 connecting a vertex in @xmath141 and a vertex in @xmath142 , 2 . the _ balancing condition _ : @xmath143 . then @xmath144 \(b ) if @xmath140 satisfies the separating condition and but @xmath145 , then @xmath146 . color all vertices of @xmath138 white , and all vertices of @xmath139 black . to prove part ( a ) of the lemma , it suffices to show that that none of the perfect matchings of @xmath64 contain any edge connecting a vertex of @xmath140 to a vertex of @xmath147 . since there are no edges of @xmath64 connecting a white vertex of @xmath140 to a vertex of @xmath147 ( by the separating condition ) , the previous assertion follows by the balancing condition . part ( b ) is perfectly analogous , the only difference is that @xmath140 has no tilings when @xmath148 the proof of theorem 2.3 and as we will see in the next section , also the proof of theorem 2.2 is based on a common generalization of the transformations @xmath149 and @xmath150 of lemmas [ t1 ] and [ t2 ] , presented below in proposition [ composite ] . the diagonals divide @xmath57 into @xmath18 parts called _ layers_. recall that @xmath40 and @xmath9 are the numbers of rows of triangular black cells of @xmath57 that point up and down , respectively . as one moves down from top , each row of black up - pointing triangles causes the length of the rows of black cells in the next layer to go up by one , and each row of black down - pointing triangles causes the length of the rows of black cells in that layer to be one less than in the previous layer . this implies that the bottom row of cells of @xmath57 consists of @xmath151 cells ( note that these are always triangular and up - pointing , but they may be either black or white ) . the _ width of a layer _ is one less than the length of its rows of blacks cells . let @xmath152 be the dual graph of the region @xmath57 . , ( b ) @xmath153 , ( c ) @xmath154 , and ( d ) @xmath155.,width=453 ] [ composite ] let @xmath1 and @xmath19 be positive integers . recall that @xmath27 is the number of black cells of @xmath57 that are either squares or up - pointing triangles , and @xmath37 is the number of rows of black cells consisting of squares or up - pointing triangles . let @xmath64 be a graph , and consider an ordered subset of its vertices @xmath156 . \(a ) if the bottom row of cells in @xmath57 is white , then @xmath157 where the connected sum acts on @xmath64 along @xmath156 , and on the other two summands along their bottom @xmath151 vertices ( ordered from left to right ) . see figures [ mass](c ) and ( d ) for an example , the white circles indicate the vertices @xmath158 @xmath159 . \(b ) if the bottom row of cells in @xmath57 is black , then @xmath160 where the connected sum acts on @xmath64 along @xmath156 , and on the other two summands along their bottom @xmath151 vertices ( ordered from left to right ) . see figures [ mass](a ) and ( b ) for an example . and ( b ) @xmath161 ; and their dual graphs ( c ) @xmath162 and ( d ) @xmath163.,width=453 ] \(a ) we prove ( [ composeq1 ] ) by induction on @xmath18 . for @xmath164 , the dual graph @xmath152 of the region is just the aztec rectangle @xmath165 , and @xmath166 . in particular , the number @xmath27 of black cells is equal to @xmath167 , and ( [ composeq1 ] ) holds . for the induction step , suppose @xmath168 and assume ( [ composeq1 ] ) holds for all regions with strictly less than @xmath18 _ layers _ ( the layers of @xmath152 correspond to the layers of @xmath57 , which are the portions of the latter between consecutive diagonals ) . note that the top layer of @xmath152 is either an aztec rectangle or a baseless aztec rectangle , based on whether @xmath169 is even or odd , respectively . we treat the two cases separately . suppose that the top layer of @xmath152 is the aztec rectangle @xmath170 , i.e. @xmath169 is even and @xmath171 . applying the transformation @xmath149 of lemma [ t1 ] with respect to this layer , we obtain @xmath172 where @xmath85 is the graph obtained from @xmath152 by replacing its top part , isomorphic to the combed aztec rectangle @xmath173 , by the baseless aztec rectangle @xmath174 . note that the graph @xmath85 is also the dual of a generalized douglas region @xmath175 with white bottom row of cells , but the number of layers of @xmath85 is @xmath176 ( see figures [ compare1](c ) and ( d ) for an example ; the subgraph above the dotted line in graph ( c ) is replaced by the subgraph above that line in graph ( d ) ) . in precise , @xmath177 . note that @xmath178 is the width of the second layer in @xmath179 , so @xmath180 . therefore , by the induction hypothesis , we have @xmath181 where the primed symbols refer to the region @xmath175 and denote quantities corresponding to their unprimed counterparts in @xmath57 . it is clear by definition of @xmath175 that @xmath182 . two regions @xmath57 and @xmath175 are different only in their top parts , and their bottom parts are the same ( see figures [ compare1](a ) and ( b ) ; the parts below the dotted line in two regions are the same ) . the number of rows of black up - pointing triangles is the same in @xmath57 and @xmath175 , but the latter has one fewer rows of black down - pointing triangles ( because the former region has such a row just below its top layer , and that is no longer present in @xmath175 ) . thus @xmath183 , and @xmath184 . moreover , the numbers of rows of black squares in @xmath185 and @xmath175 are equal , so we get @xmath186 . finally , one easily sees that @xmath187 . therefore , we obtain @xmath188 then , by ( [ composeq3])([composeq5 ] ) , we have @xmath189 which completes the induction step in this case . and ( b ) @xmath190 ; and their dual graphs ( c ) @xmath163 and ( d ) @xmath191.,width=453 ] suppose now that the top layer of @xmath152 is isomorphic to the baseless aztec rectangle @xmath192 , i.e. @xmath169 is odd and @xmath193 . applying this time the transformation @xmath150 of lemma [ t2 ] with respect to this layer , we obtain that @xmath194 where @xmath85 is now the graph obtained from @xmath152 by replacing its top part , isomorphic to the combed baseless aztec rectangle @xmath195 , by the aztec rectangle @xmath196 ( see figures [ compare2](c ) and ( d ) for an example ) . again , @xmath85 is the dual of a generalized douglas region @xmath197 of the same kind as @xmath57 ( i.e. , with a white bottom row of cells ) , but with fewer layers . therefore , by the induction hypothesis we have again the equality ( [ composeq4 ] ) . again , it is clear that @xmath198 , and two regions @xmath57 and @xmath175 are different only in their top layers ( see figures [ compare2](a ) and ( b ) for an example ) . one can verify that @xmath199 and @xmath200 in this case . moreover , the number of rows of black squares is now one fewer in @xmath185 than in @xmath175 , thus we still have @xmath186 . finally , one readily sees that @xmath201 equals now @xmath202 . this yields @xmath203 which in turn , by ( [ composeq7 ] ) , ( [ composeq4 ] ) and ( [ composeq9 ] ) , implies @xmath204 this completes the induction step in the second case , and thus also the proof of ( [ composeq1 ] ) . part ( b ) is proved by a perfectly analogous argument . the enumeration of tilings of the region @xmath57 follows easily using the above result . apply proposition [ composite ] with @xmath64 chosen to be the graph consisting of @xmath151 vertices and no edges . if the bottom row of cells in @xmath57 is black , then ( [ composeq2 ] ) implies that @xmath205 since a baseless aztec rectangle has no perfect matchings ( because it is impossible to cover all its bottom vertices by disjoint edges ) , it follows that in this case @xmath206 . assume therefore that the bottom row of cells in @xmath57 is white . then with @xmath64 chosen like above , ( [ composeq1 ] ) implies that @xmath207 it follows thus that unless @xmath208 , one again has @xmath209 this proves part ( a ) of the theorem . part ( b ) follows directly from ( [ dougeq2 ] ) and the aztec diamond theorem ( [ diamond ] ) . our proof uses the composite transformations of proposition [ composite ] to transform the dual graph of a generalized quasi - hexagon into a honeycomb graph whose perfect matchings are enumerated by macmahon s formula ( [ mcmahon ] ) . ( 0,0 ) ( left ) and @xmath210 ( right),title="fig : " ] # 1#2#3#4#5 @font ( 5994,3441)(913,-4120 ) ( 928,-1906)(0,0)[lb ] ( 1564,-3448)(0,0)[lb ] ( 3229,-4105)(0,0)[lb ] ( 3889,-3460)(0,0)[lb ] ( 3142,-1831)(0,0)[lb ] ( 1762,-1183)(0,0)[lb ] ( 3913,-1894)(0,0)[lb ] ( 4660,-3088)(0,0)[lb ] ( 6037,-3634)(0,0)[lb ] ( 6742,-2923)(0,0)[lb ] ( 6097,-1642)(0,0)[lb ] ( 4843,-889)(0,0)[lb ] and ( b ) @xmath211.,width=302 ] [ proof of theorem [ main ] . ] first , we consider the case where the bottom row of cells is white . then we have the same situation when showing that a baseless aztec rectangle does not have a perfect matching . it is also impossible to cover the bottom vertices in the dual graph by disjoint edges . therefore the dual graph has no perfect matchings , i.e. the region has no tilings . hereafter , we assume that the bottom row of cells of our region is black . there are two cases to distinguish , depending on whether the row of triangular cells just above @xmath17 and with bases resting on @xmath17 is black or white ( these two cases are illustrated in figure [ conhex ] ; the corresponding dual graphs are shown in figure [ conhex2 ] ) . suppose the row of cells just above @xmath17 consists of black triangles ( note that this is the case for our motivating problem ; see figure [ hexagonp ] ) . denote by @xmath212 the dual graph of the generalized quasi - hexagon @xmath213 apply the composite transformation in proposition [ composite](b ) separately to the portions of @xmath214 corresponding to the parts above and below @xmath17 of the region @xmath140 ( which we will call the _ upper _ and _ lower parts _ of the dual graph ; illustrated in figure [ conhex2 ] as the parts above and below the dotted lines ) . we obtain that @xmath215 where the connected sum acts along the bottom vertices of the two modified aztec rectangles ( ordered from left to right ) , and @xmath216 , @xmath217 , @xmath218 and @xmath219 are quantities having the same significance for the lower part of @xmath214 as their unprimed correspondents have for its upper part ( except that due to the geometry of @xmath32 and the resulting coloring of its fundamental regions they count _ white _ cells ) . the transformation involved in ( [ maineq1 ] ) is illustrated in figures [ symmetricfig](a ) and ( b ) . as we have seen in the second paragraph of section 4 , @xmath151 is the number of triangular cells above @xmath17 with bases resting on @xmath17 . by the same argument , @xmath220 is the number of triangular cells below @xmath17 , with bases resting on @xmath17 . these two numbers are equal by construction , and therefore we have @xmath221 clearly , a necessary condition for the graph on the right hand side of ( [ maineq1 ] ) ( which is a bipartite graph ) to have a perfect matching is that the numbers of its black and white vertices are the same . denote for brevity @xmath222 . then its upper part has @xmath223 white and @xmath224 black vertices , while its lower part has @xmath225 black and @xmath226 white vertices . the balancing condition is then @xmath227 which is equivalent to @xmath186 . this proves in particular the rest of part ( a ) of the theorem . to prove part ( b ) , we consider first the case when @xmath228 . the key idea here is to realize that the composite transformation of proposition [ composite](b ) can also be used to turn a honeycomb graph into the graph on the right hand side of ( [ maineq1 ] ) . more precisely , consider the honeycomb graph of sides @xmath229 ( clockwise from top ) , illustrated in figure [ symmetricfig](c ) . note that this is the dual graph of a generalized quasi - hexagon , in which all the inter - diagonal distances @xmath48 s and @xmath49 s are equal to one ( i.e. the lozenge hexagonal region of the same sides ) ! therefore equality ( [ maineq1 ] ) applies in this case , and the question is what is exactly its form in this particular case ( see figures [ symmetricfig ] ( c ) and ( b ) ) . by symmetry , the primed and unprimed quantities of @xmath27- , @xmath37- , @xmath1- , @xmath40- and @xmath9-type are equal . furthermore , one readily sees that the number of black regular cells above @xmath17 is @xmath230 the upper height is equal to @xmath37 and so is the @xmath40-parameter , while the @xmath9-parameter is 0 . the @xmath1-parameter equals @xmath231 . it follows that the particular form ( [ maineq1 ] ) takes is @xmath232 equations ( [ maineq1 ] ) ( in which @xmath233 by ( [ maineq2 ] ) , and @xmath234 by assumption ) and ( [ maineq3 ] ) imply the statement of part ( b ) of the theorem , provided @xmath235 . ( 0,0 ) # 1#2#3#4#5 @font ( 1917,2321)(3057,-2614 ) ( 3000,-815)(0,0)[lb ] ( 3000,-2114)(0,0)[lb ] note that if @xmath236 , then the lozenge hexagonal region of side - lengths @xmath237 is _ not _ a member of our family of quasi - hexagons anymore ( in the definition of a quasi - hexagon , we always assume that @xmath25 and @xmath28 are two distinct lattice vertices ) ; and if @xmath238 , then the lozenge hexagonal region with side - lengths above does not exist ( two side - lengths are negative ) . this means that the method used to prove ( [ maineq ] ) when @xmath228 does _ not _ work for the case when @xmath239 . to complete the proof of ( [ maineq ] ) , we need to show that the graph @xmath240 has @xmath241 perfect matchings when @xmath236 , and that it has no perfect matchings when @xmath238 . the top @xmath242 rows of its vertices induce a subgraph isomorphic to @xmath243 , and the bottom @xmath242 rows of its vertices induce also a subgraph isomorphic to @xmath243 . denote by @xmath244 and @xmath245 the two aztec rectangles ( illustrated by the subgraph above and the subgraph below two dotted lines in figure [ ben1s ] ) . if @xmath238 , then @xmath244 satisfies the conditions in graph - splitting lemma [ graphsplitting](b ) as an induced subgraph of @xmath246 , and thus @xmath247 . assume that @xmath236 , then the two aztec rectangles @xmath244 and @xmath245 become two aztec diamonds of order @xmath248 . the aztec rectangle @xmath244 satisfies the conditions in graph - splitting lemma [ graphsplitting](a ) as an induced subgraph of @xmath246 , and the rectangle @xmath249 satisfies also the conditions of this lemma as an induced subgraph of @xmath250 . therefore @xmath251 where graph @xmath252 consists of @xmath37 disjoint vertical edges ( see the circled edges in figure [ ben1s ] ) , so it has exactly one perfect matching . this finishes the proof of part ( b ) of the theorem in the case when the triangular cells above @xmath17 with bases resting on it are black . suppose now that the latter are white . apply in this case the composite transformation in proposition [ composite](a ) to the upper and lower parts of the graph @xmath253 . we obtain that @xmath254 where the connected sum acts along the bottom vertices of the two aztec rectangles ( ordered from left to right ) , and @xmath216 , @xmath217 , @xmath218 and @xmath219 are quantities having the same significance for the lower part of @xmath253 as their unprimed correspondents have for its upper part ( except that , as in the first case , they now count white cells ) . the transformation involved in ( [ maineq4 ] ) is illustrated in figures [ symmetricfig2](a ) and ( b ) . apply the vertex - splitting lemma [ vs ] to all bottom vertices of @xmath255 in the graph on the right hand side of ( [ maineq4 ] ) ( see figures [ symmetricfig2](b ) and ( c ) ) . next , apply the transformation @xmath149 of lemma [ t1 ] to the top and bottom of the resulting graph , the latter is transformed into the graph on the right hand side of ( [ maineq1 ] ) ( see figures [ symmetricfig2](c ) and ( d ) ) . then both parts ( a ) and ( b ) of the theorem are reduced to the case treated above . the generalized quasi - hexagons we treated so far had the special property that the point on the southwestern boundary where its defining property `` black on left '' changed to `` white on left '' ( i.e. vertex @xmath26 ) was on the same southwest - to - northeast lattice diagonal as the analogous point on the northeastern boundary ( i.e. vertex @xmath30 ) . therefore , the perpendicular bisector of segment @xmath256 is the symmetric axis of the quasi - hexagons . these include as a special case propp s quasi - hexagons of side - lengths @xmath1 , @xmath2 , @xmath2 , @xmath1 , @xmath2 , @xmath2 ( clockwise from top ) , and will be called from now on symmetric ( generalized ) quasi - hexagons . in order to include the general case of propp s quasi - hexagons when the side - lengths are @xmath1 , @xmath2 , @xmath3 , @xmath1 , @xmath2 , @xmath3 , we extend the definition of our generalized quasi - hexagons as follows . besides @xmath17 , we consider a second distinguished lattice diagonal @xmath257 below it , and draw in also @xmath258 additional diagonals in between them , so that the distances between successive ones ( starting from top ) are @xmath259 . as in the symmetric case , we consider @xmath18 diagonals above @xmath17 at successive distances ( starting from top ) @xmath19 , and @xmath21 diagonals below @xmath257 , at successive distances ( starting from the bottom ) @xmath22 . define our generalized quasi - hexagon as in the symmetric case , with the one change that on the northeastern boundary we make the switch from the rule `` white on left '' to `` black on left '' on the diagonal @xmath257 , rather than on @xmath17 . denote the resulting region by @xmath260 ; an illustrative example is shown in figure [ asymmetrichex ] . as mentioned in section 2 , there is no simple product formula for the number of tilings of asymmetric quasi - hexagons . however , it turns out that we can construct a conceptually simple graph so that the number of tilings of an asymmetric quasi - hexagon is equal to a certain power of @xmath12 times the number of perfect matchings of this graph . we present this construction next . roughly speaking , the graph we need is a connected sum of a baseless aztec rectangle and the bottom half of a honeycomb graph . to be precise , let @xmath1 , @xmath2 , @xmath3 be positive integers with @xmath261 , and let @xmath262 be the portion of the honeycomb graph of sides @xmath1 , @xmath263 , @xmath3 , @xmath1 , @xmath263 , @xmath3 ( clockwise from top ) that is below or on the horizontal through its center . let @xmath264 and @xmath265 be positive integers , and consider the baseless aztec rectangle @xmath266 . the latter has @xmath267 vertices on the bottom ; label them from left to right by @xmath268 . the graph @xmath262 has @xmath269 vertices at the top ; label them from left to right by @xmath270 . ( 6349,3716)(222,-3488 ) ( 1891,-2941)(0,0)[lb ] ( 2836,-2468)(0,0)[lb ] ( 592,-1996)(0,0)[lb ] ( 150,-696)(0,0)[lb ] ( 3948,-2178)(0,0)[lb ] ( 5250,-3005)(0,0)[lb ] ( 6203,-2414)(0,0)[lb ] ( 3450,-814)(0,0)[lb ] ( 1182,-224)(0,0)[lb ] ( 5080 , 12)(0,0)[lb ] on the other hand , if @xmath277 , define @xmath273 by @xmath278 where @xmath279 is the subgraph of @xmath266 obtained by deleting its vertices with labels greater than @xmath269 , and the connected sum identifies the vertices labeled @xmath275 in the two components , for @xmath280 . examples illustrating the two cases are shown in figure [ cfigure ] . for the sake of simplicity , we detail below the case when all distances between successive diagonals are odd ( recall that the original problem , problem 16 in @xcite , corresponds to the special case when all these distances are equal to 3 ) . the general case is addressed in theorem [ asymmetric1 ] . [ asymodd ] let @xmath19 , @xmath22 and @xmath281 be odd integers , and set @xmath282 let @xmath283 be an asymmetric quasi - hexagon , and let @xmath27 be the number of black cells above @xmath17 , and @xmath38 the number of white cells below @xmath257 . in order to prove this result we will need some more graph transformations , which we present in the next two lemmas . let @xmath291 be the graph obtained from the aztec rectangle @xmath95 by deleting its leftmost @xmath40 vertices , and let @xmath292 be the graph obtained from @xmath95 by deleting its rightmost @xmath40 vertices . let @xmath294 be the graph obtained from the graph @xmath291 by appending vertical edges to its top vertices , and @xmath295 the graph obtained from @xmath292 by appending vertical edges to its bottom vertices . then @xmath296 where the connected sum acts along the top and bottom vertices of @xmath294 and @xmath295 ( ordered from left to right and from top to bottom ) , and along the ordered set @xmath136 of vertices of @xmath64 . the above transformation is illustrated by figures [ lemmat3](a ) and ( e ) , for @xmath96 and @xmath297 . one can prove lemma [ t3 ] by following the argument in the proof of lemma [ t1 ] , based on figure [ lemmat3 ] . note that @xmath300 has @xmath305 vertices on the bottom , and @xmath301 has @xmath264 vertices on top . if @xmath306 , define @xmath307 where the connected sum identifies the @xmath275th vertex on the bottom in @xmath300 with the @xmath275th vertex on top in @xmath301 ( both counted from left to right ) , for @xmath308 ( the graph @xmath309 is shown in figure [ lgraph](a ) ) . on the other hand , if @xmath310 , define our graph by @xmath311 where @xmath312 is the graph obtained from @xmath313 by deleting the rightmost @xmath314 of its bottom vertices , and the connected sum identifies the @xmath275th vertex on the bottom in @xmath312 with the @xmath275th vertex on top in @xmath301 , for @xmath315 @xmath316 ( @xmath302 is pictured in figure [ lgraph](b ) ) . our second family of graphs is the result of an analogous construction , in which the role of @xmath291 is replaced by the graph @xmath317 obtained from it by deleting its top @xmath9 vertices ( i.e. , @xmath317 is obtained from @xmath318 by deleting the @xmath40 leftmost and @xmath9 topmost vertices ) . let @xmath298 be positive integers . note that @xmath319 has @xmath264 vertices on top ( just like @xmath301 ) . in analogy to ( [ lgrapheq1 ] ) , if @xmath306 , define @xmath320 where the connected sum identifies the @xmath275th vertex on the bottom in @xmath313 with the @xmath275th vertex on top in @xmath301 but now counted from _ right to left _ for @xmath308 ( the graph @xmath321 is shown in figure [ lgraph](c ) ) . finally , if @xmath310 , define @xmath322 by @xmath323 where @xmath324 is the graph obtained from @xmath313 by deleting the _ leftmost _ @xmath314 of its bottom vertices , and the connected sum identifies the @xmath275th vertex on the bottom in @xmath324 with the @xmath275th vertex on top in @xmath319 ( both counted from right to left ) , for @xmath325 ( @xmath326 is shown in figure [ lgraph](d ) ) . to state our second lemma , it will be useful to introduce the following notation . given an l - shaped graph @xmath64 from one of the two families defined above , define @xmath327 to be the subgraph obtained from @xmath64 by removing all its bottommost vertices . in the spirit of section 3 , if @xmath64 belongs to one of the above two families , or if it is obtained from a member of these families by deleting all its bottom vertices , we denote by @xmath328 ( the `` combed version '' of @xmath64 ) the graph obtained from @xmath64 by appending a vertical edge to each of its bottommost vertices . in each of the equalities ( [ lgrapheq3 ] ) , ( [ lgrapheq4 ] ) and ( [ lgrapheq5 ] ) , the connected sum acts on @xmath64 along @xmath136 , and on the other two summands along their bottommost vertices , ordered from left to right . we call the portions of an asymmetric quasi - hexagon @xmath140 above @xmath17 , below @xmath257 and between @xmath17 and @xmath257 _ the upper _ , _ the lower _ and _ the middle parts _ of the region , respectively . we also use these terms for the corresponding parts of the dual graph @xmath64 of the region . it means the upper , the lower and the middle parts of @xmath64 are the dual graphs of the upper , the lower and the middle parts of the region @xmath140 , respectively . we are now ready to present the proof of our theorem . \(a ) apply the composite transformation of proposition [ composite](b ) to the upper part of @xmath64 , and also to its lower part . this yields @xmath337 where @xmath217 is the length of the southeastern side of @xmath140 . the arguments that led to ( [ maineq2 ] ) still work , because the middle part of @xmath64 is balanced ( i.e. it has the same number of vertices in the two color classes of a proper 2-coloring ) , and they give @xmath338 furthermore , the arguments that gave ( [ maineq2 ] ) also work , because of the same reason ( the middle part of @xmath64 is balanced ) . it follows that @xmath64 is balanced if and only if @xmath186 , which proves part ( a ) . consider first the first equality in ( [ asymoddeq ] ) , provided @xmath341 . in this case @xmath342 , for @xmath343 . to prove this we use the same key idea in the proof of theorem [ main ] as follows . the lozenge hexagon @xmath344 can be viewed as an asymmetric quasi - hexagon , in which all the inter - diagonal distances @xmath48 s , @xmath49 s and @xmath345 s are equal to one . by the same argument in the equality ( [ maineq3 ] ) , we obtain @xmath346 therefore , the first equality in ( [ asymoddeq ] ) follows from ( [ asyproofeq3 ] ) and ( [ asyproofeq3 ] ) . one readily checks that , for @xmath349 , the expression @xmath350 in the exponent of 2 in the second equality of ( [ asymoddeq ] ) becomes precisely @xmath351 , and thus ( [ asyproofeq4 ] ) follows from the case @xmath352 of the second equality in ( [ asymoddeq ] ) . * stage 1 . * _ reduction to the case @xmath354 . _ if @xmath355 ( and @xmath356 ) , apply the transformation @xmath357 of lemma [ t3 ] to the graph @xmath64 , regarded as being obtained as a connected sum of type ( [ t3eq1 ] ) , in which the second summand is the portion of @xmath64 that corresponds to the portion of @xmath140 in between @xmath257 and the first diagonal above it ( see figure [ middle7 ] ; we replace the subgraph between the two dotted lines in a graph by the subgraph between those lines in the next graph ) . one readily sees that the resulting graph is isomorphic to the graph @xmath358 by lemma [ t3 ] , we have @xmath359 . if @xmath360 , then @xmath361 and the latter equality is obviously true . repeating this argument , we get that @xmath362 for @xmath363 , where @xmath364 thus @xmath365 * stage 2 . * _ transforming @xmath64 into the connected sum of an aztec rectangle with missing vertices and a half - honeycomb . _ apply the transformation @xmath150 of lemma [ t2 ] to the part of @xmath366 above @xmath17 , and then apply ( in reverse ) the composite transformation of proposition [ composite](b ) to replace the part of @xmath64 below @xmath257 by a half - honeycomb ( these transformations are illustrated in figure [ stage2 ] ) . denote the resulting graph by @xmath367 . the quoted results imply that @xmath368 note that @xmath367 is not far from being a @xmath369-type graph the only difference is in how its upper part relates to an aztec rectangle . this difference is eliminated in the remaining two stages . * stage 3 . * _ bringing the missing vertices of the aztec rectangle to the top left . _ this stage is illustrated in figures [ middle8](a)(d ) ; we replace the subgraph above the dotted line in a graph by the subgraph above this line in the next graph . note that we have the graph isomorphism @xmath370 ( recall that we are assuming @xmath352 ) . apply lemma [ ltransform](b ) to the graph @xmath367 regarded as a connected sum as shown on the right hand side of ( [ asyproofeq7 ] ) . we obtain that @xmath371 where @xmath372 shown in figures [ middle8](a ) and ( b ) . * stage 4 . * _ moving the missing vertices of the aztec rectangle to the bottom right . _ this stage is illustrated in figures [ midnew3](a)(e ) . apply lemma [ ltransform](c ) to the graph @xmath378 . we obtain that @xmath379 where @xmath380 ( see figures [ midnew3](a ) and ( b ) ) . note that for the last of these graphs we have @xmath387 the second equality in ( [ asymoddeq ] ) follows now by ( [ asyproofeq5 ] ) , ( [ asyproofeq6 ] ) , ( [ asyproofeq10 ] ) , ( [ asyproofeq12 ] ) , ( [ asyproofeq14 ] ) and ( [ asyproofeq16 ] ) . denote by @xmath388 the dual graph of @xmath349 . the top @xmath242 rows of the vertices of @xmath388 induce a subgraph isomorphic to @xmath243 , and the bottom @xmath242 rows of its vertices induce also a subgraph isomorphic to @xmath243 . denote by @xmath244 and @xmath245 the two aztec rectangles ( shown by the subgraph above and the subgraph below two dotted lines in figure [ ben2s ] ) . if @xmath288 , then @xmath244 satisfies the conditions in graph - splitting lemma [ graphsplitting](b ) as an induced subgraph of @xmath388 , so @xmath389 . assume that @xmath289 , then two aztec rectangles @xmath244 and @xmath245 are now two aztec diamonds of order @xmath390 . they satisfy the conditions in the part ( a ) of the graph - splitting lemma [ graphsplitting ] as two induced subgraphs of @xmath388 and @xmath391 , respectively . we can check that the graph @xmath393 has a unique perfect matching ( it consists of all vertical edges in the middle part of @xmath388 when @xmath341 , and its pattern is similar to the set of circled edges in figure [ ben2s ] when @xmath347 ) . therefore , we get @xmath394 from ( [ power2b ] ) . this completes the proof for part ( c ) . the main reason we chose to relate quasi - hexagons to the graphs @xmath273 as in theorem [ asymodd ] is that the number of perfect matchings of the latter can be written as a sum of quantities expressed by simple product formulas . this is presented in corollary [ sumformulas ] below , whose proof employs the following two lemmas . [ aztecdent ] label the bottom vertices of the baseless aztec rectangle @xmath98 from left to right by @xmath395 , and denote by @xmath64 the graph obtained from it by deleting the vertices with labels in the set @xmath396 , where @xmath397 are given integers . then we have @xmath398 [ hexdent ] consider the dual of the bottom half of a lozenge hexagon of side - lengths @xmath2 , @xmath1 , @xmath1 , @xmath2 , @xmath1 , @xmath1 ( clockwise from top ) on the triangular lattice . label its topmost vertices from left to right by @xmath399 , and the number of perfect matchings of the graph obtained from it by removing the vertices with labels in the set @xmath400 is equal to @xmath401 where @xmath402 are given integers . @xmath34b@xmath35 . if @xmath407 , then @xmath408 where the sum is taken over all pairs of disjoint sets @xmath25 and @xmath26 whose union is @xmath409 and whose cardinalities are given by @xmath410 and @xmath411 . @xmath34c@xmath35 . if @xmath412 , then @xmath413 where the sum is taken over all pairs of disjoint sets @xmath25 and @xmath26 whose union is @xmath414 and whose cardinalities are given by @xmath410 and @xmath415 . consider the vertices of @xmath273 that belong both to the baseless aztec rectangle part and to the half - honeycomb part ( these are the vertices on the dotted line in figure [ cfigure ] ) . in any given perfect matching , some of them are matched upward , and the remaining ones are matched downward . let @xmath18 be the number of vertices belonging to both parts , and partition the set of perfect matchings of @xmath273 into @xmath416 classes corresponding to all the possible choices for each of these vertices to be matched upward or downward . each class is then the set of perfect matchings of a disjoint union of two graphs , the top one being of the kind in lemma [ aztecdent ] , and the bottom one ( after adding some vertical forced edges that are shown as circled ones in figure [ cfigure ] ) being of the kind in lemma [ hexdent ] . part ( a ) follows from the requirement that these ( bipartite ) graphs are balanced , while parts ( b ) and ( c ) follow from lemmas [ aztecdent ] and [ hexdent ] . we end this section by presenting the general version of theorem [ asymodd ] , corresponding to the case when the distances between successive diagonals are arbitrary ( and not necessarily odd ) . to state the general result we need some more definitions . we define the upper and the lower heights of the region to be the number of rows of regular black cells above @xmath17 and the number of rows of regular white cells below @xmath257 , respectively . the diagonals divide the middle part of an asymmetric quasi - hexagon into parts , called _ middle layers_. in the case of odd distances ( as in theorem [ asymodd ] ) , the bottom row of cells is black . however , we face the same problem in investigating symmetric quasi - hexagons : the bottom row of cells may be white . by the same argument that we used in the proof of theorem [ main](a ) , when the latter happens the region has no tilings . thus , we assume from now on that the bottom row of cells is black . define the _ slope _ of the asymmetric quasi - hexagon by @xmath420 set @xmath421 since the middle part is balanced , one can verify that the numbers of black and white vertices in the dual graph of @xmath140 are equal if and only if the upper and lower heights are equal . the same arguments that we used to prove theorem [ asymodd ] can be used to obtain the following result . [ asymmetric1 ] let @xmath283 be an asymmetric quasi - hexagon . denote by @xmath27 the number of regular black cells above @xmath17 , and by @xmath38 the number of regular white cells below @xmath257 , and assume the upper and lower heights of @xmath140 are both equal to @xmath37 . let @xmath40 be the number of rows of regular black triangular cells above @xmath17 , and @xmath9 be the number of rows of irregular black triangular cells above @xmath17 . then we have @xmath422 where @xmath423 .

a novel approach to unstructured quadrilateral mesh generation for planar domains is presented . away from irregular vertices , the resulting meshes have the properties of nearly conformal grids . the technique is based on a theoretical relation between the present problem , and the inverse poisson ( ip ) problem with point sources . an ip algorithm is described , which constructs a point - source distribution , whose sources correspond to the irregular vertices of the mesh . both the background theory and the ip algorithm address the global nature of the mesh generation problem . the ip algorithm is incorporated in a complete mesh generation scheme , which also includes an algorithm for creating the final mesh . example results are presented and discussed .

0802.2399

_ boundary alignment _ is a critical feature of meshes in many applications . in a boundary aligned mesh the boundary , or some other line , is traced by the sides of high - quality cells , see fig . [ fig : bound_align ] . the definition of a well - shaped cell may be application dependent , but in many cases , cells similar in shape to squares ( for quadrilateral cells ) or equilateral triangles ( for triangles ) are preferred . characteristics of the entire mesh are also important , such as smooth cells - size and cell - shape transitions . [ ptb ] the problem of producing boundary aligned meshes with well - shaped cells has been the subject of extensive research @xcite . still , many popular algorithms are heuristic in nature , and a more general understanding of the subject is called for , especially when quadrilateral meshes are considered . a key difficulty is the problem s _ global _ character : the shape and position of every cell in the mesh is , at least in principle , related to that of any other cell . in a previous work @xcite , we described a relation between the problem of two - dimensional unstructured mesh generation , on both planar and curved surfaces , and another well - known problem , namely the _ inverse poisson _ ( ip ) problem . the ip problem is concerned with reconstructing a _ source distribution _ @xmath0 of the poisson equation @xmath1 , from information on the potential @xmath2 at the boundaries . in that work , the mesh was assumed to be _ conformal _ away from the irregular vertices ( vertices whose degree is different than four ) , like a grid mapped by a conformal mapping . such grids have the property of having square cells in the limit of an increasingly finer grid . under this assumption , the problem of mesh generation was then shown to reduce to an ip problem . the irregular vertices of the mesh correspond to point sources ( delta functions ) of @xmath0 , and @xmath2 is interpreted as the logarithm of the local resolution . this theoretical framework turns the focus to the irregular vertices of the mesh : once their distribution is fixed , the continuum properties of the mesh - local resolution and directionality - are known . note that it is also an explicitly global formulation , since the resolution at any given point is affected , via the function @xmath2 , by the locations of all irregular vertices in the mesh . in this paper the generation of planar quadrilateral meshes is discussed . resting on the results of @xcite , a new ip algorithm is presented , designed to construct source distributions of the appropriate type , which approximate the resolution and mesh directionality inputs at user - specified points , such as at the boundaries . the ip algorithm is then incorporated into a complete mesh generation scheme , which also includes a technique for generating the final mesh . an implementation is described , and shown in example cases to generate boundary aligned meshes , where well - placed point sources create smooth cell transitions and high quality cells . a similar procedure is probably applicable to triangular meshes , but is not discussed in the present work . remeshing of curved surfaces has recently attracted considerable attention ; for a review see @xcite . many of the algorithms receive an input mesh directionality throughout the surface , usually the principal curvature directions of the surface . this setting presents different challenges than those addressed here , since the mesh structure is determined , to a large extent , when mesh directionality is given everywhere in the domain . for example , the locations of critical irregular vertices are dictated by the mesh directionality in the _ vicinity _ of these points . another related subject is surface parameterization , concerned with creating mapping of surfaces to the plane . conformal surface parameterizations are created in @xcite , but boundary alignment is not addressed . the paper is organized as follows : section [ sec : ip_review ] shortly reviews the relevant theoretical background , with emphasis on the relation between the ip problem and unstructured mesh generation . sections [ sec : algorithm ] and [ sec : create_final ] describe the proposed ip algorithm , and the mesh generation scheme . section [ sec : implementation ] describes an implementation of the algorithm , and section [ sec : examples ] gives examples of meshes generated . conclusions and possible directions for future research are discussed in section [ sec : conclusions ] . in this section an overview of the background theory will be given . section [ sec : motivation ] explains the rationale underlying the mathematical formulation . in section [ sec : definitions ] , some key conformal geometry concepts are discussed , together with their relevance to the present problem . section [ sec : ip_background ] summarizes the results developed in @xcite , relating mesh generation with the ip problem . the exposition is limited to planar two - dimensional mesh generation ; a detailed account , in the more general setting of curved surfaces , can be found in @xcite . we start by considering mesh generation using conformal mappings , which can be viewed as a special case of the theory to be described . mapping techniques , in general , construct a function from one domain , for which a mesh already exists , to a second domain , which is being meshed . the mapping function is then used to map the mesh into the second domain . a key idea is that continuum properties of the mapping function control the shapes of the cells of the new mesh , at least for small enough cells . for example , if the mapping is conformal , i.e. angle preserving , a cell with right inner angles ( rectangle ) will be mapped to a target cell with approximately right inner angles . in unstructured mesh generation the connectivity of the mesh is not known in advance , and a more general framework is called for . in what follows , the interplay between the two domains which serves as a paradigm of mapping techniques , is replaced with an interplay between two definitions of distances on the input domain . instead of imagining the mesh as being the image of a mesh on a different domain , we _ redefine the distances _ on the domain to be meshed , and _ fix the cell edge length_. thus , the local mesh resolution will be proportional to the new local distance definition : a large new distance between two given points will mean more cells will be placed in between , hence a higher local resolution . distances are redefined using the concept of a _ metric _ , known from riemannian geometry . since we focus on cells which are squares in the limit of an increasingly finer mesh , we need to define only two local properties : the resolution ( inverse of cell - size ) , and the direction of the cell - edges . the new resolution , as noted before , is controlled by a new distance definition . we use a new distance which is localy a scaling of the old distances : the new distance between a point and other nearby points is equal to the old distance , multiplied by a scalar factor which is independent of the direction . thus , a small square measured in one distance definition is also ( approximately ) a square according to the new distance definition . in riemannian geometry terminology , the old and new metrics are said to be _ conformally related_. the mesh directionality is related to the resolution , as is described in the following section . it is convenient to work with a function @xmath3 , defined as the _ logarithm _ of the local scaling factor . that is , a small square of side length @xmath4 measured with the original , cartesian distance definition , is a square with side length @xmath5 as measured with the new distance definition , see fig . [ fig : lay_grid1],a .. if we imagine that the domain is covered with many small squares , all with the same side length @xmath6 in the new distance definition , the size of these cells in the original distance definition will be @xmath7 . the ( local ) resolution is the inverse of the local - cell - size , hence @xmath8 where the proportionality constant is a single number for the whole mesh . the second continuum property is the local directionality of the cell edges . the edges should ideally meet at right angles everywhere , except at _ irregular vertices _ , where the number of edges incident on the vertex is different than four . it is therefore natural to assign to every point a set of four directions , mutually parallel or perpendicular . this concept was expressed by many authors , and given various names , such as _ mesh directionality _ 4-symmetry direction field _ @xcite , and _ frame field _ , although the last may refer to a structure which also holds cell - size information @xcite . graphically , this object can be represented by a cross at every point , see fig . [ fig : lay_grid1],a . , and here will be called a _ cross - field _ @xcite . on the plane , the cross direction can be measured by the angle @xmath9 from the @xmath10-axis to one of the directions of the cross . this angle is fixed up to an addition of @xmath11 radians , i.e. @xmath12 represent the same cross iff @xmath13 , for some integer @xmath14 . [ ptb ] the function @xmath2 and the cross - field are not unrelated ; due to conformality , lines that trace the edge directions bend towards the side with smaller cells , see fig . [ fig : lay_grid1],a .. in the continuum theory , these lines are known as _ geodesics _ of the manifold . geodesics are a generalization of the concept of straight lines to non - euclidian geometries . in the original , cartesian coordinates , the geodesic obeys the following differential equation : @xmath15 where @xmath16 is the curvature of the geodesic in the cartesian coordinate system , and @xmath17 is the derivative of the function @xmath2 in the direction normal to the tangent , see fig . [ fig : lay_grid1],b .. eq . ( [ eq : kappa_dphi_dn ] ) allows one to calculate the change in the direction of the cross - field between two points connected by a geodesic . there is also a direct way of calculating the change in the direction of a cross - field between any two points , along any curve connecting the two points , known as _ parallel - transport_. let @xmath18 be some curve from point @xmath19 to point @xmath20 , and let @xmath21 be the angles of the crosses at points @xmath19,@xmath20 respectively ( as noted above , the angles are defined up to an addition of a @xmath11 radians ) . then @xmath22 where the integration denotes a line integral along the curve @xmath18 , according to the length parameter @xmath23 on @xmath18 , as measured in the original coordinate system . the differential formulation of this equation is @xmath24 where @xmath25 are a pair of perpendicular vectors that form a right hand system . where the function @xmath2 is defined , that is , at any point in the domain that is not a singularity , @xmath2 can be shown to _ harmonic _ , that is to obey the laplace equation : @xmath26 where , again , the derivatives are taken with the original coordinate system . ( [ eq : kappa_dphi_dn],[eq : phi_laplace ] ) are well - known results of conformal geometry @xcite,@xcite . ( [ eq : theta_change_par_tranport ] ) is derived in @xcite equations ( [ eq : kappa_dphi_dn],[eq : phi_laplace ] ) fully describe the relations between the cross - field and the function @xmath2 , at any regular point of the domain , that is , any point that is not a singularity of the function @xmath2 . the singularities of the function @xmath2 are a key ingredient of the theory , since they correspond to the irregular vertices of the mesh , and unstructured meshes are those which contain irregular vertices . a detailed analysis of the possible of singularities of the harmonic function @xmath2 , and their effect on the resulting mesh , was carried out in @xcite , and shows that the only type of singularity that corresponds to a mesh with a finite number of cells is of the type @xmath27 , where @xmath28 is the location of the singularity . in the ip literature , such a sigularity is known as a _ point source_. furthermore , the prefactor of the logarithm is directly related to the degree of the irregular vertex in the final mesh . more specifically , suppose there are @xmath29 singularities in a domain @xmath30 , at points @xmath31 , then the function @xmath2 can be written as * condition 1 : * @xmath32 where @xmath33 is a harmonic function , @xmath34 , and @xmath35 . the numbers @xmath36 are known as _ charges _ , so condition 1 states that the charges are integer multiples of @xmath11 . we will refer to the numbers @xmath37 simply as the @xmath38 -values of the singularities . the degree ( number of incident edges ) of the irregular vertex corresponding to the source at @xmath39 is equal to @xmath40 . so , for example , irregular vertices of degrees 3 and 5 will correspond to a singularities with @xmath41 and @xmath42 , respectively . in a geometrical context , a logarithmic singularity of @xmath2 represents a _ cone - point _ - the tip of a cone - in the new distance definition . the charge corresponds to the angle deficit of the tip of the cone . related subjects where manifolds with cone - points are considered include the study of disclinations in elastic media @xcite , and surface parameterizations @xcite . an unstructured mesh is required to be aligned with the boundary . namely , that one of the cross directions be parallel to the tangent to the boundary ( where a tangent exists ) : * boundary alignment definition : * @xmath43 where @xmath44 is a curve tracing the boundary , @xmath45 and @xmath46 . since @xmath9 is fixed up to an addition of a multiple of @xmath11 radians , this definition depends only on the cross - field itself , and not on the particular choice of @xmath9 . requiring boundary alignment , as defined in eq . ( [ eq : bound_align ] ) can be shown to be equivalent to three more conditions on the function @xmath2 @xcite . the following , condition 2 , applies to a point on a smooth section of the boundary , @xmath47 . it roughly states that a smooth section of the boundary is a geodesic . this provides information on the derivative of @xmath2 normal to the boundary : * condition 2 : * @xmath48 @xmath16 is the boundary curvature at @xmath49 . the next condition is concerned with _ junction points _ of the boundary , where the tangent to the boundary is discontinuous . it assures that the cross - field will be aligned with the boundary on both sides of the junction point . let @xmath18 be a curve from point @xmath19 to point @xmath20 on both sides of a point @xmath50 of the boundary , see fig . [ fig : cond_3_4],b .. then * condition 3 : * @xmath51 where @xmath52 is the angle of the boundary , and @xmath46 . this means that for some inner angles @xmath2 will contain a singularity at @xmath50 ; at a distance @xmath53 from @xmath50 , the singularity will be of type @xmath54 . [ ptb ] the final condition is concerned with the relation between two different boundary components . it is just a restatement of eq . ( [ eq : theta_change_par_tranport ] ) for two points on two different boundary segments . * condition 4 : * let @xmath19,@xmath20 be two points on two boundary curves @xmath55 . let @xmath18 be a curve from @xmath19 to @xmath20 . then @xmath56 conditions 1 - 4 form a complete set of conditions on @xmath2 , for a cross - field that is boundary aligned to exist . that is , conditions 1 - 3 , and condition 4 on a number of selected curves , one from a selected point to each boundary component , are sufficient for a boundary aligned cross - field to exist , see @xcite . the mesh generation scheme consists of the following steps : 1 . setup and solution of the ip problem . finds the sources locations and charges . described in sections [ sec : input]-[sec : calc_locations ] . solution of the direct poisson problem , to obtain the functions @xmath57 throughout the domain , see section [ sec : restore_harmonic ] . generation of the final mesh , see section [ sec : create_final ] . suppose that we are given an open domain @xmath30 , with boundary @xmath47 , and are given the required resolution function on @xmath47 . ( [ eq : phi_is_log_res ] ) , this is readily translated to the required value of @xmath2 on the boundary : @xmath58 condition 1 ( eq . ( [ eq : cond1 ] ) ) is a solution of the poisson equation @xmath1 , with point sources , whose locations and charges are yet unknown . the problem is to find a distribution of sources ( location and charge ) adhering to the boundary alignment definition , eq . ( [ eq : bound_align ] ) , or alternatively to conditions 2 - 4 ( eq . ( [ eq : cond2])-([eq : cond4 ] ) ) , as well as to eq . ( [ eq : phi_from_res ] ) . such a problem is known as an _ inverse poisson _ ( ip ) problem . the ip problem may be compared with the _ direct poisson _ problem , where the source distribution is given , as well as some boundary information ( e.g. , dirichlet or newmann boundary conditions ) , and the value of the function @xmath2 is to be found . in the ip problem , the source distribution is unknown , and a source distribution adhering to the known boundary information is to be found . ip problems have important applications in various areas of science and engineering @xcite,@xcite,@xcite,@xcite , @xcite . by its nature , the ip problem is ill - posed , and the solution might not be unique , and may be sensitive to small changes of the input , such as small changes in boundary conditions . in delicate problems of this type , any prior information on the source distribution may greatly affect the applicability of a specific solution procedure . in the present problem , we seek a point source distribution . a number of algorithms for solving an ip problem with point sources appear in the literature . in @xcite , an inverse problem where all sources have the same charge is solved . in our implementation , however , at least two different charge values must be incorporated ( @xmath59 ) . in @xcite,@xcite,@xcite an inverse problem where both the locations of sources and their charges are unknown , and are reconstructed . this gives more freedom in the reconstruction than we can allow , since for the present purposes the charges must be multiples of @xmath11 . another important aspect of the present problem is that the domain of the ip problem may be of any shape and topology ( i.e. , may contain holes ) , whereas the above works only deal with a simply connected region ( a circle , usually ) . it is important to note that in the present application , in contrast to other standard applications , the input to the ip problem is not generated by some existing source distribution ( perhaps with some added noise ) , but by the domain s shape and input resolution . the existence of a source distribution which reconstructs the input data , at least approximately , is therefore not obvious . this is an interesting and important subject , but is beyond the scope of the present work . as is well known , the real and imaginary parts of a complex analytic function are harmonic functions @xcite . this correspondence has been utilized in ip algorithms @xcite,@xcite,@xcite , and will be used here as well . we define the complex - valued function @xmath60 where @xmath61 , with @xmath62 the components of @xmath63 , that were defined in condition 1 . @xmath4 is a function on @xmath30 , such that @xmath64 then , recalling that @xmath65 , it follows that @xmath66 with @xmath3 defined in condition 1 , and @xmath67 , with @xmath68 . the functions @xmath69 and @xmath70 for some @xmath71 are analytic in a neighborhood of any point that is not a singularity of @xmath2 . however , as functions over the entire domain @xmath30 , they may be _ multi - valued_. multi - valued functions accept many - values at a point , depending on the path taken to that point . it is well known that the complex function @xmath72 is multi - valued . defining @xmath73 as @xmath74 , where @xmath75 is a path from @xmath76 to @xmath77 , the imaginary part of @xmath72 is only fixed up to an addition of multiples of @xmath78 , depending on the path taken . if the domain @xmath30 is not simply - connected ( i.e. , contains holes ) , then the function @xmath69 may also be multi - valued , since by following different paths around the holes , different values of @xmath79 may be obtained . the function @xmath80 , being a sum of multi - valued functions , may also be multi - valued . according to eq . ( [ eq : re_f_is_phi ] ) , the real part of @xmath81 is single - valued , as it is equal to @xmath2 at that point . we thus turn to examine the imaginary part of @xmath81 . the real and imaginary parts of a complex function are related by the cauchy - reimann ( cr ) equations @xcite . recall that for a complex differentiable function @xmath82 , where @xmath83 are real - valued , the cauchy - reimann equations read @xmath84 , and @xmath85 . this allows one to recover the imaginary part of a analytic function if its real part is given . writing eq . ( [ eq : phi_theta_right_hand_relation ] ) in the two right hand coordinate systems @xmath86 , gives the relations @xmath87 and @xmath88 . these are precisely the cr equations for the complex function @xmath89 , where @xmath90 is any real constant , hence @xmath91 the constant @xmath90 is arbitrary , and will be taken to be zero . the multi - valued nature of @xmath81 is explicit in the integral formulation in eq . ( [ eq : f_is_phi_i_theta ] ) . the right - hand - side ( rhs ) of eq . ( [ eq : f_is_phi_i_theta ] ) shows that the since the imaginary part of @xmath81 is @xmath92 , and @xmath9 must be fixed up to additions of @xmath11 , then the integral @xmath93 must also be fixed up to additions of @xmath11 . indeed , this was shown to hold if conditions 1 - 3 ( eq . ( [ eq : cond1])-([eq : cond3 ] ) ) hold , see @xcite . to summarize , the problem is restated as finding a multi - valued complex function @xmath81 , given by eq . ( [ eq : f_def ] ) , and whose value at points on the boundary of @xmath30 is : @xmath94 in the following sections , an algorithm for solving the problem as it was here restated is described . the ip algorithm first removes the contribution of @xmath69 from the boundary conditions , paying special attention to the junction points , and then constructs the source distribution . the function @xmath2 , and hence the function @xmath81 , may be singular at junction points . special procedures for addressing this behavior are taken as part of two different steps of the algorithm : a. : : choosing the input resolution near junction points . b. : : special treatment of junction points when @xmath4 is removed from the boudary conditions . the two subjects are discussed in the following subsections , [ sec : input_near_junc ] and [ sec : remove_harmonic_junc ] . except in the special cases when @xmath52 is a multiple of @xmath11 , the behavior of @xmath2 at a small neighborhood in @xmath30 of a junction point is singular , with the singularity at the junction point , see condition 3 ( eq . ( [ eq : cond3 ] ) ) . if the input resolution in a small neighbourhood of the junction point does not match this singular behavior , solution of the ip problem will feature sources at any distance from the junction , no matter how small , resulting in a distribution with an infinite number of sources . in practice , this means that an ip algorithm will cluster many sources near the junction point , in a futile attempt to reconstruct the boundary conditions there . to avoid this problem , we adjust the _ input _ @xmath2 at a neighborhood of the junction point . this adjustment should vanish rapidly at a distance larger than 1 - 2 cell - sizes , so as to have little effect on the original required resolution . condition 3 states that the gradient of @xmath2 near a junction point must diverge as @xmath95 , where @xmath53 is the distance from the junction point . such a flux is formed by a logarithmic singularity at the junction point ( see also @xcite , section 7.1 ) . consider a singular source term of the form @xmath96 , where @xmath53 is the distance from the junction point . then the flux through a circular arc @xmath97 , at distance @xmath53 from the junction point is ( see also fig . [ fig : cond_3_4],a . ) : @xmath98 where @xmath52 is the junction inner angle . then by condition 3 , @xmath99 , hence @xmath100 @xmath101 is integer , which must be positive , otherwise a mesh with an infinite number of cells will result ( @xcite , section 7.1 ) . in fact , @xmath101 has a simple interpretation : it is the number of cells incident upon the junction point in the resulting mesh ) nearby cone - points may be joined , and cone - points may be shifted to the boundary , depending on the final cell - size . in such a case , the number of cells incident on a junction point , as well as on other boundary points , may change . ] . a reasonable choice for @xmath101 is therefore @xmath102 for which the inner angles of the cells incident on the junction point are closest to @xmath11 . in order to restrict the effect of this correction of @xmath2 to a small region in @xmath30 , a source term with the opposite charge @xmath103 can be placed _ outside _ @xmath30 , at a distance of about 1 - 2 edge lengths . see for example the second and third examples in section [ sec : examples ] . in section [ sec : remove_harmonic ] a method will be described for calculating and removing the contribution of @xmath69 to the boundary conditions . at junction points , however , the function @xmath79 may be singular : as @xmath30 is an open set , boundary singularities are part of @xmath69 , so care should be taken when preforming the calculations described in section [ sec : remove_harmonic ] , especially in a numerical implementation of the technique . to avoid these problems , we subtract the junction singularities from the boundary value of @xmath80 before proceeding with removing @xmath69 . let @xmath104 be the locations of the junction points , and @xmath105 the charges as given by eq . ( [ eq : junc_charge ] ) , ( [ eq : junc_incidence ] ) . the complex function corresponding to @xmath106 is @xmath72 , so the contribution of the junctions is @xmath107 if a singularity is located on an inner boundary component , the logarithms must contain a branch cut somewhere in @xmath30 . in order to avoid adding a cut , we add another term to every hole , equal to minus the total charge in each hole . denote the inner boundaries by @xmath108 , and let @xmath109 be arbitrary points inside the respective holes . the following terms are therefore added to @xmath80 : @xmath110 in this way , the additional terms form an analytic function in @xmath30 , with no branch cut , and are a part of @xmath69 . since the contribution of @xmath69 is removed from @xmath80 as described in the next section , the exact form of this additional term , i.e. the choice of points @xmath111 , does not affect the results . in this section a technique for calculating @xmath69 is presented . after @xmath69 is calculated , its contribution to @xmath80 at the boundary can be subtracted . to simplify notation , let @xmath112 denote the sum of logarithms in @xmath113 : @xmath114 in the context of ip algorithms , the idea that the contribution of the harmonic part to the solution can be removed from the boundary information was suggested in @xcite . there , an ip problem in the unit disk @xmath115 is discussed . the value of the analytic part of @xmath80 on the boundary of @xmath115 can then be calculated by taking the fourier transform of @xmath80 , and leaving only the positive frequency components , as can be shown by considering the lourent series of @xmath80 in the unit disk . by its construction , this technique applies to functions defined on the unit disk . it can be extended to other domains , if a conformal mapping of the domain to the unit disk ( which exists according to the riemann mapping theorem ) is calculated . for our present purposes , since the domain @xmath30 in every problem input is different , using this technique would require constructing a mapping to the unit disk for each domain being meshed separately . furthermore , the domains in our problem may contain holes , which further complicates the matter . we therefore use an alternative approach , which is now described . it is based on the cauchy integral theorem , and is similar to the sum factorization step in the weiner - hopf technique @xcite , applied to a bounded domain ( @xmath116 , defined below ) in place of an infinite strip . as in @xcite , we assume there is a positive distance @xmath117 between the boundary of @xmath30 and the source closest to it , as is the case , for example , when the number of sources is finite . let @xmath116 be the set of points at a distance smaller than @xmath117 to the boundary , see fig . [ fig : cauchy],a .. as will become apparent later , the value of @xmath117 does not enter the calculation and is irrelevant , as long as there exists some @xmath118 as required . [ ptb ] we first consider a simply - connected domain @xmath30 , with boundary @xmath47 . define the flux through any curve @xmath18 : @xmath119 the flux @xmath120 through the boundary can be calculated : the value of @xmath17 at smooth boundary points is known from condition 2 ( eq . ( [ eq : cond2 ] ) ) , and the flux at junction points is calculated from condition 3 , as is explained in section [ sec : input_near_junc ] above . this flux can be shown to be a multiple of @xmath11 . [ remark : flux_calc]using conditions 2,3 and the fact that the rotation of a tangent in a simple curve is @xmath121 , the total flux through a boundary component @xmath122 can be shown to be equal to @xmath123 , where @xmath124 are the junction points , @xmath125 for each junction is given by eq . ( [ eq : junc_incidence ] ) , and @xmath126 for inner and outer boundary components , respectively . see also @xcite . to proceed with the decomposition calculation , we assume that @xmath127 , as calculated from eq . ( [ eq : bound_flux_def ] ) above . if this is not the case , following @xcite , we subtract a source term to the boundary conditions , centered at some point @xmath128 inside the domain : @xmath129 such that @xmath120 is zero . as will be shown below , the choice of @xmath128 does not affect the results . since @xmath120 was equal to a multiple of @xmath11 before , @xmath130 . when @xmath131 , the function @xmath80 is single - valued and analytic in @xmath116 , as follows from the integral representation of @xmath80 in eq . ( [ eq : f_is_phi_i_theta ] ) , along with @xmath131 . we now use the cauchy s theorem , stating that for every point @xmath132 : @xmath133 the boundary @xmath134 of @xmath116 has two boundary connectivity components : the outer boundary ( which is also the boundary of @xmath30 ) , and the inner boundary . denote them by @xmath135 and @xmath136 respectively , see fig . [ fig : cauchy],a .. eq . ( [ eq : cauchy_d_simply_con ] ) now reads @xmath137 the function @xmath138 is harmonic on @xmath116 . according to a decomposition theorem ( see @xcite , chapter 9 ) its decomposition into @xmath139 and @xmath140 , is also unique . it follows that the decomposition of the multi - valued complex function @xmath81 into @xmath141 and @xmath4 is unique , up to the arbitrary constant @xmath90 in eq . ( [ eq : f_is_phi_i_theta ] ) , chosen before to be zero . the two integrals expressions on the rhs of eq . ( [ eq : cauchy_on_omega ] ) correspond exactly to the two components of the ( unique ) decomposition of @xmath80 in @xmath116 as described in @xcite , hence @xmath142 where we have used the fact that @xmath143 . we would like to find @xmath141 on @xmath47 , but @xmath136 is not known , nor is @xmath80 on @xmath136 , so @xmath112 can not be computed directly from the second equation . the first equation can however be used , since @xmath80 is given on @xmath47 , so @xmath69 can be calculated on @xmath47 ( more precisely , since @xmath4 is defined in @xmath30 , it is the limit of @xmath4 as @xmath47 is approached ) . then @xmath112 on @xmath47 is given by @xmath144 , according to eq . ( [ eq : fp_def ] ) . finally , we add back the source term subtracted before @xmath145 we now turn to the case when @xmath30 is not simply - connected . we add source terms : one inside the domain , as discussed above ( eq . ( [ eq : source_in_d ] ) ) , and one inside every hole ( i.e. _ outside _ @xmath30 ) , such that the flux @xmath146 through every boundary connectivity element is zero , see fig . [ fig : cauchy],b .. @xmath147 the @xmath111 can be the same points used in section [ sec : remove_harmonic_junc ] , or other points . the results do not depend on the additional charges locations , as will be explained below . we now introduce cuts so that the boundary contains a single connectivity element ( the cuts play a part in the derivation , but drop out of the final calculation ) , see fig . [ fig : cauchy],b .. the cuts introduced also serve as the branch cuts of the logarithms . denote this new boundary @xmath148 . using @xmath149 , we proceed as when @xmath30 is simply connected : define @xmath116 as before , see fig . [ fig : cauchy],b . , and subtract the source term inside @xmath30 , as in eq . ( [ eq : source_in_d ] ) . according to eq . ( [ eq : cauchy_2_parts ] ) : @xmath150 the second integral on the rhs denotes the integral over the cuts introduced to form @xmath149 . note that each cut - path is traversed twice , back and forth . since the flux @xmath146 through each and every hole boundary is zero , the value of @xmath151 , when integrated along @xmath149 , is continuous across the branch cuts , and the integrations over each cut traversed in both directions cancel each other , and drop from the total integration . therefore @xmath152 as in eq . ( [ eq : cauchy_2_parts ] ) . as before , @xmath144 . finally , the source term inside @xmath30 is added back , as in eq . ( [ eq : source_in_d_back ] ) . the boundary value of @xmath141 obtained after this source term is added back may be multi - valued , with a branch cut discontinuity in the imaginary part that is a multiple of @xmath11 . this is a valid input to the next step , as explained in section [ sec : calc_locations ] below ( following eq . ( [ eq : xsi_def ] ) ) . the choice of the added singularities locations outside @xmath30 ( inside the holes ) does not affect the results : changing the location of a singularity with some charge from @xmath153 to @xmath154 is equivalent to adding a singularity with opposite charge at @xmath153 , and with the same charge at @xmath154 . since these two singularities lay outside @xmath116 , and are of opposite charge , this amounts to adding an analytic function to @xmath81 , which is removed by the cauchy integral technique described above . the result is also unaffected by the location @xmath128 of the source term subtracted inside @xmath30 , since its location only affects the @xmath141 and it is later added back . the choice of cuts in eq . [ eq : cauchy_with_cuts ] , of course , does not affect the cauchy integral calculation , since the cuts do not enter the final calculation , eq . ( [ eq : cauchy_final_calc ] ) . at this point , we are given the value of @xmath112 on the boundary @xmath47 of the domain @xmath30 . ( [ eq : f_def ] ) , we define @xmath155 = \exp\left [ { \displaystyle\sum\limits_{m=1}^{n_{c } } } k_{m}\ln\left ( z - z_{m}\right ) \right ] \text{.}\label{eq : xsi_def}\ ] ] note that while @xmath141 may be have a multi - valued imaginary part with a branch cut discontinuity : @xmath156 , for some integer @xmath14 , @xmath157 is single valued , since @xmath158 = 1 $ ] . since the @xmath38-values , @xmath37 , are integers , we can assume without loss of generality that @xmath159 . other charges may be formed by placing several sources at the same location . we can therefore write eq . ( [ eq : f_def ] ) as : @xmath160 where @xmath161 are the number of sources with @xmath162 respectively . according to the divergence theorem , the difference @xmath163 is equal to @xmath164 ( this can also be shown directly by calculating the integral of @xmath141 along @xmath47 ) . since @xmath120 is known ( see also remark [ remark : flux_calc ] ) , @xmath163 is known . therefore , to fix @xmath161 only one number remains to be chosen , e.g. @xmath165 . the choice of the total number of sources affects the accuracy of the reconstruction of the input data , see sections [ sec : ipsummary ] and section [ sec : implementation].[remark : divergence ] substituting eq . ( [ eq : fp_only_plus_minus ] ) , eq . ( [ eq : xsi_def ] ) now reads @xmath166 in eq . ( [ eq : rational_roots ] ) , the numbers @xmath167 are the roots of polynomials , and we proceed by presenting the polynomials in a different form . later , once the polynomials are found , the @xmath168 are recovered by finding the roots of the polynomials . we rewrite eq . ( [ eq : rational_roots ] ) as @xmath169 where @xmath170 are unknown coefficients . note that the prefactors of @xmath171 are indeed equal to @xmath76 , as can be seen by expanding the polynomials in eq . ( [ eq : rational_roots ] ) . to find the unknown coefficients numerically , we discretize eq . ( [ eq : rational_standard_poly ] ) , and give @xmath157 at @xmath172 points @xmath173 , @xmath174 , and write @xmath175 : @xmath176 the rhs of eq . ( [ eq : rational_discrete ] ) is a _ rational function interpolation _ of the @xmath177 data ( see e.g. @xcite ) . in the implementation described below , the unknowns @xmath178 are evaluated as follows . rearranging , eq . ( [ eq : rational_discrete ] ) reads @xmath179 these are @xmath172 linear equations for the @xmath165 unknowns : @xmath178 . since @xmath172 will typically be larger than @xmath165 , the solution will only be approximate , e.g. , a solution in the least - mean - square ( lms ) sense . once the unknown variables @xmath178 are found , the source terms locations @xmath180 are calculated by finding the roots of the two polynomials appearing in the rhs of eq . ( [ eq : rational_discrete ] ) . a solution in the lms sense , as described in eq . ( [ eq : linear_system ] ) , is not recommended if the resulting errors in the @xmath177 s are large , when compared with @xmath177 . this is due to two reasons : first , the input data @xmath177 appears in both sides of eq . ( [ eq : linear_system ] ) , so the errors in the lms solution might not reflect the errors in @xmath177 , and moreover , the lms solution may be sensitive to the choice of the coordinate system origin . secondly , an error in @xmath177 scales as a difference in the resolution ( since @xmath181 ) , rather than the more natural error definition given by the _ ratio _ of input and obtained resolution . this means that at low resolutions ( large cell - sizes ) , errors in @xmath182 might represent large relative resolution deviations . once the sources locations have been determined , the function @xmath2 can be calculated everywhere in the domain . the solution will usually be approximate , i.e. , the resulting sum of sources will only approximate the required @xmath183 , and some deviation from both boundary alignment and resolution requirements may be found . the trade - off between the two requirements can be partly controlled by choosing an appropriate harmonic @xmath33 in eq . ( [ eq : cond1 ] ) . in order to satisfy the cell - size requirements exactly , one can solve the ( direct ) poisson equation , given by eq . ( [ eq : cond1 ] ) , but this time with a known charge distribution , and with dirichlet boundary conditions , i.e. by specifying @xmath2 on the boundary . a different approach is to try and satisfy the boundary alignment conditions , by solving eq . ( [ eq : cond1 ] ) with newmann boundary conditions . this gives a new @xmath2 on the boundary . the difference between the input ( required ) @xmath184 , and the @xmath2 obtained on the boundary can be used to evaluate the quality of the source distribution calculated by the ip algorithm , and the total number @xmath165 of sources can be changed accordingly . note that when @xmath30 is not simply connected , solving @xmath33 with neumann boundary conditions ( condition 2 ) does _ not _ mean that exact boundary alignment is obtained , because condition 4 might not be exactly fulfiled . the steps followed in calculating the source distribution ( locations and charges ) can be summarized as follows : 1 . adjusting the input resolution near the junction points ( section [ sec : input_near_junc ] ) . 2 . subtracting the junction source terms ( sections [ sec : remove_harmonic_junc],[sec : remove_harmonic ] ) . @xmath185 3 . calculating the cauchy integral to find @xmath69 on the boundary . finding @xmath183 on the boundary . ( section [ sec : remove_harmonic ] ) . adding back the source inside @xmath30 ( section [ sec : remove_harmonic ] ) : @xmath186 5 . calculating the locations of sources ( section [ sec : calc_locations ] ) , using some initial total number of sources @xmath165 . 6 . solving the ( direct ) poisson problem to find the new @xmath81 approximating the original @xmath81 . according to the quality of reconstruction , step 5 can be repeated with a different total number of charges . once the function @xmath187 is set throughout the domain , the final mesh can be constructed . there are various possible approaches to this problem . we present a simple method that was used to create the examples in section [ sec : examples ] . in this method , the domain is cut along geodesics that follow the cross - field directions . every source is at the end of a cut , and every inner boundary component is connected to the outer boundary by a cut , see fig . [ fig : mapping_sketch],a .. the direction of the geodesics emanating from the sources and directed along the cross - field are calculated as explained in appendix a. the cut domain , which we denote by@xmath188 , does not contain any sources in its interior , and its boundary consists of a single connected component . a conformal mapping @xmath189 of @xmath190 into the plane can be constructed , with a conformal factor ( local scaling ) of @xmath191 . we consider @xmath190 and @xmath192 as lying in the complex plane . first , note that since @xmath190 contains no sources in its interior and is simply - connected , @xmath193 is uniquely defined in @xmath190 . denote by @xmath194 this single - valued @xmath193 . to calculate the mapped boundary @xmath192 , we calculate the function @xmath195 along the ( single ) boundary @xmath196 of @xmath190 : @xmath197 [ ptb ] if this mapping is used to create a mesh , by placing a cartesian grid on @xmath192 , and mapping it back to @xmath190 by taking the inverse of @xmath189 , the mesh obtained will be invalid : the grid will not be continuous across the cuts , will contain invalid cells at the sources , and cut cells at the boundaries . to correct this , and obtain a valid final mesh , we seek a function @xmath198 on @xmath196 approximating @xmath199 . a related problem , for surface meshes without boundary alignment , was addressed in @xcite . first , note that because the cuts are added along geodesics aligned with the cross field , @xmath199 is a piecewise straight line , whose straight segments are either horizontal or vertical . in order to obtain a valid final mesh , the new function @xmath198 should have the following properties : 1 . @xmath198 must map the closed path @xmath196 to a closed path . 2 . @xmath200 must be a piecewise straight line , composed of horizontal and vertical lines only . to each segment in @xmath201 corresponds a segment in @xmath202 , at the same order along the paths . each segment in @xmath200 must be directed in the same direction as the corresponding segment in @xmath203 , or be of zero length . if @xmath204 is a geodesic cut of @xmath196 , it is followed in both directions : @xmath204 and @xmath205 . then @xmath206 and @xmath207 must have the same length . this ensures the same number of cells on both sides of a single cut . we describe a simple algorithm for creating @xmath208 , meeting these requirements . to create @xmath200 , the path @xmath199 is followed , and the segments of @xmath209 are modified one after the other , in the order they appear in @xmath210 , see fig . [ fig : mapping_sketch ] . each segment is shifted to a parallel line of the integer grid . this is usually the nearest segment , except in the case the length of the segment directly before this segment dictates a different shift , according to properties ( 3,4 ) above , see fig . [ fig : mapping_sketch],c .. this algorithm requires that at least one junction exist in @xmath47 , which is mapped to a right angle in @xmath201 . the algorithm starts from the segment following this junction , so that when the last segment is reached and shifted , a change in the length of the first segment will be allowed . the resulting modified path is similar to @xmath211 , only with different segments lengths . @xmath200 is then defined as the composition of @xmath199 followed by a linear mapping of each segment of @xmath203 to the corresponding segment in the modified path . once @xmath212 is known , @xmath213 can be defined . in the implementation described below this is done by placing a triangular mesh in @xmath190 , and solving the laplace equation for both @xmath214 and @xmath215 on @xmath190 . @xmath198 obtained in this way is not in general analytic , since the cr conditions might not hold between @xmath214 and @xmath216 , so the final mesh will not exactly conformal , but only approximately , as @xmath198 approximates the analytical function @xmath217 . the mesh edges are then extracted by tracing the integer - valued lines of @xmath218 and @xmath219 . once the connectivity of the mesh has been established , a final smoothing procedure which involves all the interior vertices , including those on the cut path , is preformed . in the definition of @xmath198 , both @xmath214 and @xmath215 are harmonic functions . a standard finite difference approximation to a harmonic function on a uniform grid , assigns to each vertex the average of its four neighboring vertices . this is applied to all vertices inside the mesh , including those on the cuts of @xmath196 , which can be interpreted as a generalization of the defintion of @xmath198 ( to be more precise , in this smoothing procedure it is @xmath220 which is assumed to be harmonic ) . this procedure is the well - known laplacian smoothing procedure , in which the location of every vertex is equal to the average location of its neighboring vertices , and in this application it is theoretically justified if @xmath221 is approximately @xmath217 , i.e. if @xmath212 is close to @xmath199 . the algorithm described in the previous section was implemented in matlab @xcite , using standard matlab functions , such as a least - mean square solver for linear systems , an ode solver , etc .. to calculate the cauchy integral ( section [ sec : remove_harmonic ] ) , the boundary was approximated by a piecewise linear path between sampled points , with a constant value of @xmath2 on each segment . this allows the integral over each segment of the path to be calculated analytically , and the path integral there is equal to the sum of the integral over the segments . this is a low order approximation , but was sufficient for the present purposes . the linear system , eq . ( [ eq : linear_system ] ) , was solved in the least - mean - square sense , using matlab s ` mldivide ` function . note that the coefficients in this system contain powers of the @xmath173 s , the boundary data locations . this leads to ill - conditioned linear systems , which are more sensitive to the numerical round - off errors , as the number of sources @xmath161 increases , or the range of @xmath222 increases . tests indicate that when the solver issues a warning , indicating that the matrix is rank deficient to the working precision , the results of the ip algorithm become unreliable ( this was not the case in the examples in section [ sec : examples ] ) . for input domains where many sources or a high @xmath222 range are required , a different solver implementation may be needed , such as a varying precision arithmetic computation , working at higher floating - point accuracies . after sources locations are recovered by taking the roots of the polynomials , the function @xmath33 of eq . ( [ eq : cond1 ] ) is calculated , given the newmann boundary conditions in eq . ( [ eq : cond2 ] ) . this was done using the method of fundamental solutions ( mfs ) , see e.g. @xcite , where the solution to the laplace for @xmath33 is approximated by a sum of ( real ) source terms , i.e. functions of the form @xmath223 , with the locations @xmath224 lying _ outside _ the domain @xmath30 where @xmath33 is harmonic . the charges @xmath225 are then calculated to best fit the boundary conditions . once the function @xmath2 is given everywhere in @xmath30 , the directions of the star - geodesics at each source are calculated , as described in appendix a. the star - geodesics , emanating from the source were then traced by solving the differential equation , eq . ( [ eq : kappa_dphi_dn ] ) , using a runge - kutta ode solver . the cut - tree was constructed by taking one source at a time at some arbitrary order , and introducing a cut along a star - geodesic of the added source , that is closest to the cut - tree constructed so - far . this construction of the cut tree is admitantly arbitrary , and may not be optimal in some cases . lastly , the algorithm described in section [ sec : create_final ] is applied . in the first example , a domain with two boundary components is meshed , see fig . [ multi_diamond_var_res],a . the side lengths of the outer and inner squares are @xmath226 and @xmath227 respectively , centered at the origin , and the inner square is rotated at angle @xmath228 radians . the ratio of input resolution between outer and inner boundaries is @xmath229 . the inner and outer boundaries were sampled with @xmath230 and @xmath231 points , respectively , which were used both for calculating the cauchy integral , and for the sources locations calculation . singularities of @xmath2 are not required at the junctions points , since all inner angles are multiples of @xmath11 radians . as there is no flux of @xmath232 through either boundary ( because the boundaries curvature is @xmath233 , see eq . ( [ eq : cond1 ] ) , and @xmath2 is regular at the junctions ) , the total charge inside is zero ( see remark [ remark : divergence ] ) . hence an equal number of @xmath234 and @xmath235 charges were used . [ multi_diamond_var_res],a . , shows the domain boundary , together with the calculated sources locations . squares mark sources with @xmath234 , and circles sources with @xmath235 . the distribution is composed of @xmath236 sources of each type , i.e. a total of @xmath237 sources . the value of @xmath2 obtained by solving the poisson equation with this charge distribution is shown in fig . [ multi_diamond_var_res],b . , and compared to the input @xmath2 requested , denoted by @xmath238 . the difference @xmath239 is also plotted . for visual clarity , the average of the @xmath238 was subtracted from both @xmath2 and @xmath238 ( this is just a scaling of the resolution by a constant factor ) ; this was also done in the resolution comparisons below , in fig . ( [ fig : multi_ex2]),c . , fig . ( [ fig : multi_ex3 ] ) , b . and fig . ( [ fig : multi_ex4]),c .. the number of sources was chosen to be the smallest for which @xmath240 at all boundary points . this criterion is used for choosing the number of sources is also used in the following examples . the cut - tree used in creating the final mesh is shown in fig . [ multi_diamond_var_res],c . , and the final mesh in shown fig . [ multi_diamond_var_res],d . [ ptbh ] the domain in the second example is the union of three unit radius circles which pass through the origin , see fig . [ fig : multi_ex2],a . the boundary has three junctions , all with the same inner angle , @xmath241 . boundary alignment requires @xmath2 to be singular at the junctions , corresponding to sources with charge @xmath242 , or @xmath38-value @xmath243 , see eq . ( [ eq : junc_charge]),([eq : junc_incidence ] ) . as explained in section [ sec : input_near_junc ] , singular functions can be added the input resolution in order to avoid clustering of sources near the junction points . in this example , we compose the input @xmath2 from a sum of pairs of source terms , with @xmath38-values as marked in fig . [ fig : multi_ex2 ] , a .. the pairs of nearby sources with opposite charge create the desired singularities at the junctions , and have only a small effect at distances large compared with the distance between the sources of each pair . we can now calculate the total flux of @xmath2 through the boundary in this example . the ( singular ) flux through each junction is @xmath244 ( positive value when @xmath232 is directed outward ) . the flux through each of the three arcs is @xmath245 along the arc , and since for each arc @xmath246 , the total flux through the boundary is @xmath247 . this is in agreement with the formula given in remark [ remark : flux_calc ] . therefore , the sum of @xmath38-values of the sources inside the domain should be @xmath248 , so @xmath249 . hence , @xmath250 of eq . ( [ eq : source_in_d ] ) was @xmath248 ; @xmath251 was used . the lowest number of sources for which @xmath240 , shown in fig . [ fig : multi_ex2],b . , has @xmath252 , @xmath253 . the boundary was sampled at @xmath254 points . [ ptbh ] the boundary in the third example is composed of two boundary components , each with a single junction point . the inner angles at the inner and outer boundary junctions are @xmath255 respectively , so a singularity of @xmath2 will form at the inner boundary junction . as in the previous example , we add a singular function to the input @xmath2 , by adding a pair of source terms , one at the junction point with @xmath256 , as obtained from eq . ( [ eq : junc_charge]),([eq : junc_incidence ] ) with @xmath257 . no inner source term is required , and from the flux trough the hole boundary it follows that @xmath258 . after subtracting the junction source term with @xmath256 ( see section [ sec : junc_points ] ) a source with total charge of @xmath259 is added inside the hole , see fig . [ fig : multi_ex3],a .. the inner and outer boundary components are sampled at @xmath260 and @xmath261 points , respectively . the distribution with @xmath262 , the smallest number of sources for which @xmath240 , is plotted at fig . [ fig : multi_ex3],a . , with squares and circles representing charges as in previous examples . two pairs of @xmath263 charges are located inside the inner boundary , i.e. , outside the domain . their constitute a correction to the homogeneous @xmath33 , which is very small , since in each pair the opposite charges almost overlap . the obtained @xmath2 reconstruction is given in fig . [ fig : multi_ex3],b .. the final mesh , fig . [ fig : multi_ex3],c . , contains @xmath264 irregular vertices , @xmath265 of each type . this is because two pairs of sources lay outside the domain , and another pair ( the pair of sources closest to the inner junction ) was eliminated in the creation of the final mesh ( section [ sec : create_final ] ) : the two opposite charges where united , giving a zero total charge . [ ptbh ] in the final example the boundary of the square @xmath266\times\lbrack-2,2]$ ] was assigned a varying input resolution proportional to : @xmath267 the resolution thus varies by a factor of @xmath268 within the domain . as always , eq . ( [ eq : phi_from_res ] ) , @xmath269 . the total charge is zero , so @xmath270 . the boundary was sampled at @xmath271 points . a source distribution with @xmath240 was obtained with just @xmath226 charges , i.e. with @xmath272 . [ fig : multi_ex4],a . shows the source distribution , fig . [ fig : multi_ex4],b . shows the obtained @xmath2 vs. the input @xmath2 , and the final mesh is shown in fig . [ fig : multi_ex4],c .. it is instructive to compare the star - geodesics and cut - tree with the final mesh . the two are overlaid in fig . [ fig : multi_ex4],d . ( this would create a crowded appearence when more sources are present ) . the black solid line traces the cut - tree , the dashed lines are additional star - geodesics , and the mesh is drawn with the background in gray lines . [ ptbh ] cell shape quality was measured using a variant of the quality measure @xmath273 @xcite , as defined in @xcite . @xmath274 represents a square cell , while @xmath275 represents a cell with an inner angle of @xmath276 . cells with a high aspect ratio are also given low @xmath273-values . table 1 shows the minimum and average @xmath273 values , and the total number of cells , for the examples in fig . [ multi_diamond_var_res]-[fig : multi_ex4 ] . for each example , the tabulated information is given for the mesh shown in the corresponding figure , and for a finer mesh of the same domain , where the resolution function was doubled ( prodcing a mesh with about four times the number of cells ) . note that multiplying the resolution by a factor amounts to adding a constant to @xmath2 , which does not change the source distribution obtained from the ip algorithm , as this constant , which is part of @xmath69 , is readily removed when @xmath69 is subtracted . @xmath277{|c|c|c|c|}\hline & num . cells & min . $ ] @xmath278@xmath279 @xmath280 table 1 : mesh statistics . an unstructured quadrilateral mesh generation scheme in planar domains was presented . the method rests on a theoretical foundation , linking the mesh generation problem with the inverse poisson ( ip ) problem . an ip solution algorithm is presented , whose output is interpreted as the location and type ( degree ) of irregular vertices in the domain . the continuum fields obtained , describing mesh resolution and directionality , are conformal everywhere except on the irregular vertices , and fit the required input properties at the boundary , or at other user - defined locations . an algorithm for creating a valid final mesh is also presented . example meshes feature irregular vertices where they are needed , in combination with highly regular regions where possible . directions for future work include more sophisticated methods for solving the rational function interpolation equations , and for constructing the final mesh . the relations between conformal unstructured mesh generation , the ip problem and rational function interpolation raise many research questions . these may lead to a deeper understanding of the properties of high quality meshes , and to better algorithms for creating them . * acknowledgements . * the author would like to thank mirela ben - chen , shlomi hillel , dov levine , yair shokef and vincenzo vitelli for helpful discussions and critical reading of the manuscript . a cross is defined at every point @xmath49 of the domain , that is not a singularity of the function @xmath2 . geodesic curves that start at @xmath49 can be drawn in all four directions of the cross . by the definition of a cross - field , such a geodesic will be aligned with the cross - field everywhere along the curve . if @xmath49 is a singular point , with @xmath38-value @xmath281 , a cross is not defined at @xmath49 , but there are @xmath282 geodesics that are incident on @xmath49 and follow the cross - field directions elsewhere on the curve . these will be called _ star - geodesics_. for example , the geodesics drawn in fig . [ multi_diamond_var_res],c . and [ fig : multi_ex4],d . are star - geodesics . denote the angle from the @xmath10-axis around @xmath49 by @xmath283 , and the cross direction when @xmath49 is approached from direction @xmath283 by @xmath284 , see fig . [ fg : star_directions ] . to calculate the directions in which star - geodesics emanate from @xmath49 , we first calculate the cross when @xmath49 is approached from some direction , e.g. the positive @xmath10-axis . this can be done using eq . ( [ eq : theta_change_par_tranport ] ) along a curve from the boundary to @xmath49 which approaches @xmath49 from the direction @xmath285 . once @xmath286 is known , @xmath287 for any @xmath283 can be calculated by using eq . ( [ eq : theta_change_par_tranport ] ) along a small circular arc around @xmath49 at radius @xmath53 , @xmath97 . the singularity term at @xmath2 is @xmath288 , and according to eq . ( [ eq : theta_change_par_tranport ] ) @xmath289 note that contributions to @xmath2 that are regular at @xmath49 do not affect @xmath287 for @xmath290 . the star - geodesic directions are those for which the cross is directed along the ray from @xmath49 , or @xmath291 with @xmath46 . substituting eq . ( [ eq : app_a_th_calc ] ) into eq ( [ eq : app_a_star_cond ] ) and rearranging , we find @xmath292 it is easy to show that there are exactly @xmath293 different @xmath283-values for which this equation is fulfilled . they are equally distributed around @xmath49 . m. hmlinen , r. hari , r. j. ilmoniemi , j. knuutila and o. v. lounasmaa , magnetoencephalography - theory , instrumentation , and applications to noninvasive studies of the working human brain , _ reviews of modern physics _ * 65 * issue 2 ( 1993 ) 413 - 497 .

we study the dissipative quantum harmonic oscillator with general non - thermal preparations of the harmonic oscillator bath . the focus is on equilibration of the oscillator in the long - time limit and the additional requirements for thermalization . our study is based on the exact solution of the microscopic model obtained by means of operator equations of motion , which provides us with the time evolution of the central oscillator density matrix in terms of the propagating function . we find a hierarchy of conditions for thermalization , together with the relation of the asymptotic temperature to the energy distribution in the initial bath state . we discuss the presence and absence of equilibration for the example of an inhomogeneous chain of harmonic oscillators , and illustrate the general findings about thermalization for the non - thermal environment that results from a quench .

1205.2068

equilibration can be defined as the evolution of a system out of equilibrium towards a stationary state in the long - time limit . for quantum systems , the question arises how equilibration is possible in spite of the linear and unitary time evolution , how the stationary state depends on the initial conditions , and to which extent it can be described as a thermal state . general arguments relate equilibration to dephasing of quantum states @xcite . starting from the expansion of an initial state @xmath0 in the eigenstates @xmath1 of the hamiltonian @xmath2 , the time evolution of an operator expectation value is given by @xmath3 in the thermodynamic limit we can expect that only diagonal terms @xmath4 survive for @xmath5 , such that the long - time limit of the expectation value is @xmath6 \quad ( n \to \infty ) \;,\ ] ] with the density matrix @xmath7 . this argument can be justified with the riemann - lebesgue lemma @xcite that states @xmath8 for any integrable function @xmath9 ( here : the density of states @xmath10 ) . although this argument explains the origin of equilibration , not much is learned about the properties of the stationary state @xmath11 . especially the question of thermalization is left open . in this paper we study equilibration and thermalization of dissipative quantum harmonic oscillators , using the standard model of a central oscillator coupled to a harmonic oscillator bath . for this example we can determine the stationary state @xmath11 explicitly and analyze its dependence on the initial conditions completely . crucially , we allow for arbitrary non - thermal bath preparations in our study . thermalization is subject to additional conditions in this more general situation , and we show how the temperature of the asymptotic stationary state is obtained from the initial energy distribution of the oscillator bath rather than from the initial bath temperature . we also include the case study of an interaction quench in an infinite harmonic chain , where undamped oscillations can prevent equilibration at strong damping . the dissipative quantum harmonic oscillator is studied extensively in the literature @xcite , covering such diverse topics as brownian motion @xcite , quantum fluctuations @xcite , driven dissipative systems @xcite , entanglement @xcite , the existence of local temperatures @xcite , or the second law of thermodynamics @xcite . reviews are given , e.g. , in @xcite . with an exact solution this model is also an important example for the derivation of master equations @xcite , the discussion of fundamental statistical relations such as fluctuation - dissipation theorems @xcite and their connection to detailed balance and the kubo - martin - schwinger condition @xcite , or for the assessment of numerical methods that provide a perspective for non - linear models @xcite . it appears , however , that the questions addressed here have not been previously analyzed in detail , especially not for non - thermal bath preparations . to obtain our results we proceed as follows . after introduction of the model in sec . [ sec : model ] , we construct the exact solution for non - thermal initial states in sec . [ sec : eom ] , including the propagating function in sec . [ sec : prop ] . further details , including the extension to driven oscillators , are given in app . [ app : deriv ] and app . [ app : prop ] . the central results for equilibration and thermalization are formulated in sec . [ sec : equil ] . we discuss these results for the example of an infinite chain of harmonic oscillators in sec . [ sec : chain ] , before we conclude in sec . [ sec : summary ] . the hamiltonian for the dissipative quantum harmonic oscillator , @xmath12 is the sum of the contribution of the central oscillator , @xmath13 \;,\ ] ] the contribution of the harmonic oscillator bath , @xmath14 \;,\ ] ] and the linear interaction term @xmath15 in these expressions , @xmath16 , @xmath17 are position and momentum operators with canonical commutation relations , e.g. @xmath18={\mathrm i}\delta_{\mu \nu}$ ] . summations over greek indices , used for bath oscillator operators @xmath16 , @xmath17 , run from @xmath19 . we suppress an index for the central oscillator operators . the size of the coupling constants @xmath20 is restricted by the positivity condition @xmath21 it guarantees that the normal modes of the total hamiltonian @xmath22 have real frequencies , such that @xmath22 is bounded from below @xcite . a positive hamiltonian can always be obtained through addition of the term @xmath23 , which leads to renormalization of the central oscillator frequency @xcite . we prefer the present form of the hamiltonian since it allows for a more natural treatment of the harmonic chain in sec . [ sec : chain ] . of primary interest to us is the central oscillator density matrix @xmath24 \;,\ ] ] which is obtained from the initial state @xmath25 through propagation with the total hamiltonian @xmath22 and subsequent evaluation of the partial trace @xmath26 over the bath degrees of freedom . a natural choice for @xmath25 are factorizing initial conditions @xmath27 which correspond to the picture that at @xmath28 the previously isolated central oscillator is brought into contact with the oscillator bath . the restriction to factorizing initial conditions is not essential for the following derivations , especially not for the long - time limit in sec . [ sec : equil ] , but it is a natural assumption that simplifies the presentation . for example , mixed central / bath oscillator terms drop out of the expressions for the central oscillator variance ( see sec . [ sec : expvalues ] ) . the central oscillator density matrix @xmath29 can be obtained in various ways , e.g. through transformation of @xmath22 to normal modes @xcite or by using path integrals @xcite based on the feynman - vernon influence functional formalism @xcite . the arguably simplest approach is the direct solution of the heisenberg equations of motion for the operators @xmath30 , @xmath31 , which reduces to the solution of a classical equation of motion . the initial conditions @xmath32 and @xmath33 enter only the evaluation of central oscillator expectation values , such that we can allow for general initial bath states . the full solution is then given by the propagating function . as further detailed in app . [ app : deriv ] , the central piece of information is the solution @xmath34 of the classical equation of motion @xmath35 which is subject to the conditions 1 . @xmath36 solves eq . for @xmath37 , 2 . @xmath38 for @xmath39 , 3 . the initial conditions are @xmath40 , @xmath41 . we here introduced the damping kernel @xmath42 the function @xmath36 can be calculated as the fourier transform @xcite @xmath43 of the function @xmath44 writing @xmath45 . we note that the positivity condition implies that the poles of @xmath46 occur on the real axis , such that @xmath36 is a quasiperiodic function for finite @xmath47 while @xmath48 for @xmath5 is possible in the thermodynamic limit @xmath49 . an explicit example for the computation of @xmath36 is given for the harmonic chain in sec . [ sec : chain ] ( see eq . ) . to proceed , we introduce the partial fourier transforms @xmath50 @xmath51 and define the matrices @xmath52 @xmath53 { \mathop{{\rm re}}}\tilde{v}(t , \omega ) & \dfrac{{\mathop{{\rm im}}}\tilde{v}(t , \omega)}{\omega } \end{pmatrix } \;.\ ] ] we now obtain the central oscillator operators from the matrix equation @xmath54 eq . gives the operators @xmath30 , @xmath31 as linear combinations of the operators @xmath55 and @xmath56 . this allows us to express central oscillator expectation values for @xmath57 in terms of the initial expectation values at @xmath28 . the linear expectation values are given by the matrix equation @xmath58 \langle p(t ) \rangle \end{pmatrix } = { \mathbf u}(t ) { \mathbf x}(0 ) + { \mathbf i}(t ) \,,\ ] ] with the same shape as eq . . in addition to the initial expectation values @xmath59 it contains the contribution @xmath60 where we mark the initial bath expectation values @xmath61 with a breve @xmath62 as a notational convention . note that if @xmath63 , e.g. for a thermal bath , the ` noise term ' @xmath64 vanishes . then , position @xmath65 and momentum @xmath66 of the central oscillator follow the classical equation of motion . for the quadratic expectation values we define the variance of operators @xmath67 , @xmath68 as @xmath69 which simplifies to @xmath70 for @xmath71 , and write @xmath72 . we combine the central oscillator variances into the real symmetric matrix @xmath73 \sigma_{q p}(t ) & \sigma_{p p}(t ) \end{pmatrix } \;,\ ] ] and denote the initial bath variances with the matrix @xmath74 \sigma_{q_\mu p_\nu}(0 ) & \sigma_{p_\nu p_\mu}(0 ) \end{pmatrix } \;.\ ] ] note the index swap in the off - diagonal elements , and recall that mixed central oscillator / bath variances such as @xmath75 vanish for our choice of factorizing initial conditions . we now obtain with eq . the matrix equation @xmath76 similar to eq . , the first term results from the time evolution of the central oscillator according to the classical equation of motion , and appears in the same form for an isolated oscillator . only the second term @xmath77 depends on the initial bath oscillator variances @xmath78 . mixed terms in @xmath79 , @xmath80 do not appear for factorizing initial conditions . because @xmath36 is a quasi - periodic function for a finite number @xmath47 of bath oscillators , equilibration becomes possible only in the thermodynamic limit @xmath49 . we assume that for @xmath49 the density of states @xmath81 converges to a continuous function . note that @xmath82 for @xmath83 since the bath oscillator frequencies are positive . the coupling constants appear in the damping kernel @xmath84 and in eq . as @xmath85 , and must thus scale as @xmath86 . we assume that @xmath87 with a continuous function @xmath88 , and introduce the bath spectral function @xmath89 with @xmath90 for @xmath91 . the damping kernel is now given as @xmath92 and the positivity condition reads @xmath93 the function @xmath46 in eq . for @xmath36 can be written as @xmath94 under mild assumptions , the evaluation of the @xmath95-integral in this equation is possible by contour integration and results in @xmath96 for @xmath97 , where the complex function @xmath98 with @xmath99 is the analytic continuation of @xmath100 into the upper half of the complex plane ( see sec . [ sec : chain ] for an example ) . for future use in sec . [ sec : equil ] we note the relation @xmath101 that follows from this representation . the analytic properties of @xmath46 determine the behavior of @xmath36 in the long - time limit , which is essential for equilibration ( see condition ( e0 ) in sec . [ sec : equil ] ) : it is @xmath48 for @xmath5 if and only if @xmath46 has no isolated poles . the linear expectation values @xmath102 enter eq . with the prefactors @xmath103 . to obtain a finite result for the sum over @xmath47 terms , also @xmath102 has to scale as @xmath86 , which leads to the ansatz @xmath104 with a continuous vector - valued function @xmath105 . then , eq . becomes @xmath106 the variances @xmath107 enter the sum in eq . with the prefactors @xmath108 . we must now distinguish between the @xmath109 off - diagonal terms @xmath110 , which require an additional @xmath111 prefactor for convergence , and the @xmath47 diagonal terms @xmath112 . therefore , we make the ansatz @xmath113 with continuous matrix - valued functions @xmath114 and @xmath115 . then , @xmath116 from eq . is the sum of the off - diagonal term @xmath117 and the diagonal term @xmath118 if the initial bath state is uncorrelated , such as for a thermal bath or a general product state @xmath119 , the off - diagonal term @xmath120 vanishes . when we construct the propagating function in the next subsection , we will conveniently assume that the initial bath state @xmath33 is a gaussian state . for the long - time limit , the situation of interest here , this assumption can be justified in the thermodynamic limit on general grounds @xcite . the principal mechanism is illustrated with counting arguments of the following kind : consider an uncorrelated bath state , where only @xmath47 diagonal terms contribute in any sum over the bath oscillators . if we consider a higher order cumulant of bath operators , say @xmath121 as mentioned before , it appears with a prefactor @xmath122 . therefore , the total contribution of these cumulants scales as @xmath123 and vanishes in the limit @xmath49 . similar counting arguments can be given for cumulants involving two or more bath oscillators in the presence of correlations . because higher order cumulants vanish and only linear and quadratic bath expectation values survive the @xmath124 and @xmath5 limit , we can treat the bath state as gaussian in any calculation of the central oscillator density matrix . for the formulation and proof of a strict result , which is involved even under some simplifying assumptions , see @xcite . knowledge of the expectation values @xmath125 , @xmath126 does not suffice to obtain the central oscillator density matrix @xmath29 , unless we restrict ourselves completely to gaussian oscillator states ( cf . otherwise , the general solution is given by the propagating function @xmath127 that , in position representation , expresses the density matrix @xmath128 for @xmath129 as @xmath130 this expression must hold for all @xmath32 and @xmath129 , and a fixed initial bath state @xmath33 . the propagating function can be calculated using path integrals and the result for a thermal bath is given , e.g. , in @xcite . within our approach it is more natural to construct the propagating function directly , using only that an initial gaussian state of the joint central / bath oscillator system remains a gaussian state during time evolution with the bilinear hamiltonian @xmath22 . with respect to the final remarks in sec . [ sec : tdlimit ] , we assume a gaussian bath state @xmath33 . we can then consider the most general ansatz for @xmath127 that maps an initial gaussian state @xmath32 in eq . onto a gaussian state @xmath29 for @xmath129 , and will find that the parameters of this ansatz are fully specified through the linear maps , of @xmath125 , @xmath126 . the result is valid for arbitrary @xmath32 in eq . , but we do not need to consider non - gaussian @xmath29 explicitly . to translate this argument into equations we work with the wigner function @xcite @xmath131 instead of the density matrix @xmath132 in position representation ( see also refs . @xcite for a related calculation ) . the propagating function @xmath133 is defined by the relation @xmath134 where we write @xmath135 with @xmath136 and @xmath137 for abbreviation . note that @xmath138 and @xmath139 are real functions . a gaussian state to given @xmath125 , @xmath126 has the wigner function @xmath140 } { 2 \pi \sqrt{\det \mathbf{\sigma}(t ) } } \;,\ ] ] and the most general expression for @xmath141 that respects this structure is an exponential function of the 14 linear and quadratic terms in the coordinates @xmath142 . the normalization @xmath143 of wigner functions implies the condition @xmath144 on the propagating function , which fixes the prefactors of the 5 terms @xmath145 in the initial coordinates . this leaves @xmath146 free parameters that have to be fixed in accordance with the linear transformations , of expectation values . the final result is @xmath147 } { 2 \pi \sqrt { \det \mathbf{c}(t ) } } \;,\ ] ] where the @xmath148 parameters are the entries of the @xmath149 matrix @xmath79 from eq . , the symmetric and positive definite @xmath149 matrix @xmath150(t ) from eq . , and the two - dimensional vector @xmath64 from eq . . in order to check that this expression indeed reproduces the transformations , , we can express the expectation values at @xmath129 in terms of those at @xmath28 through the evaluation of simple gaussian integrals . to give an example , it is @xmath151 the integral of @xmath152 over @xmath153 is a gaussian integral with a linear term , and gives @xmath154 the final integration over @xmath155 in eq . , which now involves the right hand side of , generates the initial expectation values @xmath156 , @xmath157 therefore , we obtain the relation @xmath158 in accordance with eq . . following this recipe , we find that the given expression for the propagating function @xmath141 reproduces the entire transformations , of the expectation values @xmath125 , @xmath126 , as we required . if @xmath159 , we get a representation of the distribution @xmath160 from eq . . in particular for @xmath28 , where @xmath161 , @xmath162 in addition to @xmath163 , we have the correct result @xmath164 in eq . . we note that the conveniently simple derivation of @xmath141 relies on the use of wigner functions . of course , the expressions for @xmath165 in position representation often reported in the literature can be recovered from eq . ( see app . [ app : prop ] ) . the results from the previous section allow us to study the behavior of the central oscillator density matrix @xmath29 in the long - time limit @xmath5 . we can classify the behavior according to the general criteria of equilibration and thermalization . equilibration means convergence to a stationary state as expressed in the two conditions * the central oscillator density matrix @xmath29 converges for @xmath5 , * the stationary state @xmath166 is independent of @xmath32 . note that @xmath167 will depend on the initial bath state @xmath33 . note further that the above definition of equilibration does not distinguish between stationary equilibrium states and stationary non - equilibrium states with finite heat flows . the latter can not arise for a single bath with continuous initial conditions as in eq . such that condition ( e1 ) is sufficient for the present study . equilibration ( e1 ) implies convergence of central oscillator expectation values for @xmath5 . this , in turn , requires convergence of the matrix @xmath168 in eqs . , . because the only stationary solution of the homogeneous differential eq . is @xmath169 , convergence of @xmath168 is equivalent to @xmath170 or @xmath48 for @xmath5 . therefore , we assume in this section the condition * @xmath48 for @xmath5 as the prerequisite for equilibration ( e1 ) . under this assumption , we will be able to show convergence of expectation values and , building on this result , convergence of the central oscillator density matrix . in the weak damping limit , condition ( e0 ) is equivalent to @xmath171 ( taking the thermodynamic limit for granted ) . this expresses the basic fact that equilibration occurs through energy exchange with the environment , which is not possible for an isolated oscillator with @xmath172 . we note that a small value of @xmath173 can result in long transients that prevent equilibration over the observation time . thermalization additionally requires that the stationary state @xmath174 is a thermal state , and we have the three increasingly stronger properties * the stationary state @xmath174 is a thermal state , * the stationary state is a thermal state @xmath175 of the central oscillator , * the temperature @xmath176 of the stationary thermal state @xmath174 is independent of the central oscillator frequency . we will see that the stationary state is always gaussian , which implies property ( t1 ) . property ( t2 ) reduces to an equipartition condition on the central oscillator variances that determine the gaussian state , while property ( t3 ) leads to a strong condition on the initial bath state . the assumption @xmath170 for @xmath5 implies that the terms @xmath177 in eq . and @xmath178 in eq . drop out of the expressions for @xmath179 and @xmath180 in the long - time limit . only the terms @xmath64 and @xmath116 , which depend exclusively on the initial bath preparation , can survive the @xmath5 limit : all information about the initial central oscillator state is lost . we can not immediately draw a conclusion about the long - time behavior because the functions @xmath181 , @xmath182 from eqs . , do not converge for @xmath183 . instead , we note that @xmath181 behaves asymptotically as @xmath184 similarly , it follows @xmath185 for @xmath5 from eq . . consequently , the matrix @xmath186 behaves asymptotically as @xmath187 -\omega { \mathop{{\rm im}}}\tilde{u}_{as}(t,\omega ) & { \mathop{{\rm re}}}\tilde{u}_{as}(t,\omega ) \end{pmatrix } \quad ( t \to \infty ) \;,\ ] ] and remains oscillating for @xmath5 even if @xmath48 . the contributions to the term @xmath64 in eq . , say to @xmath188 , are of the form @xmath189 the integrand depends on @xmath190 through the factor @xmath191 from eq . , such that the integral is the fourier transform of an integrable ( by assumption even continuous ) function of @xmath95 . if we recall the riemann - lebesgue lemma we see that @xmath192 for @xmath5 . altogether , it follows that the position and momentum expectation values vanish in the long - time limit , i.e. @xmath193 for @xmath5 . for the variances , a finite contribution can survive the @xmath5 limit because the squares of the matrix elements of @xmath186 occur in @xmath116 . for example , the diagonal term @xmath194 from eq . contributes to @xmath195 the integral @xmath196 of the function @xmath197 ^ 2 \ , \breve{\sigma}_{qq}^{(1)}(\omega ) \nonumber\\ & & + \frac{2 [ { \mathop{{\rm re}}}\tilde{u}(t,\omega ) ] [ { \mathop{{\rm im}}}\tilde{u}(t,\omega ) ] } { \omega } \breve{\sigma}_{qp}^{(1)}(\omega ) \nonumber\\ & & + \frac{[{\mathop{{\rm im}}}\tilde{u}(t,\omega)]^2}{\omega^2 } \breve{\sigma}_{pp}^{(1)}(\omega ) \;.\end{aligned}\ ] ] here we write , using the notation from eq . , @xmath198 c_{q p}^{(1)}(t ) & c_{p p}^{(1)}(t ) \end{pmatrix}\ ] ] for the matrix elements of @xmath199 and @xmath200 \breve \sigma_{q p}^{(1)}(\omega ) & \breve \sigma_{p p}^{(1)}(\omega ) \end{pmatrix}\ ] ] for the matrix elements of @xmath201 from eq . . the contribution from the first term in @xmath202 is @xmath203 ^ 2 \breve\sigma^{(1)}_{qq}(\omega ) \ , { { \mathrm d}}\omega \;.\ ] ] if we expand the square @xmath204 ^ 2 $ ] according to @xmath205 ^ 2 = \frac{|z|^2}{2 } + \frac{z_r^2-z_i^2}{2 } \cos 2 \omega t - z_r z_i \sin 2 \omega t \,,\ ] ] for a complex number @xmath206 with @xmath207 , @xmath208 , we see that in the above integral a contribution @xmath209 remains finite for @xmath5 , while the oscillatory terms with @xmath210 , @xmath211 vanish according to the riemann - lebesgue lemma . similar expressions are obtained for the remaining terms in @xmath194 . the off - diagonal term @xmath120 from eq . is given by a double fourier integral and contains only oscillatory terms in the two frequencies @xmath212 , @xmath213 . therefore , @xmath214 for @xmath5 . we can now collect the finite contributions from the different terms in @xmath194 , to find that the central oscillator variances converge to stationary values @xmath215 in the long - time limit . they are given by @xmath216 @xmath217 @xmath218 where @xmath219 comparison with eqs . , gives the alternative expressions @xmath220 @xmath221 recall that @xmath222 is a continuous function according to our assumption @xmath48 . as noted before , the values @xmath223 are independent of the initial central oscillator state . furthermore , the initial bath state @xmath33 occurs only through the frequency - resolved energy distribution @xmath224 . in particular , the known equations for thermal baths @xcite are recovered whenever @xmath225 , where @xmath226 is the energy of a thermal oscillator at temperature @xmath227 . because there are no separate conditions on the two functions @xmath228 , @xmath229 , thermalization is possible also in non - thermal environments ( see below ) . follow directly if we assume a thermal bath from the outset , with initial conditions @xmath230 and @xmath231 . equipartition of energy allows us to combine the terms in eq . to @xmath232 , which depends only on the modulus of @xmath181 . we can then drop the exponential factor @xmath191 from eq . , and convergence of @xmath126 is evident . this short cut is not available in the general case . if the initial bath state @xmath33 and the central oscillator state @xmath32 are both gaussian states , the central oscillator density matrix @xmath29 is gaussian for all @xmath129 . then , @xmath29 is completely determined by the values of @xmath125 , @xmath126 , and their convergence suffices to establish equilibration ( e1 ) , and also ( e2 ) , in this case . otherwise , for non - gaussian initial states @xmath32 , we can use the propagating function @xmath233 from eq . to find @xmath29 for @xmath5 . recall that according to sec . [ sec : tdlimit ] we can assume that the initial bath state is gaussian in the thermodynamic limit , which allows for the construction given in sec . [ sec : prop ] . equilibration follows now from the observation that @xmath233 converges for @xmath5 whenever @xmath125 , @xmath126 converge . the long - time limit @xmath234 } { 2 \pi \sqrt { \det \mathbf{\sigma}^\infty } } \;.\ ] ] is obtained through substitution of @xmath235 and @xmath236 from eqs . . because @xmath237 , the result does not depend on @xmath238 . the long - time limit of the wigner function @xmath239 follows immediately with eq . : the integration over @xmath155 in the resulting expression @xmath240 evaluates to one because @xmath241 is normalized , such that @xmath242 is equal to @xmath243 . in other words , the stationary state @xmath174 is a gaussian state with parameters @xmath244 , @xmath245 . these parameters depend on the initial bath state according to eqs . , , but they are independent from the initial central oscillator state . this proves equilibration ( e1 ) and ( e2 ) for general initial central oscillator states . in particular , the stationary state is gaussian also for non - gaussian initial states . we note that the propagating function in position representation does not converge in the long - time limit ( cf . [ app : prop ] ) , which prevents an equally simple argument . because the stationary state @xmath174 in the long - time limit is a gaussian state for which only @xmath246 , @xmath247 are non - zero , it can always be interpreted as the thermal equilibrium state of some harmonic oscillator . this establishes the weakest thermalization property ( t1 ) . the effective oscillator frequency @xmath248 and temperature @xmath176 associated with @xmath174 are @xmath249 \,.\ ] ] generally , @xmath248 is not equal to the central oscillator frequency @xmath250 such that the stronger property ( t2 ) is not fulfilled . by eq . , the condition @xmath251 is equivalent to equipartition of kinetic and potential energy @xmath252 . the violation of this condition arises from the integrations over @xmath95 in eqs . , or , , which cover a finite energy range and include values @xmath253 . quantum corrections of this type are characteristic for strong damping @xcite . equipartition of energy is achieved in the limit of weak damping ( @xmath254 , when according to eq . the function @xmath255 in eqs . , converges to @xmath256 . therefore , the values @xmath257 are obtained . this establishes the stronger thermalization property ( t2 ) in the weak damping limit . for these values of @xmath258 it is ( cf . @xmath259 such that the stationary state @xmath167 is a thermal equilibrium state of the central oscillator . the temperature @xmath260 is determined by the energy @xmath261 of the bath oscillator at frequency @xmath250 in the initial state . note that the assumption ( e0 ) implies @xmath262 and @xmath263 , such that the value of @xmath261 is defined . in particular , @xmath264 and the argument of @xmath265 is equal to or greater than one . still , the asymptotic temperature @xmath266 from eq . is a function of @xmath250 . the functional dependence is determined by the choice of @xmath224 . if we demand , for the strongest thermalization property ( t3 ) , that @xmath176 is independent of @xmath250 we have to solve eq . to obtain the condition @xmath267 note that this is a condition on the particular combination @xmath224 of initial bath variances @xmath228 , @xmath229 , and not on the individual functions therefore , any initial bath preparation with @xmath268 results in the same stationary states as the thermal bath at temperature @xmath269 . one example for this additional freedom is the choice @xmath270 and arbitrary @xmath271 . it can be realized , e.g. , by superposition of coherent oscillator states at different positions . this initial bath state is not a thermal state for @xmath272 , in particular it violates equipartition of energy @xmath273 . but since @xmath274 we find that the stationary central oscillator state @xmath174 is identical to that obtained with a thermal bath at temperature @xmath269 : thermalization is well possible in non - thermal environments , even those far from thermal equilibrium . in summary , we have a hierarchy of conditions for equilibration and thermalization : 1 . the central oscillator equilibrates whenever @xmath48 for @xmath5 , 2 . the stationary state is always a gaussian and thermal state , 3 . equipartition of kinetic and potential energy occurs precisely at weak damping , 4 . the asymptotic temperature @xmath176 is independent of the central oscillator frequency under the additional condition on @xmath224 . it is a special feature of linear systems such as the one studied here that equilibration depends only on the asymptotic behavior of the solution @xmath36 of a classical equation of motion . another feature is that the stationary state always is gaussian such that equilibration implies thermalization , albeit only in the weak sense of property ( t1 ) . we noted earlier that in the situation studied here , with coupling to a single bath , a stationary state does not admit finite heat flows as would become possible for several baths with different preparations @xmath275 . therefore , conditions ( e1 ) , ( e2 ) capture the standard notion of thermodynamic equilibrium . we note that a consistent definition of thermalization requires the strong property ( t3 ) . suppose we deal with two central oscillators with frequencies @xmath276 . in the weak damping limit , the stationary state is the product state of two independent thermal states with respective temperatures @xmath277 and @xmath278 . such a state is only a thermal state of the combined system comprising the two oscillators if @xmath279 . therefore , thermalization of multiple oscillators , already in the weak sense ( t1 ) , requires the strong property ( t3 ) and thus condition ( but recall that this condition can be fulfilled also for non - thermal environments as in eq . ) . ( color online ) sketch of the infinite harmonic chain as defined in eqs . , . ] as an example for equilibration in a non - thermal environment we consider an infinite chain of harmonic oscillators ( see fig . [ fig : chain ] ) . oscillators in the right ( @xmath280 ) and left ( @xmath281 ) half of the chain , with frequency @xmath282 , are coupled to their neighbors ( @xmath283 ) with spring constant @xmath284 . they form the harmonic oscillator bath for the central oscillator at @xmath285 , with oscillator frequency @xmath250 and coupling @xmath286 to the oscillators at @xmath287 . for @xmath288 and @xmath289 we obtain a homogeneous , translational invariant chain . related examples have been studied in numerous publications , see e.g. @xcite . the behavior for thermal initial conditions , e.g. in a homogeneous chain @xcite or a chain with a single heavy mass @xcite , is well understood . equilibration in a harmonic chain with non - thermal initial conditions as discussed in refs . @xcite can be expressed in terms of our conditions from sec . [ sec : equilwithlist ] . general arguments for the appearance of gaussian states in the long - time limit are given in @xcite . still , a satisfactory and explicit analysis of equilibration and thermalization of the simple chain in non - thermal environments is missing . some studies assume too quickly that equilibration implies thermalization , in the sense of our condition ( t1 ) , failing to note , e.g. , that the appearance of gaussian states is the general behavior of linear systems and unrelated to thermalization as expressed by condition ( t3 ) . according definitions of ` temperature ' have to be taken with care . in addition we must carefully analyze the role of undamped oscillatory behavior that prevents equilibration and , therefore , thermalization . to address the harmonic chain within the formalism from secs . [ sec : model][sec : equil ] we must transform the hamilton operator @xmath290 - k_b \sum_{n=1}^\infty q_n q_{n+1}\ ] ] for the harmonic oscillator bath ( with operators @xmath291 , @xmath292 for the oscillator at site @xmath293 ) to normal modes . the same transformation has to be applied to the operator @xmath294 in the coupling term @xmath295 between the central oscillator and the chain oscillators at @xmath296 . it suffices to treat one of the two half - infinite chains explicitly , say the right chain @xmath297 as in eq . , and include a factor of two in @xmath100 to account for the left chain @xmath281 . note that in doing so we implicitly assume identical initial conditions for both sides of the chain and thus exclude the possibility of stationary non - equilibrium states with finite heat flow between the right and left half - infinite chain . the normal modes of @xmath298 are the standing wave solutions @xmath299 ( for a finite chain of length @xmath47 ) , and after a few lines of calculation we obtain the spectral function @xmath300 in the thermodynamic limit @xmath49 . it is @xmath90 for @xmath301 , and we impose the positivity condition @xmath302 to exclude negative frequencies of the bath . to proceed it is convenient to introduce the dimensionless model parameters @xmath303 and to use the normalized quantities @xmath304 note that @xmath305 . as discussed in sec . [ sec : equil ] , equilibration depends entirely on the decay of the function @xmath36 for @xmath5 , and thus on the absence of poles in @xmath46 from eq . . to obtain @xmath306 , we use the representation with the complex function @xmath307 where the branch cut of the root must be chosen along the positive real axis , and the minus ( plus ) sign applies for @xmath308 ( @xmath309 ) . note that the positivity condition , which can now be rewritten as @xmath310 , requires that @xmath311 ( color online ) real ( dashed curve ) and imaginary ( solid curve ) part of @xmath312 for @xmath313 and @xmath314 . for @xmath315 the function value is @xmath316 , respectively . ] before we can determine the function @xmath36 with eq . we must consider the possibility of isolated poles of @xmath46 . according to eq . we have to compare the functions @xmath317 and @xmath318 in regions where @xmath319 . from the qualitative behavior of @xmath312 , shown in fig . [ fig : gamz ] , we deduce that isolated poles of @xmath46 do not exist if and only if the inequalities @xmath320 are fulfilled . the first inequality excludes poles in the interval @xmath321 , the second inequality in the interval @xmath322 . another more fundamental restriction is the positivity condition , which is however less restrictive than the present condition . ( color online ) diagram of the admissible parameter space for equilibration according to condition . the white triangular region above the solid black lines is the maximal set of allowed parameter combinations . outside of this region an isolated pole exists even in the weak damping limit @xmath323 . for @xmath324 , the region of admissible parameters shrinks as depicted by the dashed black curves . the parameter combinations of homogeneous chains ( @xmath325 ) corresponds to the cusps @xmath326 of the curves , marked with red dots . the parameter combinations of chains with a single heavy mass @xcite correspond to the intersections of the curves with the @xmath327 line at @xmath328 , marked with green squares . at these points , condition coincides with the positivity condition . ] the admissible parameter combinations for equilibration of the harmonic chain that follow from condition are depicted in fig . [ fig : params ] . we note the basic restrictions @xmath329 the second inequality guarantees that the central oscillator frequency @xmath330 lies within the interval @xmath331 $ ] where @xmath332 . if this is fulfilled , equilibration is always possible for sufficiently small @xmath333 . since @xmath334 , it restricts the admissible parameters to the rectangle @xmath335\times [ 0,2]$ ] . condition is always fulfilled for the homogeneous chain ( and we note that @xmath336 requires @xmath325 ) . the chain studied by ullersma corresponds to parameters @xmath327 and @xmath337 ( @xmath338 equals the mass ratio @xmath339 in @xcite ) . condition is fulfilled if @xmath340 , i.e. only for a heavy mass . both examples lie on the boundary of the admissible parameter space , with one or two of the inequalities in becoming equalities . depending on parameters , the harmonic chain features rich dynamical behavior . for parameter combinations that fulfill condition the explicit result for @xmath36 from eq . reads @xmath341 for parameter combinations violating condition isolated poles of @xmath46 occur and additional ( undamped ) sine functions @xmath342 must be added to this expression . according to eq . , the poles of @xmath46 are the solutions of @xmath343 , which gives a quadratic equation for the harmonic chain such that zero , one , or two ( positive ) poles are possible . ( color online ) the @xmath344 parameter space of the infinite harmonic chain for @xmath345 . the solid / dotted black curves give the boundary of the two regions defined by each of the inequalities from condition . the central oscillator equilibrates for parameters in the unshaded region above the solid curve , where the condition is fulfilled ( note that @xmath334 ) . for parameters lying between the solid and dotted curves one of the two inequalities is violated and a single isolated pole of @xmath46 exists . below the dotted curve @xmath46 has two isolated poles . the dashed region to the left indicates where the positivity condition is violated , but such parameters already violate the first inequality in . both conditions coincide at @xmath327 , @xmath328 . the dashed red / green lines indicate the path followed in the next fig . [ fig : isomod ] , the crosses marked a d indicate the parameters used in fig . [ fig : ut ] . ] ( color online ) position @xmath346 and total weight @xmath347 of isolated poles of @xmath46 . we set @xmath348 and change @xmath330 , @xmath349 along the dashed path from fig . [ fig : paramsk2 ] , i.e. from @xmath350 , @xmath351 to @xmath352 , @xmath353 . the position of the poles is compared to the continuum of bath modes in the interval @xmath354 $ ] , filling the grey area around @xmath355 in the plot . between the two vertical dashed lines at @xmath325 , @xmath313 ( left ) and @xmath353 , @xmath356 ( right ) no poles exist in agreement with condition . for @xmath357 , in the dashed region to the left , the positivity condition is violated and one @xmath358 becomes negative . ] as an example let us consider the case @xmath348 . the restrictions on the parameters arising from the positivity condition and the stronger condition are summarized in fig . [ fig : paramsk2 ] . we now follow the dashed path in this figure and plot the position @xmath359 of isolated poles and their total weight @xmath360 in fig . [ fig : isomod ] . only for parameter combinations in the white unshaded area in fig . [ fig : paramsk2 ] , which corresponds to the part between the dashed vertical lines in fig . [ fig : isomod ] , condition is fulfilled . accordingly , only panel ( c ) in fig . [ fig : ut ] ( the parameter combination `` c '' in figs . [ fig : paramsk2 ] , [ fig : isomod ] ) shows a situation where @xmath361 for @xmath5 . otherwise , one ( parameter combination `` d '' ) or two ( `` a '' , `` b '' ) isolated poles exist if one or both inequalities from are violated . then , the amplitude of oscillations in @xmath362 remains finite in the long - time limit . function @xmath363 for the harmonic chain with @xmath348 . the parameters from panels ( a)(d ) correspond to the crosses in figs . [ fig : paramsk2 ] , [ fig : isomod ] . they are : ( a ) @xmath351 , @xmath364 ( two poles @xmath365 , @xmath366 , @xmath367 , @xmath368 ) , ( b ) @xmath369 , @xmath370 ( two poles @xmath371 , @xmath372 , @xmath373 , @xmath374 ) , ( c ) @xmath375 , @xmath370 ( no pole ) , ( d ) @xmath353 , @xmath376 ( one pole @xmath377 , @xmath378 ) . , title="fig : " ] function @xmath363 for the harmonic chain with @xmath348 . the parameters from panels ( a)(d ) correspond to the crosses in figs . [ fig : paramsk2 ] , [ fig : isomod ] . they are : ( a ) @xmath351 , @xmath364 ( two poles @xmath365 , @xmath366 , @xmath367 , @xmath368 ) , ( b ) @xmath369 , @xmath370 ( two poles @xmath371 , @xmath372 , @xmath373 , @xmath374 ) , ( c ) @xmath375 , @xmath370 ( no pole ) , ( d ) @xmath353 , @xmath376 ( one pole @xmath377 , @xmath378 ) . , title="fig : " ] + function @xmath363 for the harmonic chain with @xmath348 . the parameters from panels ( a)(d ) correspond to the crosses in figs . [ fig : paramsk2 ] , [ fig : isomod ] . they are : ( a ) @xmath351 , @xmath364 ( two poles @xmath365 , @xmath366 , @xmath367 , @xmath368 ) , ( b ) @xmath369 , @xmath370 ( two poles @xmath371 , @xmath372 , @xmath373 , @xmath374 ) , ( c ) @xmath375 , @xmath370 ( no pole ) , ( d ) @xmath353 , @xmath376 ( one pole @xmath377 , @xmath378 ) . , title="fig : " ] function @xmath363 for the harmonic chain with @xmath348 . the parameters from panels ( a)(d ) correspond to the crosses in figs . [ fig : paramsk2 ] , [ fig : isomod ] . they are : ( a ) @xmath351 , @xmath364 ( two poles @xmath365 , @xmath366 , @xmath367 , @xmath368 ) , ( b ) @xmath369 , @xmath370 ( two poles @xmath371 , @xmath372 , @xmath373 , @xmath374 ) , ( c ) @xmath375 , @xmath370 ( no pole ) , ( d ) @xmath353 , @xmath376 ( one pole @xmath377 , @xmath378 ) . , title="fig : " ] ( color online ) function @xmath363 for the inhomogeneous ( left panel , with @xmath325 , @xmath379 , @xmath380 ) and homogeneous ( right panel , with @xmath325 , @xmath381 , @xmath380 ) harmonic chain at weak damping . the dashed red curves indicate the exponential decay from eq . ( left panel ) and the asymptotic decay @xmath382 of the bessel function from eq . ( right panel).,title="fig : " ] ( color online ) function @xmath363 for the inhomogeneous ( left panel , with @xmath325 , @xmath379 , @xmath380 ) and homogeneous ( right panel , with @xmath325 , @xmath381 , @xmath380 ) harmonic chain at weak damping . the dashed red curves indicate the exponential decay from eq . ( left panel ) and the asymptotic decay @xmath382 of the bessel function from eq . ( right panel).,title="fig : " ] for strong damping situations ( @xmath383 ) shown in fig . [ fig : ut ] the function @xmath36 deviates significantly from an exponentially decaying function , even in the absence of poles ( panel ( c ) ) . exponential decay occurs only for weak damping @xmath384 . for @xmath385 we have @xmath386 as plotted in fig . [ fig : ut2 ] ( left panel ) . note that the case @xmath387 , with an undamped sine function in the weak damping limit , is excluded by the second inequality in eq . . for the homogeneous chain , with @xmath325 and @xmath326 , the weak damping limit gives a different result . since @xmath388 , the width of the continuum of bath states shrinks to zero for @xmath389 such that we do not obtain exponential decay of @xmath36 . instead , it is @xmath390 with the bessel function @xmath391 ( cf . refs . @xcite ) . according to condition isolated poles of @xmath46 can not occur in this situation . from the asymptotic behavior of the bessel function we find that here @xmath392 decays only as @xmath393 for @xmath394 , as shown in the right panel of fig . [ fig : ut2 ] . exponential decay in the weak damping limit is only achieved if the coupling @xmath395 of the central oscillator to the chain becomes small also in comparison to the width ( @xmath396 ) of the continuum of bath states . according to the previous discussion , the central oscillator in the harmonic chain equilibrates precisely for parameter combinations that fulfill condition . we now study , under these conditions , thermalization after a quench that generates a non - thermal environment for the central oscillator ( cf . eq . below ) . we imagine that for @xmath397 all oscillators are decoupled ( @xmath398 ) and in thermal equilibrium at temperature @xmath269 . every oscillator has the same variance @xmath399 and we do not need to specify further initial expectation values if we are only interested in the stationary state in the long - time limit . at @xmath28 we quench the system by cranking up the coupling to finite values @xmath400 . since @xmath401 , @xmath402 do not depend on @xmath403 , transformation to the normal modes of the bath results in constant functions @xmath404 for the initial bath variances at @xmath28 . the initial bath state is uncorrelated with @xmath405 . according to sec . [ sec : equil ] , the stationary state in the long - time limit depends only on the frequency - resolved energy @xmath224 of the initial bath state , which for the present example is given by the function @xmath406 this function acquires a dependence on @xmath95 through the dispersion of the bath modes after the quench , but it does not fulfill eq . . we thus see that the thermal equilibrium state of uncoupled oscillators before the quench corresponds to a non - thermal state of the coupled chain of oscillators after the quench . according to condition ( t3 ) from sec . [ sec : equilwithlist ] we expect that the temperature @xmath176 of the stationary state depends on the central oscillator frequency @xmath330 even at weak coupling . from eqs . or eqs . , , the variances in the long - time limit are obtained as @xmath407 @xmath408 we will give further results using normalized quantities @xmath409 choosing @xmath282 as the unit of energy . we recall that according to property ( t1 ) the stationary state is always a thermal state of some harmonic oscillator hamiltonian , such that we should check the stronger property ( t2 ) . from eq . , the effective frequency associated with the stationary state is @xmath410^{-1 } } \;.\ ] ] we observe that equipartition of energy , i.e. @xmath411 , can be achieved only in the weak damping limit @xmath389 . for @xmath412 , it is always @xmath413 . this confirms the conditions given for property ( t2 ) in sec . [ sec : equilwithlist ] . for weak damping , eqs . , simplify to @xmath414 equipartition of energy in the stationary state is evident , and the thermalization ( t2 ) property fulfilled . to check property ( t3 ) , we calculate the temperature @xmath415\ ] ] of the stationary state with eq . or the weak damping result . we see that @xmath416 depends explicitly on the central oscillator frequency @xmath330 , as depicted in fig . [ fig : tchain ] . it is @xmath417 only for @xmath325 . as discussed before , this results from the fact that @xmath224 after the quench violates condition . ( color online ) temperature @xmath418 of the stationary thermal state at weak damping as given in eq . . it is shown as a function of @xmath330 for different temperatures @xmath419 of the initial state , as indicated . note that @xmath418 does not depend on @xmath349 , but the admissible values of @xmath330 for which equilibration occurs are restricted by the second condition in eq . ( see also fig . [ fig : params ] ) . in particular , it must be @xmath420 . ] we note that @xmath349 does not appear in eq . . in the present example the value of @xmath349 only determines the admissible values of @xmath330 that lead to equilibration , as given by the second inequality in eq . . once equilibration has been observed , the temperature of the stationary state at weak damping depends only on the value of @xmath261 not on the functional dependence of the spectral function @xmath100 . ( color online ) frequency @xmath421 ( dashed curve ) and temperature @xmath422 ( solid curves ) for the homogeneous chain , from eqs . , and shown as a function of @xmath395 . the temperature curves are plotted for @xmath423 as indicated . ] for the homogeneous chain with @xmath388 , @xmath424 eqs . , simplify to @xmath425 @xmath426 equipartition of energy is violated for any @xmath412 , such that the effective frequency @xmath427 associated with the stationary state deviates from the central oscillator frequency ( it is always @xmath428 ) . the temperature of the stationary state is @xmath429 \;.\ ] ] it is @xmath430 for @xmath412 , for example @xmath431 for @xmath432 and @xmath433 ( see fig . [ fig : homchain ] ) . the situation simplifies again in the weak damping limit @xmath389 , where we recover from eqs . , the equilibration / thermalization result for the homogeneous chain formulated in refs . @xcite : at weak damping the central oscillator evolves into a stationary thermal state , with equipartition of energy @xmath434 . because of translational invariance this statement applies to every chain oscillator . we note , however , that thermalization of the homogeneous chain is not perfect . as discussed in sec . [ sec : equilwithlist ] , observation of a single oscillator in the homogeneous chain is not sufficient to establish thermalization of the entire chain . thermalization fails for a finite chain segment consisting of two or more oscillators , because property ( t3 ) is not fulfilled as seen in eq . . note that there is no possibility to check property ( t3 ) directly for the homogeneous chain ( @xmath325 is fixed here ) , such that results restricted to this situation have to be interpreted carefully @xcite . our study of the dissipative quantum harmonic oscillator addresses equilibration and thermalization in non - thermal environments . equilibration is the generic behavior , which is prevented only in situations where the classical oscillator equation of motion possesses undamped oscillatory solutions . the infinite harmonic chain is an example for this behavior . thermalization of the central oscillator depends on additional conditions . just as for thermal environments , equipartition of energy requires the weak damping limit but is independent of the precise initial conditions . the asymptotic temperature @xmath176 is obtained from the energy distribution @xmath224 in the initial bath state , and generally depends on the central oscillator frequency @xmath250 . if we demand that @xmath176 is independent of @xmath250 , another condition on @xmath224 follows . this condition is essential for simultaneous thermalization of several oscillators , when a thermal state of the combined system is obtained only if the same asymptotic temperature is assumed by each oscillator . part of the behavior discussed here generalizes to systems with non - linear interactions . first , we note that equilibration is possible although the linear system is integrable . equilibration occurs because , in a rough sense , the reduced density matrix of the central oscillator involves an average over conserved quantities of the joint oscillator - bath system . in other words , equilibration of small systems embedded in a large environment does not require ergodicity . second , because of the linearity and unitarity of quantum mechanical time evolution the stationary state depends explicitly on the initial ( bath ) state . but already for the linear system some properties , such as equipartition of energy , are independent of the initial conditions . furthermore , the stationary state depends only on the energy distribution @xmath224 in the initial bath state . effectively , information is lost in the long - time limit and thermalization is possible for a large class of ( non - thermal ) initial states . we did neither discuss the generalization of the fluctuation - dissipation relation to the present non - thermal setting , nor the role of stationary non - equilibrium states with finite heat flow that would require coupling to at least two baths with different preparations . multi - time correlations functions can be computed within the present formalism , which will allow for the analysis of both issues in the future . the authors wish to acknowledge helpful discussions with m. cramer , g .- ingold , and m. thorwart . this work was supported by deutsche forschungsgemeinschaft through sfb 652 ( b5 ) and al1317/1 - 2 . the solution of the dissipative quantum harmonic oscillator model through operator equations of motion instead of transformation to normal modes of @xmath22 allows for a simple treatment of general initial conditions and time - dependent coefficients . we here list the relevant steps of the derivation omitted in sec . [ sec : eom ] , and allow for a time - dependent central oscillator frequency @xmath435 ( cf . @xcite for a path integral calculation ) . the heisenberg equations of motion @xmath436 $ ] are @xmath437 for the position and momentum operator of the central oscillator , and @xmath438 for the bath oscillators . we can read eq . as an inhomogeneous linear equation for @xmath439 . using the green function for the homogeneous equation @xmath440 , we find @xmath441 inserting this result into eq . gives the equation of motion @xmath442 for the central oscillator operator @xmath30 , with the damping kernel @xmath84 from eq . and the noise term @xmath443 eq . is an inhomogeneous linear integro - differential equation , which can be solved through solution of the classical equation of motion @xmath444 we need the two solutions @xmath445 , @xmath446 to initial conditions @xmath447 , @xmath448 , and @xmath449 , @xmath450 . the solution of the operator equation of motion for @xmath30 is then given by @xmath451 and it is @xmath452 . with the partial fourier transforms @xmath453 @xmath454 and the definition of matrices @xmath455 \partial_t u_1(t,0 ) & \partial_t u_2(t,0 ) \end{pmatrix}\ ] ] and @xmath456 as in eq . the operators @xmath30 , @xmath31 are given by the matrix eq . . for constant @xmath457 , the function @xmath36 used in sec . [ sec : eom ] is recovered as @xmath458 , and it is @xmath459 ( while @xmath460 for time - dependent @xmath435 ) . then , the partial fourier transforms @xmath461 and @xmath462 are related by eq . and @xmath168 is given by the simpler expression . eqs . , and eqs . , for the calculation of expectation values and variances and eq . for the construction of the propagating function remain valid for time - dependent @xmath435 . the propagating function in position representation is the fourier transform @xmath463 of eq . . it results in the expression @xmath464 \ ; , \nonumber\\\end{aligned}\ ] ] where we write @xmath465 , @xmath466 , @xmath467 , @xmath468 for abbreviation and drop the time argument in @xmath469 . the @xmath146 real parameters @xmath470 in this expression are related to the parameters of @xmath233 in eq . through @xmath471 explicit insertion of @xmath79 from eq . or gives expressions that allow for direct comparison with the literature . for example , the expressions given in ref . @xcite are recovered for @xmath472 such that the terms @xmath473 , @xmath474 vanish . obviously , the position representation leads to less convenient expressions for the propagating function , and obscures the clear formal structure of eq . . in particular , the expressions are singular for @xmath48 , which gives a complicated representation of the @xmath475-distribution for the propagating function at @xmath28 and @xmath5 instead of the simple limit for @xmath233 ( cf . eq . ) . 48ifxundefined [ 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 @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ ( , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop _ _ , ed . ( , ) @noop * * , ( ) @noop * * , ( ) link:\doibase 10.1103/physrev.40.749 [ * * , ( ) ] @noop _ _ ( , ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( ) @noop * * , ( )

we propose a new approach for estimating the parameters of a probability distribution . it consists on combining two new methods of estimation . the first is based on the definition of a new distance measuring the difference between variations of two distributions on a finite number of points from their support and on using this measure for estimation purposes by the method of minimum distance . for the second method , given an empirical discrete distribution , we build up an auxiliary discrete theoretical distribution having the same support of the first and depending on the same parameters of the parent distribution of the data from which the empirical distribution emanated . we estimate then the parameters from the empirical distribution by the usual statistical methods . in practice , we propose to compute the two estimations , the second based on maximum likelihood principle of known theoretical properties , and the first being as a control of the effectiveness of the obtained estimation , and for which we prove the convergence in probability , so we have also a criterion on the quality of the information contained in the observations . we apply the approach to truncated or grouped and censored data situations to give the flavour on the effectiveness of the approach . we give also some interesting perspectives of the approach including model selection from truncated data , estimation of the initial trial value in the celebrate _ em _ algorithm in the case of truncation and merged normal populations , a test of goodness of fit based on the new distance , quality of estimations and data . key words and phrases : em algorithm , minimum distance , model selection from truncated data , point estimation , truncated data , grouped and censored data .

0802.2155

point estimation is the most popular forms of statistical inference ( see lehmann and casella @xcite ) . we introduce in this paper a new statistical point estimation approach which found be useful in special practical situations such as truncated and grouped and censored data . the data are said to be truncated when measuring devices fail to report observations below and/or above certain readings . for example , truncated data frequently arise in the statistical analysis of astronomical observations ( see efron and petrosian @xcite ) and in medical data ( see klein and zhang @xcite ) , and if the truncation is ignored this can cause considerable bias in the estimation . there exists in the literature many approaches of estimation from `` incomplete data '' such as maximum likelihood based approach of the em algorithm ( hartley @xcite , dempster et _ al _ @xcite ) , or nonparametric methods such as kaplan - meier ( kaplan and meier @xcite ) or lynden - bell estimators ( lynden - bell @xcite ) . the purpose of the present paper is to investigate another approach which consists on combining two new methods of estimation and to apply it in the fixed type i censored or grouped and censored data situations . in the first method , we remark that in estimation problems we deal in general with three functions : a theoretical probability law @xmath0 of a random variable @xmath1 depending on a parameter @xmath2 ( real or vector valued ) , an empirical distribution @xmath3 constructed from a sample of observations drawn from the random variable @xmath1 and an estimation @xmath4 ( from an estimation @xmath5 of @xmath2 ) obtained through the empirical law @xmath6 the empirical distribution @xmath7is considered as a representative distribution of @xmath8 but in practice it is reduced to only few of its characteristics such as the mean and variance . the variational aspect of @xmath3 is often neglected while its importance . we can easily find , for instance , two distributions having the same support , mean and variance while their variations differ significantly , or conversely having the same variations but their supports and characteristic parameters are different . but two probability distributions with same support and same variations in each subset of the support are necessarily the same . we introduce then a new distance which measures the difference between variations of two distributions on a finite number of points and to use it for estimation purposes by the method of minimum distance . since the new measure is not equivalent to classical ones it will give new insights that could not be investigated by classical distances . in the second method , we remark that the empirical distribution arising from a sample of observations can be viewed in fact as a conditional distribution as it is built from the knowledge of the data . it will be then an estimation of the theoretical conditional distribution with respect to the observations before being an estimation for the parent distribution . this theoretical conditional distribution is represented by the auxiliary distribution introduced in this paper . to determine this distribution in discrete case , we have simply to take the conditional distribution with respect to the observed values and we proceed analogously for the continuous case . it should be noted that in discrete case it is known as the truncated distribution which is the conditional distribution given a truncation ( see for example shaw @xcite ) but it is presented here in a general framework . we have to deal with two discrete probability distributions having the same finite support , a theoretical distribution and its empirical representation with respect to the observations . the parameters of the former are those of the parent distribution and the aim is to estimate them from the first instead of the parent one as commonly used . we use classical tools such as the method of moments or maximum likelihood principle . the setting that seems to us most suitable for illustrating our approach is the one of truncated or grouped and censored data . in usual practical problems , truncation can be on left or right or in either situations , and the `` cut off '' can be deterministic or random . in our approach , the truncation may be on any part of the range of the distribution so that the setting is more general . also , classical approaches for truncated data are in general custom - made depending on specific problems and distributions , or subjective based methods . instead , our approach is quite general and might be used in any situation where the underlying complete data come from a known family of distributions . we confine ourselves as a first presentation to fixed type i and grouped and censored data . in the subsequent section , we propose a variational distance between probability distributions . in section 3 , we define a truncation of data and associated empirical and theoretical distributions and we use two different methods for estimation from truncation , a first method using minimum of the new distance introduced in this paper and a second method based on traditional tools of estimations such as the method of maximum likelihood . in section 4 , we present the new approach and we illustrate the procedure by three examples : a binomial probability law , a normal distribution and a gamma density function . we present also a basic feature of the new approach which prove the accuracy of the method and some illustrative examples . in section 5 , we give some elements of comparison with the classical approach of estimation . in section 6 , we list some perspectives of the new approach : model selection from truncated data using the new distance , estimation of the first trial value in the celebrate _ em _ algorithm for incomplete data in the case of truncation and merged normal distributions , a goodness of fit test based on the new distance , decision making about the quality of estimations and data . finally , concluding remarks are made some pointing to other possible extensions and applications . as is usual , given a sample of @xmath9 independent and identically distributed observations , @xmath10 drawn from an unknown discrete random variable @xmath11 falling in a discrete family of probability laws @xmath12 depending on a parameter @xmath2 ( real or vector valued ) , i.e. , @xmath13 one can summarize the sample into @xmath14 couples @xmath15 @xmath16 where the @xmath17 are the different values taken by the sample and @xmath3 is the empirical law @xmath18 where @xmath19 represents the absolute frequency of the value @xmath20 @xmath21 usually , it is hoped that @xmath22 in a certain probabilistic sense . but if the empirical distribution arises from truncated data , we do not hope in general having @xmath23 for the values @xmath24 in the support of @xmath25 since the complete sample size @xmath9 is usually not reported . however , we expect reasonably to have approximately@xmath26 for any points in its support , only if the sample has serious irregularities . introduce the following _ distance _ _ of _ _ proportional variations _ between @xmath0 and @xmath3@xmath27 it turns out that this new distance , as we will show , measures the variations between probability distributions . in continuous case also , any sample @xmath28 is summarized into @xmath14 couples @xmath29 @xmath30 @xmath31 . this can be done uniquely , by grouping for example the sample in classes where the @xmath17 are the mid - classes ( or class means ) and @xmath32 where @xmath3 is an empirical density estimator , or the data is presented in a grouped and censored form . the proportional variational distance @xmath33 in this case , between the density @xmath34 of @xmath11 and its empirical law @xmath25 is thus defined as ( [ dv1 ] ) . one of its main powerful feature is that when using traditional distances we have to use the sample size @xmath9 through the expression of @xmath35 where @xmath36 is the size of class intervals ; but sometimes , as for truncated data situations where measuring devices fail to report even the number of sample points in certain ranges , then the real size @xmath9 is not known , but a truncated sample size @xmath37 is instead used . using the ratios @xmath38 will clear up the effect of the truncated sample size which can lead to considerable bias in the estimation . note that @xmath33 possesses the properties of symmetry and triangle inequality . but in the identity property @xmath39 the equality between @xmath40 and @xmath41 must be understood in the sense that @xmath40 and @xmath41 have the same variations on the points @xmath24 and @xmath42 it should be stressed that this new measure is not equivalent to classical ones and should then give new insights and information about other characteristics and features of probability distributions . from now on @xmath40 shall represent a theoretical probability law in both discrete or continuous cases and @xmath3 shall represent the corresponding empirical law in both cases . denote by @xmath43 the set of _ atoms _ of @xmath40 or _ support_. let @xmath44 be the @xmath45algebra generated by sets @xmath46 where the @xmath47 are the borel sets of @xmath48 and @xmath49 for all @xmath50 we have @xmath51where @xmath52 is the lebesgue measure on @xmath48 . in discrete case , we have @xmath53 for all @xmath54 we set @xmath55 @xmath56 and @xmath57 let @xmath58 @xmath59 and @xmath60 the probability space @xmath61 represents the space of samples of size @xmath9 from the random variable @xmath62 we omit the subscript @xmath9 in @xmath63 for notational convenience and shall denote the sample space as @xmath64 we will discuss now the measure theoretic aspect of the new distance introduced above . let @xmath65 and @xmath66 two probability measures defined on the same measurable space @xmath67 , @xmath40 and @xmath41 their respective probability densities , not necessarily with respect to the same measure and @xmath68 an event of this space . we say that @xmath40 and @xmath41 have the same variation on @xmath68 , if the respective restrictions of @xmath40 and @xmath41 on @xmath68 , define the same probability measure on @xmath68 endowed with the sigma algebra traces of @xmath44 on @xmath68 . let @xmath40 and @xmath41 two probability distributions positive and defined on a part @xmath68 not reduced to only one element . if in any point @xmath69 of @xmath70 , we have : @xmath71 then we say that @xmath40 and @xmath41 have same variations on @xmath68 . let @xmath40 be a density of a probability measure @xmath65 and @xmath68 an event such that @xmath72 . the restriction of @xmath40 on @xmath68 and the conditional distribution of @xmath40 with respect to @xmath68 define the same probability measure on @xmath68 and consequently they have the same variations on @xmath73 let @xmath40 and @xmath41 two probability distributions and @xmath68 an event on which they are strictly positive . if @xmath68 is discrete and not reduced to only one element , and one of the distributions @xmath40 and @xmath41 being discrete and the other may not be discrete , we call distance in variations between @xmath40 and @xmath41 on @xmath68 the quantity:@xmath74 if @xmath68 is an interval of @xmath48 and , @xmath40 and @xmath41 are probability densities on @xmath48 , with respect to lebesgue measure @xmath52 on @xmath48 , we call distance in variations between @xmath40 and @xmath41 on @xmath68 , the quantity:@xmath75 let be given a classical distance @xmath76 between two functions @xmath40 and @xmath41 which associates for points @xmath24 and @xmath77 from the intersection of their domain of definitions , the quantity @xmath78 we have the following properties for the distance @xmath79**1 . * * @xmath80 the converse is not always true.*2 . * let @xmath3 be a kernel density estimation . then @xmath81 in probability.*3 . * let @xmath40 and @xmath41 be two functions defined on @xmath48 and @xmath82 satisfying:@xmath83 if@xmath84 where @xmath52 is the lebesgue measure on @xmath48 , then@xmath85 * 1 . * follows directly from the definitions of @xmath76 and @xmath86**2 . * * follows from the fact @xmath87 in probability ( see parzen @xcite ) , then @xmath81 in the same probabilistic notion of convergence.*3 . * fix @xmath88 we have @xmath89 for all @xmath90 this implies that@xmath91 we deduce that @xmath92 and the result follows . the truncated data specification , or generally _ incomplete data , _ implies the existence of two sample spaces @xmath93 and @xmath94 , such that the complete sample space is given by @xmath95 the observed data @xmath96 where @xmath37 is the truncated sample size , are a realization from @xmath93 and the unobserved data @xmath97 where @xmath9 is the complete unknown sample size , are from @xmath98 the complete data @xmath99 is known only through the observed data @xmath100 ( see dempster , laird and rubin @xcite for further explanations about incomplete data specification)_. _ consider a sample of observations @xmath28 drawn from a theoretical probability law @xmath101 depending on a parameter @xmath102 as usual , the data are summarized , in discrete or continuous cases ( as shown in section 2 ) , into @xmath14 couples @xmath103 @xmath16 and let @xmath104 a part from the set @xmath105 @xmath106 which we will call _ truncation . _ the observed data is summarized by a truncation @xmath107 and an empirical estimation @xmath108 and assume that the unobserved data is also summarized by a set @xmath109 and @xmath110 the structure of the new distance @xmath33 allows the following decomposition property:@xmath111 the following proposition is typical for the new distance and is useful for using the minimum of distance @xmath86 let be given a truncated data @xmath112 with corresponding empirical estimation @xmath113 then @xmath114 in probability . we have from proposition 1 that @xmath115 in probability . then , from the decomposition property ( [ dec ] ) we obtain @xmath116 in probability . define the empirical distribution @xmath4 corresponding to a given truncation @xmath117 by:@xmath118{c}\widetilde{f}_{i}\text { \ \ \ \ if \ \ } x = u_{i},\text { \ \ } i=1, ... ,m,\\ 0\text { \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ otherwise , \ \ \ \ \ \ \ \ \ } \end{array } \right.\ ] ] where the @xmath119 satisfy the following set of proportional allocation equations @xmath120 for @xmath121 and @xmath122 define the following auxiliary distribution from @xmath101 which is akin to the proportional allocation procedure for missing values ( see hartley @xcite).@xmath123{c}\dfrac{f(x,\theta)}{f(u_{1},\theta)+f(u_{2},\theta)+ ... +f(u_{m},\theta)}\text { \ \ \ \ if \ \ } x = u_{i},\text { \ \ } i=1, ... ,m,\\ 0\text { \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ \ otherwise \ } \end{array } \right . \label{auxil}\ ] ] if the truncation is random , that is , there exists a random variable @xmath124 such that we observe , for example , the random variable @xmath11 only if @xmath125 or @xmath126 then the probability law used in ( [ auxil ] ) is replaced by the conditional law of @xmath11 with respect to @xmath127 or @xmath128 respectively . the auxiliary distribution @xmath129 was found be useful for estimation problems in truncated data . indeed , it is well known in classical estimation from truncated data ( see hartley @xcite ) that missing values could be recovered by `` proportional allocation '' procedures , then the auxiliary distribution @xmath130 which is already based on proportional allocation , will be an intuitive and natural tool for estimation purposes from truncated data . the function @xmath129 is a theoretical probability distribution depending on the same parameters of those of @xmath40 . it has also the same _ support _ as that of @xmath131 we call @xmath4 and @xmath132 the empirical and theoretical distributions of a given truncation @xmath133 from a sample of observations @xmath134 we will use mainly two methods of estimation . the first method is a minimum distance estimation using the metric @xmath33 between the empirical and theoretical distributions @xmath3 and @xmath135 the second is similar to traditional ones such as the method of substitution or maximum likelihood principle , by considering @xmath4 as an empirical estimation of @xmath136 the first is based on variational difference between distributions and the second in the sense of an euclidean difference and hence they treat different aspects of the sample of observations . if for a given data they give different estimations , we can not suspect the approaches but we can say that the data do not restore in a coherent way all aspects of the probability distribution from which it emanated . if on the other hand they give significantly the same estimations we can assert that the estimation is credible since through different aspects it has given the same distribution . that is the distribution which fits the best the empirical distribution . practically , we propose to calculate the estimations by the two methods and take the second one since based on maximum likelihood principle of good known theoretical properties . we use then the first as a tool of decision on whether the estimation is credible or not . the estimation will then be considered as credible in cases where the two methods give approximately the same estimation . let @xmath137 a sample with @xmath138 @xmath139 with@xmath140 @xmath141 where @xmath142 is a borel set of @xmath48 such that @xmath143 for all @xmath144 the family ( [ fam1 ] ) is very rich , one finds there , for example , the family of the normal laws , and the family of the laws of poisson . we assume that the support @xmath142 does not depend on @xmath145 denote by @xmath146 the estimator by the minimum of metric @xmath33 between the empirical and theoretical distributions @xmath147 ( based on a sample of size @xmath9 ) and @xmath101 that is@xmath148 this estimator falls into the class of m - estimators . using well known theorems on the convergence of m - estimators ( see for example amemiya @xcite ) we will prove that @xmath149 converges in probability to the true parameter . let @xmath137 be a sample from the family of distributions ( [ fam1 ] ) . if the set of natural parameters @xmath150 is covex and the true parameter @xmath2 is an interior point of @xmath151 then the estimator @xmath149 by the minimum of the distance of variations @xmath33 converges in probability to the true parameter @xmath152 i.e.,@xmath153 since we search for a minimum of the criterion function @xmath154 it suffices to show , under the assumptions of the family ( [ fam1 ] ) and the convexity of the set @xmath151 that @xmath155 seen as a function of @xmath2 is a convex function ( see amemiya @xcite ) . hence , this reduces the problem to the convexity of@xmath156 for @xmath157 with @xmath158 , and @xmath159 we have@xmath160 \left ( t_{k}(y_{i})-t_{k}(y_{j})\right ) \right\ } -a_{ij}\right\vert\ ] ] where @xmath161 and assume that @xmath162 and @xmath163we have from the convexity of the exponential function that@xmath164 \left ( t_{k}(y_{i})-t_{k}(y_{j})\right ) \right\ } & \leq\lambda\exp\left\ { \sum_{k=1}^{s}\theta_{k}^{(1)}\left ( t_{k}(y_{i})-t_{k}(y_{j})\right ) \right\ } \\ & + \mu\exp\left\ { \sum_{k=1}^{s}\theta_{k}^{(2)}\left ( t_{k}(y_{i})-t_{k}(y_{j})\right ) \right\ } , \end{aligned}\ ] ] then@xmath165 \left ( t_{k}(y_{i})-t_{k}(y_{j})\right ) \right\ } -a_{ij}\leq\]]@xmath166@xmath167 + \]]@xmath168 .\ ] ] introducing the absolute value we get@xmath160 \left ( t_{k}(y_{i})-t_{k}(y_{j})\right ) \right\ } -\left ( \lambda + \mu\right ) a_{ij}\right\vert\]]@xmath169@xmath170 hence @xmath171 is a convex function of @xmath152 which implies the convexity of @xmath155 seen as a function of @xmath2 and then the convergence in probability of the minimum of distance @xmath33 estimator . we firstly begin in a general situation , that of the one - parameter exponential family , to show how to use the procedure explained below in the case of the new method . consider the one - parameter exponential family with density@xmath172,\ ] ] where @xmath2 is the parameter , @xmath124 a statistic , @xmath173 a function of @xmath24 and @xmath174 is a function of the parameter @xmath2 . let us use the maximum likelihood principle . consider a sample of observations @xmath28 from which we derive the support @xmath175 we then construct the auxiliary distribution from the support @xmath117 , expressed in the following form@xmath176}{\sum_{i=1}^{k}k(y_{i})\times\exp[\theta t(y_{i})-a(\theta)]}.\ ] ] we have to maximize the likelihood function given in our case by@xmath177 without loss of generality , we assume that the class intervals are the same . then , we have@xmath178}{{\displaystyle\sum\limits_{i=1}^{k } } k(y_{i})\times\exp[\theta t(y_{i})-a(\theta)]}\right ] , \ ] ] taking the derivative and solving the score equation on @xmath2 we obtain an estimator of the parameter @xmath2 satisfying the relation@xmath179 the later result may be obtained directly by the method of moments , but we have presented the maximum likelihood method since it is widely used in statistical inference . in order to test the performance of the proposed approach , we use synthetic data sets which were generated by simulation from three examples of probability law : binomial law , normal density and a gamma distribution . the examples were selected from various simulation studies from different family of probability distributions and the two methods have shown their effectiveness and never deviate significantly from the true parameter . the reason for using synthetic data sets is that the true parameters for the synthetic datasets are known and the accuracy of results obtained by using the two new methods can be compared . * binomial distribution . * we generated a synthetic data set of size @xmath180 from a binomial law @xmath181 with @xmath182 and @xmath183 and denote by @xmath184 its probability mass function . the data are summarized in the following table . * table 1 . * [ c]ccccccccc@xmath17 & @xmath185 & @xmath186 & @xmath187 & @xmath188 & @xmath189 & @xmath190 & @xmath191 & @xmath192 + @xmath193 & @xmath194 & @xmath195 & @xmath196 & @xmath197 & @xmath198 & @xmath199 & @xmath200 & @xmath190 + our aim is to estimate the parameter @xmath201 with the knowledge of @xmath182 , from different truncation of data . for illustrating the two methods , consider the truncation @xmath202 with truncated sample size @xmath203 we have then a truncation proportion of @xmath204 @xmath205 in data . for the first method , we have to search the value of the parameter @xmath206 which minimizes the distance @xmath154 that is:@xmath207 using computer algebra package , we obtain the result @xmath208 for the second method , the empirical distribution @xmath4 given the truncation @xmath209 is given by @xmath210 @xmath211 , @xmath212 @xmath213 and @xmath214 if @xmath215 the auxiliary distribution @xmath216 is given by:@xmath217 by the method of substitution , the estimation of @xmath206 is obtained by solving the equation:@xmath218 using a computer algebra package we obtain the result @xmath219 . in the following table we present the estimations @xmath220 from the first method using minimum distance approach using the distance @xmath33 , and @xmath221 from the auxiliary distribution , of the parameter @xmath201 for known @xmath222 according to the truncation @xmath133 considered . * table 2 . * the estimations @xmath220 and @xmath221 by the new approach of the parameter @xmath206 of the binomial probability law @xmath181 with @xmath223 and known @xmath182 . [ c]cccccc & & truncated & proportion of & & + @xmath224 & @xmath117 & sample size @xmath37 & truncation @xmath66 @xmath225 & @xmath226 & @xmath221 + @xmath186 & @xmath227 & @xmath180 & @xmath185 & @xmath228 & @xmath229 + @xmath187 & @xmath230 & @xmath231 & @xmath232 & @xmath233 & @xmath234 + @xmath188 & @xmath235 & @xmath236 & @xmath237 & @xmath238 & @xmath239 + @xmath189 & @xmath240 & @xmath241 & @xmath194 & @xmath233 & @xmath234 + @xmath190 & @xmath242 & @xmath243 & @xmath244 & @xmath245 & @xmath239 + @xmath191 & @xmath246 & @xmath247 & @xmath248 & @xmath233 & @xmath229 + @xmath192 & @xmath249 & @xmath250 & @xmath251 & @xmath252 & @xmath253 + @xmath254 & @xmath255 & @xmath256 & @xmath257 & @xmath233 & @xmath258 + @xmath259 & @xmath260 & @xmath261 & @xmath262 & @xmath233 & @xmath263 + @xmath264 & @xmath265 & @xmath266 & @xmath267 & @xmath245 & @xmath245 + @xmath268 & @xmath269 & @xmath270 & @xmath271 & @xmath228 & @xmath228 + @xmath272 & @xmath273 & @xmath274 & @xmath275 & @xmath233 & @xmath234 + @xmath276 & @xmath277 & @xmath278 & @xmath279 & @xmath280 & @xmath281 + @xmath282 & @xmath283 & @xmath284 & @xmath285 & @xmath229 & @xmath252 + @xmath194 & @xmath286 & @xmath287 & @xmath288 & @xmath289 & @xmath233 + @xmath290 & @xmath291 & @xmath292 & @xmath293 & @xmath294 & @xmath295 + @xmath296 & @xmath297 & @xmath298 & @xmath299 & @xmath300 & @xmath263 + @xmath244 & @xmath301 & @xmath302 & @xmath303 & @xmath304 & @xmath305 + as previously said , the two estimations by the new approach , @xmath226 and @xmath306 are accurate in all cases and close to each other . furthermore , the truncation proportion has no effect on the quality of estimations . the two estimations are also not sensitive to small cell probabilities as for truncations including the value @xmath307 it should be noted that the classical estimation by maximum likelihood without truncation is @xmath308 and considering our approach we obtained the estimations @xmath309 for the first method and @xmath310 for the second.*normal distribution . * consider a sample of size @xmath311 drawn from a normal population with mean @xmath312 and standard deviation @xmath313 consider the data falling in @xmath268 fixed class intervals as shown in the following table , with mid - classes @xmath314 and absolute frequencies @xmath193**table 3.**@xmath315{cccccccccccc}\hline $ ] y_i@xmath316 - 2.581@xmath316 - 2.06@xmath316 - 1.533@xmath316 - 1.009@xmath316 - [email protected]@[email protected]@[email protected]@xmath3162.658@xmath318n_i@xmath3165@xmath3168@xmath31623@xmath31648@xmath31671@xmath31689@xmath31672@xmath31643@xmath31625@xmath31610@xmath3176@xmath319 the number of bins can be selected from an optimal procedure developed by birg and rozenholc @xcite . let the following table where we estimate simultaneously @xmath320 and @xmath321 by the minimum distance procedure with @xmath322 . we denote the estimations by @xmath323 and @xmath324 in each line of the table the estimates are made starting from the table of frequencies based on the observations indicated in the first column . the truncated sample size is denoted by @xmath325 we have then a truncation proportion of @xmath326 in data . * table 4 . * [ c]ccccc@xmath327 & @xmath37 & @xmath328 & @xmath329 & @xmath330 + @xmath331 & @xmath311 & @xmath185 & @xmath332 & @xmath333 + @xmath334 & @xmath335 & @xmath189 & @xmath336 & @xmath337 + @xmath338 & @xmath339 & @xmath340 & @xmath341 & @xmath342 + @xmath343 & @xmath344 & @xmath345 & @xmath346 & @xmath347 + @xmath348 & @xmath349 & @xmath276 & @xmath350 & @xmath351 + @xmath352 & @xmath353 & @xmath354 & @xmath346 & @xmath355 + @xmath356 & @xmath357 & @xmath358 & @xmath359 & @xmath360 + @xmath361 & @xmath362 & @xmath363 & @xmath364 & @xmath365 + @xmath366 & @xmath367 & @xmath368 & @xmath369 & @xmath370 + @xmath371 & @xmath372 & @xmath373 & @xmath346 & @xmath347 + @xmath374 & @xmath375 & @xmath376 & @xmath377 & @xmath378 + @xmath379 & @xmath380 & @xmath381 & @xmath382 & @xmath383 + in practice , the bins are in fact chosen after obtaining the truncated sample so the results should be more efficient , but this does not affect the preceding results obtained after grouping the whole sample and truncate from the bins since the aim is to give some feel about the accuracy of the estimations . also we can avoid grouping the observations by considering empirical frequencies obtained from kernel density estimations . consider a sample of size @xmath384 drawn from a gamma distribution @xmath385 with density given by@xmath386 and parameters @xmath387 and @xmath388 consider the data falling in @xmath290 fixed class intervals as shown in the following table , with mid - classes @xmath314 and absolute frequencies @xmath389 * table 5 . * @xmath315{ccccccccccc}\hline $ ] [email protected]@[email protected]@[email protected]@[email protected]@[email protected]@xmath31631.39@xmath318n_i@xmath31611@xmath31640@xmath31660@xmath316108@xmath316118@xmath316104@xmath316100@xmath31674@xmath31663@xmath31753@xmath390 [ c]cccccc@xmath391 & @xmath392 & @xmath393 & @xmath394 & @xmath395 & @xmath396 + @xmath397 & @xmath398 & @xmath268 & @xmath190 & @xmath188 & @xmath187 + in the following table we show the estimations @xmath399 from the minimum of distance @xmath33 and @xmath400 by the second method for the parameter @xmath401 with known @xmath402 according to the truncation @xmath117 considered . * table 6 . * the estimations @xmath399 and @xmath400 by the new approach of the parameter @xmath403 of the gamma probability distribution @xmath385 with @xmath404 and known @xmath387 . [ c]cccccc@xmath224 & @xmath117 & @xmath37 & @xmath66 @xmath225 & @xmath399 & @xmath405 + @xmath186 & @xmath406 & @xmath384 & @xmath185 & @xmath407 & @xmath408 + & @xmath409 & & & & + @xmath187 & @xmath410 & @xmath411 & @xmath412 & @xmath413 & @xmath414 + & @xmath415 & & & & + @xmath188 & @xmath416 & @xmath417 & @xmath418 & @xmath419 & @xmath420 + @xmath189 & @xmath421 & @xmath422 & @xmath423 & @xmath424 & @xmath425 + @xmath190 & @xmath426 & @xmath427 & @xmath264 & @xmath428 & @xmath429 + @xmath191 & @xmath430 & @xmath431 & @xmath194 & @xmath432 & @xmath433 + @xmath192 & @xmath434 & @xmath435 & @xmath436 & @xmath437 & @xmath438 + @xmath254 & @xmath439 & @xmath440 & @xmath441 & @xmath442 & @xmath407 + @xmath259 & @xmath443 & @xmath444 & @xmath445 & @xmath446 & @xmath447 + @xmath264 & @xmath448 & @xmath449 & @xmath450 & @xmath451 & @xmath452 + @xmath268 & @xmath453 & @xmath454 & @xmath455 & @xmath456 & @xmath457 + @xmath272 & @xmath458 & @xmath459 & @xmath460 & @xmath461 & @xmath462 + @xmath276 & @xmath463 & @xmath464 & @xmath465 & @xmath456 & @xmath442 + @xmath282 & @xmath466 & @xmath467 & @xmath468 & @xmath407 & @xmath469 + @xmath194 & @xmath470 & @xmath471 & @xmath472 & @xmath473 & @xmath474 + @xmath290 & @xmath475 & @xmath476 & @xmath477 & @xmath478 & @xmath479 + @xmath296 & @xmath480 & @xmath481 & @xmath482 & @xmath407 & @xmath483 + @xmath244 & @xmath484 & @xmath485 & @xmath486 & @xmath487 & @xmath488 + @xmath489 & @xmath490 & @xmath491 & @xmath492 & @xmath493 & @xmath494 + @xmath495 & @xmath496 & @xmath497 & @xmath498 & @xmath474 & @xmath499 + the estimations from the two methods are also accurate in this case of gamma distribution for the parameter @xmath500 here also the truncation proportion does not affect the quality of estimations . when we consider the complete data , the classical estimation is @xmath501 and the two new estimations are @xmath502 and @xmath503 as it was noticed in the examples above , the two methods lead to approximately the same estimation results . nevertheless , if the two estimations are significantly different , it seems related to the quality of the selected data . an important feature of this new approach is that the quality of estimations is uninfluenced by the truncation proportion . the following section will give further insights of the new approach . the preceding results have shown the effectiveness of the new approach and worked well in simulation experiments . furthermore , the proposition below will give an insight of a major feature of the new approach by considering the one parameter exponential family . we will prove that for all truncation considered formed by more than two points , from a sample of observations ; if the ratios of the relative frequencies of the @xmath314 are equal to the theoretical ones , then we may obtain the true value of the parameter . we may conjecture that when considering an arbitrary law of probability depending on @xmath504 parameters , such that we have a truncation composed by @xmath505 points having exact empirical ratios of the relative frequencies then we obtain the true values of the @xmath504 parameters . consider a probability distribution @xmath40 from the one - parameter exponential family with density@xmath172 , \label{expo - fam}\ ] ] where @xmath506 is the parameter , @xmath124 a statistic , @xmath173 a function of @xmath24 and @xmath174 is a function of the parameter @xmath145 assume that we wish to estimate the parameter @xmath145 if we consider a truncation having two points * * @xmath24 * * * * and @xmath77 * * * * with empirical frequencies * * @xmath507 and @xmath508 * * * * satisfying * * @xmath509 , then , using the approach considered here , we obtain the true value of @xmath145 \1 . if we consider the minimum of distance @xmath33 the result is immediate.2 . consider now the second method to estimate @xmath320 . consider two values @xmath24 and @xmath77 from the exponential family with density given by ( [ expo - fam])@xmath510 with @xmath5 being the estimation by the new approach , and assume that their empirical frequencies @xmath507 and @xmath508 are such that@xmath511 we obtain@xmath512 then , we solve on @xmath2 the following equation:@xmath513@xmath514 after straightforward algebra we obtain@xmath515 yielding the true value @xmath516 the proof is complete . note that the frequencies @xmath507 and @xmath508 need not be exact , that is @xmath507 may be different from @xmath34 and also @xmath517 but we require only that their ratio is equal to the theoretical one @xmath518 * examplesbinomial distribution . * consider again the binomial distribution @xmath181 with @xmath182 and @xmath223 and assume @xmath9 is known and we wish to estimate @xmath519 assume we have the following truncation with only two points @xmath520 the exact ratio of their probability distribution is given by @xmath521 which is a rational value that will simplify the example . choose the absolute frequencies of the two values considered as being @xmath522 and @xmath523 for the values @xmath524 and @xmath525 respectively , in order for having @xmath526 using the first approach , that of the minimum of distance @xmath154 we have to solve@xmath527 , \ ] ] and we get the true value @xmath528 . using the second method we have to solve the following equation on @xmath206@xmath529 and we obtain also the exact result @xmath530**gamma distribution . * * consider the gamma probability distribution @xmath385 with @xmath531 and @xmath532 assume that @xmath533 is known and we wish to estimate @xmath500 consider the truncation @xmath534 with @xmath535 and @xmath536 . we have the following value of the ratio @xmath537 ( the result is an approximate result since for probability density functions it is difficult to get an exact rational value but we will show that the estimations are very close to the true value ) . consider the absolute frequencies @xmath538 ( or @xmath539 ) and @xmath540 for the values @xmath535 and @xmath536 respectively . we have then @xmath541 using the minimum of distance @xmath154 we have to solve@xmath542@xmath543 , \ ] ] and we get the result @xmath544 from the second method , we compute @xmath545 and solve on @xmath403 the following equation@xmath546@xmath547 the result is @xmath548 now assume that the parameters @xmath533 and @xmath403 are unknown and show how to jointly estimate them using the new approach . since now there are two unknown parameters , we need to have three points from the support , so consider @xmath549 , @xmath550 and @xmath551 with their corresponding absolute frequencies @xmath552 and @xmath553 we have to find @xmath533 and @xmath403 which minimize the distance @xmath33 that is @xmath554 the result is @xmath555 and @xmath556 our aim here is not to give a detailed comparison study which needs to be investigated thoroughly , but only some elements of appreciation . a major feature which characterizes this new approach from the others is that when we have exact ratios of frequencies we obtain the true parameter and when their difference from the theoretical ratios decrease the quality of estimation increase even if we are using only a part from the sample of observations . this is not the case for classical approaches . in classical approaches , quality considerations are only viewed through mean properties of estimators or their asymptotic behaviour . by combining the two proposed methods we have in fact a point criterion . another characteristics is that the proportion of truncation has any effect on the quality of estimations . the first method uses a well known method of minimum distance but with a new one which has an important advantage of being symmetric , the property of which many traditional distances do not have . however , the estimations are obtained in this case implicitly so it is difficult to find explicit expressions and study their properties to compare them with classical ones . using the new distance we hope having fast convergent estimators since we expect that the influence of the errors in the frequencies will be slight in the new approach as we are using ratios of frequencies . consider now the second method of the new approach . we use classical procedures of estimation such as the maximum likelihood principle using the auxiliary distribution . we may obtain the estimators and study their properties as commonly used and then preserves the advantages of classical methods . in classical approach , given a sample , the estimation of certain parameters such as the mean and variance do not change according to the family of parent distributions . the latter information is not used and this disadvantages the approach . however , in the new approach the estimations of the mean and variance change according to the distribution from which the data emanated . the following two examples show the effectiveness of using the auxiliary distribution.*example . * consider the following frequency table : * table 7 . * [ c]cccc@xmath557 & @xmath187 & @xmath188 & @xmath558 + @xmath193 & @xmath559 & @xmath560 & @xmath9 + @xmath561 & @xmath562 & @xmath563 & @xmath186 + any sample of observations that satisfies the preceding frequency table may belong from one of the following distributions:@xmath564{c}\frac{x}{6}\\ 0 \end{array}\begin{array } [ c]{c}if\text { } x\in\left\ { 1,2,3\right\ } , \\ otherwise , \end{array } \right . \text { \ \ or \ \ } g_{2}\left ( x\right ) = \left\ { \begin{array } [ c]{c}\frac{x-1}{6}\\ 0 \end{array}\begin{array } [ c]{c}if\text { } x\in\left\ { 2,3,4\right\ } , \\ otherwise . \end{array } \right.\ ] ] the decision for determining which of the two distributions is more appropriate for table 7 , depends intuitively on the values @xmath559 and @xmath560 ( or @xmath507 and @xmath508 ) . however , if we use the classical maximum likelihood , we obtain that the samples of observations were generated from distribution @xmath565 whatever the values of @xmath559 and @xmath566 that is:@xmath567 we will show by using the new approach that the decision is more relevant . determine first the auxiliary distributions , @xmath565 and @xmath568 , based on the truncation @xmath569 , for @xmath570 and @xmath571 respectively . we obtain@xmath572{c}2/5\\ 3/5\\ 0 \end{array}\begin{array } [ c]{c}\text { \ \ } if\text { \ \ } x=2,\\ \text { \ \ } if\text { \ \ } x=3,\\ otherwise , \end{array } \right . \text { \ \ and \ \ } h_{2}(x)=\left\ { \begin{array } [ c]{c}1/3\\ 2/3\\ 0 \end{array}\begin{array } [ c]{c}\text { \ \ } if\text { \ \ } x=2,\\ \text { \ \ } if\text { \ \ } x=3,\\ otherwise . \end{array } \right.\ ] ] by using the maximum likelihood for @xmath565 and @xmath568 , we have to decide according to the quantities @xmath573 and @xmath574 solving the following inequality @xmath575 which is equivalent to @xmath576 where @xmath577 we obtain @xmath578 if @xmath579 the data were generated from @xmath571 and if @xmath580 , the data were generated from @xmath581 we can not make any decision about the case @xmath582**example . * * consider a binomial distribution with parameters @xmath583 and @xmath206 is unknown , from which we consider some samples of observations of size 15 given in table 8 by their absolute frequencies and chosen in order for having @xmath584 * table 8 . * [ c]|c|ccccc|cc| & & + samples & @xmath185 & @xmath186 & @xmath187 & @xmath188 & @xmath189 & @xmath585 & @xmath586 + @xmath186 & @xmath192 & @xmath254 & @xmath185 & @xmath185 & @xmath185 & @xmath587 & @xmath588 + @xmath187 & @xmath259 & @xmath190 & @xmath185 & @xmath186 & @xmath185 & @xmath587 & @xmath589 + @xmath188 & @xmath259 & @xmath189 & @xmath187 & @xmath185 & @xmath185 & @xmath587 & @xmath590 + @xmath189 & @xmath264 & @xmath188 & @xmath186 & @xmath186 & @xmath185 & @xmath587 & @xmath591 + @xmath190 & @xmath264 & @xmath189 & @xmath185 & @xmath185 & @xmath186 & @xmath587 & @xmath592 + @xmath191 & @xmath272 & @xmath185 & @xmath187 & @xmath185 & @xmath186 & @xmath587 & @xmath593 + @xmath192 & @xmath276 & @xmath185 & @xmath185 & @xmath185 & @xmath187 & @xmath587 & @xmath594 + it is clear that the information given by the samples are not the same , nevertheless the classical estimation method gives us the same estimation @xmath595 if we use the second method of the new approach , we have to solve the following equation for each sample:@xmath596 where @xmath597 is the corresponding auxiliary distribution . the estimations given by the new method differ from sample to another as shown in the latest column of table 8 , which is natural since each sample provides a different information about the parent distribution . we can also use the minimum of distance @xmath33 and we get also the same conclusion . the fact that the distance @xmath33 is a metric allow to propose various applications of this new measure . we can use it for model selection amongst different probability families . we choose two or more possible candidate parametric families of distributions , and for each alternative family , estimate the parameters to select a specific candidate . determine the distance between the specific candidate and the empirical distribution using the new metric @xmath86 finally , select the family which yields the minimum distance . in view of the new approach this can also be done in case of truncated data as opposed to classical approaches ( see for example cox @xcite , @xcite ) , taylor and jakeman @xcite ) for model selection which can be used , from the best of our knowledge , only for complete data . to investigate this perspective thoroughly , samples of various sizes from known distributions should be simulated , and the method for model selection applied , we can score the selection as correct or not after repeating the process a large number of times , the probability of correct selection could be estimated according to a given sample size . we can also use the new distance in cases where classical goodness of fit tests can not reject two candidate families . we can choose the one which yields the minimum of distance @xmath86 in the following examples , we shall select , in the first , between binomial distributions from truncated data . in the second example , we select between a weibull and a gamma distributions from right truncated data.*selection from binomial distributions . * we simulated @xmath598 samples of size @xmath599 from a binomial distribution @xmath600 and each time we retained only the observations belonging from @xmath601 with their frequencies . then we tried to identify the law simulated starting from the corresponding table of frequencies . we used the distance @xmath33 to select between the original distribution of each simulated sample and the distribution @xmath602 and we score the selection as correct if the distance between the empirical distribution and the original one is less than with the alternative one @xmath603 the correct distribution was selected @xmath604 . conversely , we simulated @xmath598 samples of size @xmath599 from a binomial distribution @xmath602 and we select with @xmath600 , the correct distribution was selected @xmath605**selection between weibull and gamma distributions . * * we simulated @xmath598 samples of size @xmath606 from the weibull distribution @xmath607 and we truncated them on right by considering only observations above the cut - off @xmath608 . each truncated sample was summarized into @xmath268 classes . we selected between @xmath607 and the gamma distribution @xmath609 the distance @xmath33 has selected the correct distribution , that is @xmath610 @xmath611 we can also find , before selecting between distribution , the best fit from the family of gamma distributions @xmath612 of the truncated data from a given probability density say @xmath613 we have then to solve an optimization problem of finding the minimum of a function of two variables , @xmath614 where @xmath3 is the empirical distribution and @xmath615 , using well known methods such as lavenberg - marquardt using a computer algebra package . also it should be better to choose the number of bins for each truncated sample by an optimal procedure , for example that of birg and rozenholc @xcite . the initial starting value is of great importance in convergence behaviour of algorithms such as _ em _ algorithm . usually , as for the latter , the initial trial value is guessed . surprisingly , we will show that our procedure gives an estimation of the starting value instead of having to guess . the approach will be illustrated by the following classical example which was the basis of the _ em _ algorithm.*example of hartley ( 1958 ) revisited . * hartley @xcite used an algorithmic procedure to estimate the parameter of a poisson distribution from data on the pollution of a sort of seeds by the presence of noxious weed seeds quoted from snedecor @xcite and truncated them by missing the frequencies of the values @xmath185 and @xmath186 as shown in the following table 9 ( table 1 in hartley @xcite ) * table 9 . * [ c]cccccccccccvalues & missing & @xmath185 & @xmath186 & & & & & & & + & observed & & & @xmath187 & @xmath188 & @xmath189 & @xmath190 & @xmath191 & @xmath192 & @xmath259 + frequencies @xmath193 & & & & @xmath616 & @xmath290 & @xmath244 & @xmath259 & @xmath188 & @xmath190 & @xmath186 + hartley @xcite has guessed the frequencies of the missing values @xmath185 and @xmath186 by taking @xmath617 and @xmath618 and after 4 steps of his algorithmic procedure , which has been the basis of the well known em algorithm for incomplete data ( dempster , laird and rubin @xcite ) , has reached the estimation @xmath619 ( see table 1 p.177 hartley @xcite ) . using the second method , we get the estimation @xmath620 and by proportional allocation procedure we can see that the frequencies we get are @xmath621 and @xmath622 which are close to the guessed values . using the distance @xmath33 we obtain the estimation @xmath623 and by removing the last value which has a small frequency @xmath624 we obtain a better result @xmath625 which are also appreciable as starting values since in practice the true parameter is unknown.*initial trial value for mixture normal populations . * we shall present an application of the previous method used for truncated data in the situation where we have a mixture population of two normal distributions . in classical methods , we use the merged distribution * * @xmath626 * * and we estimate the parameters @xmath627 @xmath628 and @xmath629 * * using for example the em algorithm which is based on maximizing the complete likelihood of the merged distribution by an algorithmic procedure from a guessed initial trial value . however , the problem of occurrence of several local maxima is well - known for the setting of em algorithm . also , seidel , mosler and alker @xcite pointed out that the likelihood - ratio test in mixture models depends on the choice of the initial trial value for the em algorithm . if the initial trial value is close to the true value it is clear that the algorithm will converge in few steps to the true local maximum . we will show that using the new approach we get an accurate estimated initial trial value . assume we have a merged sample from two samples of observations of sizes @xmath559 and @xmath560 from two normal distributions @xmath630 and @xmath631 , with @xmath632 . by assuming that @xmath633 and @xmath634 are known , our aim is to estimate the means @xmath628 and @xmath635 and also the merging proportion @xmath636 of each population . we will use a method based on truncations . the main idea being to split the range of the merged sample into three suitably chosen parts a central part where the observations are highly merged , a left and right truncated parts where the observations become mainly from one of the distributions considered . if for example @xmath637 , then to estimate @xmath628 we have to use the chosen right truncated part ( left truncation @xmath117 ) . the procedure is summarized as follows : * 1 . * we compute the sample mean @xmath638 of the merged observations . * 2 . * for determining the location of the two means @xmath628 and @xmath629 , we compute the empirical standard deviation @xmath639 of the observations less than @xmath640 and @xmath641 for those that are greater . assume that @xmath642 , in this case if @xmath643 then we deduce that @xmath628 is situated on the left of @xmath638 . otherwise , it will be assumed to be on its right . we follow the same idea for the case @xmath644 if @xmath645 we pass directly to the third step . assume that @xmath628 is on the left . it is well known that for a normal distribution @xmath646 we have @xmath647 m-\sigma , m+\sigma\right [ ) $ ] @xmath648 we hope that on the left of @xmath649 the number of observations generated from @xmath650 is negligible , and on the right of @xmath651 the number of observations generated from @xmath652 is also negligible . hence , to estimate @xmath628 , we consider only the part of observations situated on the left of @xmath653 , and to estimate @xmath629 we consider the part situated on the right of @xmath654 the following example will provide some feel for the accuracy of the procedure.*example . * we consider the case where @xmath655 consider two samples of observations generated from @xmath652 and @xmath656 where @xmath657 and @xmath658 , with known @xmath659 and sizes @xmath660 and @xmath661 we combine them to obtain a merged sample of size @xmath662 we have chosen the distributions in such a way that the histogram ( fig.1 ) of the merged sample does not show directly the existence of a mixture of two distributions . when the histogram of the merged population is bimodal the situation is more easier , since when taking a suitably left ( or right ) part we get more accurate estimation from the situation that this part will have a negligible number of observations from the second distribution . figure1-conf.eps fig 1 . merged histogram of two normal distributions @xmath663 and @xmath664 it should be stressed that the histogram is one modal and does not show at first glance any mixture situation . following the steps of the procedure we begin by calculating the mean of the resulting merged sample and we obtain @xmath665 . since the standard deviations are assumed to be equal then we compute directly @xmath666 by grouping the observations on the left of @xmath667 ( which constitute the chosen right truncated part ) in @xmath192 classes we obtain the following table : * table 10 . * [ c]ccccccc@xmath314 & @xmath668 & @xmath669 & @xmath670 & @xmath671 & @xmath672 & @xmath673 + @xmath193 & @xmath186 & @xmath188 & @xmath191 & @xmath296 & @xmath674 & @xmath675 + using the distance @xmath33 we obtain for all the truncation @xmath676 and by deleting @xmath677 we get the value @xmath678 the sample mean of the observations on the left of @xmath667 is given by @xmath679 using the second method we have to solve on @xmath320 the following formula@xmath680 + u_{2}\times\exp\left [ \frac{-\left ( u_{2}-m\right ) ^{2}}{2\sigma^{2}}\right ] + ... + u_{k}\times\exp\left [ \frac{-\left ( u_{k}-m\right ) ^{2}}{2\sigma^{2}}\right ] } { \exp\left [ \frac{-\left ( u_{1}-m\right ) ^{2}}{2\sigma^{2}}\right ] + \exp\left [ \frac{-\left ( u_{2}-m\right ) ^{2}}{2\sigma^{2}}\right ] + ... + \exp\left [ \frac{-\left ( u_{k}-m\right ) ^{2}}{2\sigma^{2}}\right ] } = \overline{u}_{l}. \label{m2}\ ] ] we obtain the estimation @xmath681 by deleting the first value @xmath677 which has a weak frequency @xmath682 that is using the truncation @xmath683 ( we compute again @xmath684 ) we obtain a better estimation @xmath685 which is very close to the true value @xmath657 . to estimate @xmath635 we consider the part situated on the right of @xmath686 grouping the observations on the right of @xmath687 ( which constitute the chosen right part ) in @xmath192 classes we obtain the following table : * table 11 . * [ c]ccccccc@xmath314 & @xmath688 & @xmath689 & @xmath690 & @xmath691 & @xmath692 & @xmath693 + @xmath193 & @xmath694 & @xmath354 & @xmath194 & @xmath259 & @xmath192 & @xmath188 + using the distance @xmath33 for all the truncation we get @xmath695 the sample mean of the observations on the right part is given by @xmath696 using formula ( [ m2 ] ) with @xmath697 , we obtain the result @xmath698 deleting the extreme values @xmath677 and @xmath699 we obtain @xmath700 the mixture proportion @xmath636 can easily be estimated using the formula @xmath701 considering the estimations obtained , which are close to the true values of @xmath628 and @xmath635 it is clear that the em algorithm will converge fastly to the unique solutions . we can obtain empirical quantile estimations of @xmath33 using montecarlo or bootstrapping technics , and use them in a test of goodness of fit for a specified probability distribution . we simulate @xmath702 samples of the same size from the specified probability distribution and calculate the distances @xmath703 we can then estimate the asymptotic distribution of @xmath33 by@xmath704 consequently , for a sample of the same size we compute @xmath705 and we reject the hypothesis that it belongs from the specified distribution if @xmath706 for a given level of significance @xmath707 the values @xmath708 may be obtained from the empirical distribution function @xmath709 of the sample . the fact that the new measure @xmath33 is not equivalent to classical ones means that it treats other aspects not investigated by the latter . this may open new perspectives such as making decision about the accuracy of an estimation in cases where the classical and new estimations are close to each others . in cases where the classical estimation and the new one using @xmath33 are significantly different then we can say that the sample of observations considered does not restore coherently all necessary information about the parent distribution from which it emanated . in the foregoing study , we have presented a new statistical point estimation method which found be useful in truncated and grouped and censored data situations . a new distance between probability distributions was introduced . it measures the difference between the variations of two given probability distributions . we introduced an auxiliary distribution based on a truncation , from a chosen family of probability distributions . this new distribution will have the same parameters to estimate as the parent one . we use then statistical methods to estimate the parameters of the random variable under study using the empirical and new auxiliary distribution in the region that captures the data , from which we determine the corresponding parent distribution . the later is the estimation by the new method . using the new distance introduced we also estimate by the minimum distance approach and use the resulting estimation as a control on the accuracy of estimation obtained by the former method . we have obtained a result which states that if we have to estimate the parameter of a probability distribution from the one parameter exponential family , then it suffices to have two points with exact ratio of frequencies , that is equal to the theoretical one expressed by the ratio of the value of the probability distribution on these two points , to obtain the true value of the parameter . we have conjectured that if we have in general @xmath504 parameters , then it suffices to have @xmath505 points with exact ratios of their frequencies to obtain the @xmath504 true parameters exactly . the later result need to be proved rigorously in a general setting for other distributions than the class considered . a large comparative study between the classical and new methods should also be investigated . we presented some perspectives of the new approach such as model selection from truncated data using the new distance , estimation of the first trial value in the celebrate _ em _ algorithm in the case of truncation and for mixture of two normal populations , a test of goodness of fit based on the new distance , decision making about the quality of estimations and data . klein , j. p. and zhang , m. j. ( 1996 ) . statistical challenges in comparing chemotherapy and bone marrow transplantation as a treatment for leukemia , _ lifetime data : models in reliability and survival analysis , n.p . jewel , 175 - 185 . _

in this paper , we characterize all links in with bridge number at least three that have a bridge sphere of distance two . we show that a link has a bridge sphere of distance at most two then it falls into at least one of three categories : * the exterior of contains an essential meridional sphere . * can be decomposed as a tangle product of a montesinos tangle with an essential tangle in a way that respects the bridge surface and either the montesinos tangle is rational or the essential tangle contains an incompressible , boundary - incompressible annulus . * is obtained by banding from another link that has a bridge sphere of the same euler characteristic as the bridge sphere for but of distance 0 or 1 .

1309.3787

following hempel s definition of distance for heegaard splittings @xcite , there have been a number of results showing that knots with high distance bridge surfaces are well behaved . for example , high distance knots are hyperbolic and do not contain essential surfaces of small euler characteristic @xcite and they have unique minimal bridge surfaces @xcite . additionally , connect sums of knots with high distance bridge surfaces have multiple distinct bridge surfaces @xcite . most recently , the cabling conjecture was established for all knots with a bridge surface of distance at least three @xcite . these results motivate us to study low distance knots and links . ( we will use the term `` link '' below to mean both one - component and multiple - component links . ) links that have a bridge surface of distance two are of particular interest because a link with a minimal bridge sphere of distance one always contains an essential meridional surface @xcite and links with bridge surfaces of distance greater than two share many nice properties that often allow them to be addressed collectively as high distance " links . in this paper , we prove the following theorem . precise definitions will be given later . [ thm : main ] let @xmath1 be a link in @xmath0 and let @xmath3 be a bridge sphere for @xmath1 of distance 2 . then at least one of the following holds : 1 . there is an essential meridional sphere in the exterior of @xmath1 , 2 . @xmath1 can be decomposed as a tangle product of a bridge - split montesinos tangle @xmath4 with an essential tangle @xmath5 such that @xmath6 is a disk bridge surface for @xmath4 and either @xmath4 is rational or the exterior of @xmath7 in @xmath8 contains an incompressible , boundary incompressible annulus , or 3 . @xmath1 can be obtained via banding from a link @xmath2 that has a bridge sphere with the same euler characteristic as @xmath9 but has distance 0 or 1 . conclusion ( 2 ) of theorem [ thm : main ] is illustrated in figure [ fig : thmexmont ] . conclusion ( 3 ) is illustrated in figure [ fig : thmexband ] . all of the conclusions are discussed in detail in section [ sect : constr ] . with a tangle @xmath5 such that the bridge surface for @xmath1 is a disk bridge surface for @xmath4 . ( each group of strands may , in general , consist of arbitrarily many strands.),width=192 ] we note that this paper is the analogue to recent results about 3-manifolds with distance - two heegaard splittings . hempel @xcite and thompson @xcite ( independently ) classified all 3-manifolds that have genus two , distance - two heegaard splittings . in work in preparation , rubinstein and thompson @xcite have proved a similar classification theorem for all manifolds that have a distance - two heegaard splitting of any genus . portions of the proof of theorem [ thm : main ] are inspired by the work of rubinstein and thompson on distance - two heegaard splittings . the analogy between the present work and these results about heegaard splittings is most easily seen in the three - bridge case . theorem [ thm : main ] implies the following : [ coro : main ] if @xmath1 is a 3-bridge knot with a minimal bridge sphere of distance two , then one of the following holds : 1 . the exterior of @xmath1 contains an incompressible meridional 4-punctured sphere . @xmath1 is obtained by banding a 3-bridge presentation of either the unknot , a 2-bridge knot , or the connected sum of two 2-bridge knots . @xmath1 is a small montesinos knot . recall that every genus - two 3-manifold is the double - branched cover of @xmath0 over a three - bridge knot and that every genus - two heegaard surface is the lift of a 6-punctured bridge sphere to the double - branched cover . additionally , every essential loop in a genus - two heegaard surface is the lift of an essential loop in the corresponding bridge sphere . this fact implies that the distance of a 6-punctured bridge sphere is equal to the distance of the corresponding genus - two heegaard surface in the double - branched cover . thus , as expected , the heegaard splittings in the hempel / thompson classification are precisely the double branched covers of the links in corollary [ coro : main ] . we describe the types of distance - two bridge spheres in section [ sect : constr ] . a number of technical lemmas are proved in section [ sect : simplifying ] , followed by the proof of theorem [ thm : main ] and corollary [ coro : main ] in section [ sect : proof ] . let @xmath1 be a link in the 3-sphere and @xmath10 a regular neighborhood of @xmath1 . let @xmath11 be the standard height function from @xmath0 to @xmath12 $ ] whose level surfaces are concentric 2-spheres . a level sphere @xmath13 is called a _ bridge sphere _ for @xmath1 if all maxima of @xmath14 are above @xmath3 and all minima of @xmath14 are below @xmath3 . in particular , if @xmath3 is a bridge sphere for @xmath1 , then @xmath1 intersects each of the 3-balls bounded by @xmath3 in boundary parallel arcs called _ bridges_. the disks of parallelism are called _ bridge disks_. more precisely , a bridge disk is a disk with interior disjoint from @xmath1 and the bridge surface , and whose boundary is the union of an arc in @xmath1 and an arc in the bridge surface . for technical reasons , we will only discuss links whose bridge spheres have bridge number at least three ( so at least 6 punctures ) . however , two - bridge knots and links are well understood @xcite and every one - bridge knot is an unknot . given a morse function @xmath15 $ ] where @xmath16 is any 3-manifold , an arc or a surface @xmath17 properly embedded in @xmath16 is _ vertical _ if @xmath18 does not have any critical points in the interior of @xmath17 . the _ curve complex _ of a compact surface @xmath17 is the simplicial complex with vertices that correspond to isotopy classes of essential ( including not boundary parallel ) simple closed curves in @xmath17 . an edge connects two vertices if the corresponding isotopy classes of curves have disjoint representatives . the _ distance _ of a bridge sphere @xmath3 for a link @xmath1 in @xmath0 is the length of the shortest path in the curve complex for @xmath9 from the boundary of a compressing disk for @xmath9 in @xmath19 above @xmath3 to the boundary of a compressing disk for @xmath9 in @xmath19 below @xmath3 . the notion of distance measured in the curve complex was first introduced by hempel in @xcite . there is also a similar notion of distance measured in the arc and curve complex introduced by bachman and schleimer in @xcite . for a bridge sphere of distance two , every compressing disk on one side of the bridge sphere intersects every compressing disk on the other side non - trivially , but there is an essential simple closed curve @xmath20 in @xmath9 that is disjoint from at least one compressing disk on each side of @xmath9 . for each of the possible conclusions of our main theorem , we consider the possibility of a converse . recall that a _ meridional surface _ for a link @xmath21 is a surface properly embedded in @xmath19 whose boundary is a set of meridional curves in @xmath22 . a meridional surface @xmath23 is _ compressible _ if there is a disk in @xmath24 with interior disjoint from @xmath23 whose boundary is contained in @xmath23 and is essential in @xmath25 . a meridional surface @xmath23 is _ incompressible _ if it is not compressible . the first possible conclusion of our main theorem is that our link has an incompressible , boundary - incompressible planar meridional surface . in general , it is likely that there are knots of distance at least 3 which have essential meridional planar surfaces in their exterior . however , if there is an essential planar surface c for the link which intersects the bridge surface @xmath3 in a single simple closed curve which is essential on both surfaces , such that @xmath23 is disjoint from compressing discs for @xmath3 on both sides of @xmath3 , then the link will have distance at most 2 . the intersection curve @xmath26 is the middle vertex in a length two path in the curve complex of @xmath3 . in some cases , the planar surface we construct in the proof of the main theorem will be of this type . in what follows , it will be natural to study tangles from the perspective of the height function on @xmath27 $ ] . let @xmath28 be the projection map @xmath27 \rightarrow [ -1,1]$ ] . given a tangle @xmath29 properly embedded in @xmath30 $ ] such that @xmath31 , we will say that @xmath29 is in _ bridge position _ if all of the maxima of @xmath32 project into @xmath33 and all of the minima of @xmath32 project into @xmath34 . a disk @xmath35 is a _ disk bridge surface _ for @xmath29 if @xmath36 is a regular value of the height function @xmath32 , and all the minima of @xmath29 lie below and all the maxima lie above it . recall that an @xmath37-strand tangle @xmath38 is _ rational _ if each of the @xmath37 strands can be simultaneously isotoped into @xmath39 while fixing their endpoints . an @xmath37-strand tangle @xmath38 is _ essential _ if @xmath39 is incompressible when viewed as a meridional , planar surface . let @xmath40 for @xmath41 be a rational tangle and let @xmath42 be an incompressible @xmath43-punctured disc for some @xmath44 . the result @xmath38 of gluing @xmath45 and @xmath46 along @xmath47 and @xmath48 is a _ montesinos tangle _ if the image of @xmath47 and @xmath48 is not boundary - parallel in @xmath38 . the disc @xmath49 in @xmath38 is a _ montesinos disk _ for @xmath38 . given a montesinos tangle @xmath29 properly embedded in @xmath30 $ ] such that @xmath31 , we say that @xmath29 is _ bridge - split _ if @xmath32 has both maxima and minima , the foliation induced by @xmath28 on @xmath50 consists of parallel arcs and @xmath50 separates the maxima of @xmath29 from the minima of @xmath29 , as in figure [ fig : bridgesplit ] . a knot @xmath51 in @xmath0 is a _ tangle product _ of tangles @xmath45 and @xmath46 if @xmath0 is the result of gluing @xmath52 to @xmath8 along their boundary and @xmath53 . suppose a link @xmath51 has a tangle decomposition such that one of the tangles @xmath38 is a bridge - split montesinos tangle . if the bridge surface @xmath3 meets @xmath54 in a disk bridge surface for @xmath29 , then @xmath3 has distance at most two . a disk bridge surface for a bridge - split montesinos tangle is depicted in green in figure [ fig : bridgesplit ] . a third way to construct a bridge surface of distance at most two is as follows : let @xmath2 be any link such that some local maximum @xmath55 lies below some local minimum @xmath56 with respect to the standard height function @xmath11 . notice that if we monotonically isotope all maxima of @xmath2 above all minima of @xmath2 , there will be a compressing disk for one of the resulting bridge spheres @xmath3 that surrounds @xmath57 and a second disjoint compressing disk around @xmath58 , on the other side of @xmath3 . thus , the level sphere corresponding to a bridge sphere for @xmath2 will have distance at most 1 . we can also construct a distance - one bridge surface for any link by perturbing a given bridge surface , i.e. adding a new pair of canceling bridges . this new bridge surface will no longer be minimal . choose any vertical arc @xmath59 with one endpoint at @xmath57 and the other endpoint at a local maximum above @xmath58 , as in figure [ fig : banding ] . let @xmath60 be a band along @xmath59 , i.e. the image of a square @xmath12 \times [ 0,1]$ ] such * the interior of @xmath61 is disjoint from @xmath2 , * @xmath61 is vertical with respect to @xmath11 , * @xmath59 is the image of @xmath62 $ ] and * the arcs @xmath12 \times \{0\}$ ] and @xmath12 \times \{1\}$ ] are contained in small neighborhoods of @xmath63 . let @xmath1 be the result of replacing the arcs @xmath12 \times \{0\}$ ] and @xmath12 \times \{1\}$ ] of @xmath2 with the arcs @xmath64 $ ] and @xmath65 $ ] . note that even after we choose @xmath59 , there are infinitely many different possible bands @xmath61 , related by twisting around @xmath59 . in particular , if @xmath2 is a knot , but our initial choice of @xmath61 produces a link , we can add a half twist to ensure that @xmath1 is a knot . we will say that @xmath1 is obtained from @xmath2 by _ more generally , the banding operation described here is also known as _ integer - sloped rational tangle replacement_. however , for our purposes it suffices to note that the resulting link has a bridge surface of distance at most two . to see this , let @xmath66 be the level sphere immediately above the highest minimum of @xmath2 and let @xmath67 be the boundary of a regular neighborhood in @xmath66 of @xmath68 . any maximum of @xmath2 disjoint from @xmath61 gives rise to a compressing disk for @xmath66 contained above @xmath66 and disjoint from @xmath67 . similarly , any minimum of @xmath2 that does not cobound ( with @xmath57 ) a monotone subarc of @xmath2 will give rise to a compressing disk for @xmath66 contained below @xmath66 and disjoint from @xmath67 . thus , @xmath66 has distance at most two . by choosing a `` sufficiently complicated '' arc @xmath59 , we expect that one can always construct a link of distance exactly two , but proving this is beyond the scope of the present paper . in this section , we introduce several results that will be needed in the proof of the main theorem . we will have several occasions to consider a tangle embedded in a cylinder and a disjoint collection of boundary compressing disks for its disk bridge surface . in the next lemma , we show that we can always assume that such boundary compressing disks for the disk bridge surface are vertical with respect to the natural height function on @xmath27 $ ] given by projection onto the second component . we can in fact assume so while also assuming that the tangle is polygonal and composed only of vertical and horizontal components . [ lem : vertical boundary compressing disks ] let @xmath29 be a tangle properly embedded in a cylinder @xmath69 $ ] such that all endpoints of @xmath29 are in @xmath70 . suppose that @xmath71 is a disk bridge surface for @xmath29 . let @xmath72 and @xmath73 be collections of pairwise disjoint boundary compressing disks for @xmath71 above and below @xmath71 respectively such that @xmath74 . then there is an isotopy during which the tangle remains transverse to @xmath71 and after which all disks in @xmath72 and @xmath73 are disjoint and vertical . furthermore , we may assume that after the isotopy , the tangle @xmath29 is a set of polygonal curves composed only of horizontal and vertical arcs such that all horizontal arcs are contained in @xmath75 or @xmath76 and each component of @xmath77 contains at most one horizontal arc . let @xmath78 $ ] and @xmath79 $ ] . the proof follows the following basic structure . we begin by using standard techniques to construct an ambient isotopy so that @xmath80 , @xmath81 and subsequently @xmath82 have the desired properties . we then analyze the intersection between @xmath83 and the vertical bridge disks for @xmath82 to construct an ambient isotopy that results in @xmath72 being vertical and disjoint from @xmath73 while preserving the desired properties of @xmath80 , @xmath81 and @xmath82 . first , perform an isotopy supported in a neighborhood of @xmath84 after which each arc of @xmath81 that is parallel into @xmath85 consists of two vertical arcs and one horizontal arc in @xmath76 and all other arcs of @xmath81 are @xmath86 fibers of @xmath87 $ ] . because the boundary of each disk of @xmath73 consists of two vertical and two horizontal arcs , each disk is isotopic to a vertical disk . by the isotopy extension theorem , these isotopies extend to a single ambient isotopy in a neighborhood of @xmath84 that fixes @xmath81 and results in @xmath73 being vertical . note that this ambient isotopy preserves @xmath74 . next , perform an isotopy supported in @xmath88 ( where @xmath89 is a regular neighborhood of @xmath84 ) after which each arc of @xmath82 that is parallel into @xmath85 consists of two vertical arcs and one horizontal arc in @xmath75 and every other arc of @xmath90 is an @xmath86 fiber of @xmath91 $ ] . this isotopy does not affect @xmath73 or @xmath92 . after these isotopies , @xmath29 meets the conclusion of the theorem , with the exception that it remains to be shown that we can now perform an isotopy that guarantees that @xmath72 is a collection of vertical disks without affecting the conditions on @xmath29 and @xmath73 that we have already imposed . let @xmath93 be a complete collection of vertical bridge disks for @xmath90 . consider @xmath94 . after an isotopy of @xmath72 supported in the interior of @xmath95 which does not affect the tangle or @xmath93 , we may assume that this intersection consists only of arcs . let @xmath20 be an outermost arc of intersection on some disk @xmath96 . this arc cuts off a subdisk @xmath97 whose interior is disjoint from all disks in @xmath93 and a subdisk @xmath17 from some disk in @xmath93 . let @xmath54 be the ball cobounded by @xmath98 , @xmath17 and a subdisk @xmath99 of @xmath71 . by standard transversality arguments , we can assume that there is @xmath100 so that @xmath101)$ ] is vertical . let @xmath102 be the set of all bridges contained in @xmath54 . since @xmath93 is disjoint from @xmath98 , the bridge disk of each @xmath103 is disjoint from @xmath98 so there is a vertical isotopy of @xmath102 relative to @xmath93 and supported in the interior of @xmath54 after which arcs in @xmath102 are contained in @xmath104 $ ] . now the interior of @xmath105)$ ] is a ball that is completely disjoint from the link and thus there is an isotopy shrinking @xmath105)$ ] so it lies in a neighborhood of @xmath106 , as in figure [ fig : isotopy ] . and @xmath72 to remind the reader that @xmath20 is not necessarily outermost on @xmath93 . we have dropped these arcs in the rest of the figures to simplify the picture.,title="fig : " ] ( -255,290)@xmath98 ( -308,270)@xmath17 ( -273,246)@xmath99 ( -282,275)@xmath107 ( -60,275)@xmath54 ( -270,50)@xmath23 let @xmath23 be the cylinder @xmath108 $ ] together with the ball @xmath54 after the above isotopies . the cylinder @xmath108 $ ] may , of course , intersect @xmath29 both in its boundary and in its interior . perform a horizontal isotopy of @xmath23 , first shrinking it horizontally until @xmath109 is contained in a neighborhood of @xmath17 and then pushing @xmath23 through @xmath17 , thus reducing @xmath94 , as in figure [ fig : isotopy ] . this sequence of isotopies preserves the property that @xmath29 is composed of horizontal and vertical arcs and that @xmath73 is vertical . note also that these ambient isotopies preserve the property that @xmath74 . thus , we may assume that @xmath110 . however , if the collection of disks @xmath72 is contained in the complement of all vertical bridge disks then they are in fact contained in a product region and thus can be isotoped to be vertical via an isotopy that is supported in @xmath111 $ ] , does not affect @xmath29 , and preserves the fact that @xmath74 . the following is a standard sweepout argument similar to that of @xcite . [ lem : weakly reducible ] let @xmath38 be a tangle with @xmath112)$ ] such that @xmath113 and let @xmath3 be a bicompressible disk bridge surface for @xmath29 . let @xmath114 \rightarrow [ -1,1]$ ] be the natural projection map . suppose @xmath50 is a compressing disk for @xmath115 in @xmath116 such that @xmath117 can not be isotoped to be disjoint from @xmath118 . then there is an isotopy of @xmath50 that fixes @xmath117 and the height and number of saddles of @xmath50 after which one of the following occurs : 1 . there exists a value @xmath119 of @xmath120 such that @xmath121 is isotopic to @xmath3 and there exists a compressing disk for @xmath121 that is disjoint from a compressing disk or boundary compressing disk for @xmath121 on the opposite side of @xmath121 , 2 . there exists a regular value @xmath119 of @xmath120 such that @xmath121 is isotopic to @xmath3 and there are two disjoint boundary compressing disks for @xmath121 with boundary contained in @xmath122 and on opposite sides of @xmath121 , or 3 . there exists a critical value @xmath119 of @xmath120 such that @xmath121 is isotopic to @xmath3 and there are two disjoint boundary compressing disks for @xmath121 on opposite sides of @xmath121 with the property that the boundaries of these disks meet @xmath121 in the boundary of a regular neighborhood of the component of @xmath122 that contains a saddle . consider the projection map @xmath114 \rightarrow [ -1,1]$ ] . we have @xmath123 for each @xmath124 $ ] and we will assume @xmath125 . isotope all maxima of @xmath29 up above all critical values of @xmath126 into a neighborhood of @xmath127 and isotope all minima of @xmath29 down below all critical values of @xmath126 into a neighborhood of @xmath128 . this isotopy fixes @xmath117 and fixes the height and number of saddles of @xmath50 . after this isotopy , each component of the intersection between @xmath50 and a neighborhood of @xmath129 is either a vertical boundary compressing disk or a compressing disk with level boundary and a single critical point corresponding to a maximum . similarly , @xmath50 meets a neighborhood of @xmath128 in a collection of vertical boundary compressing disks and compressing disks with level boundary and a single critical point corresponding to a minimum . let @xmath130 $ ] be the level of the highest local minimum of @xmath29 ( with respect to @xmath28 ) and let @xmath61 be the level of the lowest local maximum . then for every @xmath131 , the disk @xmath121 is isotopic to @xmath3 by an isotopy transverse to @xmath29 . if for any @xmath131 @xmath121 meets @xmath50 in a collection of arcs and loops that are all inessential in @xmath121 , then @xmath50 can be isotoped to be disjoint from @xmath121 . hence , we will assume that for every @xmath131 @xmath121 meets @xmath50 in at least one component that is essential in @xmath121 . for each fixed @xmath119 , an innermost loop essential in @xmath121 or an outermost arc essential in @xmath121 of @xmath132 in @xmath50 will bound a subdisk of @xmath50 that can be isotoped to be either a compressing or boundary compressing disk for @xmath121 . for @xmath119 near @xmath133 , at least one of these disks will be below @xmath121 , while for @xmath119 near @xmath61 , at least one of these will be above @xmath121 . if there is a value @xmath131 with loops or arcs of intersection defining disks both above and below then these disks are disjoint ( since @xmath50 is embedded ) and we have one of conditions ( 1 ) or , in the case that both disks are boundary compressing disks , conclusion ( 2 ) . otherwise , there must be a critical point of @xmath126 at which the disks switch from below to above . however , in this case the loops or arcs above and below the critical point correspond to disjoint arcs or loops in @xmath3 , so we obtain conclusion ( 1 ) or , in the case that both disks are boundary compressing disks , conclusion ( 3 ) . [ lem : weakly reducible2 ] let @xmath38 be a tangle with @xmath112)$ ] such that @xmath113 and let @xmath3 be a bicompressible disk bridge surface for @xmath29 . let @xmath114 \rightarrow [ -1,1]$ ] be the natural projection map . suppose @xmath50 is a compressing disk for @xmath115 in @xmath116 such that @xmath117 can be isotoped to be disjoint from @xmath118 . assuming @xmath134 , there is an isotopy of @xmath50 that fixes @xmath117 and the height and number of saddles of @xmath50 after which one of the following occurs : 1 . there exists a regular value @xmath119 of @xmath120 such that @xmath121 is isotopic to @xmath3 and there are two disjoint compressing disks for @xmath121 on opposite sides of @xmath121 , 2 . there exists a critical value @xmath119 of @xmath120 such that @xmath121 is isotopic to @xmath3 and there are two disjoint compressing disks for @xmath121 with boundaries contained in the boundary of a regular neighborhood of the component of @xmath122 that contains a saddle and on opposite sides of @xmath121 , 3 . for some @xmath135 , @xmath136 , 4 . there is a strict subdisk of @xmath3 that is bicompressible . without loss of generality , we can isotope @xmath117 to be contained in @xmath137 via an isotopy supported in a regular neighborhood of @xmath39 . note that there can be no arcs of intersection between @xmath50 and @xmath121 for any @xmath119 . isotope all maxima of @xmath29 up above all critical points of @xmath126 into a neighborhood of @xmath127 and isotope all minimum of @xmath29 down below all critical points of @xmath126 into a neighborhood of @xmath128 . this isotopy fixes @xmath117 and fixes the height and number of saddles of @xmath50 . after this isotopy , the collection of components of the intersection between @xmath50 and a neighborhood of @xmath129 is the union of a single vertical annulus with compressing disks , each having level boundary and a single critical point which is a maximum . similarly , @xmath50 meets a neighborhood of @xmath128 in a collection of compressing disks with level boundary and a single critical point corresponding to a minimum . if the intersection between @xmath50 and a neighborhood of @xmath129 contains a compressing disk , then by the proof of lemma [ lem : weakly reducible ] we can conclude ( 1 ) or ( 2 ) holds . if the intersection between @xmath50 and a neighborhood @xmath138 $ ] of @xmath129 is a single vertical annulus @xmath139 , then @xmath50 meets the portion of @xmath54 below @xmath140 in a single compressing disk @xmath141 . if @xmath142 is isotopic to @xmath143 , then ( 3 ) holds , so we can assume that @xmath142 is not isotopic to @xmath143 . since @xmath3 is bicompressible , let @xmath144 be a compressing disk for @xmath140 above @xmath140 and let @xmath145 be a compressing disk for @xmath140 below @xmath140 . boundary compress @xmath144 along @xmath139 and boundary compress @xmath145 along @xmath141 to produce two new compressing disks for @xmath140 each of which is disjoint from @xmath142 . if the boundaries of these disks lie in distinct components of the complement of @xmath142 in @xmath140 , then conclusion ( 1 ) holds . if the boundaries of these disks lie in a common component of the complement of @xmath142 in @xmath140 , then conclusion ( 4 ) holds . the following is known as the pop - over lemma and has been used in a number of papers . we refer the reader who is interested in the proof to lemma 2.3 of @xcite . [ lem : popover ] suppose that a tangle @xmath146 $ ] has @xmath147 and , under the projection to @xmath12 $ ] only has maxima . then there is an isotopy @xmath148 fixing @xmath149)$ ] so that @xmath150 has the same number of maxima as @xmath29 and so that there is a strand @xmath20 of @xmath29 that contains the highest maximum of @xmath29 , meets a collar neighborhood @xmath151 of @xmath149)$ ] in two vertical arcs and a horizontal arc and meets the complement of @xmath151 in @xmath91 $ ] in a vertical arc as in figure [ fig : popover ] . let @xmath95 and @xmath84 be the two balls bounded by @xmath3 in @xmath0 . let @xmath152 be a sequence of essential simple closed curves in @xmath3 such that @xmath153 is disjoint from @xmath154 for @xmath155 and such that @xmath156 bounds a compressing disk in @xmath95 and @xmath157 bounds a compressing disk in @xmath84 . the curve @xmath158 separates @xmath3 into two disks . if @xmath156 and @xmath157 are in distinct disk components of @xmath159 then they are disjoint and @xmath160 . thus @xmath156 and @xmath157 must be contained in a single disk @xmath161 with boundary @xmath158 . let @xmath162 be the closure of the complement of @xmath17 in @xmath3 . in general , there may be many different distance - two paths for @xmath3 and we will assume that @xmath156 , @xmath158 and @xmath157 have been chosen amongst all such triples of curves so that @xmath163 is minimal . let @xmath23 be the boundary of a regular neighborhood @xmath164 of @xmath17 . isotope @xmath1 relative to @xmath3 so as to minimize @xmath165 without creating any new minima or maxima , as in figure [ fig : general form ] . let @xmath166 be the closure of the complement in @xmath0 of @xmath164 . note that @xmath164 and @xmath166 are both balls bounded by the sphere @xmath23 such that @xmath164 contains @xmath17 and @xmath167 is a tangle in @xmath164 . moreover , we can naturally identify @xmath164 with @xmath30 $ ] so that @xmath17 is a disk bridge surface for @xmath168 . ( -35,150)@xmath3 ( -30,230)@xmath95 ( -30,70)@xmath84 ( -180,200)@xmath23 ( -130,150)@xmath17 ( -100,150)@xmath162 ( -230,180)@xmath164 ( -280,224)@xmath166 * claim 0 : * @xmath162 is incompressible in @xmath166 . _ proof of claim 0 : _ suppose that @xmath98 is a compressing disk for @xmath162 . since both @xmath156 and @xmath158 are disjoint from @xmath162 then there is a compressing disk for @xmath3 that is disjoint from @xmath98 and on the opposite side of @xmath3 from @xmath98 . hence , @xmath3 is distance at most one , a contradiction . ( claim 0 ) * claim 1 : * @xmath109 is incompressible in @xmath95 and @xmath169 is incompressible in @xmath84 . _ proof of claim 1 : _ assume for contradiction @xmath50 is a compressing disk for @xmath23 in @xmath95 . then @xmath117 also bounds a disk @xmath93 in @xmath170 . the union @xmath171 is a ball @xmath172 and each arc of @xmath1 in @xmath172 contains exactly one maximum since @xmath3 is a bridge sphere for @xmath1 . hence , @xmath173 is a trivial tangle in @xmath172 . moreover , the endpoints of each strand of @xmath173 are in @xmath93 , so we can isotope a strand @xmath59 of @xmath174 into @xmath93 . after pushing @xmath59 just past @xmath93 and into or out of @xmath164 , @xmath23 will meet @xmath1 in two fewer points , contradicting our minimality assumption . the proof that @xmath169 is incompressible in @xmath84 is essentially the same . ( claim 1 ) * claim 2 : * as a properly embedded surface in @xmath164 , @xmath17 does not have a compressing disk on one side that is disjoint from a compressing or a boundary compressing disk on the opposite side . _ proof of claim 2 : _ if there are two disjoint compressing disks on opposite sides of @xmath17 then @xmath3 is a bridge sphere of @xmath1 with distance at most one , contrary to our hypothesis . suppose there is a compressing disk @xmath175 for @xmath17 on one side , say in @xmath176 , that is disjoint from a boundary compressing disk @xmath93 for @xmath17 in @xmath177 , as in figure [ fig : reduction ] . let @xmath178 be the compressing disk that @xmath157 bounds in @xmath84 . by claim 1 , we may assume that this disk is disjoint from @xmath23 and therefore contained in @xmath177 . we can eliminate intersections between @xmath93 and @xmath178 by compressing and boundary compressing @xmath178 along @xmath93 . one of the resulting components will be a compressing disk @xmath141 for @xmath17 in @xmath177 that is disjoint from @xmath93 . if @xmath175 is on the opposite side of @xmath93 from @xmath141 , then @xmath141 and @xmath179 are disjoint so the distance of @xmath3 is at most one , contradicting the assumption . thus , @xmath175 and @xmath141 are on the same side of @xmath93 . ( -212,90)@xmath175 ( -133,40)@xmath93 ( -185,43)@xmath141 ( -132,105)@xmath180 ( -60,110)@xmath181 consider the curve @xmath180 that is contained in the boundary of a regular neighborhood of @xmath182 in @xmath17 and separates @xmath158 from @xmath142 , as in figure [ fig : reduction ] . let @xmath183 be the disk in @xmath3 bounded by @xmath184 that contains @xmath185 and @xmath142 . then @xmath186 , contradicting the minimality assumption on the path @xmath152 . ( claim 2 ) [ minsandmaxes ] note that if @xmath17 has disjoint boundary compressing disks @xmath187 and @xmath188 on opposite sides of @xmath17 , then we know where the critical points of the tangle @xmath189 occur with respect to @xmath187 and @xmath188 . if the tangle has maxima on both sides of @xmath187 then we can find compressing disks associated to each of these maxima that are disjoint from @xmath187 and occur on both sides of @xmath187 by taking the frontier of regular neighborhoods of bridge disks that have been chosen to be disjoint from @xmath187 . since @xmath190 is on one side or the other of @xmath191 in @xmath17 , this implies that at least one of these compressing disks is disjoint from @xmath188 . thus by claim 2 , every maximum of the tangle must be on the same side of @xmath187 as @xmath188 . symmetrically , every minimum of the tangle must be on the same side of @xmath188 as @xmath187 . * claim 3 : * if @xmath23 is compressible in @xmath166 , then @xmath1 is obtained from a link @xmath2 via banding where @xmath2 has a bridge sphere with the same bridge number as @xmath3 , but with distance at most one . _ proof of claim 3 : _ suppose @xmath50 is a compressing disk for @xmath23 contained in @xmath166 that intersects @xmath3 minimally amongst all such disks . if @xmath192 contains a simple closed curve , then an innermost loop of intersection in @xmath50 bounds a compressing disk for @xmath3 that is disjoint from @xmath156 and @xmath157 , so the distance of @xmath3 would be at most one , contradicting our hypothesis . also , by claim 1 , @xmath193 . thus , we can conclude that @xmath192 is a non - empty collection consisting entirely of arcs . ( -343,225)@xmath194 ( -200,225)@xmath195 ( -310,200)@xmath52 ( -227,200)@xmath8 let @xmath93 be any compressing disk for @xmath196 or for @xmath197 that decomposes @xmath198 or @xmath199 into two 3-balls @xmath52 and @xmath8 such that both @xmath200 and @xmath201 have non - trivial intersection with @xmath202 . in the current situation , we could let @xmath93 be a disk cut from @xmath50 by an outermost arc @xmath20 of @xmath192 . however , we will need the full generality of the definition of @xmath93 in claim 5 . choose a collection of bridge disks for the arcs of @xmath1 above @xmath3 which is disjoint from @xmath93 . use these bridge disks to lower all bridges of @xmath1 above @xmath203 below any critical points of @xmath93 . now we can isotope @xmath93 to have a single maximum by an isotopy that does not affect the link . we can then isotope @xmath1 to its original position without introducing any new critical points in @xmath93 . therefore we may assume that after an isotopy @xmath93 has a single maximum in its interior . let @xmath194 and @xmath195 be two copies of @xmath93 with the point at infinity between them , as in the top row of figure [ fig : boundarycompression ] . after thickening @xmath93 and an isotopy , @xmath52 and @xmath8 are the balls indicated in the top left of the figure bounded by @xmath194 and @xmath195 , respectively , and disjoint disks in @xmath3 . then @xmath204 is a tangle that only has maxima so we can apply lemma [ lem : popover ] to pull one strand with at least one endpoint in @xmath3 out of the braid in each of @xmath52 and @xmath8 as indicated in the upper right picture of figure [ fig : boundarycompression ] . our definition of @xmath93 guarantees that such strands exist . these two strands then form a band as indicated in the last picture . the link @xmath2 produced by undoing this band , the inverse of the process depicted in figure [ fig : banding ] , continues to have bridge surface @xmath3 . however , @xmath3 now has an obvious pair of disjoint compressing disks on opposite sides . therefore @xmath2 has a bridge surface with the same bridge number as @xmath1 , but distance at most one . ( note that this bridge surface may be the result of perturbing a bridge surface with strictly lower bridge number . ) ( claim 3 ) both claim 4 and claim 5 derive consequences from the assumption that @xmath23 is compressible in @xmath164 . it is necessary to use the conclusions of these claims in tandem to prove the theorem . * claim 4 : * if @xmath23 is compressible in @xmath164 , then @xmath205 is a bridge - split montesinos tangle or @xmath1 is obtained from a link @xmath2 via banding where @xmath2 has a bridge sphere with the same bridge number as @xmath3 , but with distance at most one . _ proof of claim 4 : _ suppose @xmath50 is a compressing disk for @xmath23 in @xmath164 . if @xmath117 can be isotoped to be disjoint from @xmath206 , then we can apply lemma [ lem : weakly reducible2 ] and we arrive at a contradiction to claim 2 if conclusion ( 1 ) or ( 2 ) holds . if ( 3 ) of lemma [ lem : weakly reducible2 ] holds , then we contradict claim 0 . if ( 4 ) of lemma [ lem : weakly reducible2 ] holds , then we contradict how we chose the curves @xmath152 by finding a smaller bicompressible subdisk of @xmath3 . thus , we can assume that @xmath117 can not be isotoped to be disjoint from @xmath206 . apply lemma [ lem : weakly reducible ] to conclude that there are two cases to consider . if conclusion ( 1 ) of lemma [ lem : weakly reducible ] holds , then we contradict claim 2 . thus , we may assume conclusion ( 2 ) or ( 3 ) holds . hence , @xmath17 has a pair of disjoint boundary compressing disks on opposite sides . let @xmath72 and @xmath73 be these disks such that @xmath72 lies above @xmath17 and @xmath73 lies below @xmath17 . by lemma [ lem : vertical boundary compressing disks ] , we may assume that these disks are vertical and by remark [ minsandmaxes ] , every compressing disk disjoint from @xmath72 must be contained in the component of @xmath207 that contains @xmath73 and vice versa . let @xmath208 be the result of boundary compressing @xmath17 along @xmath209 . by the above argument , some component @xmath210 of @xmath208 must separate all the minima from all the maxima of the tangle . assume @xmath210 is boundary parallel in @xmath164 . this boundary parallelism induces a product structure of the form @xmath211 on at least one of the two components of @xmath212 . since @xmath17 is bicompressible , some strand of @xmath167 contains a maximum and some strand contains a minimum . without loss of generality , the component of @xmath212 that contains maxima , @xmath213 inherits the product structure . let @xmath20 be a strand of @xmath214 that contains a maximum . the product structure of @xmath213 allows for a level preserving isotopy of @xmath1 that fixes @xmath1 away from @xmath20 and results in @xmath20 being pulled out of @xmath164 and into @xmath166 , as in leftmost isotopy of figure [ fig : r_0boundarypara ] . apply lemma [ lem : popover ] to pull one strand with at least one endpoint in @xmath162 out of @xmath215 , as in the middle isotopy in figure [ fig : r_0boundarypara ] . this strand and @xmath20 cobound a band , as illustrated in the right most isotopy in figure [ fig : r_0boundarypara ] . hence , @xmath1 is obtained from a link @xmath2 via banding where @xmath2 has a bridge sphere with the same bridge number as @xmath3 , but with distance at most one . since we arrive at the same conclusion independent of which component of @xmath212 illustrates the boundary parallelism , we can assume that @xmath210 is not boundary parallel in @xmath164 . is boundary parallel , then @xmath1 is banded.,width=480 ] we can think of @xmath210 as a subdisk of @xmath17 together with vertical flaps corresponding to @xmath72 and @xmath73 , as in figure [ fig : extending ] . hence , after a small tilt of the portion of @xmath210 in @xmath17 , the foliation of @xmath210 induced by the restriction of the projection of @xmath216 onto its @xmath86-factor is a collection of parallel arcs . to demonstrate that @xmath205 is a bridge - split montesinos tangle , it only remains to show that @xmath210 is incompressible in @xmath164 . suppose @xmath98 is a compressing disk for @xmath210 . after an isotopy , @xmath98 becomes a compressing disk for @xmath17 with boundary disjoint from @xmath217 and @xmath218 . without loss of generality , assume @xmath98 is above @xmath17 . since @xmath17 is bicompressible in @xmath164 by construction , there is a compressing disk for @xmath17 contained below @xmath17 . by compressing and boundary compressing this compressing disk along subdisks of @xmath73 , we obtain a compressing disk @xmath219 for for @xmath17 contained below @xmath17 , with boundary disjoint from @xmath59 . @xmath220 must be contained in the same component of @xmath221 as @xmath20 , otherwise we contradict claim 2 . since the boundary of @xmath219 and @xmath222 both are contained in a subdisk of @xmath17 that meets @xmath1 in strictly fewer points , we arrive at a contradiction to how we chose @xmath152 . thus , @xmath210 is incompressible . ( -50,120)@xmath210 ( claim 4 ) * claim 5 : * if @xmath23 is compressible in @xmath164 , then at least one of the following holds : 1 . the exterior of @xmath1 contains an incompressible meridional planar surface , 2 . @xmath205 is rational , 3 . @xmath1 is obtained from a link @xmath2 via banding where @xmath2 has a bridge sphere with the same bridge number as @xmath3 , but with distance at most one , 4 . the exterior of @xmath1 in @xmath166 contains an incompressible , boundary incompressible annulus . the proof of the following claim is inspired by the work of rubinstein and thompson @xcite on the classification of distance - two heegaard splittings and the proof of lemma 3.2 in @xcite . _ proof of claim 5 : _ by claim 3 , we can assume that @xmath23 is incompressible in @xmath166 . let @xmath223 be a maximal collection of compressing disks for @xmath23 in @xmath164 . let @xmath151 be the regular neighborhood of @xmath224 . if any of the boundary components of @xmath151 are 2-spheres bounding balls in @xmath164 disjoint from @xmath1 or are boundary parallel twice punctured 2-spheres , then fill these components with the corresponding 3-balls or 3-balls containing unknotted arcs to form the 3-manifold @xmath16 . let @xmath225 . if @xmath226 , then @xmath205 is rational and the claim holds . hence , we can assume that @xmath227 . since @xmath223 was maximal , @xmath228 is incompressible to the side not containing @xmath23 . @xmath16 is a punctured 3-ball that meets @xmath1 in a collection of arcs isotopic into @xmath23 or connecting @xmath23 to @xmath228 . thus , @xmath228 is incompressible in @xmath229 . suppose that @xmath228 is compressible in the exterior of @xmath1 with compressing disk @xmath50 . isotope @xmath50 to be transverse to @xmath230 so that @xmath231 consists of loops and 3-valent graphs in @xmath50 . assume that we have isotoped @xmath50 to meet @xmath230 in a minimal number of edges , where we think of each loop component as consisting of a single vertex and a single edge . since @xmath228 is incompressible in @xmath229 , @xmath232 . a component @xmath233 of @xmath231 in @xmath50 is innermost in @xmath50 if all but one of the boundary components of a closed regular neighborhood of @xmath233 in @xmath50 bound disks in @xmath50 that are disjoint from @xmath230 . let @xmath233 be an innermost component of @xmath231 in @xmath50 . * case 1 : * suppose that @xmath233 is a loop . since @xmath233 is innermost it bounds a disk @xmath141 in @xmath50 which is disjoint from @xmath234 . if @xmath141 is contained in @xmath166 , then @xmath233 is disjoint from @xmath235 since it is a loop and not a 3-valent graph . if @xmath233 is contained in @xmath162 , then we obtain a contradiction to claim 0 . thus , we can assume that @xmath236 . if @xmath141 is contained in @xmath166 , then @xmath141 is a compressing disk for @xmath109 in @xmath95 or @xmath169 in @xmath84 , a contradiction to claim 1 . hence , @xmath141 is contained in @xmath164 . given the natural identification of @xmath164 with @xmath237 and the fact that @xmath233 is disjoint from @xmath206 , then we can assume without loss of generality that @xmath141 has boundary in @xmath238 . apply lemma [ lem : weakly reducible2 ] to @xmath141 and analyze each possible conclusion . if conclusions ( 1 ) or ( 2 ) of lemma [ lem : weakly reducible2 ] hold , then we contradict the fact that @xmath3 is distance two . if conclusion ( 3 ) of lemma [ lem : weakly reducible2 ] holds , then we contradict claim 0 . if conclusion ( 4 ) of [ lem : weakly reducible2 ] holds , we contradict how we chose @xmath152 . thus , @xmath233 can not be a loop . * case 2 : * suppose that @xmath233 is a 3-valent graph . since @xmath233 is embedded in the interior of the disk @xmath50 , the formula for euler characteristic gives @xmath239 where @xmath240 is the number of vertices of @xmath233 , @xmath241 is the number of edges of @xmath233 and @xmath242 counts the number of components of the complement of @xmath233 in @xmath50 . since @xmath233 is 3-valent , @xmath243 . thus , @xmath244 . from this equality follows the inequality @xmath245 since every edge of @xmath233 borders exactly two faces and since the closure of each component of @xmath242 has boundary a cycle in @xmath233 consisting of an even number of edges , then the above inequality implies the existence of a disk component of @xmath242 with closure @xmath141 such that @xmath142 consists of exactly two edges and two vertices of @xmath233 ( @xmath141 is a bigon ) , or @xmath142 consists of exactly four edges and four vertices of @xmath233 ( @xmath141 is a square ) . suppose @xmath141 is a bigon properly embedded in @xmath164 . since @xmath141 is a bigon , @xmath141 meets @xmath17 in exactly one arc @xmath20 that is essential in @xmath17 . identify @xmath164 with @xmath246 and isotope @xmath142 to consist of two vertical arcs in @xmath247 and two horizontal arcs in @xmath248 . hence @xmath141 meets every level surface @xmath249 in exactly one arc . apply lemma [ lem : weakly reducible ] and the proof of claim 4 to the compressing disk @xmath141 to conclude that @xmath17 has a pair of disjoint boundary compressing disks on opposite sides , @xmath72 and @xmath73 such that the arc @xmath250 is isotopic to the arc @xmath250 . since @xmath17 is bicompressible in @xmath164 , we can boundary compress along @xmath72 to find a compressing disk for @xmath17 above @xmath17 which is disjoint from @xmath72 . similarly , we can find a compressing disk for @xmath17 below @xmath17 which is disjoint from @xmath73 . if these compressing disks have boundary on the same side of @xmath250 in @xmath17 , then we contradict how we chose the curves @xmath152 . if these compressing disks have boundary on opposite sides of @xmath250 in @xmath17 , then @xmath3 has distance at most one , a contradiction . hence , @xmath141 is not a bigon properly embedded in @xmath164 . suppose @xmath141 is a bigon properly embedded in @xmath251 or @xmath252 . then @xmath141 satisfies the definition of @xmath93 in the proof of claim 3 and we conclude that @xmath1 is obtained from a link @xmath2 via banding where @xmath2 has a bridge sphere with the same bridge number as @xmath3 , but with distance at most one . suppose @xmath141 is a square properly embedded in @xmath164 . identify @xmath164 with @xmath246 and , since @xmath141 is a square , we can isotope @xmath142 to consist of four vertical arcs in @xmath253 , two horizontal arcs in @xmath238 and two horizontal arcs in @xmath85 . recall that @xmath28 is the natural height function on @xmath164 . at any regular value @xmath119 of @xmath254 , @xmath255 consists of exactly two arcs and some number of loops . since @xmath141 is connected , there must be some saddle singularity of @xmath254 at which two arcs are banded together . since @xmath141 is simply connected , there is at most one saddle where two arcs are banded together to form two new arcs . let @xmath133 be the height of this singularity , then the graph @xmath256 consists of a single `` x '' and possibly a collection of loops . if any of the loops of intersection of @xmath256 are inessential in @xmath140 , then we can eliminate them via an isotopy of @xmath141 that leaves the saddle singularity fixed . as in the proof of claim 4 , we can apply lemma [ lem : weakly reducible ] to find an isotopy of @xmath141 after which conclusion ( 2 ) or ( 3 ) of lemma [ lem : weakly reducible ] holds . let @xmath61 be the value for which @xmath257 has a pair of disjoint boundary compressing disks on opposite sides , @xmath72 and @xmath73 . suppose conclusion ( 2 ) of lemma [ lem : weakly reducible ] holds . if @xmath258 , then the arc @xmath259 separates @xmath141 into two disks . let @xmath260 be the possibly immersed disk that is the union of @xmath73 and the portion of @xmath141 that is cut off by @xmath59 and lies above @xmath261 when near @xmath59 , see figure [ fig : dprime ] . since @xmath258 , @xmath262 consists of two vertical arcs in @xmath253 , one horizontal arc in @xmath238 and one horizontal arc in @xmath85 . since the self - intersections of @xmath260 correspond to the transverse intersections of embedded disks , then after surgering @xmath260 along any loops of self intersection , we can assume @xmath260 is an embedded compressing disk . in this case @xmath260 is a bigon in @xmath164 and we can apply the above argument to derive a contradiction . similarly , we derive a contradiction if @xmath263 . hence @xmath264 . however , @xmath264 implies conclusion ( 3 ) rather that conclusion ( 2 ) of lemma [ lem : weakly reducible ] . , title="fig : " ] ( -108,110)@xmath265 ( -104,84)@xmath59 ( -108,60)@xmath73 suppose conclusion ( 3 ) of lemma [ lem : weakly reducible ] holds . if @xmath258 let @xmath266 for @xmath267 small . we arrive at a contradiction when we consider the immersed disk @xmath260 that is the union of @xmath73 below @xmath268 and the portion of @xmath141 that is cut off by @xmath59 and lies above @xmath268 when near @xmath59 exactly as before . similarly , we derive a contradiction if @xmath263 . if @xmath264 , then there are no loops in @xmath269 as we have already isotoped @xmath141 to eliminate loops that are inessential in @xmath140 and if any such loops are essential in @xmath140 we derive a contradiction as in claim 4 . hence , @xmath256 consists of a single `` x. '' identify @xmath17 with @xmath140 and let @xmath270 be the four properly embedded arcs in @xmath17 that lie in the boundary of a regular neighborhood of @xmath271 in @xmath17 . by the above argument , each of these arcs bound a boundary compressing disk for @xmath17 . in particular , @xmath272 and @xmath273 bound boundary compressing disks above @xmath17 and @xmath274 and @xmath275 bound boundary compressing disks below @xmath17 . by remark [ minsandmaxes ] , all maxima of @xmath167 must be contained in the 3-ball between the upper boundary compressing disks in @xmath276 and all minima of @xmath167 must be contained in the 3-ball between the lower boundary compressing disks in @xmath277 . hence @xmath141 decomposes @xmath278 into two rational tangles glued together along an unpunctured disk . see figure [ fig : monkeyclasp2 ] . thus , @xmath278 is a rational tangle . suppose @xmath141 is a square properly embedded in @xmath251 or @xmath252 . we consider the case that @xmath141 is a square properly embedded in @xmath251 and the case when @xmath141 is a square properly embedded in @xmath252 will follow by a symmetric argument . recall that @xmath162 is the closure of the complement of @xmath17 in @xmath3 . since @xmath141 is a square , it meets @xmath162 in exactly two essential arcs . if these arcs are not isotopic in @xmath162 then @xmath141 meets the definition of @xmath93 in the proof of claim 3 and we conclude that @xmath1 is obtained from a link @xmath2 via banding where @xmath2 has a bridge sphere with the same bridge number as @xmath3 , but with distance at most one . hence , we can assume that the two arcs in @xmath279 together with two arcs in @xmath235 bound a disk @xmath280 in @xmath162 which is disjoint from @xmath1 . after a small isotopy pushing @xmath280 into the interior of @xmath95 , @xmath281 is an annulus properly embedded in @xmath251 with boundary in @xmath109 . if @xmath139 is compressible in the exterior of @xmath1 in @xmath166 , then , after compressing @xmath139 , each component of @xmath282 bounds a compressing disk for @xmath109 in @xmath166 , a contradiction to the incompressibility of @xmath23 in @xmath166 . hence , @xmath139 is incompressible . @xmath139 is not boundary parallel since @xmath282 partitions @xmath23 into three subsurfaces , each of which has non - trivial intersection with @xmath1 . lastly , @xmath139 is boundary incompressible since boundary compressing @xmath139 would result in a compressing disk for @xmath23 in @xmath166 which was ruled out as a possibility at the beginning of the proof of this claim . ( claim 5 ) to conclude , we note that if @xmath23 is incompressible , then conclusion ( 1 ) holds ; if @xmath23 is compressible into @xmath166 , then , by claim 4 , conclusion ( 3 ) holds ; and if @xmath23 is compressible into @xmath164 but incompressible into @xmath166 , then , by claim 3 and claim 5 , at least one of conclusions ( 1 ) , ( 2 ) or ( 3 ) holds . let @xmath1 be a 3-bridge knot with a 6-punctured bridge sphere @xmath3 of distance two . let @xmath23 be the boundary of a regular neighborhood @xmath164 of @xmath17 as defined at the beginning of the proof of theorem [ thm : main ] . recall that after isotopying @xmath1 relative to @xmath3 so as to minimize @xmath165 , @xmath165 is at most @xmath283 since @xmath17 is compressible in @xmath95 and in @xmath84 . since @xmath206 is essential in @xmath3 , @xmath284 2 , 3 or 4 . if @xmath285 , then @xmath286 and @xmath3 is distance zero , a contradiction . if @xmath287 , then @xmath288 and @xmath289 . let @xmath20 be the strand of @xmath290 that is disjoint from @xmath164 and let @xmath98 be a bridge disk for @xmath20 . by an outermost arc , innermost loop argument , we can assume that @xmath98 is disjoint from @xmath23 . the boundary of a regular neighborhood of @xmath98 in @xmath95 is a compressing disk for @xmath162 , this contradicts the incompressibly of @xmath162 . if @xmath291 , then @xmath292 0 , 2 or 4 . in the case that @xmath292 0 or 2 , one of @xmath170 or @xmath293 is disjoint from @xmath1 and @xmath162 is compressible , a contradiction . hence , @xmath291 and @xmath294 . if @xmath23 is incompressible , then conclusion ( 1 ) of the corollary holds . if @xmath23 is compressible into @xmath164 , then @xmath278 is a rational tangle or at least one strand of @xmath278 is knotted . if @xmath278 is a rational tangle , @xmath1 can be decomposed as the union of three rational tangles @xmath278 , @xmath295 and @xmath296 and conclusion ( 3 ) of the corollary holds . if at least one strand of @xmath278 is knotted and @xmath23 is compressible into @xmath164 , then @xmath1 is composite or @xmath1 is 2-bridge , a contradiction . in the case that @xmath1 is composite , then the distance of @xmath3 is at most one by theorem 1.6 of @xcite , a contradiction . the careful reader will notice that theorem 1.6 of @xcite only implies that @xmath3 has a compressing disk on one side that is disjoint from a properly embedded disk on the opposite side such that this disk has essential boundary and meets @xmath1 in at most one point . however , it is an easy exercise to show that any bridge sphere with this property has distance at most one . if @xmath23 is compressible in @xmath166 , then , by claim 3 of the proof of theorem [ thm : main ] , @xmath1 is obtained from a link @xmath2 via banding where @xmath2 has a bridge sphere with the same bridge number as @xmath3 , but with distance at most one . hence , @xmath2 is the unknot , a 2-bridge knot , or the connected sum of two 2-bridge knots and conclusion ( 2 ) of the corollary holds . theorem [ thm : main ] raises a number of interesting questions . [ [ decompositions - of - bridge - surfaces - into - surfaces - of - higher - distance ] ] decompositions of bridge surfaces into surfaces of higher distance ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ as motivation , recall that casson and gordon @xcite showed that the existence of a distance 1 heegaard surface @xmath3 for a closed 3-manifold implies the existence of a closed incompressible surface in the 3-manifold . scharlemann and thompson @xcite elaborated on the casson and gordon construction to show that , in fact , @xmath3 can be decomposed into a collection of incompressible surfaces and heegaard surfaces ( for the complement of the incompressible surfaces ) each of distance at least 2 . similarly , thompson @xcite showed ( roughly speaking ) that if a knot @xmath297 has a bridge sphere @xmath3 which is distance 1 then there is an incompressible meridional surface in the exterior of the knot . hayashi and shimokawa @xcite developed that idea and showed that , in fact , a distance @xmath298 bridge surface can be decomposed into essential meridional surfaces and bridge surfaces , each of distance at least @xmath299 , for the complementary tangles . [ q : dist 2 decomp ] if @xmath300 is a bridge sphere of distance 2 , does @xmath301 have a tangle decomposition such that @xmath3 can be decomposed into bridge spheres , each of distance at least 3 , for the tangles in the decomposition ? the proof of theorem [ thm : main ] suggests that the sphere @xmath23 is a possible starting point for such a tangle decomposition . however , in the cases when @xmath23 is compressible , the bridge sphere induced by @xmath3 on the side of @xmath23 to which @xmath23 compresses will likely have distance 0 . thus , much needs to be done to obtain a tangle decomposition answering question [ q : dist 2 decomp ] . if @xmath3 is a bridge surface ( of any genus ) for a properly embedded 1manifold @xmath51 in a closed 3-manifold @xmath16 of distance @xmath302 , is there a decomposition of @xmath303 into ( 3-manifold , 1-manifold ) submanifolds and a decomposition of @xmath3 into bridge surfaces , each of distance at least @xmath304 , for the submanifolds ? for a knot @xmath297 we can define @xmath305 to be the minimal distance of a minimal bridge sphere for @xmath51 . letting @xmath306 denote the bridge number of @xmath51 , the pair @xmath307 is an invariant of @xmath51 . both conclusion ( 3 ) of theorem [ thm : main ] and corollary [ coro : main ] suggest the following : theorem [ thm : main ] was inspired , in part , by our studies @xcite of the relationship between distance and exceptional surgeries on knots . we showed , in particular , that if a knot @xmath297 of bridge number at least 3 has a reducing surgery or a toroidal surgery then @xmath311 . we , therefore , wonder :

spectral properties and the confinement phenomenon for the coupling are studied , where is the free dirac operator in and is a measure - valued potential . the potentials under consideration are given in terms of surface measures on the boundary of bounded regular domains in . a criterion for the existence of point spectrum is given , with applications to electrostatic shell potentials . in the case of the sphere , an uncertainty principle is developed and its relation with some eigenvectors of the coupling is shown . furthermore , a criterion for generating confinement is given . as an application , some known results about confinement on the sphere for electrostatic plus lorentz scalar shell potentials are generalized to regular surfaces .

1401.1297

in this article we study the coupling of the free dirac operator with measure - valued potentials developed in @xcite . the aim of the present work is to give spectral properties of these couplings and to investigate the phenomenon of confinement . given @xmath4 , the free dirac operator in @xmath2 is defined by @xmath5 where @xmath6 , @xmath7 is the family of _ pauli matrices_. although one can take @xmath8 in the definition of @xmath9 , throughout this article we always assume @xmath10 . the work @xcite contains some results concerning @xmath0 for quite general singular measures @xmath11 in @xmath2 and suitable @xmath12-valued potentials @xmath3 . however , in that article , most of the interest is focused on the case that @xmath11 is the surface measure on the boundary of a bounded regular domain in @xmath2 , and this is our setting for the present paper . in order to make an understandable exposition of the main results , let us first recall the basic ideas used in @xcite . the ambient hilbert space is @xmath13 with respect to the lebesgue measure , and @xmath9 is defined in the sense of distributions . given a bounded regular domain @xmath14 with boundary @xmath15 and surface measure @xmath11 , in @xcite we find domains @xmath16 in which @xmath17 is an unbounded self - adjoint operator , where @xmath3 is a suitable @xmath12-valued potential . the construction of @xmath18 relies on the following simple argument : by assumption , @xmath3 is @xmath12-valued . thus , given @xmath19 , we can write @xmath20 in the sense of distributions for some @xmath21 . moreover , since @xmath22 , we can also write @xmath23 for some @xmath24 . therefore , @xmath25 in the sense of distributions , and so @xmath26 should be the convolution @xmath27 , where @xmath28 is a fundamental solution of @xmath9 . in particular , @xmath29 to guarantee that @xmath0 is self - adjoint on @xmath30 , in @xcite we impose some relations between @xmath31 and @xmath32 with the aid of bounded self - adjoint operators @xmath33 . that is to say , given suitable @xmath34 s , following ( [ intro 1 ] ) we find domains @xmath35 where @xmath0 is self - adjoint . we consider the potential @xmath3 given by ( [ intro 1 ] ) as a `` generic '' potential since it seems to be prescribed from the begining , i.e. , @xmath20 for all @xmath36 , so @xmath3 is independent of @xmath34 . for an a priori given potential @xmath37 , the key idea in @xcite to construct a domain where @xmath38 is self - adjoint is to find a precise bounded self - adjoint operator @xmath39 so that @xmath40 for all @xmath41 , and the self - adjointness of @xmath38 on @xmath42 follows directly from the one of @xmath0 . the generic potential @xmath3 given by ( [ intro 1 ] ) reflects the following rough idea : if we know that the gradient of a function @xmath26 has an absolutely continuous part @xmath31 and a singular part @xmath32 supported on @xmath15 ( in our setting , @xmath43 and @xmath22 ) , then @xmath26 must have a jump across @xmath15 , and this jump completely determines the singular part of the gradient ( in our setting , the jump determines the value @xmath44 ) . for a given potential @xmath37 , one manages to define a suitable domain @xmath18 such that , for any @xmath19 , the singular part which comes from the gradient on the jump of @xmath26 across @xmath15 agrees with @xmath45 . as outlined above , in @xcite we show that any @xmath46 has non - tangential boundary values @xmath47 when we approach to @xmath15 from @xmath48 , where @xmath49 . this enables us to consider , for example , the electrostatic shell potential @xmath50 for @xmath51 . following the argument above , we construct a domain @xmath52 where @xmath53 is self - adjoint for all @xmath54 . other similar potentials are also treated in @xcite . at this point a remark is in order . in @xcite ( see also ( * ? ? ? * section 2 ) ) the author provides , in a very general framework , all self - adjoint extensions of symmetric operators obtained by restricting a self - adjoint operator @xmath55 with domain @xmath56 to a dense ( and closed with respect to the graph norm ) subspace @xmath57 . typically , @xmath55 is a differential operator and @xmath58 is the kernel of some trace operator along a null set . in particular , @xcite can be used to provide all the domains where @xmath0 is self - adjoint and , in this sense , some of the results in @xcite follow from the ones in @xcite . however , as regards applications , we make use of some explicit layer potentials on @xmath15 derived from our approach in @xcite which turn out to be very fruitful both in the description of the domains and in the study of the properties of the couplings under consideration . concerning the results in this article , in section [ s spec ] we show a general criterion for the existence of eigenvalues in @xmath59 for @xmath0 , namely proposition [ spec p1 ] , which has strong connections with ( * ? ? ? * theorem 2.5 ) . this criterion relates the existence of eigenvalues , which is a problem in the whole @xmath2 , with a spectral property of certain bounded operators in @xmath12 , which is a problem settled exclusively on @xmath15 . afterwards , we show some applications to the case of electrostatic shell potentials @xmath60 . in particular , theorem [ spec t1 ] shows that @xmath53 and @xmath61 have the same eigenvalues in @xmath59 for all @xmath62 , which is a special case of the more general isospectral transformation , and that @xmath53 has no eigenvalues in @xmath59 if @xmath63 is too big or too small . this is an interesting feature , since it shows that there are lower and upper thresholds on the possible values of @xmath63 in order to have non - trivial eigenvalues in @xmath59 for @xmath53 , unlike what happens to the coupling of @xmath9 with similar potentials ( see remark [ threshold ] ) . theorem [ symm a -a ] is another consequence of the general criterion , where we show that , under a symmetry assumption on @xmath11 , @xmath53 has an eigenvalue @xmath64 if and only if @xmath65 has @xmath66 as an eigenvalue . for completeness , we also show in theorem [ spec t2 ] that if @xmath67 is connected then @xmath53 has no eigenvalues in @xmath68 $ ] . section [ s sphere ] is devoted to the spectral study of @xmath53 when @xmath15 is the sphere @xmath69 our interest is to characterize the eigenvalues finding sharp constants of some appropriately chosen inequalities . in principle , this is far from obvious for dirac hamiltonians because they are not semibounded operators . however , this procedure has been successfully used in @xcite with hamiltonians with coulombic singularities . in that case , the sharp constants appear as strictly positive lower bounds for the absolute value of the commutator of two operators , being @xmath9 one of them . therefore , this approach can be seen as another use of the uncertainty principle . in the current article we proceed in the same vein and we give an uncertainty principle that concerns some @xmath70 bounded operators also related to @xmath9 . in theorem [ thmineq ] we obtain a sharp inequality on @xmath71 which turns out to be connected to the eigenvalues of @xmath53 . for its proof , we strongly use the classical spherical harmonics , so a generalization to other surfaces @xmath15 seems an interesting and challenging question . as a consequence of the above - mentioned inequality , we recover a well - known sharp lower bound for the @xmath72-dimensional riesz transform on the sphere ( see corollary [ riesz ] and @xcite ) . in lemma [ eigenfunction ] we provide a specific criterion ( based on proposition [ spec p1 ] ) to generate eigenvectors of @xmath53 . section [ further com ] contains some comments on the relation between the uncertainty principle and the eigenvectors of @xmath53 , positive results on the existence of eigenvalues , and an open question ( as far as we know ) and some consequences of an affirmative answer . finally , in section [ s confinement ] we deal with the confinement phenomenon on regular surfaces . roughly speaking , one says that an @xmath12-valued potential @xmath3 generates confinement with respect to @xmath9 and @xmath15 if the particles modelized by the evolution operator associated to @xmath0 never cross @xmath15 over time . that is , if a function @xmath73 verifies the equation @xmath74 and @xmath75 is supported in @xmath48 , that @xmath3 generates confinement means that @xmath76 is supported in @xmath48 for all @xmath77 , so @xmath15 becomes impenetrable for the particles . similarly to section [ s spec ] , in section [ s confinement ] we first show a general criterion on @xmath0 to generate confinement , namely theorem [ conf t1 ] . this criterion is stated in terms of an algebraic property of certain bounded operators in @xmath12 , so as before we convert a problem in the whole @xmath2 to a problem exclusively settled in @xmath15 . an application to electrostatic and lorentz scalar shell potentials is also shown . in particular , in theorem [ conf appl ] we prove that , for @xmath78 , the potential @xmath79 generates confinement if and only if @xmath80 . this was already known for the case of the sphere ( see @xcite ) , and we generalize it to sufficiently regular surfaces . for the reader only interested on confinement , we mention that section [ s confinement ] can be read independently of sections [ s spec ] and [ s sphere ] . finally , we also want to recall @xcite , where the authors studied the confinement phenomenon for singular perturbations of general self - adjoint hamiltonian operators . it is worth mentioning that , although all the applications in this article are concerned to the potentials @xmath60 and @xmath81 above - mentioned , the general criteria stated in proposition [ spec p1 ] and theorem [ conf t1 ] can be used as a first step to study the spectrum and confinement for the coupling of @xmath9 with other shell potentials . in a sense , once a potential @xmath37 is given , one must find the suitable operator @xmath34 ( mentioned in the beginning of the introduction ) so that @xmath40 for all @xmath82 , and then one must check the criteria for that specific @xmath34 . since this article is a continuation of the study developed in @xcite , we assume that the reader is familiar with the notation , methods and results in there . however , in this section we recall some basic rudiments for the sake of completeness . given a positive borel measure @xmath83 in @xmath2 , set @xmath84 and denote by @xmath85 and @xmath86 the standard scalar product and norm in @xmath87 , i.e. , @xmath88 and @xmath89 for @xmath90 . we write @xmath91 or @xmath92 interchangeably to denote the identity operator on @xmath87 . we say that @xmath83 is @xmath93-dimensional if there exists @xmath94 such that @xmath95 for all @xmath96 , @xmath97 . we denote by @xmath98 the lebesgue measure in @xmath2 . let @xmath15 be the boundary of a bounded lipschitz domain @xmath14 , let @xmath11 and @xmath58 be the surface measure and outward unit normal vector field on @xmath15 respectively , and set @xmath49 , so @xmath99 . note that @xmath11 is 2-dimensional . since we are not interested in optimal regularity assumptions , for the sequel we assume that @xmath15 is of class @xmath100 ( see remark [ r regularity ] for the lipschitz case ) . finally , we introduce the auxiliary space of measures @xmath101 observe that @xmath9 , which is symmetric and initially defined in @xmath102 ( @xmath103-valued functions in @xmath2 which are @xmath104 and with compact support ) , can be extended by duality to the space of distributions with respect to the test space @xmath102 and , in particular , it can be defined on @xmath105 . the following lemma is concerned with the resolvent of @xmath9 , which will be very useful for the results below . as usual , we denote by @xmath106 the complex conjugate of the transpose of @xmath107 , that is , @xmath108 and @xmath109 for all @xmath110 . [ resolvent ] given @xmath64 , a fundamental solution of @xmath111 is given by @xmath112 i.e. , @xmath113 in the sense of distributions , where @xmath114 denotes the dirac measure on @xmath115 . furthermore , @xmath107 satisfies @xmath116 , @xmath117 , and @xmath118 of ( * ? ? ? * section 2.2 ) , that is , * @xmath119 for all @xmath110 , * @xmath120 for all @xmath121 such that @xmath122 , * there exist @xmath123 such that * * @xmath124 for all @xmath125 , * * @xmath126 for all @xmath127 , * * @xmath128 where @xmath129 denotes the fourier transform in @xmath2 . it is well known that @xmath130 is a fundamental solution of the helmhotz operator @xmath131 in @xmath2 . note that @xmath132 , so @xmath133 and if we set @xmath134 then @xmath135 in the sense of distributions . the explicit formula for @xmath107 follows by a straightforward computation . the rest of the proof is analogous to ( * ? ? ? * lemma 3.1 ) . given a positive borel measure @xmath83 in @xmath2 , @xmath136 , and @xmath96 , we set @xmath137 whenever the integral makes sense . actually , by lemma [ resolvent ] and ( * ? ? ? * lemma 2.1 ) , if @xmath83 is a @xmath93-dimensional measure in @xmath2 with @xmath138 , then there exists some constant @xmath94 such that @xmath139 for all @xmath140 . in what follows we use a non standard notation , @xmath141 , to define the convolution of measures in @xmath105 with the fundamental solution of @xmath111 , @xmath107 . capital letters , as @xmath142 or @xmath31 , in the argument of @xmath141 denote elements of @xmath143 , and the lowercase letters , as @xmath144 or @xmath32 , denote elements in @xmath12 . despite that this notation is non standard , it is very convenient in order to shorten the forthcoming computations . given @xmath145 , we define @xmath146 then , the above inequality shows that @xmath147 for some constant @xmath94 and all @xmath145 , so @xmath148 . moreover , following ( * ? ? ? * section 2.3 ) one can show that @xmath149 in the sense of distributions . this allows us to define a `` generic '' potential @xmath3 acting on functions @xmath150 by @xmath151 so that @xmath152 in the sense of distributions . for simplicity of notation , we write @xmath153 . in order to construct a domain of definition where @xmath0 is self - adjoint , in @xcite we had to consider the trace operator on @xmath15 . for @xmath154 , one defines the trace operator on @xmath15 by @xmath155 . then , @xmath156 extends to a bounded linear operator @xmath157 ( see ( * ? ? ? * proposition 2.6 ) , for example ) , where @xmath158 denotes the sobolev space of @xmath103-valued functions such that all its components have all their derivatives up to first order in @xmath159 . from lemma [ resolvent ] , we have @xmath160 for some @xmath94 and all @xmath161 ( see ( * ? ? ? * lemma 2.8 ) ) , thus we can define @xmath162 and it satisfies @xmath163 for all @xmath161 . note that , for @xmath164 , the above definitions recover the ones in ( * ? ? ? * section 2.3 ) . the next lemma , which is a generalization of ( * ? ? ? * lemma 3.3 ) , will be used in the sequel . [ l jump ] given @xmath165 , @xmath166 and @xmath167 , set @xmath168 where @xmath169 means that @xmath170 tends to @xmath166 non - tangentially . then @xmath171 and @xmath172 are bounded linear operators in @xmath12 . moreover , the following holds : * @xmath173 ( plemelj sokhotski jump formulae ) , * @xmath174 for @xmath175 $ ] . the proof of the lemma is completely analogous to the one of ( * ? ? ? * lemma 3.3 ) , so we omit it . concerning the second term in @xmath117 , once we know that @xmath176 , then , multiplying the equation by @xmath177 from the left and from the right and using that @xmath178 , we obtain @xmath179 . in accordance with the notation introduced in @xcite , for the case @xmath164 , we write @xmath180 , @xmath181 , @xmath182 and @xmath183 instead of @xmath184 , @xmath185 , @xmath186 and @xmath187 , respectively . finally , we recall our main tool to construct domanis where @xmath0 is self - adjoint , namely ( * ? ? ? * theorem 2.11 ) . actually , the following theorem is a direct application of ( * ? ? ? * theorem 2.11 ) to @xmath0 , and we state it here in order to make the exposition more self - contained . given an operator between vector spaces @xmath188 , denote @xmath189 [ pre t1 ] let @xmath33 be a bounded operator . set @xmath190 where @xmath191 and @xmath23 for all @xmath192 . if @xmath34 is self - adjoint and @xmath193 is closed , then @xmath194 is an essentially self - adjoint operator . in that case , if @xmath195 is closed in @xmath143 , then @xmath196 is self - adjoint . in particular , if @xmath34 is self - adjoint and semi - fredholm ( see for example ( * ? ? ? * definition 1.40 ) ) , then the operator @xmath196 in theorem [ pre t1 ] is self - adjoint . recall also that any bounded , semi - fredholm and self - ajoint operator on a hilbert space is actually fredholm . [ r regularity ] all the results in this article which are proved without the use of fredholm s theorem are valid when @xmath15 is just lipschitz ( but not necessarily of class @xmath100 ) , or when it is the graph of a lipschitz function from @xmath197 to @xmath198 . actually , the smoothness and boundedness assumptions on @xmath15 are exclusively required for compactness purposes , in order to use fredholm s theory . in this section , we show a criterion for the existence of eigenvalues in @xmath59 for @xmath0 , namely proposition [ spec p1 ] . this criterion relates the eigenvalues with a spectral property of certain bounded operators in @xmath12 . afterwards , we show some applications to the case of electrostatic shell potentials . [ spec p1 ] let @xmath196 be as in theorem [ pre t1 ] . given @xmath64 , there exists @xmath192 such that @xmath199 if and only if @xmath200 and @xmath201 . therefore , @xmath202 if and only if @xmath203 . let @xmath64 and assume that @xmath199 for some @xmath192 . then , @xmath204 and therefore @xmath205 in the sense of distributions . this yields @xmath206 , and applying @xmath141 we conclude that @xmath201 . in particular , we have seen that @xmath207 , and thus lemma [ l jump]@xmath116 yields @xmath208 summing these equations , we obtain @xmath209 , which is equivalent to @xmath200 when @xmath210 . for the case @xmath164 , from ( [ spec p1 eq1 ] ) we have that @xmath211 and , since @xmath212 , @xmath213 . on the contrary , assume that there exists @xmath165 such that @xmath200 . set @xmath201 and @xmath214 . as we did in ( [ spec p1 eq1 ] ) , @xmath201 implies @xmath206 , so @xmath215 and @xmath216 in particular , @xmath217 , thus lemma [ l jump]@xmath116 once again gives @xmath218 for @xmath210 , this and the assumption on @xmath32 imply @xmath219 , thus @xmath212 . as before , the case @xmath164 can be easily treated separately . the proposition is proved . in ( * ? ? ? * theorem 3.8 ) we proved that , if @xmath220 and @xmath196 is the operator defined by @xmath221 where @xmath222 and @xmath223 for @xmath192 , then @xmath224 is self - adjoint . moreover , we also showed that @xmath225 on @xmath226 for all @xmath62 , so the self - adjointness was a consequence of theorem [ pre t1 ] . [ spec l1 ] set @xmath227 then , we have @xmath228 we write @xmath229 note that @xmath230 and by the mean value theorem we have the same estimate for @xmath231 . using ( [ spec l1 eq1 ] ) , that @xmath11 is 2-dimensional and rather standard arguments ( essentially , that @xmath15 is bounded and the generalized young s inequality ) , it is easy to show that the convolution operator with kernel @xmath232 is bounded in @xmath12 uniformly on @xmath64 . finally , the @xmath12 boundedness of the singular intergal operator with convolution kernel @xmath233 follows , for example , by ( * ? ? ? * theorem 20.15 ) , working component by component . note that this last operator does not depend on @xmath234 , so the lemma is proved . [ spec t1 ] let @xmath235 , let @xmath236 be as in @xmath237 , and @xmath64 . then @xmath202 if and only if @xmath238 . in particular , @xmath53 and @xmath61 have the same eigenvalues in @xmath59 . moreover , if @xmath239 $ ] then @xmath234 is not an eigenvalue of @xmath53 , and if @xmath240 $ ] then @xmath53 has no eigenvalues in @xmath59 , where @xmath241 . that @xmath202 if and only if @xmath238 is a direct consequence of the definition of @xmath34 and proposition [ spec p1 ] . assume that @xmath238 . then there exists a non - trivial @xmath165 such that @xmath242 . using lemma [ l jump]@xmath117 we deduce that @xmath243 . this easily implies that @xmath244 for @xmath245 , so @xmath246 . the same arguments actually show that @xmath238 if and only if @xmath246 , and by the first part of the theorem , @xmath53 and @xmath61 have the same eigenvalues in @xmath59 . for the last part of the theorem , since @xmath178 , we easily have @xmath247 for all @xmath165 . combining this estimate with lemma [ l jump]@xmath117 , we obtain @xmath248 hence , @xmath249 if @xmath250 . by the first part of the theorem , if there exists @xmath212 such that @xmath251 for some @xmath64 , then there exists a non - trivial @xmath165 such that @xmath242 . thus , ( [ spec t1 eq1 ] ) easily implies @xmath252\subset[1/c,4c]$ ] , and the theorem follows . [ threshold ] theorem [ spec t1 ] shows that the coupling of the free dirac operator @xmath9 with electrostatic shell potentials @xmath60 relative to @xmath15 does not generate eigenvalues either for big or small values of @xmath63 . recall that the coupling of @xmath9 with the coulomb potential @xmath253 generates eigenvalues for any small @xmath63 ( see ( * ? ? ? * theorem 1 ) , for example ) and is not essentially self - adjoint if @xmath63 is big enough . on the other hand , it is not hard to see that there exists a sequence @xmath254 with @xmath255 for @xmath256 such that the coupling of @xmath9 with the potentials @xmath257 generates eigenvalues . [ lambda - a ] if we define @xmath258 , by lemma [ l jump]@xmath117 we have @xmath259 where @xmath260 and @xmath261 . following the arguments of ( * ? ? ? * lemma 3.5 ) , one can show that @xmath262 is a compact operator , as well as @xmath263 . moreover , @xmath263 is easily seen to be self - adjoint , and hence it has a non - trivial eigenfunction . therefore , for any @xmath64 there exists some @xmath264 such that @xmath238 by ( [ spec t1 eq2 ] ) , so the second part of theorem [ spec t1 ] is meaningful . note that ( [ spec t1 eq1 ] ) yields @xmath265 for all @xmath64 . in particular , this lower bound of @xmath266 does not depend on @xmath15 . for an upper bound , this type of result may not be expected because , roughly speaking , the abruptness of @xmath15 may play a role in questions concerning upper bounds for the norm of singular integral operators on lipschitz surfaces ( see for example ( * ? ? ? * chapter 20 ) for related topics ) . the following theorem generalizes some results of ( * ? ? ? * theorem 3.8@xmath117 ) . [ symm a -a ] assume that @xmath267 , where @xmath268 for @xmath96 and @xmath269 is the image measure of @xmath11 with respect to @xmath270 . let @xmath235 and @xmath53 be as in @xmath237 . if @xmath53 has some eigenvalue @xmath64 , then @xmath65 has @xmath66 as an eigenvalue . from theorem [ spec t1 ] we see that , if @xmath53 has @xmath234 as an eigenvalue , there exists a nontrivial @xmath165 such that @xmath271 . set @xmath272 , where @xmath273 and @xmath274 denotes the identity operator in @xmath70 . obviously , @xmath275 and , since @xmath267 , we have @xmath276 . it is straightforward to check that @xmath277 for all @xmath278 . therefore , using the assumptions on @xmath11 and on @xmath32 , @xmath279 that is , @xmath280 for some @xmath275 . by theorem [ spec t1 ] once again , @xmath65 has @xmath66 as an eigenvalue . [ spec t2 ] let @xmath235 and let @xmath236 be as in @xmath237 . if @xmath67 is connected , then @xmath196 has no eigenvalues in @xmath68 $ ] . let @xmath281 $ ] and @xmath212 such that @xmath199 . we will see that @xmath282 . note that @xmath283 in @xmath67 and hence @xmath284 . therefore , @xmath285 in @xmath67 for @xmath286 . since @xmath212 , then @xmath214 for some @xmath145 such that @xmath287 . using that @xmath10 , it is not hard to show that @xmath288 where @xmath289 and @xmath290 denotes the surface measure on @xmath291 . therefore , rellich s lemma yields @xmath282 in a neighbourhood of infinity ( see @xcite , for example ) , and thus @xmath282 in @xmath67 by unique continuation and the connectivity assumption . in particular @xmath292 in @xmath15 , and so lemma [ l jump]@xmath116 and the definition of @xmath34 give @xmath293 this means that @xmath294 , so @xmath295 and then @xmath296 in @xmath15 . it only remains to check that @xmath282 in @xmath297 , but this goes along well known lines . since @xmath199 , then @xmath283 in @xmath297 . if one integrates by parts on the identities @xmath298 where @xmath299 and @xmath300 is the ball centered at @xmath301 and with radius @xmath302 , and then one takes @xmath303 , one can show that @xmath304 where @xmath305 since @xmath296 in @xmath15 , we conclude from ( [ spec t2 eq1 ] ) that @xmath26 vanishes in @xmath297 , and thus @xmath282 . although the definition of @xmath34 does not make sense for @xmath306 , one can replace @xmath287 by @xmath307 in the definition of @xmath226 in ( [ domain ] ) , in order to recover the free dirac operator @xmath9 when @xmath306 . it is well known that @xmath9 does not have eigenvalues in @xmath68 $ ] , so theorem [ spec t2 ] also holds for @xmath306 . with the same arguments , one can check that theorem [ spec t2 ] also holds for other @xmath12-valued potentials different from @xmath60 , as far as they yield a suitable @xmath34 for which a relation like ( [ spec t2 eq2 ] ) implies that @xmath296 in @xmath15 . this section is focused on the coupling @xmath53 given in section [ ss electro ] ( see ( [ domain ] ) ) , and mostly in the case that @xmath15 is the sphere . however , the following two lemmata hold for any @xmath15 and @xmath11 as in section [ s preliminaries ] . first , we need some definitions . for @xmath175 $ ] and @xmath308 , where the @xmath309 s are the family of pauli matrices ( see ( [ freedirac ] ) ) , define @xmath310 for @xmath311 . given @xmath312 and @xmath313 , set @xmath314 that @xmath315 and @xmath316 are bounded operators in @xmath70 can be verified similarly to the case of @xmath317 in @xmath12 , we omit the details . moreover , note that @xmath318 [ sphere l1 ] for @xmath175 $ ] , the following hold : * the anticommutator @xmath319 vanishes identically , * @xmath320 . from lemma [ l jump]@xmath117 and ( [ sphere eq1 ] ) we have @xmath321 and the lemma follows . [ sphere l2 ] @xmath315 is a positive operator in @xmath70 for all @xmath64 . we want to verify that @xmath322 for all @xmath312 . if we set @xmath323 for @xmath96 , it is not hard to check that @xmath324 and that it satisfies @xmath325 moreover , since @xmath326 for all @xmath311 , a proof analogous to the one of ( * ? ? ? * lemma 3.3@xmath116 ) shows that @xmath327 where @xmath328 denote the boundary values of @xmath329 when one approaches to @xmath15 non - tangentially from @xmath48 . therefore , using ( [ sphere l2 eq1 ] ) , ( [ sphere l2 eq2 ] ) , and the divergence theorem , we conclude @xmath330 throughout this section we set @xmath331 , @xmath332 , @xmath11 denotes the surface measure on @xmath71 , and @xmath333 for @xmath334 . we also use the notation of ( * ? ? ? * section 4.6.4 ) . let @xmath335 be the usual spherical harmonics . they are defined for @xmath336 , and @xmath337 , and they satisfy @xmath338 , where @xmath339 denotes the usual spherical laplacian . moreover , @xmath335 form a complete orthonormal set in @xmath340 . for @xmath341 and @xmath342 , set @xmath343 then @xmath344 form a complete orthonormal set in @xmath70 , and @xmath345 where @xmath346 ( see ( * ? ? ? * equation ( 4.121 ) and the remark in page 127 ) ) . [ sphere l3 ] given @xmath64 , there exist positive numbers @xmath347 and purely imaginary numbers @xmath348 for all @xmath341 and @xmath342 , such that : * @xmath349 and @xmath350 . moreover , @xmath351 * @xmath352 and @xmath353 . moreover , @xmath354 for any @xmath355 and @xmath356 , we identify the spherical harmonic @xmath335 with its homogeneous extension of degree @xmath115 to @xmath357 . that is , @xmath358 for all @xmath311 . in particular , @xmath359 is a homogeneous polynomial of degree @xmath355 which is harmonic . we use the same identification for @xmath360 . _ proof of @xmath116_. in order to prove the first identity in @xmath116 , fix @xmath361 and @xmath362 and set @xmath363 given @xmath302 , we define @xmath364 . it is easy to verify that @xmath365 converges to @xmath11 in the weak@xmath366 topology when @xmath367 . in particular , since @xmath11 and @xmath365 have compact support and @xmath368 is continuous in @xmath357 and has exponential decay at infinity , it is not hard to show that actually @xmath369 in @xmath370 , where @xmath371 the term on the right hand side of last equality in ( [ uncert l1 eq6 ] ) denotes the usual convolution of ( matrix and vectors of ) functions in @xmath159 . applying the fourier transform to ( [ uncert l1 eq6 ] ) and using that @xmath368 is a fundamental solution of @xmath372 , we obtain @xmath373 note that , for any @xmath374 , @xmath375 is a bounded radial function with compact support , thus ( * ? ? ? * corollary in page 72 ) shows that @xmath376 for some radial function @xmath377 depending on @xmath378 but not on @xmath362 . hence , @xmath379 since @xmath380 is also a radial function , we can use that @xmath381 is the identity operator and ( * ? ? ? * corollary of page 72 ) in ( [ uncert l1 eq7 ] ) to deduce that @xmath382 for some radial function @xmath383 depending on @xmath378 but not on @xmath362 . finally , using that @xmath369 and ( [ uncert l1 eq8 ] ) , we conclude that @xmath384 for some radial function @xmath385 . we already know that @xmath315 is a bounded operator in @xmath70 , and since @xmath386 for @xmath387 , one can check that @xmath388 which implies that @xmath389 is continuous at @xmath390 . then , by setting @xmath391 , ( [ uncert l1 eq9 ] ) shows that @xmath392 concerning the second statement in @xmath116 , since @xmath386 for @xmath387 , it is easy to check that @xmath315 is a compact operator in @xmath70 , so the eigenvalues of @xmath315 form a bounded sequence which has @xmath393 as the only possible accumulation point ( see ( * ? ? ? * fredholm s theorem ( 0.38)@xmath394 ) , for example ) . therefore , @xmath350 . let us now prove the last statement in @xmath116 . from lemma [ sphere l2 ] , @xmath315 is a positive operator , which implies that @xmath395 for all @xmath361 . moreover , @xmath396 . following ( * ? ? ? * generalized young s inequality ( 0.10 ) ) and since @xmath71 is invariant under rotations , it is easy to see that @xmath397 where @xmath398 ( we have identified the matrix @xmath368 with its scalar version ) . consider the change of variables to polar coordinates in @xmath71 @xmath399 then , we have @xmath400 where we used the change of variables @xmath401 in the last equality above . to finish the proof of @xmath116 , it only remains to check that @xmath402 . note that @xmath403 where @xmath404 is some constant . therefore , @xmath405 and we are done . _ proof of @xmath117_. fix @xmath361 and @xmath362 . recall from ( [ uncert l1 eq11 ] ) and ( [ uncert l1 eq10 ] ) that , for @xmath96 , we have @xmath406 for some function @xmath407 , where @xmath408 and @xmath409 . furthermore , @xmath410 and , by similar arguments to the ones that prove that @xmath389 is continuous at @xmath390 , one can show that @xmath411 and @xmath412 exist ; we omit the details . since @xmath413 , similarly to ( [ sphere l2 eq2 ] ) one can check that @xmath414 where @xmath415 denote the boundary values of @xmath416 when one approaches to @xmath15 non - tangentially from @xmath48 . it is well known and quite easy to see that @xmath417 , where @xmath418 and @xmath346 . combining this with ( [ uncert l1 eq2 ] ) , ( [ uncert l1 eq1 ] ) and ( [ uncert eq1 ] ) , we obtain @xmath419 i.e. , @xmath352 for some @xmath420 . by @xmath421 , lemmata [ sphere l1]@xmath117 and [ sphere l3]@xmath116 , @xmath422 which yields @xmath423 from the last statement in @xmath116 , we have @xmath424 thus @xmath425 by ( [ uncert l1 eq3 ] ) , and that means that @xmath348 are purely imaginary numbers . the last statement in @xmath117 follows by ( [ uncert l1 eq3 ] ) and ( [ uncert l1 eq4 ] ) , so it only remains to prove that @xmath353 . for that purpose , we use the first identity in @xmath117 and that @xmath426 and @xmath316 are symmetric operators to see that @xmath427 since @xmath428 and we already know that @xmath348 are purely imaginary , we obtain from ( [ uncert l1 eq5 ] ) that @xmath429 , which implies that @xmath353 . the lemma is finally proved . note that , if we know that @xmath430 for all @xmath361 , the relation @xmath353 also follows from lemma [ sphere l1]@xmath116 . the following theorem is based on an uncertainty principle on the sphere and goes on the lines of ( * ? ? ? * theorem 1 ) . [ thmineq ] given @xmath431 and @xmath64 , the operator @xmath432 is invertible in @xmath70 . furthermore , for any @xmath433 and any @xmath434 , we have @xmath435 where @xmath436 if @xmath437 is such that @xmath438 , then equality in @xmath439 holds by taking @xmath440 for any @xmath441 and @xmath442 recall that @xmath315 is a positive operator by lemma [ sphere l2 ] , thus @xmath432 is positive and invertible in @xmath70 for all @xmath431 by ( * ? ? ? * theorem 12.12@xmath443 ) , and the inverse is also a positive operator . since @xmath395 for all @xmath361 and @xmath350 by lemma [ sphere l3]@xmath116 , lemma [ sphere l3]@xmath117 shows that @xmath444 for all @xmath361 and @xmath445 . therefore , there exists some @xmath437 such that @xmath446 , and thus @xmath447 is well defined . the estimate of @xmath447 from below is stated in lemma [ sphere l3]@xmath117 . given @xmath361 and @xmath362 , we define @xmath448 we will prove ( [ ineq ] ) by estimating @xmath55 from above and from below . from lemma [ sphere l3 ] we know that @xmath449 are purely imaginary . therefore , arguing as in ( [ uncert l1 eq5 ] ) , we have @xmath450 to estimate @xmath55 from above , we use lemma [ sphere l3 ] , ( [ uncert eq1 ] ) , cauchy - schwarz inequality , and that @xmath451 for all @xmath452 , to deduce that @xmath453 from ( [ eq2 ] ) and ( [ eq3 ] ) , we see that ( [ ineq ] ) holds for @xmath454 . the functions @xmath344 with @xmath341 and @xmath342 form a complete orthonormal system in @xmath70 . hence , to prove ( [ ineq ] ) in full generality , we first write any @xmath312 as a linear combination of the @xmath344 s and we expand the left and right hand side of ( [ ineq ] ) in terms of this basis . then , using the orthogonality and , for the right hand side of ( [ ineq ] ) , that @xmath455 from lemma [ sphere l3 ] , we conclude that ( [ ineq ] ) holds for all @xmath312 if and only if it holds for all @xmath344 . therefore , ( [ ineq ] ) is finally proved . it only remains to check the last statement of the theorem . let @xmath437 be such that @xmath438 ( we already know that such @xmath437 exists ) . then , for the functions @xmath456 the inequality in ( [ eq2 ] ) becomes an equality . furthermore , for @xmath457 satisfying ( [ delta ] ) , we have @xmath458 which implies that the inequality in ( [ eq3 ] ) is an equality for @xmath459 . recall from lemma [ sphere l3]@xmath117 that @xmath460 for all @xmath361 . hence , for any @xmath437 such that @xmath438 , we have two possible elections of the subindex , say @xmath461 and @xmath462 , and therefore two possible values of @xmath457 for which equality in ( [ ineq ] ) holds . hence , we get two ( a priori different ) sharp inequalities . the same observation applies if such @xmath437 is not unique . theorem [ thmineq ] has an interesting consequence concerning a lower bound for the @xmath72-dimensional riesz transform on the sphere . given a finite borel measure @xmath83 in @xmath2 , @xmath463 and @xmath96 , one defines the @xmath72-dimensional riesz transform of @xmath464 as @xmath465 whenever the limit makes sense . it is well known that @xmath466 is a bounded operator for 2-dimensional uniformly rectifiable ad regular measures @xmath83 in @xmath2 ( see @xcite for a deep study on this subject ) . in particular , @xmath466 is a bounded operator when @xmath83 is the surface measure of a bounded lipschitz domain . this means that @xmath467 for some constant @xmath94 and all @xmath463 . much less is said about lower @xmath468-bounds for the riesz transform . however , in the case of the sphere , from the results in @xcite ( see also ( * ? ? ? * equation ( 4.6.9 ) ) ) one can easily show that the riesz transform ( multiplied by a suitable constant ) is an isometry , providing sharp constants to the above - mentioned inequalities . the following corollary of theorem [ thmineq ] yields the sharp constant of the inequality from below on @xmath71 . set @xmath469 for @xmath311 , and @xmath470 [ riesz ] the following inequalities hold and they are sharp : * @xmath471 for all @xmath312 . * @xmath472 for all real - valued @xmath473 . set @xmath164 and @xmath474 in theorem [ thmineq ] . given @xmath312 , ( [ ineq ] ) yields @xmath475 where @xmath476 . notice that , by lemma [ sphere l3 ] , @xmath347 are uniformly bounded and @xmath477 uniformly in @xmath361 when @xmath478 . in particular , @xmath479 when @xmath478 . recall that @xmath480 and @xmath481 are defiend by means of the convolution kernels @xmath482 we define @xmath483 for @xmath311 , and @xmath484 that @xmath263 and @xmath485 are bounded operators in @xmath70 follows essentially as in the case of @xmath480 and @xmath481 . moreover , it is not hard to show that @xmath486 ( see the proof of lemma [ spec l1 ] for a related argument ) . roughly speaking , in @xmath71 , @xmath480 and @xmath481 are compact perturbations of @xmath263 and @xmath485 which depend on @xmath487 continuously . therefore , if we take @xmath478 in ( [ sphere c1 eq1 ] ) and we use that @xmath488 , we obtain @xmath489 for all @xmath434 . minimazing in @xmath457 , i.e. , taking @xmath490 , we get @xmath491 which is the inequality in corollary [ riesz]@xmath116 . corollary [ riesz]@xmath117 follows from ( [ sphere c1 eq2 ] ) by taking @xmath492 for any real - valued @xmath473 . that the inequalities are sharp is a consequence of the fact that ( [ sphere c1 eq1 ] ) is sharp for @xmath457 as in ( [ delta ] ) . since @xmath479 and @xmath493 for @xmath478 ( recall that we have set @xmath164 and @xmath474 ) , using ( [ sphere c1 eq1 ] ) and ( [ pirris ] ) one can check that ( [ sphere c1 eq2 ] ) is sharp , and the corollary follows . the following lemma gives a specific criterion based on proposition [ spec p1 ] to generate eigenvectors of @xmath53 . [ eigenfunction ] let @xmath494 be as in @xmath237 with @xmath332 . if @xmath431 and @xmath64 satisfy @xmath495 then , for any @xmath362 , @xmath344 gives rise to an eigenfunction of @xmath494 with eigenvalue @xmath234 . let @xmath264 , @xmath234 and @xmath361 be as in the lemma . since @xmath431 , @xmath496 is invertible by theorem [ thmineq ] . hence we can define @xmath497 in particular , we have the relation @xmath498 using lemma [ sphere l3 ] , we have @xmath499 if @xmath431 and @xmath64 satisfy ( [ a ] ) , then @xmath500 which by ( [ eigenfunction eq1 ] ) implies that @xmath501 finally , combining ( [ sphere eq1 ] ) , ( [ eigenfunction eq2 ] ) and ( [ eigenfunction eq3 ] ) , we obtain @xmath502 which means that @xmath503 . following proposition [ spec p1 ] ( see also theorem [ spec t1 ] ) , if we set @xmath201 and @xmath214 , then @xmath504 . the lemma is proved . for the case @xmath505 , one can develop results analogous to theorem [ thmineq ] and lemma [ eigenfunction ] by repeating the arguments involved in the proofs but by using the invertibility of @xmath506 instead of @xmath496 . we leave the details for the reader . from lemma [ spec l1 ] we have that @xmath241 , and in theorem [ spec t1 ] we proved that if @xmath507\ ] ] then @xmath53 has no eigenvalues in @xmath59 ( recall that @xmath508 by ( [ spec t1 eq1 ] ) ) . furthermore , in remark [ lambda - a ] we showed that , for any @xmath64 , there exists some @xmath264 such that @xmath53 has @xmath234 as an eigenvalue . from these results , it is not clear what can be said positively about the set of @xmath264 s for which there exist an eigenvalue of @xmath53 . thanks to lemma [ eigenfunction ] , we can give a bit more of information in the case of the sphere , but first we need to do some computations . lemma [ sphere l3]@xmath116 gives a precise value for @xmath509 in terms of @xmath10 and @xmath64 , that is , @xmath510 following a similar argument , we are going to prove that @xmath511 where @xmath512 corresponds to @xmath513 with @xmath514 . from ( [ uncert eq1 ] ) and ( [ d0 eq ] ) , we have @xmath515 therefore , if we identify the matrix @xmath368 with its scalar version and we set @xmath398 , lemma [ sphere l3]@xmath116 yields @xmath516 that @xmath517 follows from ( [ d1 eq1 ] ) , but it can be also verified using the change of variables ( [ polar ] ) and noting that the resulting integrals contain a @xmath518 or a @xmath519 , which integrated on @xmath520 $ ] vanish . on the other hand , using ( [ polar ] ) , @xmath521 where we used the change of variables @xmath401 and integration by parts in the last equality above . combining ( [ d1 eq1 ] ) and ( [ d1 eq2 ] ) , we get ( [ d1 eq ] ) , as desired . recall that lemma [ sphere l3]@xmath116 states that @xmath395 for al @xmath361 . it is an exercise to check directly from ( [ d1 eq ] ) that @xmath522 for all @xmath64 . we now turn to lemma [ eigenfunction ] , which can be used to provide eigenfunctions of @xmath53 under some assumptions on @xmath431 and @xmath64 . for @xmath514 , ( [ a ] ) reads as @xmath523 using ( [ d0bis eq ] ) and ( [ d1 eq ] ) , it is not difficult to see that if @xmath264 is very big or very small , then ( [ pepaa ] ) and ( [ pepab ] ) do not hold for any @xmath64 , so lemma [ eigenfunction ] can not be used . this agrees with the above - mentioned result on non - existence of eigenvalues given by ( [ pepa2 ] ) . let us take @xmath524 for simplicity . figure [ fig1 ] shows the set of points @xmath525 such that ( [ pepaa ] ) and ( [ pepab ] ) hold . we see that for any @xmath526 we can take a @xmath264 for which ( [ pepaa ] ) holds , hence there exists an eigenfunction with eigenvalue @xmath234 , by lemma [ eigenfunction ] . this agrees with remark [ lambda - a ] . however , from figure [ fig1 ] we also see that for any @xmath264 in some interval there exists an @xmath526 such that ( [ pepaa ] ) or ( [ pepab ] ) hold . more precisely , using the computer one can show that if @xmath527 then there exists @xmath526 such that ( [ pepaa ] ) holds , and if @xmath528 then there exists @xmath526 such that ( [ pepab ] ) holds . in particular , if @xmath529 then there exists @xmath526 such that either ( [ pepaa ] ) or ( [ pepab ] ) hold . by lemma [ eigenfunction ] , this means that the set of @xmath264 s for which there exists an eigenfunction of @xmath53 contains the interval @xmath530 ( recall that we have set @xmath524 for these calculations ) . the combination of theorem [ thmineq ] and lemma [ eigenfunction ] yields an interesting relation between the minimizing functions of ( [ ineq ] ) and some eigenfunctions of @xmath53 . more precisely , let @xmath437 be such that @xmath531 ( recall that @xmath460 for all @xmath361 ) . assume that the given @xmath431 and @xmath64 satisfy @xmath532 i.e. , the election of the first sign in ( [ a ] ) holds for this particular @xmath437 . then lemma [ eigenfunction ] shows that @xmath533 gives rise to an eigenfunction of @xmath494 with eigenvalue @xmath234 , for any @xmath534 . fix @xmath457 as in ( [ delta ] ) , but we chose the first sign on the possible definitions of @xmath457 , i.e. , @xmath535 then the functions @xmath533 are minimizers of ( [ ineq ] ) , that is , they attain the equality in ( [ ineq ] ) . once @xmath264 , @xmath234 and @xmath457 are fixed depending on @xmath437 , let @xmath536 be set of @xmath361 s such that @xmath537 and that one sign election in ( [ delta ] ) is satisfied for @xmath361 . in particular , @xmath538 . given @xmath539 , since @xmath540 and both @xmath437 and @xmath361 satisfy ( [ delta ] ) , we easily deduce that either @xmath541 an inspection of the proof of theorem [ thmineq ] shows that , * if @xmath542 then @xmath543 are minimizers of ( [ ineq ] ) , and * if @xmath544 , then @xmath545 are minimizers of ( [ ineq ] ) . actually , because of the orthogonality , it can be seen that any minimizer of ( [ ineq ] ) must be a linear combination of these functions indexed by @xmath539 and by a choose of the sign in @xmath378 depending on ( [ cases ] ) . similarly , for any @xmath539 , ( [ cases ] ) and ( [ a1 ] ) show that @xmath546 holds for a suitable election of the sign in @xmath378 , and the corresponding functions @xmath547 give rise to eigenfunctions of @xmath494 with eigenvalue @xmath234 . combining the above - mentioned arguments we can conclude that , once @xmath264 , @xmath234 , and @xmath457 are properly chosen , then any function which attains the equality in ( [ ineq ] ) give rise to an eigenfunction of @xmath494 with eigenvalue @xmath234 . roughly speaking , the minimizers of ( [ ineq ] ) provide eigenfunctions of @xmath494 . from lemma [ sphere l3](i ) , we know that @xmath548 for all @xmath549 and @xmath350 , so it is to be expected that the following question has a positive answer . [ ques ] is it true that @xmath550 for all @xmath551 ? if so , from lemma [ sphere l3]@xmath117 we see that the minimum in the definition of @xmath447 in ( [ ineq ] ) would be attained only at one particular @xmath361 , namely @xmath514 . actually , @xmath447 could be calculated explicitely using ( [ d0bis eq ] ) and ( [ d1 eq ] ) , that is , @xmath552 the same could be said about the two possible values of @xmath457 in ( [ delta ] ) , say @xmath553 if question [ ques ] has a positive answer , the argument of section [ delib ] becomes much more transparent , since in this case @xmath554 . furthermore , it would yield the following result : _ let @xmath64 and @xmath431 . then , for any @xmath312 , @xmath555 where @xmath447 is given by @xmath556 . the equality in @xmath557 is only attained at linear combinations of @xmath558 for @xmath559 . moreover , if @xmath560 then the minimizers of @xmath557 give rise to eigenfunctions of @xmath53 . _ these conclusions also hold if we exchange the roles of @xmath509 and @xmath512 in @xmath557 and @xmath561 and we replace @xmath558 by @xmath562 ( that is , we exchange the roles of @xmath563 and @xmath564 for @xmath514 ) . in this section , we show a criterion on @xmath0 to generate confinement , namely theorem [ conf t1 ] . this criterion is stated in terms of an algebraic property of certain bounded operators in @xmath12 . an application to electrostatic and lorentz scalar shell potentials is also shown . but before , we need some auxiliary lemmata . if @xmath567 is a function which is regular in @xmath568 and @xmath569 denote the boundary values of @xmath567 on @xmath15 when we approach from @xmath48 , then @xmath570 in the sense of distributions . the proof of this formula , which follows essentially by stoke s theorem , is an easy exercise left for the reader . for any @xmath161 , since @xmath571 in the sense of distributions , by ( [ conf l1 eq1 ] ) we have @xmath572 which implies that @xmath573 . recall that @xmath574 in @xmath575 , so the boundary values @xmath576 vanish on @xmath15 . combining this fact with lemma [ l jump]@xmath116 , we obtain @xmath577 given @xmath192 , let @xmath578 denote the boundary values of @xmath26 on @xmath15 . since @xmath579 in @xmath568 and @xmath287 , using ( [ conf l1 eq1 ] ) and lemma [ l jump]@xmath116 we obtain @xmath580 where @xmath581 . this implies that @xmath582 . hence , @xmath583 if and only if @xmath584 which , by ( [ conf l1 eq2 ] ) , the definition of @xmath585 and that @xmath287 , is equivalent to @xmath586 . therefore , @xmath583 for all @xmath212 if @xmath587 which is equivalent to @xmath588 ; the lemma is proved . note that , since @xmath226 is a vector space , in the statement of the lemma one only needs to require that @xmath589 for all @xmath212 . [ rem ult ] from the last part of the proof of lemma [ conf l1 ] , we actually see that @xmath583 for all @xmath212 if and only if @xmath588 holds on the set of functions @xmath165 such that there exists @xmath161 with @xmath590 . let us prove @xmath597 . since @xmath598 are bounded operators , @xmath599 . by @xmath116 , @xmath600 thus @xmath601 . if @xmath192 then @xmath602 , but from ( [ conf l1 eq4 ] ) we have seen that @xmath582 for some @xmath603 , so @xmath604 therefore , @xmath605 on @xmath606 , and @xmath117 is proved . the implication @xmath607 is straightforward . in order to prove @xmath608 , recall the well known fact that , if @xmath196 is self - adjoint , then @xmath609 is invertible for all @xmath77 . this assertion can be easily verified using the arguments in the proof of ( * ? ? ? * theorem viii.3 ) , for example . by definition , @xmath117 is equivalent to @xmath610 which , by writting @xmath611 , is further equivalent to @xmath612 in conclusion , @xmath613 it is well know that , if @xmath614 is a solution of @xmath615 then one can write @xmath616 ( see ( * ? ? ? * theorem 7.9 ) for a similar result ) . assume that @xmath617 , so @xmath618 . by ( [ conf l2 eq1 ] ) , we have @xmath619 therefore , we have proved that if @xmath117 holds then @xmath620 for all @xmath77 , which is a restatement of @xmath118 . the implication @xmath621 is left for the reader . that @xmath0 makes @xmath15 impenetrable means that the particles under consideration which are initially confined either in @xmath297 or @xmath67 at time @xmath623 , remain confined in @xmath297 or in @xmath67 for all @xmath77 under the evolution given by @xmath624 , i.e. , that lemma [ conf l2]@xmath118 holds . thus , the theorem is a straightforward application of lemmata [ conf l1 ] and [ conf l2 ] . [ conf appl ] assume that @xmath15 is @xmath100 . given @xmath78 such that @xmath625 , let @xmath196 be the operator defined by @xmath626 and @xmath627 on @xmath226 , where @xmath628 and @xmath223 for @xmath192 . if @xmath629 then @xmath196 is self - adjoint . in that case , @xmath196 makes @xmath15 impenetrable if and only if @xmath80 . for proving the self - adjointness of @xmath196 when @xmath629 , we follow the arguments of the proof of ( * ? ? ? * theorem 3.8 ) . set @xmath632 so @xmath633 , and observe that @xmath634 are self - adjoint on @xmath12 because @xmath635 and @xmath183 also are . using that @xmath636 and lemma [ l jump]@xmath117 we have @xmath637 where @xmath638 and @xmath639 . in ( * ? ? ? * lemma 3.5 ) we proved that @xmath640 is a compact operator in @xmath12 , and since @xmath635 anticommutes with the @xmath641 s , we easily have @xmath642 thus @xmath643 is a compact operator by ( * ? ? ? * proposition 3.11 ) and hence @xmath263 is also compact . if @xmath629 then @xmath644 and fredholm s theorem applies to @xmath645 ( see , ( * ? ? ? * theorem 0.38@xmath443 ) ) . if for example @xmath646 , using ( [ conf t2 eq2 ] ) we can follow the proof of ( * ? ? ? * lemma 3.7 ) to show that @xmath34 has closed range . moreover , as we did in the first part of the proof of ( * ? ? ? * theorem 3.8 ) , fredholm s theorem also shows that @xmath647 is closed , we omit the details . in any case , that @xmath193 and @xmath195 are closed for all @xmath629 follows by ( [ conf t2 eq2 ] ) , fredholm s theorem and ( * ? ? ? * theorem 1.46@xmath117 ) , so the restriction @xmath646 is not necessary . these properties of @xmath34 together with ( [ conf t2 eq1 ] ) allow us to apply theorem [ pre t1 ] , which proves that @xmath196 is self - adjoint for @xmath629 . let us finally check the impenetrability condition relative to @xmath196 and @xmath15 . by theorem [ conf t1 ] , @xmath196 makes @xmath15 impenetrable for the particles if @xmath34 satisfies @xmath588 . by a straightforward computation using lemma [ l jump]@xmath117 , that @xmath636 , and that @xmath648 and @xmath635 anticommute , we have @xmath649 therefore , @xmath34 satisfies @xmath588 if and only if @xmath80 , and in such case @xmath15 becomes impenetrable . furthermore , since the right hand side of ( [ kpopkosa ] ) is a constant times @xmath91 , from remark [ rem ult ] and lemma [ conf l2 ] we actually deduce that @xmath196 makes @xmath15 impenetrable if and only if @xmath80 . we have seen that @xmath650 makes @xmath15 impenetrable if and only if @xmath651 , which is precisely equation @xmath652 of ( * ? ? ? * section v ) . hence , for the potentials @xmath81 , our results on confinement generalize the ones stated in @xcite to regular surfaces . s. hofmann , e. marmolejo - olea , m. mitrea , s. prez - esteva and m. taylor , _ hardy spaces , singular integrals and the geometry of euclidean domains of locally finite perimeter , _ geometric and functional analysis ( forthcoming ) .

we provide necessary and sufficient conditions for admissible transformations in the projectivised dual quaternion model of rigid body displacements and we characterise constraint varieties of dyads with revolute and prismatic joints in this model . projective transformations induced by coordinate changes in moving and/or fixed frame fix the quadrics of a pencil and preserve the two families of rulings of an exceptional three - dimensional quadric . the constraint variety of a dyad with two revolute joints is a regular ruled quadric in a three - space that contains a `` null quadrilateral '' . if a revolute joint is replaced by a prismatic joint , this quadrilateral collapses into a pair of conjugate complex null lines and a real line but these properties are not sufficient to characterise such dyads . we provide a complete characterisation by introducing a new invariant , the `` fiber projectivity '' , and we present examples that demonstrate its potential to explain hitherto not sufficiently well understood phenomena . keywords : kinematic map , dual quaternion , study quadric , null cone , revolute joint , prismatic joint , fiber projectivity , vertical darboux motion .

1607.08119

a common technique in theoretical and applied kinematics is the use of a point model for the group @xmath0 of rigid body displacements . one prominent example is the projectivised dual quaternion model of @xmath0 which exhibits particularly nice geometric and algebraic properties @xcite . in this article , we revisit some fundamental concepts related to this model , the transformation group generated by coordinate changes in the moving and the fixed frame and the kinematic images of dyads with revolute and prismatic joints . while numerous necessary conditions on these objects are well - known , we contribute sets of provably sufficient conditions . our characterisation of the transformation group generated by coordinate changes in fixed and moving frame ( ) is based on the pencil of quadrics spanned by the study quadric and null cone . these are quadrics corresponding to dual quaternions of real norm and purely dual norm , respectively . admissible transformations fix each member of this pencil and , in addition , preserve the two families of rulings on a further quadric in a subspace of dimension three . this leads to the important distinction between `` left '' and `` right '' rulings . at this point we also introduce a further invariant , the `` fiber projectivity '' , which will be crucial in our latter characterisation of dyads with prismatic joints in . the relative position of two rigid bodies can be constrained by a link or a sequence of links . fixing one of the two bodies , the collection of all possible poses ( position and orientation ) of the other is called a _ constrained variety . _ these are important objects in the study of open and closed kinematic chains , in linkage synthesis or analysis and other fields . in , we characterise constraint varieties generated by dyads of two revolute joints ( `` rr dyads '' ) as regular ruled quadrics in the study quadric that contain four complex rulings of the null cone . in we extend this result to dyads containing one prismatic and one revolute joint ( `` rp dyads '' and `` pr dyads '' ) . it is tempting to view them as limiting cases of rr dyads where one joint axis becomes `` infinite '' ( lies in the plane at infinity ) . indeed , their kinematic image is a regular ruled quadric in the study quadric that intersects the null cone in two complex lines and a real transversal line . nonetheless , this viewpoint is not complete because of the possibility of commuting r and p joints ( `` cylindrical joints '' ) . a closer investigation leads us to a more refined concept involving the fiber projectivity which allows to distinguish between the rp , the pr , and the cylindrical case . we conclude this paper with an application of our results to a recently presented non - injective extension of the classical kinematic map @xcite . here , commuting rp dyads appear naturally as kinematic images of straight lines . we use this to prove that the extended kinematic image of a straight line is , in general , a vertical darboux motion . some parts of this paper , mostly and the computations in the appendix , overlap with a previously published conference paper @xcite . the investigation on the group of admissible transformation in , the characterisation of rp and pr dyads in , and the relation of straight lines in extended kinematic image space to vertical darboux motions in are new . this article s scene is the projectivised dual quaternion model of spatial kinematics . here , we give a very brief introduction to this model for the purpose of settling our notation . more details will be introduced in the text as needed . for more thorough introductions to dual quaternions and there relations to kinematics we refer to section 3 in klawitter ( 2015 ) @xcite or section 11 of selig ( 2005 ) @xcite . the dual quaternions , denoted by @xmath1 , form an associative algebra in @xmath2 where multiplication of the basis elements @xmath3 , @xmath4 , @xmath5 , @xmath6 , @xmath7 , @xmath8 , @xmath9 , and @xmath10 is defined by the rules @xmath11 an element @xmath12 may be written as @xmath13 with quaternions @xmath14 ( angled brackets denote linear span ) . in this case the _ quaternions _ @xmath15 and @xmath16 are referred to as _ primal _ and _ dual _ part of @xmath17 , respectively . the conjugate dual quaternion is @xmath18 and conjugation of quaternions is done by multiplying the coefficients of @xmath4 , @xmath5 , and @xmath6 with @xmath19 . it satisfies the rule @xmath20 for any @xmath21 . the dual quaternion norm is defined as @xmath22 . we readily verify that it is a _ dual number , _ that is , an element of @xmath23 . we identify linearly dependent non - zero dual quaternions and thus arrive at the projective space @xmath24 . writing @xmath25 $ ] for the point in @xmath26 that is represented by @xmath12 , the _ study quadric _ is defined as @xmath27 \in p^7\colon { \vert{q}\vert } \in { \mathbb{r}}\}$ ] . with @xmath13 , the algebraic condition for @xmath25 \in { { \mathcal{s}}}$ ] is @xmath28 . identifying @xmath29 with the projective subspace generated by @xmath30 , a point @xmath25 = [ p + { \varepsilon}d ] \in { { \mathcal{s}}}$ ] with non - zero primal part acts on @xmath31 \in p^3 $ ] via @xmath32 \mapsto [ y ] \coloneqq [ ( p + { \varepsilon}d ) x ( { \overline{p } } - { \varepsilon}{\overline{d}})].\ ] ] the map is the projective extension of a rigid body displacement in @xmath33 . composition of displacements corresponds to dual quaternion multiplication . the map that takes a point @xmath25 = [ p + { \varepsilon}d ] \in { { \mathcal{s}}}\setminus { [ { \varepsilon}{\mathbb{h}}]}$ ] to the rigid body displacement is an isomorphism between the factor group of dual quaternions of non - zero real norm modulo the real multiplicative group and @xmath0 . we refer to it as _ study s kinematic map _ or simply as _ kinematic map._. it provides a rich and solid algebraic and geometric environment for investigations in kinematic . provided @xmath34 , always describes a rigid body displacement , even if the study condition is not fulfilled @xcite . in this case , the map that sends @xmath35 to the rigid body displacement is no longer a group isomorphism but a homomorphism . we call it _ extended kinematic map . _ in we will characterise the extended kinematic image of straight lines in @xmath26 . the geometry in kinematic image space @xmath26 is invariant with respect to coordinate changes in three - dimensional euclidean space . more precisely , it is invariant with respect to coordinate changes in both , the frame of @xmath31 $ ] in ( the _ moving frame _ ) and the frame of @xmath36 $ ] in ( the _ fixed frame _ ) . the former correspond to right - multiplications , the latter to left - multiplications with dual quaternions of real norm and non - zero primal part . both transformations induce projective transformations in @xmath26 and our aim in this section is a geometric characterisation of the transformation group they generate . necessary geometric conditions on this group are already known but we are not aware of a formal proof of sufficiency for a set of these conditions . this we will provide in this section . it will refer to an important new geometric invariant , the _ fiber projectivity , _ whose usefulness we demonstrate in an example . later , it will re - appear in our characterisation of the kinematic image of cylinder spaces . for a quaternion @xmath37 we define @xmath38 with this notation , the projective maps @xmath31 \mapsto [ px]$ ] and @xmath31 \mapsto [ xp]$ ] on @xmath39}= p^3 $ ] may be written as @xmath40^{\intercal}\mapsto { \varrho(p ) } \cdot [ x_0,x_1,x_2,x_3]^{\intercal}$ ] and @xmath40^{\intercal}\mapsto { \lambda(p ) } \cdot [ x_0,x_1,x_2,x_3]^{\intercal}$ ] , respectively . both leave invariant the quadric @xmath41 whence the geometry in @xmath39}$ ] , induced by coordinate changes in moving and fixed frame , is that of elliptic three - space . in fact , the matrices in describe the well - known clifford right and left translations that generate the transformation group of this space , see coxeter ( 1998 ) , p. 140 . @xcite clifford right ( left ) translations leave fixed every member of one family of ( complex ) rulings of @xmath42 and we call those rulings _ right ( left ) rulings _ , respectively . similarly , left - multiplication by a dual quaternion @xmath43 and right - multiplication by a dual quaternion @xmath44 can be effected by multiplication with matrices @xmath45 and @xmath46 , respectively . these @xmath47 matrices are conveniently described in terms of blocks of dimension @xmath48 : @xmath49 here , @xmath50 denotes the zero matrix of dimension @xmath48 . two matrices of above shape commute and their product @xmath51 is the matrix of a general coordinate transformation which we want to characterise geometrically . for that purpose , we introduce two further invariant quadrics : the _ null cone @xmath52 _ is the quadric defined by the quadratic form @xmath53 . the points of the null cone @xmath52 are characterised by having purely dual norm ( @xmath54 ) . the null cone is a singular quadric with three - dimensional vertex space @xmath55}\coloneqq \{[{\varepsilon}d]\colon d \in { \mathbb{h}}\}$ ] which we call the _ exceptional generator_. it is contained in the study quadric @xmath56 and the real points of @xmath52 are precisely those of @xmath55}$ ] . by @xmath57 we denote the regular quadric in @xmath55}$ ] defined by the quadratic form @xmath58}\mapsto d{\overline{d}}$ ] . the quadrics @xmath59}$ ] , @xmath52 , @xmath56 , and @xmath60}$ ] are all invariant under transformations of the shape . [ th:1 ] the transformation group described by matrices of shape where @xmath43 and @xmath44 satisfy the study condition is characterised by the following properties : 1 ) it fixes any quadric in the pencil spanned by study quadric @xmath56 and null cone @xmath52 and 2 ) the restriction to @xmath55}$ ] preserves the two families of ( complex ) rulings of the quadric @xmath57 . to begin with , it is elementary to verify that the conditions of the theorem are necessary for transformations given by . for the proof of sufficiency , we again employ block matrix notation . assume that @xmath61 is the matrix of a projective transformation @xmath62 with @xmath48 blocks @xmath63 , @xmath64 , @xmath65 , and @xmath66 . the matrices of the pencil spanned by @xmath56 and @xmath52 are of the shape @xmath67 where @xmath68 and @xmath69 denote the @xmath48 identity and zero matrix , respectively , and @xmath70 and @xmath71 are real numbers . the only singular quadric in this pencil is @xmath52 and @xmath72 must fix its vertex space @xmath55}$ ] ( the only three - dimensional space contained in @xmath52 ) . hence @xmath73 and @xmath72 transforms the matrix of @xmath52 to @xmath74 whence @xmath63 is the scalar multiple of an orthogonal matrix . the transformed matrix of @xmath56 is @xmath75 from this we infer that @xmath76 for some @xmath77 and @xmath78 . this latter matrix equation has the general solution @xmath79 with an arbitrary skew - symmetric matrix @xmath80 of dimension @xmath48 . finally , we study the action on a third quadric in the pencil : @xmath81 this implies @xmath82 whence @xmath83 the restriction of @xmath72 to @xmath55}$ ] transforms the matrix @xmath84 $ ] of @xmath60}$ ] to @xmath85 = [ { i}]$ ] and hence fixes @xmath57 . the condition on the rulings implies @xmath86 . by well - known results of three - dimensional elliptic geometry , @xmath63 is uniquely expressible as product of a clifford left translation and a clifford right translation ( page 140 of coxeter 1998 @xcite ) . in our notation , this means that there exist quaternions @xmath87 , @xmath88 such that @xmath89 . hence , the upper left and lower right corners of and match . for given @xmath63 , the set of possible matrices @xmath79 is a real vector space of dimension six . the same is true for the @xmath48 sub - matrices in the lower left corner of the matrix ( recall that @xmath90 ) . hence , there exist unique quaternions @xmath91 , @xmath92 such that @xmath93 and the theorem is proved . examining the proof , it is easy to see that the system of invariants in is minimal . invariance of left and right rulings , respectively , of @xmath57 is an important property that is , for example , responsible for different kinematic properties of motion and inverse motion . for the planar case , this is mentioned in bottema and roth ( 1990 ) , chapter 11 , 14 @xcite . the map @xmath94 , @xmath25 \mapsto [ { \overline{q}}]$ ] ( quaternion conjugation ) is a projective transformation of @xmath26 leaving invariant the quadrics of the pencil spanned by @xmath56 and @xmath52 ( lets call it the _ absolute pencil _ ) but it interchanges the left and right rulings of @xmath57 . given a motion @xmath95 ( a curve in @xmath56 ) , the inverse motion is @xmath96 . it is of the same algebraic degree and has the same relative position to absolute pencil ( in projective sense ) but not necessarily the same position to the left and right rulings of @xmath57 . in order to demonstrate this at hand of an example , we introduce a further invariant concept . [ def:1 ] the projective map @xmath97},\quad [ x'+{\varepsilon}x '' ] \mapsto [ { \varepsilon}x']\ ] ] that assigns to each point in @xmath26 the projection on its primal part times @xmath98 is called the _ fiber projectivity . _ the name `` fiber projectivity '' is motivated by a recently introduced non - injective extensions of study s kinematic map @xcite which reads as in but drops the study condition @xmath28 . its fibers are obtained by connecting a point @xmath31 $ ] with @xmath99)$ ] . the fiber projectivity is invariant with respect to transformations @xmath72 given by in the sense that @xmath100 ) = \varphi \circ \tau([x])$ ] for every point @xmath31 \in p^7 $ ] . the restriction of the fiber projectivity to @xmath39}$ ] is a bijection in which @xmath42 and @xmath57 correspond so that we may speak of _ right ( left ) rulings _ of @xmath57 as well . [ ex:1 ] the _ darboux motion _ is the only spatial motion with planar trajectories , see bottema and roth , 1990 , equation ( 3.4 ) @xcite . a parametric equation in dual quaternions reads @xmath101 ( see li et al . , 2015 @xcite ) . the motion parameter is @xmath102 while @xmath103 , @xmath104 , and @xmath105 are constant real numbers . this rational cubic curve intersects @xmath55}$ ] in the two points @xmath106 \coloneqq [ { \varepsilon}(\pm c{\mathrm{i}}+b - a{\mathbf{i}}\pm a{\mathrm{i}}{\mathbf{j}}+(\pm b{\mathrm{i}}-c){\mathbf{k}})]\ ] ] which even lie on @xmath57 . ( note that `` @xmath107 '' denotes a complex number which must not be confused with the quaternion unit `` @xmath4 '' . ) the fiber projection of @xmath108 $ ] is the point @xmath109 ) = [ ( 1+t^2){\varepsilon}({\mathbf{k}}t + 1 ) ] = [ { \varepsilon}({\mathbf{k}}t + 1)].\ ] ] for varying @xmath102 , these points vary on a straight line which intersects @xmath57 in the two points @xmath110 \coloneqq [ { \varepsilon}(1 \pm { \mathrm{i}}{\mathbf{k}})]$ ] . it is now easy to verify that the lines @xmath111 \vee [ f_1]$ ] and @xmath112 \vee [ f_2]$ ] are rulings of @xmath57 . moreover , @xmath111 = [ pf_1]$ ] and @xmath112 = [ pf_2]$ ] where @xmath113 . this means that a darboux motion intersects @xmath57 in two points @xmath111 $ ] , @xmath112 $ ] and its fiber projection intersects @xmath57 in two points @xmath114 $ ] , @xmath115 $ ] such that the lines @xmath111 \vee [ f_1]$ ] and @xmath112 \vee [ f_2]$ ] are _ left rulings _ of @xmath57 . the inverse motion @xmath116 ( _ mannheim motion _ ) has similar properties as curve in @xmath26 but different kinematic properties . most notably , the trajectories of @xmath117 are rational of degree _ two _ while those of @xmath118 are rational of degree _ four_. the deeper reasons for this is the non - invariance of left and right rulings with respect to the quaternion conjugation map @xmath119 . hence , the intersection points of @xmath120 and @xmath121 with @xmath57 span _ right _ rulings . the case of a _ vertical darboux motion _ ( @xmath122 ) is special . here we have @xmath111 = [ f_1]$ ] and @xmath112 = [ f_2]$ ] and the distinction between left and right rulings vanishes . indeed , the inverse motion is again a vertical darboux motion . given two non co - planar lines @xmath87 , @xmath91 in euclidean three - space , we consider the set of all displacements obtained as composition of a rotation around @xmath91 , followed by a rotation about @xmath87 . its kinematic image is known to lie in a three - space ( selig 2005 , section 11.4 ) @xcite which we call a _ 2r space_. it always contains the displacement corresponding to zero rotation angle around both axes . it is no loss of generality to view it as identity displacement @xmath123 $ ] which we will often do in proofs ( and consistently did in a previous publication@xcite ) . however , we avoid this in our theorems because it is not invariant with respect to the transformations characterised in . fixing one rotation angle and varying the other yields a straight line in @xmath56 . the thus obtained two families of lines form the rulings of a quadric surface in @xmath56 . in order to characterise 2r spaces among all three - dimensional subspaces of @xmath26 with these properties , we introduce the following notions : a _ spatial quadrilateral _ is a set of four different lines @xmath124 in projective space such that the intersections @xmath125 , @xmath126 , @xmath127 , and @xmath128 are not empty . the lines @xmath129 , @xmath87 , @xmath91 , and @xmath130 are called the quadrilateral s _ edges_. a _ null line _ is a straight line contained in the study quadric @xmath56 and the null cone @xmath52 . null quadrilateral _ is a spatial quadrilateral whose elements are _ null lines . _ [ th:2 ] a three - space @xmath131 is a 2r space if and only if it * intersects the study quadric in a regular ruled quadric @xmath132 , * does not intersect the exceptional three - space @xmath55}$ ] , and * contains a null quadrilateral . the first and second item in exclude exceptional cases with co - planar , complex , or `` infinite '' revolute axes . the latter correspond to prismatic joints and will be treated later . the crucial point is existence of a null quadrilateral . we split the proof of into a series of lemmas . it will be finished by the end of this section . [ lem:1 ] the straight line @xmath31 \vee [ y]$ ] is contained in @xmath133 if and only if @xmath134 . we omit the straightforward computational proof of . note that the left - hand sides of each of the three conditions in this lemma are dual numbers . hence , they give _ six _ independent linear equations for the real coefficients of @xmath135 and @xmath136 . [ lem:2 ] the conditions of are necessary for 2r spaces . the quadric @xmath132 contains a regular point @xmath137 $ ] and , without loss of generality , we may assume @xmath137 = [ 1]$ ] . then , the constraint variety of a 2r chain can be parameterised as @xmath138 with two dual quaternions @xmath139 that satisfy @xmath140 , @xmath141 and @xmath142 . the first condition ensures that the dual quaternions are suitably normalised and @xmath143 , [ h_2 ] \in { { \mathcal{s}}}$ ] . the second condition means that @xmath144 and @xmath145 describe half - turns ( rotations through an angle of @xmath146 ) . the third condition is true because the axes of these half turns are not co - planar . ( two half turns commute if and only if their axes are identical or orthogonal and co - planar . ) equation describes a composition of two rotations for all values of @xmath147 , @xmath148 in @xmath149 with @xmath150 corresponding to zero rotation angle . even if their kinematic meaning is unclear , we also allow complex parameter values . expanding yields @xmath151 . we see that the kinematic image of the 2r dyad lies in the three - space spanned by @xmath123 $ ] , @xmath143 $ ] , @xmath152 $ ] , @xmath153 $ ] . in a suitable projective coordinate frame with these points as base points , we use projective coordinates @xmath40 $ ] . then , the surface parameterisation reads @xmath154 , @xmath155 , @xmath156 , @xmath157 . it describes the quadric with equation @xmath158 which is indeed regular and ruled . the intersection of @xmath159 with the exceptional three - space @xmath55}$ ] is non empty if and only if the primal part of @xmath160 vanishes for certain parameter values @xmath147 , this can only happen if the primal parts of @xmath144 and @xmath145 are linearly dependent over @xmath161 but then the revolute axes are parallel and @xmath159 , contrary to our assumption , is contained in @xmath56 . the other possibility for @xmath162 , intersecting revolute axes , has been excluded as well . clearly , @xmath159 is not contained in the null cone @xmath52 either . we claim that the intersection of @xmath159 and @xmath52 consists of the four lines given by @xmath163 , @xmath164 . indeed , they are null lines . set , for example , @xmath165 , @xmath166 and @xmath167 . in view of , we have to verify @xmath168 . but this follows from @xmath169 here , we used the fact that @xmath170 is a real number and thus commutes with all other factors . the cases @xmath171 , @xmath172 are similar so that we have verified all conditions of . the proof of sufficiency is more involved . we need two additional lemmas from projective geometry which are formulated and proved in . [ lem:3 ] a three - space @xmath159 that satisfies all conditions of is a 2r space . denote the vertices of the null quadrilateral by @xmath173 $ ] , @xmath174 $ ] , @xmath175 $ ] , @xmath176 $ ] such that consecutive points span the quadrilateral s edges . again , it is no loss of generality to assume @xmath123 \in { { \mathcal{q}}}$ ] . we denote the tangent hyperplane of @xmath56 at @xmath123 $ ] by @xmath177 . a point @xmath178 $ ] is contained in @xmath177 if and only if @xmath179 . since @xmath123 $ ] does not lie on the null quadrilateral , there exist ( possibly not unique ) points @xmath180 \in [ u_1 ] \vee [ v_1],\quad [ n_1 ] \in [ v_1 ] \vee [ u_2],\\ [ m_2 ] \in [ u_2 ] \vee [ v_2],\quad [ n_2 ] \in [ v_2 ] \vee [ u_1 ] \end{gathered}\ ] ] in the intersection of @xmath159 and @xmath177 such that * @xmath181 $ ] , @xmath182 $ ] and @xmath183 $ ] , @xmath184 $ ] are pairs of complex conjugate points ( whence their respective joins are real lines ) and * the quadrilateral with these points as vertices is planar and non - degenerate ( ) . the real points of @xmath181 \vee [ m_2]$ ] and @xmath183 \vee [ n_2]$ ] correspond to rotations about fixed axes . pick real rotation quaternions @xmath144 , @xmath185 such that @xmath143 \in [ m_1 ] \vee [ m_2]$ ] and @xmath152 \in [ n_1 ] \vee [ n_2]$ ] . the axes of these rotations are the only candidates for our 2r dyad . hence , we have to show that either @xmath153 \in u$ ] or @xmath186 \in u$ ] . it is easy to see that this is the case if and only if either @xmath187 \in u$ ] or @xmath188 \in u$ ] . in fact , we will even show that one of these products equals @xmath189 . we claim that both @xmath190 \vee [ m_1n_1]$ ] and @xmath191 \vee [ n_1m_1]$ ] are null lines . in order to show this , we have to verify the conditions of : @xmath192 similarly , we see that also @xmath193 \vee [ m_1n_1]$ ] and @xmath194 \vee [ n_1m_1]$ ] are null lines . thus , we are in the situation depicted in where we have three null quadrilaterals with respective vertices @xmath195 , [ v_1 ] , [ u_2 ] , [ v_2];\quad [ m_1n_1 ] , [ m_2n_1 ] , [ m_2n_2 ] , [ m_1n_2];\\ [ n_1m_1 ] , [ n_1m_2 ] , [ n_2m_2 ] , [ n_2m_1 ] . \end{gathered}\ ] ] the second and third quadrilateral are different because @xmath196 and @xmath197 do not commute ( otherwise they would lie on the same line through @xmath123 $ ] which contradicts the regularity of the quadric @xmath198 ) . our proof will be finished as soon as we have shown that the first and the second or the first and the third quadrilateral are equal . for this , it is sufficient to show that @xmath174 = [ m_1n_1]$ ] or @xmath174 = [ n_1m_1]$ ] . at first , we argue that the primal part of @xmath174 $ ] equals the primal part of @xmath187 $ ] or of @xmath188 $ ] . because @xmath159 does not intersect @xmath55}$ ] , the projection on the primal part is a regular projectivity @xmath199 $ ] with centre @xmath55}$ ] . we denote projected objects by a prime , that is , we write @xmath200 , @xmath201 , @xmath202 etc . , for the primal parts of @xmath203 , @xmath189 , @xmath196 etc . the quadric @xmath132 is regular and ruled and so is its primal projection @xmath204 . hence , the point @xmath205 \in { { \mathcal{q}}}'$ ] is incident with precisely two lines contained in @xmath204 , say @xmath206 and @xmath207 . but @xmath205 \vee [ v'_1]$ ] is contained in @xmath204 . hence @xmath208 \in m'_1 $ ] or @xmath208 \in m'_2 $ ] . similarly , @xmath208 $ ] is also incident with one of the two lines contained in @xmath204 and incident with @xmath209 $ ] . thus , @xmath208 = [ m'_1n'_1]$ ] or @xmath208 = [ n'_1m'_1]$ ] . now we have to lift this result to the dual part and show that @xmath174 = [ m_1n_1]$ ] or @xmath174 = [ n_1m_1]$ ] . the alternative being similar , we assume @xmath208 = [ m'_1n'_1]$ ] . this means that we have two null quadrilaterals , one with vertices @xmath173 $ ] , @xmath174 $ ] , @xmath175 $ ] , @xmath176 $ ] and one with vertices @xmath187 $ ] , @xmath210 $ ] , @xmath211 $ ] , @xmath212 $ ] such that their primal projections are equal and corresponding sides intersect in the vertices @xmath181 $ ] , @xmath183 $ ] , @xmath182 $ ] , @xmath184 $ ] of a planar quadrilateral . now we wish to apply in with @xmath213}$ ] , @xmath214}$ ] , and @xmath215 in order to conclude @xmath174 = [ m_1n_1]$ ] . this is admissible if and only if the plane @xmath216 \vee [ h_1 ] \vee [ h_2 ] = [ 1 ] \vee [ m_1 ] \vee [ n_1]$ ] does not intersect the four - spaces @xmath173 \vee { [ { \varepsilon}{\mathbb{h}}]}$ ] , @xmath174 \vee { [ { \varepsilon}{\mathbb{h}}]}$ ] , @xmath175 \vee { [ { \varepsilon}{\mathbb{h}}]}$ ] , and @xmath176 \vee { [ { \varepsilon}{\mathbb{h}}]}$ ] . we consider the four - space @xmath174 \vee { [ { \varepsilon}{\mathbb{h}}]}= [ m_1n_1 ] \vee { [ { \varepsilon}{\mathbb{h}}]}$ ] . it does not intersect @xmath217 if and only if the linear combination @xmath218 with some basis @xmath219 of @xmath220 is trivial . by considering only primal parts , we see that implies @xmath221 but the points @xmath123 $ ] , @xmath205 $ ] , @xmath209 $ ] , @xmath222 $ ] are vertices of a quadrilateral contained in a regular ruled quadric and therefore not co - planar . thus @xmath223 and @xmath224 follows . the arguments for the remaining four - spaces are similar . now we proceed with a geometric characterisation of the images of rp and pr dyads . it is tempting to view them as limiting case of rr dyads where one of the revolute axes becomes `` infinite '' . however , this is not entirely justified because of the possibility of commuting joints a phenomenon which is trivial for rr dyads but relevant for rp or pr dyads . it happens precisely if the direction of the revolute axis and translation direction are linearly dependent . these dyads are special enough to deserve a name of their own . since their transformations are usually modelled by a `` cylindrical joint '' , we define : the projective span of the kinematic image of an rp or pr dyad is called an _ rp _ or _ pr space _ respectively . if rotation and translation commute , we speak of a _ cylinder space _ or _ c space . _ the following example demonstrates that rp and pr spaces can not be characterised by a limiting configuration of . [ ex:2 ] consider the three - space @xmath159 spanned by @xmath123 $ ] , @xmath181 $ ] , @xmath183 $ ] and @xmath225 $ ] where @xmath226 it is straightforward to verify that the line @xmath181 \vee [ s_1]$ ] is a null line contained in @xmath55}$ ] and the line @xmath183 \vee [ s_1]$ ] and its conjugate complex line are null lines not contained in @xmath55}$ ] . the three - space @xmath159 contains all rotations about @xmath4 and all translations in direction of @xmath4 . the corresponding rp space is generated by the regular quadric with parametric equation @xmath227 where @xmath228 and @xmath229 both range in @xmath149 . this quadric is not contained in @xmath159 which is hence neither an rp nor a pr spac . it is crucial to above example that @xmath196 and @xmath197 commute . this can also be seen in our characterisation of rp and pr spaces where c spaces require a special treatment . [ th:3 ] a three - space @xmath131 is an rp but not a c space if and only if it * intersects the study quadric in a regular ruled quadric @xmath132 , * intersects the exceptional three - space @xmath55}$ ] in a straight line @xmath230 , and * contains a pair of conjugate complex null lines @xmath87 , @xmath91 that intersect @xmath230 in points @xmath225 $ ] , @xmath231 $ ] respectively , such that @xmath232 $ ] and @xmath233 $ ] are right rulings of @xmath57 . rp spaces that are not c spaces are characterised by the same conditions but with `` left rulings '' instead of `` right rulings '' . necessity of the first and second condition of have been proved in the master thesis by stigger @xcite by means of grbner basis computations . here , we give an independent proof . [ lem:4 ] the conditions of are necessary for an rp or pr space that is not a c space . assuming without loss of generality @xmath123 \in u$ ] , the constraint variety of an rp chain can be parameterised as @xmath234 with a non - zero quaternion @xmath15 with @xmath235 and a dual quaternion @xmath236 that satisfies @xmath237 , @xmath238 . since it is not the constraint variety of a pr chain , the primal part @xmath239 of @xmath236 and @xmath15 do not commute , that is @xmath240 \neq [ p]$ ] . by the same arguments as in , this constraint variety is a regular quadric : the parameters @xmath228 and @xmath229 range in @xmath149 ( with @xmath150 corresponding to zero rotation angle or translation distance ) but we will also admit complex values for them . expanding yields @xmath241 . we see that the kinematic image of this rp dyad lies in the three - space spanned by @xmath123 $ ] , @xmath242 $ ] , @xmath243 $ ] and @xmath244 = [ { \varepsilon}h'p]$ ] and is a regular quadric . clearly , this quadric it contains the line @xmath245 \vee [ { \varepsilon}hp ] \subset { [ { \varepsilon}{\mathbb{h}}]}$ ] . moreover , the lines @xmath87 , @xmath91 parameterised by @xmath246 are conjugate complex null lines . the intersection points @xmath183 $ ] , @xmath184 $ ] of these lines with @xmath177 correspond to @xmath247 , their intersection points @xmath225 $ ] , @xmath231 $ ] with @xmath55}$ ] correspond to @xmath248 : @xmath249 ( ) . in view of the null line property follows from @xmath250 and similar with @xmath107 replaced by @xmath251 . moreover @xmath252 ) = { \varphi}([n_1 ] ) = [ { \varepsilon}({\mathrm{i}}- h')]$ ] and @xmath225 = [ { \varepsilon}({\mathrm{i}}-h')p]$ ] . because of @xmath240 \neq [ p]$ ] , these points do not coincide and , indeed , span a right ruling of @xmath57 . the statements for pr spaces follow from by application of the conjugation map @xmath119 . it interchanges rp spaces with pr spaces and right rulings with left rulings but fixes the study quadric and the null cone . the proof of sufficiency of the conditions in is , in large parts , a copy of . however , at one point the third condition comes into play . [ lem:5 ] a three - space @xmath159 that satisfies all conditions of is a pr or rp space but not a c space . once more , we assume @xmath123 \in u$ ] . the intersection of @xmath159 and @xmath56 is a regular quadric @xmath132 that contains the straight line @xmath253}$ ] and the two conjugate complex null lines @xmath87 , @xmath91 . there exist points @xmath181 \in e_1 $ ] and @xmath183 \in \ell_1 $ ] such that @xmath123 \vee [ m_1]$ ] and @xmath123 \vee [ n_1]$ ] are rulings of @xmath132 . the points of the line @xmath123 \vee [ m_1]$ ] correspond to translations with fixed direction and the points of @xmath123 \vee [ n_1]$ ] to rotations around a fixed axis . there exists an intersection point @xmath225 \coloneqq e_1 \cap \ell_1 $ ] and we have to show that either @xmath225 = [ m_1n_1]$ ] or @xmath225 = [ n_1m_1]$ ] . in general , this can be done just as in our proof of : denote by @xmath254 the complex conjugate to @xmath196 and by @xmath255 the complex conjugate of @xmath197 . then @xmath123 = ( [ m_1 ] \vee [ m_2 ] ) \cap ( [ n_1 ] \vee [ n_2])$ ] , the quadrilaterals with respective vertices @xmath187 $ ] , @xmath210 $ ] , @xmath211 $ ] , @xmath212 $ ] and @xmath188 $ ] , @xmath256 $ ] , @xmath257 $ ] , @xmath258 $ ] are contained in @xmath132 , and the points @xmath181 $ ] , @xmath183 $ ] , @xmath182 $ ] , and @xmath184 $ ] form a planar quadrilateral . because the intersection of @xmath55}$ ] and @xmath159 is not empty , we have to use a projection with three - dimensional centre @xmath259 that does not intersect @xmath159 ( and hence differs from @xmath55}$ ] ) and is such that the plane @xmath123 \vee [ m_1 ] \vee [ n_1]$ ] is complementary to the four - spaces @xmath260 \vee z,\quad [ v_1 ] \vee z,\quad [ u_2 ] \vee z,\quad [ v_2 ] \vee z\ ] ] when appealing to . such a choice is possible and then the differences to the proof of are irrelevant _ unless @xmath196 and @xmath197 commute . _ in this case our argument fails because the two spatial quadrilaterals coincide and we may no longer conclude that the projection of @xmath225 $ ] equals the projection of @xmath187 $ ] or of @xmath188 $ ] . fortunately , this case is already special enough to allow a straightforward computational treatment . the dual quaternions @xmath196 and @xmath197 commute ( @xmath261 ) if the translation direction of @xmath196 and the direction of the rotation axis of @xmath197 are linearly dependent . without loss of generality , we may set @xmath262 and @xmath263 so that @xmath264 . now @xmath159 equals the span of @xmath123 $ ] , @xmath181 $ ] , @xmath183 $ ] , and @xmath225 = [ m_1n_1 ] = [ n_1m_1]$ ] but @xmath265 ) = [ { \varepsilon}({\mathrm{i}}+{\mathbf{i } } ) ] = [ m_1n_1]$ ] a contraction to the assumption that @xmath225 $ ] and @xmath266)$ ] span a straight line . above proof already contains the basic idea for the still missing characterisation of c spaces . [ th:4 ] a three - space @xmath131 is a c space if and only if it * intersects the study quadric in a regular ruled quadric @xmath132 , * contains a pair of conjugate complex null lines @xmath87 , @xmath91 and a real null line @xmath230 , and * satisfies @xmath267 . lets assume that @xmath159 is a c space . without loss of generality , we may assume that it is spanned by @xmath123 $ ] , @xmath181 = [ { \varepsilon}{\mathbf{i}}]$ ] , @xmath183 = [ { \mathrm{i}}+ { \mathbf{i}}]$ ] and @xmath225 = [ m_1n_1 ] = { \varphi}([n_1])$ ] . clearly , it contains the real line @xmath268 \vee [ s_1]$ ] , the null line @xmath269 \vee [ n_1]$ ] , and its complex conjugate @xmath91 . a parametric equation for the intersection quadric @xmath132 of @xmath159 and @xmath56 is @xmath270 it is a regular ruled quadric and its fiber projection equals @xmath271 is parameterized by @xmath272 . this is indeed the straight line @xmath230 . conversely , assume that the condition of the theorem hold and , without loss of generality , @xmath123 \in u$ ] . as in the proof of , we can find points @xmath181 $ ] , @xmath183 $ ] , @xmath225 \in u$ ] such that @xmath268 \vee [ s_1]$ ] , @xmath273 \vee [ s_1]$ ] , and the lines @xmath123 \vee [ m_1]$ ] and @xmath123 \vee [ n_1]$ ] are rulings of @xmath132 . after a suitable choice of coordinates , we may assume @xmath274 whence @xmath275 ) \vee { \varphi}([n_1 ] ) = [ { \varepsilon } ] \vee [ { \varepsilon}({\mathrm{i}}+ { \mathbf{i } } ) ] = [ { \varepsilon } ] \vee [ { \varepsilon}{\mathbf{i}}].\ ] ] but then @xmath181 = [ { \varepsilon}{\mathbf{i}}]$ ] because @xmath181 \in e_1 $ ] and @xmath123 \vee [ m_1]$ ] is contained in @xmath56 . in particular , @xmath187 = [ n_1m_1 ] = [ { \varepsilon}({\mathrm{i}}{\mathbf{i}}- 1 ) ] = [ { \varepsilon}({\mathrm{i}}+{\mathbf{i}})]$ ] . moreover , there exist scalars @xmath276 and @xmath277 such that @xmath278 . the condition @xmath279 implies @xmath280 so that @xmath225 = [ { \varepsilon}({\mathrm{i}}+ { \mathbf{i } } ) ] = [ m_1n_1 ] = [ n_1m_1]$ ] and @xmath159 is indeed a c space . in this section we illustrate the usefulness of at hand of an example . in pfurner et al . @xcite the authors consider the map but without imposing the study condition ( `` extended kinematic map '' ) . the action of a dual quaternion on points is still a rigid body transformation but the map is no longer an injection . pfurner et al . ( 2016 ) show that the fibers of ( the set of pre - images of a fixed rigid body displacement ) is a straight line . more precisely , the fiber incident with a point @xmath281 = [ p ' + { \varepsilon}p '' ] \in p^7 \setminus { [ { \varepsilon}{\mathbb{h}}]}$ ] is the straight line @xmath282 \vee [ { \varepsilon}p '' ] = { \varphi}([p])$ ] . this line intersects the study quadric in @xmath243 $ ] and one further point which gives the corresponding displacement in the classical kinematic map ( which also the underlying concept in this paper ) . pfurner et al . ( 2016 ) also mention that the motion corresponding to a generic straight line in @xmath26 is a darboux motion @xcite , that is , the motion obtained by composing a planar elliptic motion with a harmonic oscillation in orthogonal direction . this is fairly straightforward to see . by virtue of , all trajectories are rational curves degree two or less which already limits possible candidates to darboux motions , rotations with fixed axis , and curvilinear translations along rational curves of degree two or less . we will now prove that the motion of a generic straight line @xmath283 is actually a _ vertical _ darboux motion . in this limiting case of the generic darboux motion the elliptic translation is replaced by a rotation with fixed axis . our proof is based on the fact that the three - space spanned by @xmath283 and its fiber @xmath284 is a c space . [ th:5 ] if a straight line @xmath285 contains a point @xmath281 \notin { [ { \varepsilon}{\mathbb{h}}]}$ ] and is neither contained in the null cone @xmath52 nor in the four - dimensional subspace @xmath286}\coloneqq [ p ] \vee { [ { \varepsilon}{\mathbb{h}}]}$ ] , the span of @xmath283 and @xmath284 is a c space . without loss of generality we assume @xmath281 = [ 1]$ ] . then @xmath286}$ ] is the space of all translations . the straight line @xmath283 is neither contained in nor tangent to the null cone @xmath52 ( because it is real and does not intersect @xmath55}$ ] ) . hence , it intersects @xmath52 in two points @xmath287 = [ a ' + { \varepsilon}a'']$ ] , @xmath288 = [ b ' + { \varepsilon}b'']$ ] . they satisfy the null cone condition @xmath289 . because @xmath123 \in [ a ] \vee [ b]$ ] , we may write @xmath290 . note that @xmath103 and @xmath104 are complex conjugates but we will not make direct use of this , thus avoiding possible confusion arising from mixing of complex and quaternion conjugation . if the vectors @xmath291 and @xmath292 are linearly dependent , there exists a linear combination @xmath293 of @xmath103 and @xmath104 with @xmath294 . but then @xmath123 \neq [ { \varepsilon}c '' ] \in { [ t]}$ ] and , contrary to our assumption , @xmath295 \vee [ { \varepsilon}c '' ] \subset { [ t]}$ ] . hence , @xmath291 and @xmath292 are linearly independent . this implies that the points @xmath287 = [ a'+{\varepsilon}a'']$ ] , @xmath288 = [ { \overline{a ' } } - { \varepsilon}a'']$ ] , @xmath296 ) = [ { \varepsilon}a']$ ] , and @xmath297 ) = [ { \varepsilon}{\overline{a'}}]$ ] span a three - dimensional space @xmath159 . now we compute the intersection of @xmath159 with @xmath56 and @xmath52 . we set @xmath298 and solve the equation @xmath299 for @xmath300 $ ] . the left - hand side is a dual number . equating its primal part with zero yields @xmath301 note that @xmath302 is a real number and it is different from zero because of linear independence of @xmath291 and @xmath292 . hence @xmath303 or @xmath304 . we further abbreviate @xmath305 ( these are real numbers as well ) and equate the dual part of @xmath299 with zero : @xmath306 in case of @xmath304 , this boils down to @xmath307 while we get @xmath308 if @xmath303 . from these considerations we infer that @xmath309 consists of precisely three straight lines : the line @xmath230 given by @xmath300 = [ 0,0,\gamma,\delta]$ ] which is contained in the exceptional generator @xmath55}$ ] , and the two lines @xmath87 , @xmath91 given by @xmath310 = [ -f,0,\gamma , g_1 ] , \quad\text{and}\quad [ \alpha,\beta,\gamma,\delta ] = [ 0,f , g_2,\delta]\ ] ] respectively . because of @xmath311 , the lines @xmath87 and @xmath91 do not intersect and they are complex conjugates by construction . this already implies that @xmath312 is a ruled quadric but neither a quadratic cone nor a double plane ( but we still have to argue that it is not a pair of planes ) . for @xmath313 and @xmath314 or @xmath315 we obtain the intersection points @xmath316 = [ { \varepsilon}a ' ] \quad\text{and}\quad [ s_2 ] = [ { \varepsilon}{\overline{a'}}]\ ] ] of @xmath87 and @xmath91 , respectively , with @xmath230 . there is a unique line @xmath317 through @xmath123 $ ] that intersects both @xmath87 and @xmath91 . its intersection points @xmath183 $ ] , @xmath184 $ ] with @xmath87 and @xmath91 , respectively , are given by @xmath310 = [ -f,0,g_2,g_1 ] \quad\text{and}\quad [ \alpha,\beta,\gamma,\delta ] = [ 0,f , g_2,g_1].\ ] ] the line @xmath183 \vee [ n_2]$ ] is a fourth ruling in the intersection of @xmath159 and @xmath56 and does not intersect @xmath318 $ ] . hence @xmath159 and @xmath56 can not intersect in a pair of planes . the quadric @xmath319 is given by the equations @xmath298 where the parameters @xmath276 , @xmath277 , @xmath95 , and @xmath320 satisfy @xmath321 . from this we see that that its fiber projection is given by @xmath322 $ ] . it , indeed , coincides with @xmath230 so that also the last condition of is fulfilled . [ cor:1 ] a straight line @xmath283 satisfying the assumptions of corresponds , via , to a _ vertical _ darboux motion . since the map generically doubles the degree , the trajectories of the motion in question are at most quadratic and the motion itself is either a curvilinear translation along a curve of degree two or less , the rotation about a fixed axis or a darboux motion . pure translations are excluded by the assumptions of , non - vertical darboux motions by its conclusion . is also valid for straight lines contained in the study quadric @xmath56 unless they intersect the exceptional generator @xmath55}$ ] . these lines are known to describe rotations about a fixed axis ( selig 2005 , section 11.2 ) @xcite , that is , a vertical darboux motion with zero amplitude . in this sense , we may say that _ all straight lines in extended kinematic image space that intersect the null cone in two distinct points describe a vertical darboux motion . _ we provided necessary and sufficient geometric conditions on the transformation group induced by coordinate changes in the fixed and moving frame and on the kinematic image of 2r , rp , and pr dyads . while many properties of these objects are already known , sufficiency of certain collections of conditions seems to be new . we payed proper attention to the case of commuting joints ( which is trivial for rr dyads ) , highlighted the role of left and right rulings , and introduced the fiber projectivity . it is important in our characterisation of rp and pr spaces but also for the kinematic images of darboux motions , as illustrated in and . in this appendix we prove two technical results of purely projective nature and with no direct relations to kinematics . the formulation of could be simplified but in its present form its application in the proof of is apparent . [ lem:6 ] given a three - space @xmath323 and four points @xmath324 $ ] , @xmath208 $ ] , @xmath325 $ ] , @xmath326 $ ] that span a three - space @xmath327 with @xmath328 , set @xmath329 \vee e$ ] , @xmath330 \vee e$ ] , @xmath331 \vee e$ ] , @xmath332 \vee e$ ] and consider four projections @xmath333 with respective centres @xmath181 $ ] , @xmath183 $ ] , @xmath182 $ ] , and @xmath184 $ ] . if the centres are pairwise different and span a plane @xmath217 that is complementary to @xmath334 , @xmath335 , @xmath336 , and @xmath337 , respectively , the composition @xmath338 of the four projections is the identity . note that @xmath339 so that the projection @xmath340 ( and also @xmath341 , @xmath342 , @xmath343 ) from a suitable point is well defined . take an arbitrary point @xmath173 \in u_1 $ ] but not in @xmath344 and set @xmath174 \coloneqq \zeta_1([u_1])$ ] , @xmath175 \coloneqq \eta_1([v_1])$ ] , @xmath176 \coloneqq \zeta_2([u_2])$ ] . then the projective space @xmath345 \vee [ v_1 ] \vee [ u_2 ] \vee [ v_2 ] = [ u_1 ] \vee [ m_1 ] \vee [ n_1 ] \vee [ m_2 ] \vee [ n_2]$ ] ( ) is of dimension three and does not intersect @xmath344 because of the dimension formula @xmath346 \vee l \vee e ) = 3 + 3 - 7 = -1 . \end{aligned}\ ] ] therefore , we may take @xmath173 $ ] , @xmath174 $ ] , @xmath175 $ ] and @xmath176 $ ] in that order as base points of a projective coordinate system in @xmath347 and complement them by four further base points in @xmath344 and a suitable unit point to a projective coordinate system of @xmath26 . we select the unit point such that its projection into @xmath347 from @xmath344 gives @xmath348 \vee [ m_2 ] ) \cap ( [ n_1 ] \vee [ n_2])$ ] . then , the projection centres are @xmath180 = [ 1,1,0,0,0,0,0,0],\quad [ n_1 ] = [ 0,1,1,0,0,0,0,0],\\ [ m_2 ] = [ 0,0,1,1,0,0,0,0],\quad [ n_2 ] = [ 1,0,0,1,0,0,0,0 ] . \end{gathered}\ ] ] we pick an arbitrary point @xmath349 \in u_1 $ ] and compute @xmath350 \eqqcolon z_1,\\ \eta_1(z_1 ) & = [ 0,0,x_0,0,x_4,x_5,x_6,x_7 ] \eqqcolon h_1,\\ \zeta_2(h_1 ) & = [ 0,0,0,-x_0,x_4,x_5,x_6,x_7 ] \eqqcolon z_2,\\ \eta_2(z_2 ) & = [ x_0,0,0,0,x_4,x_5,x_6,x_7 ] = h_2 . \end{aligned}\ ] ] this concludes the proof . [ lem:7 ] let @xmath344 , @xmath327 be non - intersecting three - spaces , the latter spanned by points @xmath324 $ ] , @xmath208 $ ] , @xmath325 $ ] , @xmath326 $ ] . consider further a regular quadric @xmath132 containing @xmath344 and four vertices @xmath181 $ ] , @xmath183 $ ] , @xmath182 $ ] , @xmath184 \in { { \mathcal{q}}}$ ] of a quadrilateral in a plane @xmath217 that is complementary to @xmath351,\quad e \vee [ v'_1],\quad e \vee [ u'_2],\quad\text{and}\quad e \vee [ v'_2].\ ] ] then there exists a unique spatial quadrilateral with vertices @xmath173 $ ] , @xmath174 $ ] , @xmath175 $ ] , @xmath176 $ ] that is contained in @xmath132 , whose sides @xmath173 \vee [ v_1]$ ] , @xmath174 \vee [ u_2]$ ] , @xmath175 \vee [ v_2]$ ] , @xmath176 \vee [ u_1]$ ] are , in that order , incident with @xmath181 $ ] , @xmath183 $ ] , @xmath182 $ ] , @xmath184 $ ] and whose projection from @xmath344 into @xmath352 is the quadrilateral with vertices @xmath324 $ ] , @xmath208 $ ] , @xmath325 $ ] , and @xmath326 $ ] . denote the quadratic form associated to @xmath132 by @xmath353 . given @xmath324 $ ] , @xmath208 $ ] , @xmath325 $ ] , @xmath326 $ ] , we have to reconstruct @xmath173 $ ] , @xmath174 $ ] , @xmath175 $ ] , @xmath176 $ ] subject to the constraints @xmath354 and @xmath355 \in [ u_1 ] \vee [ v_1],\quad [ n_1 ] \in [ v_1 ] \vee [ u_2],\\ [ m_2 ] \in [ u_2 ] \vee [ v_2],\quad [ n_2 ] \in [ v_2 ] \vee [ u_1 ] . \end{gathered}\ ] ] if @xmath173 $ ] is given , we find @xmath174 $ ] by projecting @xmath173 $ ] from centre @xmath181 $ ] onto @xmath356 $ ] . tells us that we can find @xmath175 $ ] and @xmath176 $ ] in similar manner such that is satisfied . then , some of the conditions in become redundant and it is sufficient to consider only @xmath357 in a projective coordinate system with base points @xmath324 $ ] , @xmath208 $ ] , @xmath325 $ ] , @xmath326 \in f$ ] and further base points in @xmath344 , the quadratic form @xmath353 of the quadric @xmath132 is described by a matrix of the shape @xmath358 where @xmath63 , @xmath64 and @xmath69 are matrices of dimension @xmath48 , @xmath69 is the zero matrix and @xmath64 is regular . now gives rise to a linear system for the unknown coordinates of @xmath173 $ ] , @xmath174 $ ] , @xmath175 $ ] , and @xmath176 $ ] . we have @xmath359 with certain scalars @xmath340 , @xmath341 , @xmath342 that can be computed from @xmath200 , @xmath201 , @xmath360 , @xmath361 and @xmath196 , @xmath197 , @xmath254 , @xmath255 alone . now we write @xmath173 = [ 1,0,0,0,x_0,x_1x_2,x_3]$ ] with unknown @xmath362 , @xmath363 , @xmath364 , @xmath365 . then , yields @xmath366 \cdot b [ x_0,x_1,x_2,x_3]^{\intercal},\\ 0 & = [ 0,1,0,0 ] \cdot b ( [ x_0,x_1,x_2,x_3 ] + \zeta_1m''_1)^{\intercal},\\ 0 & = [ 0,0,1,0 ] \cdot b ( [ x_0,x_1,x_2,x_3 ] + \zeta_1m''_1 + \eta_1 n''_1)^{\intercal},\\ 0 & = [ 0,0,0,1 ] \cdot b ( [ x_0,x_1,x_2,x_3 ] + \zeta_1m''_1 + \eta_1 n''_1 + \zeta_2m''_2)^{\intercal}\end{aligned}\ ] ] where the double prime denotes projection on the last four coordinates . is a linear system for @xmath362 , @xmath363 , @xmath364 , @xmath365 with regular coefficient matrix @xmath64 . hence , the solution is unique . this work was supported by the austrian science fund ( fwf ) : p26607 ( algebraic methods in kinematics : motion factorisation and bond theory ) . m. pfurner , h .- schrcker , and m. husty : `` path planning in kinematic image space without the study condition . '' in j. lenari and j .- p . merlet , editors , `` proceedings of advances in robot kinematics , '' ( 2016 ) . url https://hal.archives-ouvertes.fr/hal-01339423 . rad and h .- schrcker : `` the kinematic image of 2r dyads and exact synthesis of 5r linkages . '' in `` proceedings of the i m a conference on mathematics of robotics , '' ( 2015 ) . isbn 978 - 0 - 90 - 509133 - 4 .

the results of vlbi total intensity ( ) and linear polarization ( ) observations at cm are presented for ten radio bright bl lacertae objects . these images complete first - epoch polarization observations for the 1-jy sample of northern bl lacertae objects defined by khr and schmidt . estimates of superluminal speeds are presented for several sources , bringing the total number of sources in the sample for which such estimates are available to 16 . second epoch observations currently being reduced should yield speed estimates for vlbi features in essentially all the sources in the sample . the jet magnetic fields of these bl lacertae objects are usually transverse to the local jet direction , but a sizeable minority ( about 30% ) have vlbi jet components with longitudinal magnetic fields . this may suggest that the conditions in the vlbi jets of bl lacertae objects are favorable for the formation of relativistic shocks ; alternatively , it may be that the toroidal component of the intrinsic jet magnetic field is characteristically dominant in these sources . bl lacertae objects : general bl lacertae objects : individual : ( 0003066 , 0814 + 425 , 0820 + 225 , 0823 + 033 , 1334127 , 1732 + 389 , 2131021 , 2150 + 173 , 2155152 , 2254 + 074 polarization radio sources : galaxies

astro-ph0307192

bl lacertae objects are active galactic nuclei whose most characteristic distinguishing property is their relatively low - luminosity optical line emission ; in many cases , their optical continua are completely featureless . like many high - polarization quasars , bl lacertae objects have strong and variable polarization in wavebands ranging from optical through radio ; they usually have compact , flat - spectrum radio structure , and point - like optical structure . for some , luminous elliptical host galaxies are observed ( angel and stockman 1980 ; miller 1981 ; kollgaard 1994 ) , though the optical images of many bl lacertae objects remain unresolved , even in high - resolution observations ( e.g. falomo 1996 ) . the radio emission and much of the optical emission is believed to be synchrotron radiation . historically , bl lacertae objects were first detected primarily via radio surveys , and strong radio emission was earlier thought to be characteristic of this type of agn . more recently , large numbers of bl lacertae objects with much weaker radio emission have been discovered by x - ray surveys . the relationship between these so - called `` radio '' and x - ray " bl lacertae objects is not entirely clear ; the most popular current hypotheses are ( 1 ) that they are similar objects whose spectral energy distributions peak in the infrared and x - ray , respectively , with the radio emission in radio bl lacertae objects experiencing a larger intrinsic relativistic enhancement ( giommi & padovani 1994 ; padovani & giommi 1995 ; fossati et al . 1997 ) , and ( 2 ) that they are intrinsically identical objects whose jets are oriented at different characteristic angles to the line of sight , with x - ray bl lacertae objects being viewed further `` off - axis '' ( stocke et al . 1985 ; maraschi et al . 1986 ) . in the broader context of unified schemes , it is usually thought that the `` parent population '' of bl lacertae objects is primarily fr i radio galaxies ( browne 1983 ; wardle , moore & angel 1984 ) . previous vlbi polarization observations of radio bl lacertae objects at 6 cm and 3.6 cm ( gabuzda & cawthorne 1996 , gabuzda et al . 1999 , and references therein ) have revealed a tendency for the electric vector @xmath3 in polarized knots in the vlbi jets to lie nearly along the local jet direction . the degrees of polarization in the jet components of bl lacertae objects have been observed to be as high as @xmath4 , with typical values @xmath5 , indicating that these components are optically thin and that in at least some cases the magnetic field is very highly ordered . assuming the jet components to be optically thin , the observed typical @xmath3 orientation implies that the associated magnetic fields are perpendicular to the direction of the jet . one natural interpretation of this transverse magnetic field structure is that the visible jet components are associated with relativistic shocks that compress an initially tangled magnetic field , enhancing the magnetic field transverse to the compression ( laing 1980 ; hughes , aller and aller 1989 ) . as images for more bl lacertae objects have become available , it has become increasingly clear that there are also a sizeable minority of these sources in which , in contrast , longitudinal magnetic fields dominate in at least some parts of their vlbi jets . lccccc antenna & diameter & t@xmath6 & t@xmath7 & @xmath8 & epoch + & ( m ) & ( k ) & ( k ) & ( k / jy ) + medicina & 32 & 60 & 50 & 0.160 & 1 , 2 + effelsberg & 100 & 60 & 50 & 0.991 & 1 , 2 + wsrt & @xmath925 & 120 & 120 & & 2 + st . croix & 25 & 45 & 35 & 0.102 & 2 + hancock & 25 & 55 & 60 & 0.092 & 2 + haystack & 37 & 75 & 65 & 0.130 & 1 + green bank & 43 & 35 & 40 & 0.227 & 1 , 2 + north liberty & 25 & 45 & 45 & 0.120 & 1 , 2 + phased vla & @xmath1025 & * * & * * & * * & 1 + kitt peak & 25 & 65 & 65 & 0.120 & 1 + owens valle & 25 & 110 & 110 & 0.120 & 1 , 2 + brewster & 25 & 35 & 35 & 0.088 & 2 + mauna kea & 25 & 35 & 35 & 0.090 & 2 + + at 6 cm , the degrees of polarization of the cores of bl lacertae objects ( @xmath11 ) are , on average , higher than those of quasars ( @xmath12 ) . recent 6 cm space vlbi polarization observations of the bl lacertae object 1803 + 784 ( gabuzda 1999 ) indicate that this is due to the fact that the observed core polarizations include a substantial contribution from newly emerging knots , as suggested by gabuzda et al . ( 1994b ) . previous observations have shown the distribution of core @xmath3 orientations at 6 cm to be bimodal , with @xmath3 either aligned with or transverse to the inner jet direction ; there is some evidence that @xmath13 is roughly perpendicular to the jet when the cores are quiescent , and aligns with the jet at epochs when emission from newly emerging shock components dominate the observed `` core '' polarization ( gabuzda et al . 1994b , gabuzda & cawthorne 1996 ) . to elucidate the nature of the characteristic vlbi total intensity ( @xmath0 ) and polarization ( @xmath1 ) structures observed for radio - bright bl lacertae objects , we are engaged in an ongoing project to obtain multi - epoch , multi - frequency vlbi @xmath0 and @xmath1 images for all sources in the complete sample of bl lacertae objects defined by khr and schmidt ( 1990 ) . these sources have 6 cm fluxes of at least 1 jy , radio spectral indices @xmath14 ( @xmath15 ) , rest frame equivalent widths of the strongest emission lines less than 5 , and optical counterparts on the palomar sky survey plates with brightness greater than 20@xmath16 . vlbi @xmath0 and @xmath1 images for most of the sample sources are presented by roberts , gabuzda & wardle ( 1987 ) ; gabuzda , roberts & wardle ( 1989b ) ; gabuzda et al . ( 1992 , 1994b ) ; and gabuzda , pushkarev & cawthorne ( 1999 ) . the ten sets of @xmath0 and @xmath1 images presented here complete our first - epoch images for the 34 sources in the original sample ( one , 1334127 , was subsequently reclassified as a quasar , so that the final number of sources in the complete bl lac sample is 33 ) . these are the first @xmath1 images for these ten sources , and in a number of cases , the first 6 cm @xmath0 images as well . [ tab : rm ] lccc source & rm & rm ref & rot . at @xmath176 cm + & ( rad / m@xmath18 ) & & ( deg ) + 0003@xmath19066 & + 13 & 2 & + 3 + 0814 + 425 & + 23 & 1 & + 5 + 0820 + 225 & + 81 & 2 & + 17 + 0823 + 033 & 1 & 2 & + 1 + 1334@xmath19127 & @xmath20 & 2 & @xmath21 + 1732 + 389 & + 67 & 2 & + 14 + 2131@xmath19021 & + 46 & 2 & + 9 + 2150 + 173 & @xmath22 & 2 & -8 + 2155@xmath19152 & + 19 & 2 & + 4 + 2254 + 074 & @xmath23 & 2 & @xmath24 + + + the observations of all sources except 0003066 were made in march 1992 ( 1992.23 ) , using an 8-element global vlbi array . the observations for 0003066 were made in may 1995 ( 1995.41 ) using a 10-element global array . the antennas used in both experiments are listed in table [ tab : ants ] . the observations were made under the auspices of the us and european vlbi networks . the data were recorded using the mkiii system , and the data were subsequently correlated using the mk iiia correlators at haystack observatory ( 1992.23 ) and the max - planck - institut - fr - radioastronomie in bonn ( 1995.41 ) . the polarization calibration of these data was performed as described by roberts , wardle and brown ( 1994 ) . the instrumental polarizations ( `` d - terms '' ) were determined using observations of the unpolarized sources 3c84 and oq208 ; the overall d - terms residuals are at a level @xmath25 . the vlbi polarization position angles were calibrated by comparing the total vlbi - scale and vla core polarizations for the especially compact and relatively strongly polarized sources 0823 + 033 and 1749 + 096 . the rotations required to align @xmath3 for the total vlbi - scale polarizations of these sources with @xmath3 for their vla core polarizations agreed to within less than one degree , suggesting that the overall vlbi polarization position angle calibration is good to about this level of precision . images of the distribution of total intensity @xmath0 were made using a self - calibration algorithm similar to that described by cornwell and wilkinson ( 1981 ) . maps of the linear polarization , where @xmath26 is the polarized intensity , @xmath27 is the fractional linear polarization , and @xmath3 is the position angle of the electric vector on the sky , measured from north through east . ] @xmath1 were made by referencing the calibrated cross - hand fringes to the parallel - hand fringes using the antenna gains determined in the hybrid mapping , fourier transforming the cross - hand fringes , and performing a complex clean . one byproduct of this procedure is to register the @xmath0 and @xmath1 maps to within a small fraction of a beamwidth , so that corresponding @xmath0 and @xmath1 images may be directly superimposed . cccccccccc & @xmath0 & @xmath28 & @xmath29@xmath30 & @xmath27 & @xmath31 & @xmath32 & @xmath33 & @xmath34 & fwhm + & ( mjy ) & ( mjy ) & ( deg ) & ( % ) & ( mas ) & ( mas ) & ( deg ) & ( deg ) & ( mas ) + + c & 922 & 63.3 & 79 & 6.8 & & & & & 0.25 + k6 & 824 & @xmath35 & & @xmath36 & 0.60 & & @xmath37 & & 0.23 + k5 & 372 & 55.8 & 21 & 13.6 & 1.46 & & @xmath38 & & 0.46 + k4 & 150 & @xmath35 & & @xmath39 & 2.51 & & @xmath40 & & 1.94 + k3 & 11 & @xmath35 & & @xmath39 & 4.31 & & @xmath41 & & 0.02 + k2 & 15 & @xmath35 & & @xmath42 & 5.23 & & @xmath43 & & 0.33 + k1 & 97 & @xmath35 & & @xmath44 & 6.10 & & @xmath45 & & 1.15 + + c & 620 & @xmath46 & & @xmath47 & & & & & 0.48 + k5 & 200 & 22.2 & 26 & 11.1 & 0.99 & 0.01 & 128 & 1 & 0.50 + k4 & 35 & 9.2 & 84 & 26.2 & 2.29 & 0.02 & 111 & 1 & 0.92 + k3 & 11 & @xmath46 & & @xmath48 & 3.56 & 0.03 & 108 & 1 & 0.74 + k2 & 19 & @xmath46 & & @xmath49 & 4.89 & 0.07 & 101 & 2 & 1.48 + k1 & 10 & @xmath46 & & @xmath50 & 6.34 & 0.13 & 106 & 2 & 1.08 + + c & 198 & @xmath46 & & @xmath51 & & & & & 0.20 + k11 & 119 & 7.4 & @xmath52 & 6.2 & 2.16 & 0.07 & @xmath53 & 1 & 0.80 + k10 & 43 & 8.0 & @xmath54 & 18.6 & 4.26 & 0.17 & @xmath55 & 1 & 1.03 + k9 & 47 & @xmath46 & & @xmath56 & 5.84 & 0.13 & @xmath57 & 1 & 1.51 + k8 & 68 & 12.2 & 16 & 17.9 & 8.86 & 0.09 & @xmath58 & 1 & 2.28 + k7 & 22 & 9.2 & @xmath21 & 41.8 & 11.33 & 0.16 & @xmath59 & 1 & 1.52 + k6 & 113 & 7.2 & @xmath60 & 6.4 & 12.84 & 0.07 & @xmath61 & 1 & 2.95 + k5 & 90 & 20.8 & @xmath62 & 23.1 & 17.45 & 0.08 & @xmath59 & 1 & 2.87 + k4 & 13 & @xmath46 & & @xmath63 & 19.99 & 0.10 & @xmath64 & 1 & 0.24 + k3 & 165 & 24.1 & @xmath40 & 14.6 & 21.70 & 0.06 & @xmath65 & 1 & 1.76 + k2 & 60 & 9.7 & @xmath66 & 16.2 & 23.38 & 0.08 & @xmath59 & 1 & 1.27 + k1 & 180 & @xmath46 & & @xmath67&24.81 & 0.07 & @xmath68 & 1 & 2.69 + + c & 1494 & 60.6 & @xmath69 & 4.0 & & & & & 0.07 + k3 & 415 & @xmath70 & & @xmath71 & 0.71 & 0.03 & 30 & 3 & 0.18 + k2 & 29 & @xmath70 & & @xmath72 & 3.16 & 0.31 & 31 & 4 & 0.47 + k1 & 25 & @xmath70 & & @xmath73 & 4.81 & 0.23 & 28 & 2 & 0.23 + + c & 3822 & 67.0 & 83 & 1.8 & & & & & 0.14 + k3 & 201 & 37.7 & 54 & 18.8 & 2.25 & 0.06 & 147 & 1 & 0.20 + k2 & 56 & @xmath44 & & @xmath74 & 2.26 & 0.14 & 110 & 9 & 0.01 + k1 & 26 & @xmath44 & & @xmath75 & 7.18 & 0.23 & 120 & 3 & 0.20 + + c & 923 & 18.7 & 5 & 2.0 & & & & & 0.32 + k3 & 116 & 8.0 & 33 & 6.8 & 0.80 & 0.07 & 103 & 7 & 0.08 + k2 & 21 & @xmath76 & & @xmath77 & 2.48 & 0.30 & 112 & 3 & 0.15 + k1 & 10 & @xmath76 & & @xmath78 & 3.67 & 0.19 & 108 & 5 & 0.66 + + c & 1011 & 18.822.5 & @xmath79 & 1.82.2 & & & & & 0.11 + k5 & 344 & @xmath80 & & @xmath67 & 0.41 & 0.03 & 100 & 6 & 0.13 + k4 & 19 & @xmath80 & & @xmath81 & 1.77 & 0.30 & 70 & 25 & 0.14 + k3 & 18 & @xmath80 & & @xmath82 & 2.82 & 0.09 & 75 & 6 & 0.14 + k2 & 12 & @xmath80 & & @xmath83 & 5.39 & 0.09 & 86 & 10 & 0.14 + k1 & 17 & @xmath80 & & @xmath84 & 6.58 & 0.04 & 88 & 3 & 0.12 + + c & 207 & 21.446.9 & 52.0 & 10.322.6 & & & & & 0.16 + k & 844 & @xmath85 & & @xmath86 & 1.55 & 0.04 & @xmath87 & 1 & 0.28 + k & 897 & 49.3 & @xmath88 & 5.5 & 4.52 & 0.05 & @xmath89 & 1 & 1.50 + k1 & 39 & @xmath85 & & @xmath74 & 9.41 & 0.20 & @xmath90 & 1 & 0.05 + + + + + + + cccccccccc & @xmath0 & @xmath28 & @xmath29@xmath30 & @xmath27 & @xmath31 & @xmath32 & @xmath33 & @xmath34 & fwhm + & ( mjy ) & ( mjy ) & ( deg ) & ( % ) & ( mas ) & ( mas ) & ( deg ) & ( deg ) & ( mas ) + + c & 222 & 11.5 & @xmath91 & 5.2 & & & & & 0.13 + k5 & 137 & 3.2 & @xmath20 & 2.3 & 0.76 & 0.02 & @xmath92 & 3 & 0.79 + k4 & 26 & @xmath93 & & @xmath85 & 1.51 & 0.10 & @xmath45 & 5 & 0.47 + k3 & 35 & 4.5 & @xmath94 & 12.8 & 3.03 & 0.05 & @xmath95 & 3 & 1.86 + k2 & 40 & @xmath93 & & @xmath96 & 5.73 & 0.03 & @xmath97 & 1 & 0.85 + k1 & 45 & 2.7 & 52 & 6.0 & 6.51 & 0.03 & @xmath98 & 1 & 1.01 + + c & 104 & @xmath99 & & @xmath99 & & & & & 0.25 + k3 & 20 & @xmath99 & & @xmath100 & 0.85 & 0.04 & @xmath101 & 9 & 0.34 + k2 & 25 & @xmath99 & & @xmath102 & 2.70 & 0.06 & @xmath103 & 3 & 1.34 + k1 & 7 & @xmath99 & & @xmath104 & 3.69 & 0.30 & @xmath105 & 6 & 0.55 + + + results for the ten sources are discussed below . in each of the images , the restoring beams are shown as crosses in a corner of the images . for the linear polarization maps , the contours are those of polarized intensity @xmath28 , and the plane of the electric vector is indicated by the polarization position angle vectors that are superimposed . there is always the possibility that the observed polarization position angles include some contribution due to faraday rotation along the line of sight to the emission region . multi - frequency vlbi studies of a number of quasars have revealed the presence of non - uniform rotation - measure distributions on parsec scales in some sources ( taylor 1998 , 2000 ; nan et al . 1999 ) . at the same time , other sources show no clear evidence for substantial local ( non - galactic ) rotation measures ( gabuzda & gmez 2000 ) . in the absence of simultaneous multi - wavelength vlbi polarimetry , the only faraday correction that may be applied is one based on integrated measurements . for reference , the rotation measures for the ten sources considered here are summarized in table [ tab : rm ] . all the rotations produced at @xmath2 cm by the integrated rotation measures in table [ tab : rm ] are less than about @xmath106 ; all except that for 1732 + 389 and 0820 + 225 are less than @xmath107 ( table [ tab : rm ] ) . therefore , application of these rotations is probably unlikely to introduce large errors in the resulting @xmath3 values . the @xmath3 values for all sources except for 0820 + 225 have accordingly been corrected using the rotation measures in table [ tab : rm ] ; the possibility of offsets in @xmath3 due to local rotation measures that have not been accounted for should be borne in mind , however . the case of 0820 + 225 is discussed below : for this source , quasi - simutaneous multi - frequency vlba polarization images ( gabuzda , pushkarev & garnich , in prep . ) have revealed a non - uniform rotation - measure distribution on milliarcsecond scales , so that it is clearly not appropriate to apply the integrated rotation measure to each of the vlbi components . models for the source structures were derived by fitting the complex @xmath0 and @xmath1 visibilities that come from the hybrid mapping process as described by roberts , gabuzda and wardle ( 1987 ) and gabuzda , wardle and roberts ( 1989b ) . the model fits are shown in table [ tab : models ] . we denote distance from the core by @xmath31 and the vlbi jet direction by @xmath33 . in all cases , we checked for consistency between the model - fitting results , the distribution of clean components , and the visual appearance of the images . we included components in the models only if there was clear evidence for them in the clean components derived during the imaging process . the @xmath0 and @xmath1 visibilities were fit separately , in order to allow for small differences in the positions and sizes of corresponding @xmath0 and @xmath1 components , either intrinsic to the source structure or associated with residual calibration errors . when @xmath1 components have been identified with specific @xmath0 components , the two positions agree to within their 3@xmath108 errors . when calculating the degrees of polarization @xmath27 for individual features , we have not taken into account possible small physical offsets between the corresponding @xmath0 and @xmath1 components , so that the @xmath27 values in table [ tab : models ] represent averages for each component . the errors of the separations of jet components from the core given in table [ tab : models ] are formal 1@xmath108 errors , corresponding to an increase in the best - fit @xmath109 by unity . the smallest of these formal errors almost certainly underestimate the actual errors ; realistically , the smallest 1@xmath108 errors in component separations are probably no less than @xmath110 mas . in our 1992.23 vlbi run , the polarizations of three of the eighteen sources1334127 , 2131021 , and 2155152varied during the vlbi observations . we were able to detect these variations since the phased vla was included in the vlb array , so that we had measurements of the integrated total intensity and polarization during each of the vlbi scans . summary information about the polarizations of these sources is given below ; the polarization variations are analysed in detail in an accompanying paper by gabuzda et al . ( 2000 ) . we assume throughout a friedmann universe with hubble constant of @xmath111kmsec@xmath112mpc@xmath112 and @xmath113 . the source 0003066 has a redshift of @xmath114 . the 2 cm image of kellermann et al . ( 1998 ) and 13 and 4 cm images of fey & charlot ( 1997 ) show a jet extending toward the northwest . our @xmath0 image ( fig . [ fig:0003 ] ) clearly shows the straight jet in position angle @xmath115 , in which we have modelled six components ( fig . [ fig:0003 ] ) . the linear polarization map shows a double structure , where one component corresponds to the core and the other to the jet feature k5 . the polarization position angle @xmath3 in the jet is nearly perpendicular to the jet axis . observations by wills & wills ( 1976 ) provided evidence that the redshift of this object is @xmath116 . subsequent observations ( stickel et al . 1993a ) have not directly confirmed this value , though the cfht observations of wurtz et al . ( 1996 ) showed the host galaxy to be resolved , consistent with a redshift of 0.258 . the 2 cm image of kellermann et al . ( 1998 ) for epoch 1995 april shows a jet that emerges to the east , then appears to bend sharply to the south . our 6 cm images ( fig . [ fig:0814 ] ) suggest a more gently curving jet structure that initially emerges in position angle @xmath117 , then turns more nearly east . our images do not show clear evidence for the sharp bending observed by kellermann et al . ( 1998 ) , even in the innermost part of the jet . vlba images for epoch 1997 february at 2 , 4 , and 6 cm currently being analyzed ( pushkarev et al . , in prep . ) clearly show the predominant jet direction to be nearly due east , with the 2 cm image in good qualitative agreement with that of kellermann et al . this suggests that the compact eastern extention in the 2 cm images referred to above may have emerged after the 6 cm observations presented here . we detect polarization in the two innermost jet components in our images ; @xmath3 appears to be transverse to the local jet direction in k5 and aligned with the local jet direction in k4 . the redshift of this source is @xmath118 ( stickel et al . 1993a ) , making it one of the most distant objects in the sample . as far as we are aware , ours are the first vlbi images of this source . it displays a rather extraordinary vlbi jet , with very rich @xmath0 and @xmath1 structure . there is much more extended emission than observed in most of the sample objects ; this is particularly striking given its large redshift . the jet extends nearly 30 mas to the southwest , and only a small fraction of the total flux in the vlbi image is contained in the vlbi core . we have modelled the image as a core plus 11 jet components . we expect that this is a good description of the source within 1520 mas from the core , but our jet components k1k4 represent only an approximate model for the bright , extended region at the end of the visible jet . the observed orientation of @xmath3 is transverse to the jet ; however , multi - frequency vlbi polarization observations ( gabuzda and pushkarev , in preparation ) indicate the presence of a large local rotation measure in the inner jet ( near k10 ) , and the derotated @xmath3 values in this region are well aligned with the local jet direction . the redshift of this source is @xmath119 ( stickel et al . 1993a ) . the 2 cm image of kellermann et al . ( 1998 ) and 13 and 4 cm images of fey & charlot ( 1997 ) show a jet extending to the northeast . the relatively weak jet is also visible in our images ( fig . [ fig:0823 ] ) ; we detected polarization only from the vlbi core . this extremely bright and compact source has a redshift of @xmath120 ( wilkes 1986 ; stickel et al . though this source was included in the khr & schmidt ( 1990 ) bl lac sample , it was subsequently reclassified as a quasar by stickel et al . ( 1993b ) . the 2 cm image of kellermann et al . ( 1998 ) shows a jet toward the southeast . the dominant orientation for the inner vlbi jet in our image is @xmath121 . the polarization of this source varied during the vlbi observations . the @xmath1 map in fig . [ fig:1334]b was obtained for the time subinterval considered in the intraday variability analysis ( gabuzda et al . 2000 ) that had the best @xmath122 coverage . the polarization position angle @xmath3 in the brightest jet component is roughly perpendicular to the inner - jet direction . the orientation of @xmath3 in the core relative to the jet direction is not clear ; it is interesting , however , that this orientation is nearly the same as that observed by lister et al . ( 1998 ) in the inner vlbi jet at 7 mm ( where it is also oblique ) , suggestive that our 6 cm `` core '' emission may be associated with this same region . the 6 cm jet is fairly highly polarized , @xmath123 . the redshift of this object is @xmath124 , making it one of the most distant sources in the sample . the 6 cm vlbi image of xu et al . ( 1995 ) showed a compact vlbi jet toward the east . our images ( fig . [ fig:1732 ] ) show the jet more clearly . the jet contains three components , with polarization detected in the core and innermost jet component k3 . the @xmath3 in k3 is roughly perpendicular to the jet direction . this source has the largest redshift among the sample objects , @xmath125 ( drinkwater et al . the structure is very compact , with a jet extending nearly directly east ( fig . [ fig:2131 ] ) ; the faint jet is also visible in the 2 cm image of kellermann et al . ( 1998 ) and the 13 and 4 cm images of fey & charlot ( 1997 ) . we detected polarization only in the vlbi core , and this polarization varied during the vlbi observations . the @xmath1 map in fig . [ fig:2131]b was obtained for one time subinterval considered in the intraday variability analysis ( gabuzda et al . the degree of polarization was only mildly variable , @xmath126 , while @xmath13 varied by about @xmath106 over the course of the vlbi run , remaining roughly perpendicular to the direction of the vlbi jet ( to within @xmath127 ) . this object is one of comparatively few in the sample whose spectra are so featureless that no redshift has been determined ( stickel et al . 1993b ; veron - cetty & veron 1993 ) . the 13 and 4 cm vlbi images of fey & charlot ( 1997 ) show a prominent jet extending toward the northwest , which is also clearly visible in our images ( fig . [ fig:2150 ] ) . we detected polarization in the core and several jet components ; @xmath3 is predominantly transverse to the jet direction in both the core and jet . the redshift of this object is @xmath128 ( stickel et al . its total intensity structure ( fig . [ fig:2155 ] ) is somewhat unusual , with structure present on both sides of the brightest feature at the phase center . we believe that the strongest feature is a bright knot in the inner jet , and that the feature to the northeast of this component is the core ; multi - frequency vlba observations currently being analysed ( pushkarev et al . , in prep . ) confirm that the source has a core jet structure with the jet extending toward the southwest . we detected polarization in the core to the northeast and the second jet component to the southwest . the polarized flux of the core varied during the vlbi run , while that of the jet component was constant ( gabuzda et al . 2000 ) , with no detectable variability in @xmath3 for either of these components . the map in fig . [ fig:2155]b corresponds to one time subinterval considered in the variability analysis . @xmath3 in the core is well aligned with the inner - jet direction , and is perpendicular to the jet direction in the sw component , suggesting that the dominant magnetic field in this part of the jet is longitudinal . the host galaxy of this source , which has a redshift @xmath129 ( stickel et al . 1988 ) , is clearly resolved in direct images ( stickel et al . 1993a ; falomo 1996 ) . the bl lac object is surrounded by a number of galaxies , one of which appears to be a physical companion ( stickel et al . 1993a ) . to our knowledge , no previous vlbi images are available . we detected three components in the vlbi jet ( fig . [ fig:2254 ] ) ; the polarization of this source was not sufficiently strong to be detected . having completed our first - epoch 6 cm polarization observations of all sources in the khr & schmidt sample of bl lacertae objects , we are now able to evaluate the vlbi properties of the sample as a whole . we summarize our results for the vlbi polarization properties of the sample sources in table 4 . the columns present ( 1 ) source name ( 1950.0 ) , ( 2 ) redshift @xmath130 , ( 3 ) epoch for the vlbi polarization observations , ( 4 ) total flux density on vlbi scales @xmath131 , ( 5 ) fraction of vla core flux present on vlbi scales @xmath132 , ( 6 ) total polarized flux density on vlbi scales @xmath133 , ( 7 ) fraction of vla core polarized flux present on vlbi scales @xmath134 , ( 8) position angle for the total polarization on vlbi scales @xmath135 , ( 9 ) position angle for the polarization of the vla core @xmath136 , ( 10 ) degree of polarization in the vlbi core @xmath137 , ( 11 ) degree of polarization in the vlbi jet @xmath138 , ( 12 ) offset between @xmath13 and the jet direction @xmath139 , ( 13 ) offset between @xmath140 and the jet direction @xmath141 , ( 14 ) integrated rotation measure rm , and ( 15 ) references . separate rows are given for each epoch for which vlbi polarization data have been analyzed . when polarization was detected in more than one jet component at a given epoch , multiple entries separated by commas are given . the @xmath135 and @xmath136 values are as observed , i.e. , not corrected for faraday rotation , but @xmath139 and @xmath141 have both been determined after correction for the integrated rotation measure listed in column ( 16 ) . the one exception is 0820 + 225 , for which multi - frequency vlbi polarization observations have revealed a non - uniform distribution of the rotation measure on milliarcsecond scales ( gabuzda , pushkarev & garnich , in prep . ) ; in this case , the values in columns ( 14 ) and ( 15 ) have been determined after correction for the local rotation measure in the vlbi jet . previous observations ( gabuzda et al . 1994b , 1999 , and references therein ) have indicated that the vlbi core polarizations of bl lac objects are typically @xmath142 . after completing our first - epoch polarization observations for the entire khr & schmidt bl lac sample , we have measurements of the vlbi core polarizations @xmath137 for all but five of the 33 sample sources . these @xmath137 values , collected in table 4 , with few exceptions range from 1.8 to 6.8% , demonstrating that the behaviour suggested by the earlier incomplete observations is , indeed , characteristic of bl lac objects . another tendency indicated by the earlier observations was for the core polarization position angle @xmath13 to be either aligned with or transverse to the direction of the inner 6 cm vlbi jet . we now have measurements of @xmath143 for 24 of the 33 sample sources . the @xmath143 values in table 4 and in fig . [ fig : corehist ] clearly show this behaviour : in 16 cases , @xmath143 is between 0 and @xmath144 , and in 12 cases @xmath143 is between 60 and @xmath145 . this indicates that the @xmath143 distribution for this @xmath146 of the sample is bimodal , with the numbers of sources displaying either of the two preferred values for @xmath143 being roughly equal . it appears that the peak at small misalignments may be offset from @xmath147 ; the origin of this behaviour is not clear , though it could be associated with uncertainty in the jet direction on scales slightly smaller than the resolution of our 6 cm global observations . earlier observations indicated a tendency for @xmath3 in the vlbi jets of bl lacertae objects to be parallel to the local jet direction , which has usually been interpreted as evidence for the presence of relativistic plane shocks . we now have measurements of @xmath148 for 25 of the 33 sample sources . the @xmath148 values in table 4 and in fig . [ fig : jethist ] clearly show a predominance of sources with @xmath140 and @xmath33 aligned to within @xmath127 ( @xmath149 ) . at the same time , the histogram shows a weaker secondary peak of sources with @xmath148 near @xmath145 . thus , it is clear that , as hinted by previous observations ( gabuzda et al . 1999 and references therein ) , a sizeable fraction ( @xmath150 ) of sources in the khr & schmidt ( 1990 ) sample of bl lacertae objects have longitudinal magnetic fields in their vlbi jets . at this time , it is not clear how jet components with transverse and longitudinal magnetic fields physically differ ; multi - frequency vlba polarization observations for all the sample sources currently under analysis will hopefully help eludicate this question in the near future . in the remaining sources , @xmath140 bears no obvious relation to @xmath33 , possibly due to the presence of faraday rotation with substantially different values than suggested by the integrated rotation measures . multi - frequency polarization observations currently being analyzed should shed light on this issue . among the sources for which images are presented and analysed here , to our knowledge , previous vlbi images at 6 cm or nearby wavelengths are available for five of them : 0003066 , 0823 + 033 , 1732 + 389 , 2131021 , and 2150 + 173 . 1732 + 389 was observed at epoch 1990.18 as part of the 6 cm caltech jodrell bank survey ( xu et al . the other four sources were observed at 4 cm by fey & charlot ( 1997 ; hereafter fc ) . in their 6-cm image of 1732 + 389 , xu et al . ( 1995 ) detected a single jet component at @xmath151 mas . it is natural to tentatively identify this feature with k3 in table [ tab : models ] . our second - epoch observations at epoch 1995.41 support this identification , and indicate that the apparent proper motion of this feature is 0.24 mas / yr , which corresponds to an apparent speed @xmath152 ( pushkarev & gabuzda 1999a ) . the feature k2 is also present in our 1995.41 image ; a comparison suggests a tentative proper motion of 0.61 mas / yr , corresponding to an apparent speed @xmath153 ( pushkarev & gabuzda 1999a ) . fc obtained images of a large number of compact extragalactic radio sources at 13 and 3.6 cm . we will be interested primarily in the 3.6 cm images , since they have higher resolution and the observing wavelength is closer to that for our data . we must be aware of possible small frequency - dependent shifts in the apparent separations of jet components from the core when comparing these 3.6 cm images with our own 6 cm images . however , these two wavelengths are close enough that we would not expect such shifts to be significant in most cases , so that the 3.6 cm data of fc can be used to derive at least tentative estimates of the structural changes in the sources observed in both projects . below , we will refer to the 3.6 cm images of fc unless explicitly stated otherwise . fc s 3.6 cm observations of 0003066 were less than five months after our first - epoch observations . there is a reasonably close correspondance between the jet structures detected in the two observations , but the interval between them is too short to reliably identify systematic proper motions . most features appear to be stationary to within the errors . we therefore postpone estimation of the apparent speeds in the vlbi jet of this source until the results of our second - epoch observations become available . fc detected only a single jet component in 0823 + 033 at epoch 1995.28 , at @xmath154 mas . there are no components in our image near this position . if we assume expansion of the source structure with time , the simplest joint interpretation of the two images is that the feature detected by fc is k3 in table [ tab : models ] . in this case , its tentative proper motion is 0.39 mas / yr , which corresponds to an apparent speed @xmath155 . two jet components were detected by fc in 2131021 : one at @xmath156 mas in @xmath157 and another at @xmath158 mas in @xmath159 . a comparison with the model for this source in table [ tab : models ] shows that these positions nearly coincide with those for our components k5 and k4 . in both cases , the positions of these features in the two images agree to within the errors . thus , the simplest joint interpretation of the two images is that k4 and k5 were stationary from epoch 1992.23 to epoch 1995.78 . there is a very good correspondance between the structures in our image of 2150 + 173 and the image of fc , in which features were detected at @xmath160 and 6.9 mas . it seems likely that these four components should be identified with k5 , k4 , k3 , and k2 in table [ tab : models ] . in this case , the implied proper motions for these features are @xmath161 and 0.32 mas / yr , respectively ; we can see that this suggests acceleration with distance from the core , but further observations are required to test this hypothesis . unfortunately , the redshift of this source is unknown , so that it is not possible to directly translate these proper motions into apparent speeds . finally , we have one more tentative speed to report here . at the time of the analysis for our first - epoch 6 cm observations of 0138097 ( gabuzda et al . 1999 ) , we were not aware that fc had also observed this source at 3.6 cm . they detected a single jet component 1.9 mas from the core . if this feature should be identified with k2 from gabuzda et al . ( 1999 ) , this implies a proper motion of 0.35 mas / yr , which corresponds to a tentative apparent speed @xmath162 . our observations have substantially increased the number of objects in the khr & schmidt sample with at least tentative ( two - epoch ) speed estimates . such estimates are now available for 16 of the 33 sample sources , and are summarized in table [ tab : speeds ] . the histogram in figure [ fig : speeds ] shows the distribution of superluminal speeds in these 16 objects . when making this histogram , we included more than one entry per source only if apprecially different speeds were observed for different components ( i.e. , only if the speeds fell in different bins ) . we can see that there is a concentration toward relatively low values , and the apparent speeds are fairly uniformly distributed in the range from @xmath163 , with a possible weak peak near @xmath164 . we see no evidence from this distribution that the typical jet lorent factors exceed @xmath165 . there is some evidence for the occurrence of higher superluminal speeds in some bl lacertae objects on the smaller scales probed by higher - frequency observations ( e.g. , marscher & marchenko 1998 ) ; it remains unclear whether the jets of bl lacertae objects in general tend to display higher superluminal speeds on smaller scales . figure [ fig : speeds_cdq ] shows the distribution of superluminal speeds for components in 19 core dominated quasars taken from vermeulen & cohen ( 1994 ) . here , as for the distribution in fig . [ fig : speeds ] , we have not plotted more than one entry per source per bin . we can see that the quasar speed distribution has fewer values below @xmath166 and appreciably more values exceeding @xmath167 than the bl lac distribution . the overall centroid of the distribution is shifted toward higher speeds : the median and average speeds for the bl lacertae objects are both 3.7 , while the corresponding values for the quasars are both 5.8 . a kolmogorov smirnov test indicates that the probability that the two distributions are the same is less than 1% . this confirms previous evidence that the superluminal speeds in bl lac objects are systematically lower than those observed in core - dominated quasars ( gabuzda et al . 1994b ; britzen et al . 1999 ) . cccccc source & @xmath130 & n & @xmath168 & @xmath169 & ref . + & & & ( mas / yr ) & & + 0138097 & 0.44 & 2 & 0.35 & 5.5 & 12 + 0454 + 844 & 0.11 & 5 & 0.14 & 0.6 & 1,3 + 0716 + 714 & @xmath170 & 4 & 0.07 & @xmath170 & 1,8 + & @xmath170 & 2 & 0.11 & @xmath170 & 8 + 0735 + 178 & @xmath171 & 4 & 0.48 & 7.4 & 5.6 + & & 4 & 0.33 & 5.0 & 6 + & & 3 & 0.28 & 4.2 & 6 + 0823 + 033 & 0.51 & 2 & 0.39 & 6.9 & 12 + 0828 + 493 & 0.55 & 2 & 0.34 & 6.3 & 9 + 0851 + 202 & 0.31 & 3 & 0.20 & 2.4 & 2 + & & 3 & 0.27 & 3.2 & 2 + 0954 + 658 & 0.37 & 2 & 0.37 & 5.2 & 7 + & & 2 & 0.44 & 6.2 & 7 + 1418 + 546 & 0.15 & 3 & 0.56 & 3.6 & 9,10 + & & 2 & 0.08 & 0.5 & 9,10 + 1652 + 398 & 0.03 & 3 & 0.27 & 0.4 & 7 + & & 2 & 0.55 & 0.8 & 7 + 1732 + 389 & 0.97 & 3 & 0.24 & 6.5 & 10,12 + & & 2 & 0.61 & 16.6 & 10,12 + 1749 + 701 & 0.77 & 3 & 0.14 & 3.2 & 7 + & & 3 & 0.08 & 2.0 & 7 + & & 2 & 0.09 & 2.1 & 7 + 1803 + 784 & 0.68 & 2 & 0.08 & 1.8 & 7 + 1823 + 568 & 0.34 & 3 & 0.18 & 3.8 & 7 + & & 2 & 0.20 & 4.3 & 7 + 2007 + 777 & 0.34 & 5 & 0.22 & 2.9 & 1,7 + 2200 + 420 & 0.07 & 4 & 1.14 & 3.7 & 4 + & & 5 & 1.12 & 3.6 & 4 + & & 4 & 1.12 & 3.6 & 4 + & & 2 & 1.00 & 3.3 & 4 + 2150 + 173 & @xmath170 & 2 & 0.01 & @xmath170 & 12 + & & 2 & 0.11 & @xmath170 & 12 + & & 2 & 0.28 & @xmath170 & 12 + & & 2 & 0.34 & @xmath170 & 12 + 2254 + 074 & 0.19 & 2 & 0.46 & 3.7 & 11,12 + & & 2 & 0.30 & 2.4 & 11,12 + & & 2 & 0.55 & 4.3 & 11,12 + + + + + + + this paper completes analysis of our first - epoch observations of all sources in the complete sample of bl lacertae objects defined by khr & schmidt ( 1990 ) . certain tendencies were present even in the earliest vlbi polarization observations of bl lac objects ; however , without systematic observations of a complete sample of objects , it was not possible to say whether these properties were relevant for radio - loud bl lac objects in general , or only to a handful of the best studied sources . on the whole , our analysis of the complete sample has confirmed the results of these early observations . however , it is also clear that the behaviour shown by the sample sources is not entirely uniform . we summarize our results below . the vlbi core polarizations of bl lacertae objects are appreciable , with values typically ranging from @xmath172 , and occasionally reaching values as high as @xmath173 . gabuzda et al . ( 1994b ) suggested that these relatively high values reflected the dominant contribution of newly emerging jet components . in the case of one source in the sample 1803 + 784 6 cm space vlbi polarization observations directly showed that the polarized flux was dominated by a compact component in the inner vlbi jet , which was unresolved from the core in ground - based observations ( gabuzda 1999 ) . the 6 cm vlbi core polarizations of quasars are much lower , typically @xmath12 ( cawthorne et al . 1993 ) . this suggests that either the cores of quasars are depolarized ( see , e.g. , taylor 1998 , 2000 ) , or that they are considerably less likely to be dominated by emission from compact new jet components . if the vlbi cores of quasars are depolarized , we expect that the observed degrees of polarization of the cores of quasars and bl lac objects will become more similar at higher frequencies ; thus far , there is no clear evidence for this , but more systematic studies are required . thus , it remains a possibility that the vlbi core polarizations of bl lac objects are more often dominated by the contribution of emerging jet components than are the vlbi core polarizations of quasars . the origin of this systematic difference is not obvious . one possibility is that the birth rate for new jet components is higher in bl lacertae objects than in quasars , increasing the probability of observing a bl lac object core harbouring new jet components . there is some evidence from the university of michigan monitoring database that outbursts in bl lacertae objects may be more frequent and well resolved than those in quasars , suggesting that bl lac objects may generate new components more frequently ( m. aller , private communication ) . since the superluminal speeds observed in bl lacertae objects are , on average , slower than those in quasars , it may be that the jet components of bl lacertae objects spend more time in the unresolved core region before becoming detectable as distinct jet knots . another possibility is that highly - polarized jet components close to the core are often long - lived stationary components whose polarization blends with the true core polarization . quasi - simultaneous multi - frequency vlba observations for all sources in the khr & schmidt sample currently being analyzed should help distinguish between these various possibilities . one of the striking tendencies noted in the earliest vlbi polarization results was for the jets of bl lacertae objects to have transverse magnetic fields . our first - epoch images for the complete khr & schmidt sample confirm that this is the predominant behaviour for the sample as a whole : among the 25 sources in which jet polarization was detected , some 6070% have transverse magnetic fields . at the same time , a sizeable minority of about 30% have longitudinal jet magnetic fields . thus far , transverse magnetic fields have usually been interpreted as manifestations of relativistic shocks in the vlbi jets of these sources , while longitudinal fields have been taken to reflect the presence of shear between the jet and the surrounding medium . in this picture , the common presence of transverse magnetic fields reflects the existence of conditions favorable for the formation of transverse shocks ; the jet components with dominant longitudinal fields would be those in which shocks did not form , or did form but were dominated by the effect of shear . though it seems likely that many individual , compact , highly - polarized features with transverse magnetic fields are associated with shocks , this does not necessarily imply that _ all _ the observed transverse jet fields should be identified with shock components . another possibility is that , in at least some cases , we are detecting the toroidal component of an intrinsic helical jet magnetic field ( see , e.g. , gabuzda 1999 ) . if the dominant magnetic field in the jet is helical , the net observed field can be either transverse or longitudinal , depending on the pitch angle of the field and the viewing angle , though it is more likely that the net observed field will be transverse . in addition , a longitudinal field component could develop due to interaction between the edges of the jet and the surrounding medium ( e.g. aaron 1998 , laing et al . 1999 , aloy et al . thus , it could be that the dominant magnetic - field component in the vlbi jets of bl lacertae objects is characteristically toroidal . this could be consistent with the fact that the observed jet magnetic fields are most often transverse , but occasionally longitudinal . the characteristically modest superluminal speeds observed in the khr & schmidt sources suggest that bl lacertae objects differ from quasars in either the characteristic angles of their jets to the line of sight , or the characteristic intrinsic velocities of components in their jets , or both . the observed apparent speed for a vlbi feature has a peak for motion at an angle to the line of sight of about @xmath174 , where @xmath175 is the lorentz factor of the motion . therefore , if the intrinsic velocities in the two types of sources were essentially the same , the bl lacertae objects could , in principle , have smaller apparent velocities if their jets were significantly further from or nearer to the line of sight than the jets in quasars . since bl lacertae objects are obviously highly beamed sources ( kollgaard 1994 , for example ) , it is not reasonable to suppose that their jets could typically be at significantly larger angles to the line of sight than those in quasars . on the other hand , if their jets were significantly _ closer _ to the line of sight than quasar jets , we would expect bl lacertae objects to be significantly more highly beamed than quasars , and there is no evidence for this ( ghisellini et al . thus , the most straightforward interpretation of the more modest superluminal speeds observed in the vlbi jets of bl lacertae objects compared to quasars is that the intrinsic velocities in the bl lac jets are lower . this seems quite natural in the context of unified schemes linking bl lacertae objects with fr i and quasars with fr ii radio galaxies . this work was supported by an american astronomical society henri chretien international research grant and small grant ( dcg ) . abp acknowledges support from the russian foundation for basic research . dcg acknowledges support from the european commission under the ihp programme ( ari ) contract no . hpri - ct-1999 - 00045 . we would like to thank r. m. campbell for useful discussions of some aspects of our analysis . we also thank the staff at the participating observatories and correlation facillities who made these observations possible . this research has made use of data from the university of michigan radio astronomy observatory , which is supported by the national science foundation and by funds from the university of michigan . aaron s. , 1999 , in _ the bl lac phenomenon _ , asp conf . 159 , 427 . aloy m .- a . , gmez j .- , ibaez j .- , mart j .- m . , & mller e. , 2000 , apj , 528 , l85 . angel j.r.p . , stockman h.s . , 1980 , ara&a , 8 , 321 . bth l.b . & zhang f.j . , 1991 , a&a , 243 , 328 . britzen s. , vermeulen r.c . , taylor g.b . , readhead c.s . , pearson t.j . , henstock d.r . , & wilkinson p.n . , 1999 , in _ the bl lac phenomenon _ , asp conf . ser . , 159 , 431 . browne i. w. a. , 1983 , mnras , 204 , 23p . cawthorne t.v . , wardle j.f.c . , roberts , d.h . , gabuzda d.c . , & brown , l.f . , 1993 , apj , 416 , 496 . cornwell t.j . , wilkinson p.n . , 1981 , mnras , 196 , 1067 . drinkwater m.j . , webster , r.l . , francis , p.j . , condon , j.j . , ellison s.l . , jauncey d.l . , lovell j. , peterson b.a . , & savage a. , 1997 , mnras , 284 , 85 . falomo , r. 1996 , mnras , 283 , 241 . fey a.l . & charlot p. , 1997 , apjs , 111 , 95 . fossati , g. , celotti , a. , ghisellini , g. , & maraschi , l. 1997 , mnras , 289 , 136 . gabuzda d.c . 1999 , new astronomy reviews , 43 , 691 . gabuzda d.c . & cawthorne t.v . 1996 , mnras , 283 , 759 . gabuzda d.c . , cawthorne t.v . , roberts d.h . , wardle j.f.c . , 1989 , apj , 347 , 701 . gabuzda d.c . , cawthorne t.v . , roberts d.h . , wardle j.f.c . , 1992 , apj , 388 , 40 . gabuzda d.c . , kochenov p.yu . , & cawthorne t.v . 2000 , mnras , submitted . gabuzda d.c . , mullan c.m . , cawthorne t.v . , wardle j.f.c . , roberts d.h . , 1994b , apj , 435 , 140 . gabuzda d.c . , pushkarev a.b . , & cawthorne t.v . , 1999 , mnras , 307 , 725 . gabuzda d.c . , sitko , m.l . , smith , p.s . , 1996 , aj , 112 , 1877 . gabuzda d.c . , wardle j.f.c . , & roberts d.h . , 1989a , apj , 336 , l59 . gabuzda d.c . , wardle j.f.c . , & roberts d.h . , 1989b , apj , 338 , 743 . gabuzda d.c . , wardle j.f.c . , roberts d.h . , aller m.f . , & aller h.d . , 1994a , apj , 435 , 128 . ghisellini g. , padovani p. , celotti a. , & maraschi l. , 1993 , apj , 407 , 65 . giommi , p. & padovani , p. 1994 , mnras , 268 , p51 . hughes p.a . , aller h.d . , aller m.f . , 1989 , apj , 341 , 68 . kellermann k. i. , vermeulen r. c. , zensus j. a. , & cohen m. h. , 1998 , aj , 115 , 1295 . kollgaard r.i . , 1994 , vistas in astronomy , 38 , 29 . khr , h. & schmidt , g.d . 1990 , aj , 90 , 1 . laing r. , 1980 , mnras , 193 , 439 . laing r.a . , parma p. , de ruiter h.r . , & fanti r. , 1999 , mnras , 306 , 513 . lister m.l . , marscher a.p . , & , 1998 , apj , 504 , 702 . maraschi , l. , ghisellini , g. , tanzi , e. g. , & treves , a. 1986 , apj , 310 , 325 . marscher a.p . & marchenko s.g . , 1999 , in _ the bl lac phenomenon _ , asp conf . ser . , 159 , 417 . miller j.s . , 1981 , pasp , 93 , 681 . mutel r.l . , su bumei , bucciferro r.r . , & phillips r.b . , 1990 , apj , 352 , 81 . padovani , p. & giommi , p. 1995 , apj , 444 , 567 . pushkarev , a. b. & gabuzda , d. c. 1999a , in _ the bl lac phenomenon _ , asp conf . ser . , 159 , 458 . pushkarev , a. b. & gabuzda , d. c. 1999b , new astronomy reviews , 43 , 695 . roberts d.h . , gabuzda d.c . , wardle j.f.c , 1987 , apj , 323 , 536 . roberts d.h . , wardle j.f.c . , brown l.f . , 1994 , apj , 427 , 718 . rudnick l. , jones t.w . , 1983 , aj , 88 , 518 . rusk r. , 1988 , ph.d . thesis , university of toronto . stickel m. , fried j.w . , khr h. , 1988 , a&a , 191 , 16 . stickel m. , fried j.w . , khr h. , 1989 , a&as , 80 , 113 . stickel m. , fried j.w . , khr h. , 1993a , a&as , 98 , 393 . stickel m. , khr h. , & fried j.w . , 1993b , a&as , 97 , 393 . stocke , j.t . , liebert , j. , schmidt , g. , gioia , i.m . , maccacaro , t. , schild , r.e . , maccagni , d. , & arp , h. c. 1985 , apj , 298 , 619 . vermeulen , r.c . & cohen , m.h . 1994 , apj , 430 , 467 . veron - cetty m.p . & veron p. , 1993 , a&as , 100 , 521 . wardle j. f. c. , moore r. l. , & angel , j. r. p. , 1984 , apj , 279 , 93 . wilkes , b.j . , 1986 , mnras , 218 , 331 . wills , d. & wills , b.j . , 1976 , apjs , 31 , 143 . witzel a. , schalinski c.j . , johnston k.j . , biermann p.l . , krichbaum t.p . , hummel c.a . , & eckart a. , 1988 , a&a , 206 , 245 . wurtz r. , stocke , j.t . , & yee , h.k.c . , 1996 , apjs , 103 , 109 . xu , w. , readhead , a.c.s . , & pearson , t.j . 1995 , apjs , 99 , 297 .

it has recently been proposed in refs . that the dark energy could be attributed to the cosmological properties of a scalar field with a non - standard dispersion relation that decreases exponentially at wave - numbers larger than planck scale ( ) . in this scenario , the energy density stored in the modes of trans - planckian wave - numbers but sub - hubble frequencies produced by amplification of the vacuum quantum fluctuations would account naturally for the dark energy . the present article examines this model in detail and shows step by step that it does not work . in particular , we show that this model can not make definite predictions since there is no well - defined vacuum state in the region of wave - numbers considered , hence the initial data can not be specified unambiguously . we also show that for most choices of initial data this scenario implies the production of a large amount of energy density ( of order ) for modes with momenta , far in excess of the background energy density . we evaluate the amount of fine - tuning in the initial data necessary to avoid this back - reaction problem and find it is of order . we also argue that the equation of state of the trans - planckian modes is not vacuum - like . therefore this model does not provide a suitable explanation for the dark energy .

hep-th0212027

recently , it has been claimed in a series of articles @xcite that the cosmic dark energy component could be explained naturally by the trans - planckian energy of a scalar field with a suitable non - linear dispersion relation in the trans - planckian regime . such dispersion relations , which relate the frequency @xmath4 to the wave - number @xmath5 of a scalar field wave - packet , and which depart from the standard linear dispersion relation in the trans - planckian regime are a way of modeling phenomenologically the unknown physics for sub - planckian wavelengths . they have been used extensively in the recent literature in the context of black - hole physics @xcite and of the inflationary trans - planckian problem @xcite . in the particular case considered in refs . @xcite , the dispersion relation departs from its standard linear form and approach a decreasing exponential at large wave - numbers . this type of dispersion relation could possibly emerge from string theory @xcite . it has been argued that the energy density of the modes of sub - planckian wavelengths and sub - hubble frequencies ( referred to as `` tail '' modes ) is naturally of the same order as the critical energy density today and has the same equation of state as a cosmological constant . hence , it could account without fine - tuning for the dark energy . the energy density contained in the tail today has been calculated in ref . @xcite by solving for the time evolution of a test quantum scalar field evolving in the curved cosmological background , assuming its initial state at the onset of inflation is the vacuum . the equation of state of the tail modes has been calculated in ref . @xcite and its cosmological evolution has been solved to argue that the cosmic coincidence problem ( why does the dark energy dominate now ? ) is solved . in this paper we argue that this model does not and can not work for several reasons . we first argue that the energy density and equation of state of the `` tail '' modes depend directly on the choice of initial data for the scalar field , and that this latter can not be specified _ unambiguously _ since there is no preferred initial vacuum state ( section [ sec : ii ] ) . this implies that any cosmological consequence derived depends directly on the ad - hoc choice of initial data ( initial quantum state ) . we then show that the violation of the wentzel - kramers - brillouin ( wkb ) approximation in the remote past for all co - moving wave - numbers , which is inevitable in the present scenario , implies the continuous production _ at all times _ of a large amount of quanta with physical wave - number @xmath2 ( section [ sec : iii ] ) . this finding is in agreement with general arguments given by starobinsky @xcite ( this latter work did not however study the present scenario ) . we evaluate the amount of energy density produced for modes of physical wave - number and frequency @xmath6 for various choices of initial data and conclude that it is generically of order @xmath7 . this process takes place at all times , and since the energy density produced is much larger than the background energy density , it implies that the semi - classical perturbative framework on which the model of refs . @xcite rests breaks down . in an earlier study , devoted to constructing an effective stress - energy tensor for theories with non - linear dispersion relations @xcite , we already criticized this model by arguing that it led generically to the wrong equation of state . we revisit this issue further in section [ sec : iv ] , where we prove that the effective energy - momentum tensor we derived earlier is well - behaved , thus disproving an improper claim of ref . @xcite and confirming our earlier criticisms . we provide a summary of our conclusions in section [ sec : v ] . refs . @xcite consider a scalar field , @xmath8 , with a non - linear dispersion relation that is linear in the sub - planckian regime and approaches a decreasing exponential at trans - planckian wave - numbers ( i.e. for @xmath9 , @xmath10 being a fundamental characteristic scale ) . this dispersion relation is shown in fig . this scalar field is assumed to describe the density ( scalar ) perturbations and/or the primordial gravitational waves . the `` tail '' modes are thus interpreted as a bath of gravitons of super - planckian wavelengths and sub - hubble frequencies . this scalar field is treated as a test - field ( its back - reaction on the background is neglected ) and is quantized on the curved cosmological background . assuming that the `` tail '' modes of this field are initially in a well chosen vacuum state as @xmath11 ( @xmath12 denoting conformal time ) , the occupation number at late times ( @xmath13 ) of quanta extracted out of the vacuum by the dynamical background has been calculated in ref . this occupation number can then be used to calculate the energy density stored today in the `` tail '' . this is the thread of the calculation performed in ref . @xcite , which we now follow in some detail . this discussion will take us to the two main arguments that we bring forward against this model ( given in this section and the following ) . _ vs _ @xmath5 . region i ( hashed area ) corresponds to the `` tail '' modes for which @xmath14 [ and @xmath15 . region ii corresponds to the sub - horizon modes ( @xmath16 ) that are outside of the tail , while region iii corresponds to the super - horizon modes @xmath17 . the fundamental scale @xmath18 is also indicated.,scaledwidth=70.0% ] the equation of motion of a scalar field @xmath19 in a friedmann - lematre - robertson - walker ( flrw ) space - time with scale factor @xmath20 reads : @xmath21\mu_k=0,\ ] ] where @xmath22 is a coupling parameter to gravity and a prime denotes differentiation with respect to conformal time . @xmath23 is the co - moving wave - number and @xmath24 is the co - moving frequency . @xmath25 for tensor perturbations degrees of freedom and @xmath26 for a conformally coupled field . the scale factor is taken to be a power law in conformal time , @xmath27 , and the following dispersion relation , parameterized by two parameters @xmath28 and @xmath29 with @xmath30 in order to insure that the dispersion relation is linear for small wave - numbers , is introduced : @xmath31,\ ] ] with @xmath32 with @xmath33 . a problem with this dispersion relation is that it depends on the power - law index @xmath34 of the scale factor . if taken literally , this means that the dispersion relation , or the physical frequency of a given mode , changes as the scale factor power - law index @xmath34 changes between various cosmological eras ( e.g. inflation / radiation domination / matter domination ) . more importantly one easily sees that the above dispersion relation has a pathological behavior in the radiation ( @xmath35 ) or matter ( @xmath36 ) dominated eras . in fact , for @xmath37 , it implies @xmath38 as @xmath39 , whereas one should instead reach the linear dispersion relation in that regime with @xmath40 . since ref . @xcite focused on the case of de sitter space - time with @xmath41 , we set @xmath41 in the above dispersion relation , i.e. , the above @xmath42 should be understood as @xmath43 . this reformulated dispersion relation thus coincides with that used in ref . @xcite for de sitter space . however , in the matter dominated era , for instance , we have @xmath44 and the general class of solutions to the field equation obtained in ref . @xcite does not hold anymore . the linear dependence of @xmath42 on @xmath12 is lost for background metrics other than de sitter , but the linear dependence of @xmath4 on @xmath5 is preserved in the small @xmath5 limit for all metrics , which is obviously an imperative . the field equation can finally be rewritten as : @xmath45 -(1 - 6\xi ) \frac{\beta(\beta+1)}{\eta ^2}\biggr\}\mu _ k=0\ , , \ ] ] with , again , @xmath46 $ ] . the solution to this equation depends on the value of @xmath22 and @xmath34 . in ref . @xcite , the contribution of the @xmath47 $ ] term is assumed to be negligible at early times . however the above equation shows that this is not the case ; denoting by @xmath48 the term in curly brackets in eq . ( [ eq:2 ] ) , one has @xmath49 as @xmath50 and the term @xmath51 is always dominant in that limit if @xmath52 ( @xmath53 , @xmath54 ) . therefore , in the limit @xmath52 and @xmath11 the two independent solutions to the field equation are power - laws in @xmath12 : @xmath55 this is an important point as it implies that the mode function does not behave as a plane wave in the limit @xmath11 when @xmath52 . the solution to the field equation in this limit is reminiscent of the mode freezing in inflationary theories for fields with linear dispersion relations and @xmath25 when the mode exits the horizon . it is also argued that the term in @xmath56 can be absorbed at late times in a redefinition of the dispersion relation . however this can not be correct since by construction , the dispersion relation @xmath57 depends only on @xmath5 , i.e. time only enters via @xmath5 . therefore , one can absorb a term @xmath58 into @xmath59 only if @xmath41 ( de sitter space - time ) , as inspection of eq . ( [ eq:2 ] ) reveals . in effect , the curly bracket of eq . ( [ eq:2 ] ) can then be rewritten as @xmath60 times a function of @xmath61 . but , in that particular case , the redefined modified dispersion relation does not have an exponential shape anymore , since @xmath4 approaches a constant ( @xmath62 ) as @xmath63 . however , this should not give the impression that the corresponding solution is a plane wave since , evidently , the co - moving frequency @xmath64 which enters eq . ( [ eq:1 ] ) still behaves as @xmath65 . moreover in the case of a matter or radiation dominated cosmology , one can not absorb the scale factor term in the dispersion relation . nevertheless one can also assume @xmath66 . in that case , it is possible to find an exact solution to the equation of motion in de sitter space - time . indeed , for a conformally coupled field , the term @xmath56 disappears from the field equation and the equation becomes simpler . note however that the scalar field can not correspond to tensor perturbations degrees of freedom since these are minimally coupled to the metric . let us consider @xmath26 for the moment . a solution to the field equation , given in ref . @xcite reads : @xmath67 where @xmath68 is an hypergeometric function and @xmath69 and @xmath70 are expressed in terms of @xmath28 and @xmath29 as : @xmath71 this solution is valid only for de sitter space - time with @xmath72 ( @xmath73 ) . as already mentioned , this is due to the fact that , with the reformulated dispersion introduced above , the linear dependence of @xmath42 in the conformal time is lost for other scale factors . however similar solutions for other metrics can be obtained if the dispersion relation is tuned to the power - law evolution of the scale factor , i.e. if the parameter @xmath42 remains linear in @xmath12 ( possibly at the expense of linearity of @xmath4 in the small @xmath5 limit , see above ) . equation ( [ eq:2 ] ) has in fact two independent solutions ( see below ) and the choice ( [ eq:4 ] ) represents only one branch of the solution , which is moreover written on the branch cut of the hypergeometric function @xmath68 . at early times ( @xmath54 , i.e. @xmath74 ) , this solution ( [ eq:4 ] ) does not oscillate and it blows up is singular at @xmath75 . one can also solve eq . ( [ eq:2 ] ) for @xmath66 and @xmath41 in the limit @xmath50 . in this case , the equation reduces to @xmath76 and the solution can be written as @xmath77\ , , \ ] ] where @xmath78 and @xmath79 are bessel functions and where @xmath80/2 $ ] . the neumann function diverges in the limit @xmath81 ( @xmath82 ) . in the tail , the corresponding behavior for the scalar field itself is given by @xmath83 and @xmath84 . ] . it is more convenient to write the general solution to the field equation , with @xmath26 , as @xmath85 , where @xmath86 and @xmath87 are two independent solutions given by : @xmath88 since @xmath69 is pure imaginary , and @xmath70 is real , one concludes easily that @xmath89 . the wronskian of these two solutions is non - zero , and can be used to relate the coefficients @xmath90 and @xmath91 so as to obtain canonical commutation relations for the field operator and its adjoint . since only one branch of the solution was given in ref . @xcite , the canonical commutation relations for the field and its adjoint could not be satisfied . more precisely , it can be checked that the solution given in eq . ( [ eq:4 ] ) is real . this is due to the fact that it involves an hypergeometric function of the form @xmath92 with @xmath93 and @xmath94 in that case and @xmath95 . therefore , one has @xmath96 and the mode function is indeed real . it follows that the wronskian of the solution considered in ref . @xcite vanishes : @xmath97 . using the properties of hypergeometric functions , one can check that both independent solutions @xmath86 and @xmath87 behave as @xmath98 in the limit @xmath54 , i.e. these mode functions blow up . this result is consistent with eq . ( [ footeq ] ) since , in the tail , the two branches @xmath86 and @xmath87 are linear combinations of the bessel functions @xmath78 and @xmath79 . therefore , we have shown that neither in the case @xmath99 nor in the case @xmath66 does the mode function behave as a plane wave in the tail . thus the initial state of the field can not reduce to the bunch - davies adiabatic vacuum , contrary to the claim @xcite : `` _ we show that there is no ambiguity in the correct choice of the initial vacuum state . the only initial vacuum is the adiabatic vacuum obtained by the solution to the mode equation _ '' . the usual prescription to remove the ambiguity on the choice of vacuum state in curved space - time , i.e. for constructing a vacuum state which is closest to the definition of vacuum in minkowski , is indeed to rely on the wkb approximation to construct vacua of successively higher adiabatic order @xcite . in this scenario @xcite , this construction can not be performed for a simple reason : the wkb approximation , which quantifies the adiabaticity of the quantum mode evolution is violated at all times for modes contained in the tail , i.e. modes with @xmath100 and @xmath101 . the wkb condition can be written in the form @xmath102 @xcite , where @xmath48 denotes the term in curly brackets in eq . ( [ eq:2 ] ) as before , and @xmath103 . in effect the wkb solution @xmath104 exactly verifies the following equation @xmath105 . therefore it is a good approximation to the solution of the actual mode equation @xmath106 if the above inequality @xmath102 is satisfied ( see also ref . @xcite for more details ) . the expression for @xmath107 is cumbersome , but since we are interested in the regime @xmath108 , we may use the limiting form of the dispersion relation : @xmath109 if @xmath26 , then @xmath110 and @xmath111 . in order to understand the behavior of @xmath112 , it is convenient to introduce the physical wave - number @xmath113 such that @xmath114 [ in the case @xmath25 , one has @xmath115 . this wavenumber @xmath116 , where @xmath117 is the physical wavenumber that gives the lower limit of the tail , as indicated in fig . using eq . ( [ eq : omega - approx ] ) , one easily derives : @xmath118.\ ] ] this formula is written to zeroth order in @xmath119 but can be expanded to arbitrary order in a straightforward way . the meaning of the physical wave - number @xmath120 is the following ( see fig . [ fig0 ] ) . if @xmath121 but @xmath122 ( i.e. , within region ii ) , the mode is outside of the tail with @xmath123 . if however @xmath124 , the mode is in the tail with @xmath125 ( region i of fig . [ fig0 ] ) . then @xmath124 means that @xmath126 which implies in turn @xmath127 , hence the wkb approximation is violated at all times in the tail . note that outside of the tail , i.e. for @xmath121 and @xmath108 ( region ii of fig . [ fig0 ] ) , the wkb approximation becomes valid . in effect @xmath128 , hence @xmath129 . if @xmath99 , then for @xmath125 ( region i or tail in fig . [ fig0 ] ) , the dominant term is @xmath130 in the expression of @xmath131 , namely @xmath132 , hence @xmath133^{-1}$ ] , which for @xmath41 ( de sitter ) and @xmath134 ( minimal coupling ) reduces to @xmath135 . in this case , it can be shown that the wkb approximation does not give the right behavior for the mode function even though @xmath136 is smaller than unity @xcite , and that the wkb approximation is not valid either . again , note that outside of the tail ( region ii of fig . [ fig0 ] ) the wkb approximation is valid . the calculation is the same as in the previous paragraph , since for @xmath121 and @xmath108 , one has @xmath137 since @xmath123 . thus one finds @xmath138 outside the tail even for @xmath139 . to summarize this discussion the wkb condition is violated by the present dispersion relation in the tail ( region i in fig . [ fig0 ] ) at all times and an initial vacuum state can not be constructed unambiguously . outside of the tail ( region ii of fig . [ fig0 ] ) , for @xmath123 or @xmath121 , the wkb approximation is a good approximation . one can also verify that the construction of an initial vacuum state by minimization of the energy content does not work in this case , see ref . this point is one major obstacle to the scenario proposed in ref . since there is no preferred initial vacuum state , all cosmological conclusions drawn depend directly on the particular choice of the initial state , hence on the choice of initial data . at the very best , one has to fine - tune the initial conditions to obtain a given amount of energy in a given part of the spectrum . the standard calculation of the amount of energy contained at late times in a given co - moving wave - number mode is done by decomposing the solution at late times ( outside the tail ) in terms of positive and negative frequency plane waves , as @xmath140 the squared modulus of the bogoliubov coefficient @xmath141 then will give the occupation number of quanta produced in the mode of co - moving wave - number @xmath142 . note that , in principle , the coefficients @xmath143 and @xmath141 can be slowly varying functions of time , and the above expression implicitly involves a wkb approximation to first order in which the time evolution of @xmath143 and @xmath141 is neglected . the corresponding vacuum is an adiabatic vacuum to first order . in ref . @xcite @xmath141 is calculated in the limit @xmath144 as @xmath145 . however the limit @xmath13 does not hold in an inflationary universe with @xmath146 and @xmath53 since @xmath147 is singular as @xmath148 . one needs to match the background evolution to a decelerated universe as @xmath149 . in effect , if one wishes to calculate the contribution of the tail modes to the energy density _ today _ , it is necessary to calculate the evolution of the modes from the inflationary era up to today . note that the dynamical evolution of the tail modes depends _ a priori _ strongly on the background scale factor dynamics . this calculation could not be performed in ref . @xcite , since the solution given in terms of the hypergeometric function is not valid at late times in the radiation dominated or matter dominated eras unless the parameter @xmath42 of the dispersion relation is tuned to the evolution of the scale factor , but the dispersion relation would become pathological as we saw before for @xmath150 with @xmath37 . furthermore , as explained above , the solution to the field equation given in ref . @xcite [ see eq . ( [ eq:4 ] ) ] describes only one branch of the solution . finally , one can not compute @xmath141 for modes still contained in the `` tail '' at late times by matching the solution to plane waves as done in ref . @xcite since for those modes , the wkb approximation is never valid so that the out solution can not be decomposed in a sum of plane waves . thus the calculation of the bogoliubov coefficient @xmath141 performed in ref . @xcite can not apply to modes contained in the `` tail '' today . in this section we calculate the amount of energy density created in quanta that redshift out of the `` tail '' , and show that it leads to a severe back - reaction problem . in ref . @xcitethe energy density contained in the tail is calculated as @xmath151 where @xmath152 is the physical wave - number such that @xmath153 today . this expression for @xmath154 is ill - defined due to the double integration element @xmath155 in the absence of a dirac function on the mass shell . the total energy density @xmath156 is defined analogously but the lower bound is extended to @xmath157 . then , it is argued that @xmath158 during inflation . note that if @xmath159 and @xmath154 is constant ( corresponding to a acuum - like equation of state as suggested ) in order to account for the dark energy then the above statement yields @xmath160 . if this holds during inflation , one faces a severe back - reaction problem since the background energy density during inflation is @xmath161 orders of magnitude below @xmath162 , and the overall calculation framework ( a test quantum scalar field on a classical background ) breaks down . as we argue in this section , it is actually a generic prediction of this model that @xmath163 _ at all times_. this result @xmath163 is in agreement with a recent work by starobinsky @xcite , which showed that models with dispersion relations such that the wkb approximation is not valid in the far past when the physical wave - number @xmath164 implies a very substantial amount of particle production . in the following we calculate the amount of energy density stored in modes with physical wave - number @xmath165 . the calculation follows the line of thought indicated in the previous section . since in the range @xmath166 the wkb approximation is valid , one can decompose the solution to the field equation in terms of plane waves as in eq . ( [ eq : planew ] ) when modes enter this regime . as long as @xmath138 one can neglect the time evolution of @xmath141 , and it is natural to interpret @xmath167 as the occupation number of particles in mode @xmath142 . as argued earlier this decomposition in plane waves can not be made for modes that are still contained in the tail . one can then calculate the amount of energy density @xmath168 stored in the log interval around the physical wave - number @xmath5 and the corresponding fractional density parameter @xmath169 in units of the background energy density : @xmath170 using @xmath171 . the fractional density parameter must be smaller than unity _ at all times _ and _ for all physical wave - numbers _ , otherwise back - reaction is significant and all semi - classical first order calculations are unreliable . in the following we calculate this quantity @xmath172 for a physical wave - number @xmath173 , i.e. , once the wavelength becomes larger than the fundamental scale . it can be expressed _ via _ @xmath141 in terms of the constants that parametrize the choice of initial data . our goal here is to study the dependence of the amount of energy density created in modes of physical momenta @xmath6 on the initial data , for which there is no definite prescription as we argued in the previous section . in the case of conformal coupling @xmath26 there exists an exact solution to the field equation written in terms of the two independent solutions @xmath86 and @xmath87 in eq . ( [ eq : mu1mu2 ] ) . one can then calculate the bogoliubov coefficient deep in the region where the wkb approximation is valid , for instance around @xmath174 by decomposing this exact solution in plane waves . however the coefficients of the hypergeometric function in term of which the exact solution hence @xmath141 are written are of order @xmath175 . for values of these coefficients that are relevant for our cosmological applications ( i.e. @xmath176 during inflation ) , the numerical calculation of the hypergeometric function turns out to be too involved and we have been unable to calculate @xmath141 in a reasonable amount of time for @xmath177 . therefore we take a different approach and approximate the exact solution in the tail @xmath178 by the solution derived in terms of bessel functions in eq . ( [ footeq ] ) , and that in the region @xmath179 by the plane wave solution . the bogoliubov coefficient @xmath141 of the plane wave solution is obtained by matching the two solutions and their first derivatives at the transition point @xmath180 . of course , it gives an approximation to @xmath141 , but as we show in the following the deviation from the overall behavior of @xmath141 away from its minimum and on its behavior around its minimum are negligible . we thus proceed as follows : in the following section , we calculate the bogoliubov coefficient denoted @xmath181 by solving for the bessel functions in the remote past and performing the matching at @xmath120 . in the subsequent section , we calculate the bogoliubov coefficient @xmath182 analytically using the exact solution and demonstrate that @xmath183 is a good approximation for values of @xmath184 as high as @xmath185 . finally we examine the behavior of @xmath181 and evaluate the amount of energy density produced by the non - adiabatic evolution of modes in the `` tail '' for realistic values of @xmath184 . this calculation is entirely analytical , only the verification of the accuracy of the approximation is numerical . as already mentioned above , see eq . ( [ footeq ] ) , the mode function in the tail can be approximatively expressed in terms of the bessel and neumann functions @xmath78 and @xmath79 as @xmath186\ , , \quad y\equiv \frac{h}{2k_{{\rm c}}}k\eta + \frac12\ln \biggl[\frac{4k_{{\rm c}}^2(4-\epsilon _ 1)}{h^2}\biggr]\ , .\ ] ] this solution is valid if the scale factor is that of the de sitter space - time : @xmath187 . the mode function must satisfy the relation @xmath188 . using the above equation , one finds that the wronskian is equal to @xmath189 . as a consequence , if one represents the coefficient @xmath190 in polar form , @xmath191 , one has @xmath192 , where we have chosen @xmath193 to be real . the parameters @xmath194 and @xmath195 will characterize in the following the choice of initial data . in the region where the wkb approximation is valid , i.e. for @xmath196 , one has @xmath197 where @xmath198 . in order to express the bogoliubov coefficient @xmath199 in terms of the constants parameterizing the choice of the initial data in the tail , @xmath200 and @xmath201 , we use the continuity of the mode function @xmath202 and of its derivative at the transition between the two regions at @xmath203 , for which @xmath204 . the result reads @xmath205 + a_2(k)\biggr[-\gamma _ kn_0({\rm e}^{y_{\rm m } } ) + \frac{h}{2}\frac{k}{k_{{\rm c } } } { \rm e}^{y_{\rm m}}n_1({\rm e}^{y_{\rm m } } ) \biggr]\biggr\ } \ , , \ ] ] where @xmath206 . working out this last expression , one obtains @xmath207 + a_2(k ) \biggr[-\biggr(\frac{k_+}{4k_{{\rm c}}}+i\sqrt{2}\biggl ) n_0({\rm e}^{y_{\rm m}})+\sqrt{2}n_1({\rm e}^{y_{\rm m}})\biggr]\biggr\}\ , .\ ] ] we are now in a position where we can compute the @xmath208 using the parametrization of the coefficients @xmath193 and @xmath190 introduced above . the final result reads @xmath209 where we have defined the rescaled variable @xmath210 by @xmath211 and where the coefficients @xmath212 , @xmath213 and @xmath214 can be expressed as @xmath215 the bessel and neumann functions are evaluated at the matching point for which their argument reads @xmath216 . a direct calculation shows that @xmath217 . the bogoliubov coefficient @xmath218 can be viewed as a two - dimensional surface parametrized by the polar coordinates @xmath219 . before studying the above bogoliubov coefficient in greater detail , one must check that the approximation is well - controlled . for this purpose , it is interesting to calculate the bogoliubov coefficient using the exact solution expressed in terms of hypergeometric functions @xmath220,\ ] ] where the functions @xmath221 and @xmath222 have been defined in eq . ( [ eq : mu1mu2 ] ) above , and the ( dimensionless ) functions @xmath223 and @xmath224 are related to each other by the wronskian normalization condition @xcite : @xmath225 \frac{{\rm e}^{2i\pi b}}{2k_{\rm c}\vert\eta_0\vert}{\cal f},\ ] ] with the numerical factor @xmath226 is written in terms of the parameters @xmath69 and @xmath70 , as : @xmath227 this solution is valid at all times since it is an exact solution of the field equation . in this case , one can calculate the bogoliubov coefficient at any time provided the wkb approximation is then valid , using : @xmath228 where , in the last expression , @xmath202 is given by eq . ( [ exact ] ) . notice that this procedure differs from the previous calculation of the bogoliubov coefficient . here , we do not perform a matching at the transition between the tail and the wkb region but rather use the exact solution ( [ exact ] ) all the way through and calculate its `` overlap '' with the wkb solution deep in the wkb region . the initial conditions enter this expression via the two constants @xmath223 and @xmath224 . we need to compare @xmath229 with @xmath230 for the same initial conditions . since @xmath231 is expressed in terms of the constants @xmath200 and @xmath201 , one needs to re - express @xmath200 and @xmath201 in terms of @xmath223 and @xmath224 . this can be done by matching the asymptotic behaviors of the two solutions deep in the tail , i.e. in the limit @xmath11 . there , the approximate solution given by eq . ( [ soljn ] ) reduces to @xmath232\biggr\}\ , , \ ] ] where @xmath233 is the euler constant , @xmath234 . on the other hand , the exact solution of eq . ( [ exact ] ) can be written as @xmath235 \nonumber \\ & & -g^*c_2\biggl[2\gamma _ { { \rm e } } + \psi \biggl(-b+d+\frac12\biggr ) + \psi \biggl(-b - d+\frac12\biggr ) \biggr ] \biggr\}\ , , \end{aligned}\ ] ] where the coefficient @xmath236 is given in terms of the euler beta function as : @xmath237 $ ] , and satisfies , since @xmath69 is pure imaginary and @xmath70 is real , @xmath238 . this relation stems from the asymptotic behavior of the hypergeometric functions for large values of their argument given by eq . ( 15.3.13 ) of ref . the di - gamma function @xmath239 function is defined by @xmath240 . if one identifies the constant term and the linear term in conformal time of the two previous relations , we obtain : @xmath241 -\psi \biggl(b+d+\frac12 \biggr)-\psi \biggl(b - d+\frac12 \biggr ) \biggr\ } \nonumber \\ & & + g^*c_2(k ) \biggl\ { \ln \biggl [ \frac{k_{{\rm c}}^2(4-\epsilon _ 1)}{h^2}\biggr ] -\psi \biggl(-b+d+\frac12 \biggr)-\psi \biggl(-b - d+\frac12 \biggr ) \biggr\}\ , , \\ a_2(k ) & = & -\pi [ g c_1(k ) + g^*c_2(k)]\ , .\end{aligned}\ ] ] then , it is sufficient to use the above relations in eq . ( [ betaka12 ] ) to obtain @xmath231 in terms of @xmath223 and @xmath224 and compare it to @xmath242 . in order to characterize the accuracy with which the bogoliubov coefficient is calculated , we plot the following quantity @xmath243 for various values of the coefficients @xmath223 and @xmath224 . more precisely , we use a polar representation and take @xmath244 while @xmath223 is real and calculated in terms of @xmath224 using the wronskian relation eq . ( [ eq : wronskian - exact ] ) . in fig . [ 3d ] , we have plotted @xmath245 for @xmath246 and @xmath247 . defined in eq . ( [ error ] ) that quantifies the deviation of the approximate @xmath248 with respect to its exact form . @xmath249 is depicted as a function of @xmath250 that parameterize the initial conditions ( @xmath251 ) of the exact solution . right panel : zoom around the origin in order to unveil the two peaks structure.,title="fig:",scaledwidth=49.0% ] defined in eq . ( [ error ] ) that quantifies the deviation of the approximate @xmath248 with respect to its exact form . @xmath249 is depicted as a function of @xmath250 that parameterize the initial conditions ( @xmath251 ) of the exact solution . right panel : zoom around the origin in order to unveil the two peaks structure.,title="fig:",scaledwidth=49.0% ] we see that the error for large values of @xmath252 is less than @xmath253% and is constant , i.e. the offset between the two bogoliubov coefficients does not depend on @xmath252 and @xmath195 in a first approximation . for @xmath254 , the error increases to 1 ; this artefact comes from the fact that the minima where the two bogoliubov coefficients vanish are slightly offset one from the other . if one coefficient vanishes while the other remains finite and non - zero , then the value of @xmath249 is pushed toward one , @xmath255 . this error however is of no consequence for what follows . indeed we will not be interested in the location of the minimum but in the behavior of @xmath141 around the minimum and far from the minimum . as is obvious from fig . [ 3d ] , these behaviors match closely in both cases and our approximation will be sufficient for our purposes . we have checked that the function @xmath249 remains the same for other values of @xmath256 which allow numerical calculations , i.e. @xmath257 $ ] . the situation is in fact very similar to the standard calculation of the power spectrum in an inflationary theory : in principle , one can not match two different branches at a point where the approximation breaks down ( for standard inflation this occurs at first horizon crossing ) . however , since the approximation is only violated in a small region one expects the corresponding result to be correct at leading order . this is indeed the case for inflation for which the amplitude of the spectrum is predicted up to a factor of order unity and the spectral slope is unchanged by the matching . here we also find that @xmath258 . since we have demonstrated that the approximation to @xmath259 is quite reasonable , we now study the behavior of eq . ( [ betakrphi ] ) for more realistic values of the ratio @xmath260 . the bogoliubov coefficient possesses an absolute minimum with @xmath261 located at @xmath262 using the notations defined previously . one should not be surprised to find a minimum with @xmath261 since one can express the matching of the two branches of the solution and their first derivatives at @xmath263 as two equations relating the coefficients @xmath200 and @xmath201 as a function of @xmath143 and @xmath141 and find a solution with @xmath261 . the wronskian normalization condition is always satisfied by both branches of the solution . this solution with initial conditions ( @xmath264 , @xmath265 ) corresponds to a choice of initial data such that at late times , when modes have exited from the tail , their quantum state is that of an adiabatic vacuum . note therefore that there is no naturalness in choosing these initial conditions since the adiabatic vacuum is only a late time consequence of such initial data . furthermore one can show that for generic initial data , the state of the quantum field at late times is not an adiabatic vacuum , hence quanta have been produced . indeed the behavior of @xmath266 around this absolute minimum can easily be established . from a taylor expansion , one obtains @xmath267 for a crude order of magnitude estimate , one can develop the bessel functions to first order in the small argument limit @xmath268 ( more exactly , for de sitter inflation and @xmath269 , one has @xmath270 ) . this leads to @xmath271 and @xmath272 . thus in order to avoid a back - reaction problem , the initial conditions @xmath252 and @xmath195 must not differ too much from @xmath264 and @xmath265 which lead to @xmath261 ( hence a zero amount of energy density created ) . more precisely , the energy density produced is of the order of the background energy density , i.e. @xmath273 , when @xmath252 or @xmath195 respectively depart from the minimum by amounts @xmath274 or @xmath275 given by : @xmath276\ , , \quad \delta \phi \simeq { \cal o}\biggl[\frac{h}{m_{\rm pl}}\ln ^{-2}\biggl ( \frac{m_{\rm pl}}{h}\biggl)\biggr]\ , .\ ] ] here we assumed @xmath269 . hence the corresponding fine - tuning of the initial conditions is of order @xmath277 ( if one assumes a uniform measure in @xmath252 and @xmath195 in parameter space ) . one should note that the above constraint is valid for a given co - moving wave - number @xmath142 and has been calculated at a time when @xmath278 . since the fractional density parameter of quanta extracted out of the vacuum @xmath279 must be smaller at all times during inflation , i.e. for a range of co - moving wave - numbers @xmath142 since @xmath142 and @xmath12 can be related by the above constraint @xmath278 , the above constraint on @xmath264 and @xmath265 rather applies to a continuum of values of co - moving wave - numbers . in other words one does not have to fine - tune two parameters characterizing the initial data but a whole continuum of parameters , i.e. the functions @xmath280 and @xmath281 themselves . the dependence in @xmath142 of these functions is hidden in the argument of the bessel and neumann functions @xmath282 , since @xmath283 depends on @xmath142 . one can also perform a similar calculation of the bogoliubov coefficient when the coupling is no longer conformal @xmath52 . in this case the calculation can be performed analytically for all background scale factors . for the sake of simplicity we choose minimal coupling @xmath25 but this can be trivially expanded to various choices of the coupling to gravity , and does not modify the conclusions we derive below . if @xmath25 , the evolution of the modes is dominated by @xmath56 in the tail , i.e. when @xmath125 ( @xmath285 ) , and the solution can be written as : @xmath286,\ ] ] where @xmath287 and @xmath288 are two dimensionless @xmath289dependent constants , and @xmath290 is some initial conformal time . one can check that this solution and the power - laws given in eq . ( [ eq : power - law ] ) are the same . here one can not choose the time of matching @xmath263 to the wkb solution , since the matching has to be done when @xmath291 , i.e. when @xmath292 or @xmath180 . in the region @xmath123 , i.e. for @xmath293 , the wkb approximation is valid as we have seen before , and the matching to the wkb form is justified at @xmath294 . for a given wave - number @xmath142 , we are free to set @xmath295 , since this amounts to a redefinition of the constants @xmath287 and @xmath288 by a function of @xmath142 . the matching at @xmath263 then gives : @xmath296 , \quad\quad \beta_k \ , = \ , \frac{-i}{\sqrt{4k\omega_k}}\left [ c_+ \biggl(\gamma_k + \frac{a'}{a}\biggr ) -\frac{c_-}{\eta_{\rm m}}\right]\ , , \ ] ] with the function @xmath297 as above [ see eq . ( [ eq : beta - approx1 ] ) ] and where all quantities in the above two equations are understood to be taken at @xmath298 . in particular , at time @xmath263 , @xmath299 . since @xmath108 at @xmath263 one can use the limiting form of @xmath4 given in eq . ( [ eq : omega - approx ] ) , hence : @xmath300 the constants @xmath287 and @xmath288 are related to one another by the normalization of the mode functions : @xmath301 , which gives : @xmath302 and as before we keep @xmath303 and @xmath304 $ ] as the two independent parameters characterizing the choice of initial data . one finally derives the squared modulus of the bogoliubov coefficient @xmath141 as : @xmath305 + \frac{r^2}{4\beta ^2}\sqrt{\frac{\beta}{1+\beta}}\frac{h_{\rm m}}{k_+ } -\frac12\sqrt{\frac{\beta}{1+\beta}}\left(1+\frac{k_+}{4k_{\rm c}}\right){\rm cot}\phi - \frac{1}{2}.\ ] ] since @xmath306 is a large number , in the following we use the rescaled variable @xmath307 instead of @xmath194 . equation ( [ eq : beta - mu - a ] ) above is particularly attractive because it has exactly the same functional shape as eq . ( [ betakrphi ] ) . it allows us to understand analytically the behavior of the amount of energy density produced in modes with @xmath173 as a function of the initial data . the occupation number @xmath308 has an absolute minimum located at : @xmath309 ^ 2 + ( 1+\beta ) /\beta}}.\ ] ] as before the occupation number vanishes exactly at this minimum , but the back - reaction problem can not be avoided for generic initial conditions . in the present case it is not possible to make a sensible contour plot of @xmath310 since this function changes by many orders of magnitude over very small intervals of @xmath311 . therefore , we take a conservative approach in which we calculate @xmath312 as a function of @xmath210 for the values of @xmath313 that minimize this quantity at each @xmath210 . we also evaluate @xmath314 as a function of @xmath195 for the values of @xmath252 that minimize this quantity at each @xmath195 . in other words , we solve @xmath315 for @xmath313 as a function of @xmath210 and @xmath316 for @xmath252 as a function of @xmath195 : @xmath317 ^ 2+(1+\beta)/\beta } { 1+k_+/(4k_{\rm c})}\biggr\ } \ , , \quad \rho_{\rm min}^4(\phi)\,=\,\frac{\beta ^4}{\sin^2\phi}\left[\left(1+\frac{k_+}{4k_{\rm c}}\right)^2 + \frac{1+\beta}{\beta}\right],\ ] ] $ ] plotted as a function of the rescaled variable @xmath210 characterizing the initial data . this plot corresponds to an inflationary era with a de sitter metric , and hubble parameter @xmath318 . the other parameter of initial data is @xmath319 . allowed regions correspond to @xmath320 , and are peaked around a particular value of @xmath210 . the minimum is in fact @xmath321 , corresponding to @xmath322 , but can not be seen in the figure due to insufficient resolution . the dotted line provides a continuation of the numerical result to the analytical value at that point . in most of parameter space , the energy density is too large by @xmath32310 orders of magnitude . right panel : same as left panel for @xmath324 in a matter dominated era . in nearly all of parameter space the energy density is too large by @xmath325 orders of magnitude [ fig : f1_ds].,title="fig:",scaledwidth=49.0% ] $ ] plotted as a function of the rescaled variable @xmath210 characterizing the initial data . this plot corresponds to an inflationary era with a de sitter metric , and hubble parameter @xmath318 . the other parameter of initial data is @xmath319 . allowed regions correspond to @xmath320 , and are peaked around a particular value of @xmath210 . the minimum is in fact @xmath321 , corresponding to @xmath322 , but can not be seen in the figure due to insufficient resolution . the dotted line provides a continuation of the numerical result to the analytical value at that point . in most of parameter space , the energy density is too large by @xmath32310 orders of magnitude . right panel : same as left panel for @xmath324 in a matter dominated era . in nearly all of parameter space the energy density is too large by @xmath325 orders of magnitude [ fig : f1_ds].,title="fig:",scaledwidth=49.0% ] $ ] plotted as a function of the phase characterizing the initial data . this plot corresponds to an inflationary era with a de sitter metric , and hubble parameter @xmath326 . as indicated in the text , the other parameter of initial data ( a modulus ) has been eliminated for the phase @xmath195 by minimizing @xmath167 . allowed regions correspond to @xmath320 , and are seen to be peaked around a particular value of @xmath195 . the plot is symmetric in the interchange @xmath327 . note that the ordinate scale is in @xmath328 : in most of parameter space , the energy density is too large by @xmath32310 orders of magnitude . right panel : same as left panel for @xmath324 in a matter dominated era . in nearly all of parameter space the energy density is too large by @xmath325 orders of magnitude , i.e. @xmath329[fig : f1_minimal_ds].,title="fig:",scaledwidth=49.0% ] $ ] plotted as a function of the phase characterizing the initial data . this plot corresponds to an inflationary era with a de sitter metric , and hubble parameter @xmath326 . as indicated in the text , the other parameter of initial data ( a modulus ) has been eliminated for the phase @xmath195 by minimizing @xmath167 . allowed regions correspond to @xmath320 , and are seen to be peaked around a particular value of @xmath195 . the plot is symmetric in the interchange @xmath327 . note that the ordinate scale is in @xmath328 : in most of parameter space , the energy density is too large by @xmath32310 orders of magnitude . right panel : same as left panel for @xmath324 in a matter dominated era . in nearly all of parameter space the energy density is too large by @xmath325 orders of magnitude , i.e. @xmath329[fig : f1_minimal_ds].,title="fig:",scaledwidth=49.0% ] and plot @xmath279 for @xmath330\vert^2 $ ] in figs . [ fig : f1_ds ] and @xmath331\vert^2 $ ] in figs . [ fig : f1_minimal_ds ] ( for respectively de sitter inflation and today ) . one clearly sees from these figures that @xmath332 for most values of @xmath195 , which corresponds to our previous expectations , i.e. the amount of energy density created in quanta with @xmath333 is of order @xmath7 . the behavior of @xmath167 around the local minimum can be studied in the same way as before and one obtains : @xmath334 ^ 2\left(\phi-\phi_{\rm min}\right)^2\ , , \quad\quad ( \phi\sim \phi_{\rm min}).\end{aligned}\ ] ] if we write the value @xmath335 such that @xmath336 ( i.e. similar amount of energy created in quanta with @xmath173 than in the background ) , then this value is reached if @xmath210 and @xmath195 departs from @xmath337 and @xmath265 by an amount @xmath338 , @xmath275 , with : @xmath339.\ ] ] in other words , the fine - tuning necessary to avoid back - reaction is of order @xmath340 ( neglecting the ln term ) . during inflation @xmath341 , and today @xmath342 . although the fine - tuning can be considered as not too severe during inflation , we see that , during the matter epoch , it is as severe as the usual fine - tuning of the cosmological constant problem . therefore , the scenario of refs . @xcite does not improve the situation . at this point , it should be emphasized that the above problem is a generic feature of the dispersion relation used in refs . in order for the trans - planckian effects to modify the power spectrum of the fluctuations , the physical modes of interest must spend some time in a region where the wkb approximation is violated . as already mentioned , this implies production of particles and the energy density associated to these particles must not exceed the background energy density . this implies some constraints on the occupation number @xmath343 . it has been shown in ref . @xcite that these constraints are quite stringent if the production is taking place today . usually , these tight constraints can be avoided by requiring that the violation only occurs during inflation where the problem is less severe . this can be achieved if the dispersion relation is such that @xmath344 for all trans - planckian wave - numbers , with @xmath345 the hubble constant some unspecified time after inflation . such an example is provided in fig . 2 of ref . then as was argued in the discussion following eq . ( [ eq : omega - approx ] ) the wkb approximation should be valid at all times after inflation and adiabaticity restored . here however , since @xmath346 as @xmath347 , there is at all times a region where the wkb is violated . therefore the class of dispersion relations used in refs . @xcite suffer in a generic way from the problem discussed in ref . @xcite . furthermore , we also note as previously that the above calculation of the fine - tuning holds for a given co - moving wave - number @xmath142 . similar constraints apply for other wave - numbers but the values of @xmath348 and @xmath264 are shifted from the above [ notably because of the choice @xmath349 . therefore one must not only pick one right value of @xmath195 to one part in @xmath350 , but a whole continuum of values of @xmath351 such that back - reaction is avoided for all of these values . up to now we have argued that : _ ( i ) _ in the scenario proposed in refs . @xcite there is no preferred initial vacuum state , hence all conclusions drawn depend on the ad - hoc choice of initial data ; _ ( ii ) _ for arbitrary values of the two parameters characterizing this choice of initial data one finds that energy in excess of the background energy density is produced due to the non adiabatic evolution of modes in the tail . in a separate publication , we have constructed an effective energy - momentum tensor for scalar field theories with non - linear dispersion relations , and we have shown that dispersion relations of the form of that proposed in refs . @xcite generically led to the wrong equation of state . this finding has been challenged by bastero - gil and mersini recently , who argue that the energy - momentum tensor we have constructed is ill - defined as it ( supposedly ) is not divergenceless . explicitly , in ref . @xcite , the _ vev _ for the energy density and pressure are given by @xmath352 , \label{rhoeff}\\ \langle p\rangle&=&\frac{1}{4\pi^2a^4}\int { \rm d } k k^2 \left[a^2\left|\left(\frac{\mu_k}{a}\right)'\right|^2 + \left ( \frac{2}{3}k^2\frac{{\rm d}\omega^2}{{\rm d}k^2}-\omega^2\right ) \left|\mu_k\right|^2\right ] , \label{peff}\end{aligned}\ ] ] and the integrals extend from @xmath353 to @xmath354 . reference @xcite claims that @xmath355 , with @xmath356 . however , noting that the co - moving frequency @xmath357 , a straightforward calculation gives : @xmath358\left(\mu_k^{*\prime } - { \cal h}\mu_k^*\right ) + \left(\mu_k^{\prime } - { \cal h}\mu_k\right ) \left[\mu_k^{*\prime\prime } - { \cal h}\mu_k^{*\prime } + \left({\cal h}^2 - { a''\over a}\right)\mu_k^*\right ] \nonumber\\ & & - 4{\cal h}\left\vert\mu_k ' - { \cal h}\mu_k\right\vert^2 + \omega^2(k)\left(\mu_k'\mu_k^*+ \mu_k\mu_k^{*\prime}\right ) - 2{\cal h}k^2 { { \rm d}\omega^2 \over { \rm d}k^2}\vert\mu_k\vert^2 -2{\cal h}\omega^2(k)\vert\mu_k\vert^2\biggr\ } \nonumber\\ & = & \frac{1}{4\pi^2a^4}\int { \rm d } k k^2 \biggl\ { \left(\mu_k '' - { a''\over a}\mu_k\right)\left(\mu_k^{*\prime } - { \cal h}\mu_k^*\right ) + \left(\mu_k^{\prime } - { \cal h}\mu_k\right ) \left(\mu_k^{*\prime\prime } - { a''\over a}\mu_k^*\right ) -6{\cal h}\left\vert\mu_k ' - { \cal h}\mu_k\right\vert^2 \nonumber \\ & & + \omega^2(k)\left[\mu_k\left(\mu_k^{*\prime } - { \cal h}\mu_k^*\right ) + \mu_k^*\left(\mu_k^\prime - { \cal h}\mu_k\right)\right ] - 2{\cal h}k^2 { { \rm d}\omega^2 \over { \rm d}k^2}\vert\mu_k\vert^2\biggr\ } \nonumber\\ & = & \frac{1}{4\pi^2a^4}\int { \rm d } k k^2 \biggl ( -6{\cal h}\left\vert\mu_k ' - { \cal h}\mu_k\right\vert^2 - 2{\cal h}k^2 { { \rm d}\omega^2 \over { \rm d}k^2}\vert\mu_k\vert^2\biggr),\end{aligned}\ ] ] where the field equation @xmath359 has been used in the last step . since @xmath360 the energy conservation condition @xmath361 is trivially satisfied . we take advantage of this section to point out that the energy - momentum tensor we have constructed in a previous publication @xcite is well behaved and the construction is entirely consistent . in effect we constructed a generally covariant lagrangian for a scalar field with a non - linear dispersion relation . the breaking of the lorentz invariance is explicited by introducing a dynamical four - vector @xmath362 in the lagrangian whose role is to define the preferred rest frame while preserving general covariance at the same time , following previous work by jacobson and mattingly @xcite ( see also references therein ) . the energy - momentum tensor derived by varying the action with respect to the metric is the sum of the energy - momentum tensor of the scalar field and that of @xmath362 . we have restricted ourselves to flrw space - times by choosing @xmath363 . this approach is consistent as @xmath362 satisfies its field equation [ see eq . ( 15 ) in ref . @xcite ] , and the scalar field also satisfies its field equation . in frlw space - time , the energy - momentum tensor of the four - vector ( i.e. the part which depends only on @xmath362 ) vanishes since @xmath362 is constant @xcite , hence the remainder is the scalar field energy - momentum tensor , which is conserved as shown above . we consider this scalar field as a test field , which we quantize on the curved background . the dynamics of the background can be specified by adding matter fields to the action without modifying our approach . indeed the matter field energy - momentum tensor will be separately conserved . therefore our earlier criticisms on the equation of state of the trans - planckian modes apply . we also note that a recent paper by frampton @xcite argues that by adding higher order terms of the form @xmath364 in the effective lagrangian , one can obtain a correct equation of state , i.e. similar to that of vacuum . here @xmath365 denotes a three - dimensional laplacian expressed in a generally covariant way on space - like hyper - surfaces ( see refs . @xcite for explicit definitions ) . the choice @xmath363 then reduces this term to the usual three - dimensional laplacian on constant time hyper - surfaces for flrw space - times . however , as should be obvious , the only terms that can enter the energy - momentum tensor _ via _ this new term always contain derivatives of the form @xmath366 when @xmath367 . those terms vanish when one picks @xmath362 to be constant as above . for @xmath368 , i.e. the lowest order term of the expansion , one can check that the variation of @xmath369 with respect to the metric or its first derivative always induces a term proportional to a derivative of @xmath362 that is either spatial or temporal . this calculation can be found in eqs . ( b3 ) and ( b4 ) of ref . @xcite ( the substitution @xmath370 in this calculation can be made without changing the result since the equation is written in a generally covariant way ) . therefore we conclude that the addition of terms proposed by frampton can not lead to any difference with respect to the energy - momentum tensor we proposed earlier . hence such terms can not account straightforwardly for a vacuum - like equation of state in this scenario . the difference with the conclusion of ref . @xcite probably lies in the choice of normalization of @xmath362 : ref . @xcite claims that , in a cosmological context , @xmath371 ( @xmath147 is the scale factor ) . however , @xmath372 is a scalar , while @xmath60 is the component of a rank two tensor so that this choice of normalization is manifestly not covariant . of course , it can be made covariant at the expense of the introduction of a second vector field but this does not seem to be the case in ref . this implies that the lagrangian proposed by frampton is generally not covariant . of course the correct normalization for @xmath362 in a cosmological context is always @xmath373 ( up to the sign convention ) , and @xmath362 can be written as @xmath374 if the frlw metric is written in terms of conformal time , or @xmath375 if the metric is written in terms of cosmic time . to finish , let us also note that it is claimed in ref . @xcite that `` _ the tail modes are still frozen at present ... thus the energy of the tail is a contribution to the dark energy of the universe : up to the present it has the equation of state of a cosmological constant term _ '' . here , that the tail modes are frozen means @xmath376 , the solution to the field equation when the term @xmath56 dominates ( for @xmath25 ) . but there is no logical relationship between these modes being frozen and the equation of state being that of a cosmological constant . as a matter of fact the equation of state of the modes of wavelengths larger than the horizon size for a scalar field with a linear dispersion relation takes the form @xmath377 . the derivation of the equation of state requires to define correctly a stress - energy tensor when such non - linear dispersion relations are taken into account ( and thus when lorentz invariance is broken ) as we did in ref . @xcite . note also that there is a contradiction between the above claim of ref . @xcite and the whole content of the recent paper by bastero - gil and mersini @xcite which calculates the equation of state of the trans - planckian modes and finds that it is not the equation of state of the vacuum but that it approaches it only at late times . in this article , we have examined in detail the scenario proposed in refs . @xcite which attributes the dark energy to the properties of a scalar field with a dispersion relation that decreases exponentially with trans - planckian wave - number @xmath378 . we have demonstrated that this mechanism does not work , mainly for two reasons . ( i ) the mode function of the scalar field does not behave as a plane wave as @xmath54 in the so - called `` tail '' ( i.e. , the part of the non - linear dispersion relation where @xmath125 and @xmath379 ) . this implies that there is no definite prescription for constructing a well - defined initial vacuum state , hence that the choice of initial data is entirely arbitrary . this situation is similar to the problem of setting initial data for cosmological perturbations in the absence of an accelerated expansion era ( but with linear dispersion relations ) , in which case the data would have to be specified on super - horizon scales where the mode function does not oscillate and is frozen by the expansion . in the scenario of refs . @xcite the wkb approximation is not valid at all times for modes in the tail . this explains that the notion of adiabatic vacuum can not be used to set up initial data and the initial state proposed in ref . @xcite is thus ad - hoc . it also brings us to the second objection against this scenario : ( ii ) since all modes originate in the tail of the dispersion relation , the breakdown of the wkb approximation for a given physical wave - number at early times implies the continuous production of a substantial number of quanta with physical wave - numbers @xmath380 . the breakdown of the wkb approximation can indeed be seen as the signal of a strongly non - adiabatic evolution which is generically associated with particle production in expanding space - times . we have calculated the amount of energy density produced in modes of physical wave - number @xmath18 as a function of the two parameters that characterize the ( arbitrary ) choice of initial data . we have shown that this energy density is generically of order @xmath7 . the production of energy density in quanta with wave - numbers @xmath6 in excess of the background energy density @xmath381 implies the breakdown of the perturbative semi - classical framework used for the calculation , and renders all claims irrelevant . there is a small region of parameter space in which this energy density can be tuned down to zero , but the fine - tuning in the choice of initial data is of order @xmath3 . such a fine - tuning at the time of inflation is of order @xmath382 and is probably acceptable , but today , it is of the same order as the celebrated fine - tuning of the cosmological constant problem .

characteristics of the depression contribution functions are studied for the stokes line profiles formed in a magnetic field . the form of the depression functions depends mainly on the strength of splitting and the zeeman component intensity , and is of a complicated character with a distinctly pronounces asymmetry . the depths of formation of magnetically sensitive lines are found by means of these contribution functions . the calculations reveal that the steep section of the line profile is formed higher than the profile center when a strong longitudinal magnetic field is present . the stokes profiles that describe the polarization characteristics are formed only several kilometers higher than the stokes profile that specifies the general depression of the unpolarized and polarized radiation . the averaged depth of formation of the whole line profile is practically independent of the magnetic field strength . the depths of formation of 17 photospheric lines usually used in magnetospectroscopic observations are calculated for the models of the quiet photosphere , a flux tube , and the sunspot umbra . [ multiblock footnote omitted ] main astronomical observatory , national academy of sciences of ukraine + zabolotnoho 27 , 03689 kyiv , ukraine + e - mail : [email protected]

1210.4023

the determination of depths of formation of the observed stokes parameter profiles for magnetically sensitive fraunhofer lines still remains a topical problem of the spectral analysis which is used for obtaining information on the fine structure of magnetic field in the solar photosphere . the problem of the depth of line formation may be considered completely solved for the fraunhofer lines which are formed in the regions without a magnetic field , in the so - called quiet photosphere , while the present state of the problem dealing with the depths of formation of the stokes line profiles still remains unsatisfactory . in observational programs for studying the magnetic field structure in the photospheric layers of the sun , approximate estimates are usually used for selecting lines with necessary characteristics . thus , for example , to determine the depth of formation of a fraunhofer line , its equivalent width is estimated . it is considered that the larger the equivalent width the higher in the atmosphere the line is formed . the magnetic and temperature sensitivity of a line is determined by the value of such atomic parameters of the line as the excitation potential , ionization potential , and land factor . the smaller the sum of potentials the more sensitive the line is to the temperature , and the greater the land factor the more magnetically sensitive the line is . though it is possible to get approximate estimates from such reasonings , more reliable information on the depths of formation of the fraunhofer line profiles and the stokes parameter profiles is necessary for a detailed investigation of relationships between the physical parameters of the atmosphere and the depth . it is not long ago that we have begun the theoretical investigation of the stokes parameters of magnetically sensitive fraunhofer lines . the calculation algorithm for the stokes profiles , for contribution and response functions was developed on the basis of the theory of line formation in a magnetic field for the condition of the local thermodynamic equilibrium ; the theory was developed first by unno and improved by rachkovskii , beckers , and landi deglinnocenti . the algorithm and structure of the calculation program spansatm can be found in our paper @xcite . this program was produced on the basis of the spansat program @xcite intended for calculation and analysis of line profiles in stellar atmospheres . the program system spansatm enables to carry out the complete analysis of magnetically active photospheric lines in the solar spectrum . our paper @xcite gives the results of investigations of the effect of physical conditions in a magnetic medium and of atomic parameters of lines on the shape of the stokes profiles and on the magnetic broadening of the fraunhofer lines . in the present paper , the attention is concentrated mainly on the study of properties of the depression contribution functions and on determination of the depths where the stokes profiles of photospheric lines are formed . the results of analysis of the response functions of the stokes parameters will be given in a next paper . when the depths of line formation are calculated , the contribution functions which describe the contribution of atmospheric layers to the observed value of relative absorption in the spectral line are usually used . besides the contribution functions , the response functions are used in the interpretation of observations . they serve for determination of the response of an observed quantity to a given disturbance . these two functions supplement well each other and together they can give a complete enough information about physical conditions in the medium where spectral line profiles are formed . depression contribution functions are the functions which show what part of the radiation emerging at the surface ( relative to the continuum intensity @xmath0 ) is absorbed by each of the atmospheric layers in a given region @xmath1 of the spectral line . this function defines the value of the contribution to the relative absorption @xmath2 { / } { i_c } $ ] . here @xmath3 is the intensity of the radiation emerging in the region @xmath1 of the line . the depression contribution function is equal to zero when the absorption coefficient in the line is zero and the source function in the line @xmath4 is not equal to the intensity of continuous radiation , i.e. , @xmath4 should be greater or smaller than @xmath0 . in other words , the depression contribution function is significant if there are atoms which absorb in the given line and the intensities of re - emitted and absorbed radiation are different . it is no mere chance that we have chosen the depression contribution functions as a tool for investigating the regions where the line absorption processes are localized and for determining the depths of formation of absorption lines . this choice is based on the results of gurtovenko and sheminova @xcite and conclusions of magain @xcite , where a formal solution of the radiation transfer equation was obtained for a relative line depression in the integral form and where the integrand was shown to be the only correct depression contribution function fit for determining the depth of formation of an absorption line . gurtovenko and sarychev @xcite demonstrate that a depression contribution function proposed earlier which was derived using the unsld - pecker procedure of weight functions is identical to the depression contribution function from @xcite . at present , the depression contribution function proposed first by gurtovenko , ratnikova , and de jager @xcite and substantiated theoretically by magain @xcite is widely used in calculating the depths of formation of absorption lines and seems to be beyond any doubt by now . very few studies have been dedicated to the contribution functions of the stokes parameters and to depths of their formation due to difficulties in calculations of the stokes profiles . the depths of formation of the stokes profiles were studied for the first time by rachkovskii @xcite , who compared the variations with depth of the emergent radiation intensities in continuum and in lines and drew the conclusion that the polarization parameters of radiation refer to higher layers than the line intensity does . the contribution functions were already used by staude @xcite to determine the depth of formation of magnetically sensitive lines , but only the contribution to line emission was considered . these studies had some drawbacks the depths found in them were not the depths of absorption line formation , but characterized the regions of effective radiation in a given line . in order to determine the depth of line formation in the presence of a magnetic field , it is necessary to obtain the depression contribution functions for the stokes parameters and to calculate the depth of formation , using these functions . the progress in studying the depths of formation of the stokes parameters began with paper @xcite by van ballegooijen , though the depths of formation were not determined directly there . the method proposed in @xcite for solving the transfer equation for polarized radiation enabled to write a single equation in a matrix form instead of four transfer equations for the stokes parameters ; in that equation , the polarized radiation was represented as the matrix @xmath5 where @xmath6 , @xmath7 , @xmath8 , @xmath9 are the stokes parameters describing the intensity and polarization characteristics of radiation . a formal solution of the transfer equation was obtained in the integral form : @xmath10 where the integrand @xmath11 is a matrix whose elements define the contribution functions for the stokes parameters . these contribution functions are called the emission contribution functions of the stokes parameters . they describe the contribution of radiation of atmospheric layers to the polarized radiation emerging in the line , and the integral of the contribution function along height gives the value of the stokes parameters . thus , the author of @xcite succeeded in representing each of the stokes parameters as an integral of an ordinary function along height . this paper stimulated the deriving of the depression contribution functions for the stokes parameters of the fraunhofer lines . grossman - doerth et al . @xcite , using the method of @xcite and the procedure of @xcite , were the first to compose the transfer equation for the relative depression of propagating polarized radiation , this depression being represented as the matrix @xmath12 the matrix elements in @xmath13 contain the stokes parameters of relative depression : @xmath14 ; @xmath15 ; @xmath16 ; @xmath17 . a formal solution of the transfer equation for the matrix @xmath13 is the integral @xmath18 where the integrand @xmath19 is also a matrix . the depression contribution functions ( similar to the emission ones ) for each of the stokes parameters are defined in terms of the @xmath19 matrix elements . mathematical expressions of the contribution functions for the stokes parameters can be found in @xcite . we used the depression contribution functions @xcite for the stokes parameters in our study to determine the depths of formation of the stokes profiles of magnetically active absorption lines , denoting them as @xmath20 , @xmath21 , @xmath22 , @xmath23 in accordance with the stokes parameters . in a general case , a spectral line in a magnetic field splits into three groups of components . the first group is characterized by absorption of the radiation polarized linearly in the direction of the magnetic field ( the @xmath24 components ) . the second group ( the @xmath25 components ) is characterized by absorption of the radiation right - hand circularly polarized in the plane perpendicular to the direction of the field . the components of the third group ( the @xmath26 components ) are left - hand circularly polarized in the same plane . the observed relative line depression described by the stokes parameter @xmath27 which represents the absorption of both unpolarized and polarized light in a certain section of the line profile contains information on all the components . the profile of @xmath27 is to be considered as a superposition of profiles of all three component groups . in consequence of this , the function of contribution to the relative depression @xmath20 may be of a complex character . as the contribution function @xmath20 contains contributions of all components , its shape will depend on the spacing between the groups of components and on their intensities . since the @xmath28 components remain undisplaced in the magnetic field , the spacing between the @xmath29 groups which is determined according to the splitting rule is of importance . the intensities of each group of components depend on the relative selective absorption coefficient at the line center , on the inclination of magnetic field , and on the value of anomalous dispersion . thus , the contribution function @xmath20 is determined by the effective land factor @xmath30 , magnetic field strength , wavelength , number of absorbing atoms , doppler width , damping constant , magnetic field inclination , etc . the number of parameters is so great that a detailed investigation of contribution functions is very laborious . it seems reasonable to study only characteristic properties of the contribution functions . with that end in view , we calculated the contribution functions for the stokes parameters of a line and separately for each group of components depending on the profile section @xmath31 , the inclination angle @xmath32 , and the magnetic field strength @xmath33 . while calculating contribution functions for one group of components , we assumed other functions to be equal to zero . the spectral line fe i 643.085 nm was selected for calculations . as a model of the region where the line absorption is formed , we took the empirical model of the magnetic flux tube @xcite which was obtained by comparing calculated profiles of the stokes parameters @xmath6 and an observed line profile in a plage area . the microturbulent velocity was assumed to be 1 km / s , the damping constant was 1.5@xmath34 , and the iron abundance was 7.64 dex . [ fig1 ] analysis of the results obtained revealed a clear dependence of the contribution function profiles for the stokes parameters on the intensity of the zeeman components . figures 1 and 2 show the profiles for the parameters @xmath27 , @xmath35 , @xmath36 and the corresponding contribution functions @xmath20 , @xmath21 , and @xmath23 for the line and for its groups of components at different profile sections @xmath31 which are formed in a magnetic flux tube with a magnetic field strength of 0.1 t , an inclination of @xmath37 , and an azimuth of @xmath38 . it is evident that the contribution functions for the parameter @xmath27 are asymmetric . a second maximum is outlined in those functions @xmath20 which were calculated for a steep part of the line profile , for @xmath39 pm , for example . the contribution functions @xmath20 have a shape close to a symmetric one only for the center and for distant wings of the line . the contribution functions of parameters @xmath27 , @xmath35 , and @xmath36 can take on the negative values also , as they characterize the difference in absorptions of polarized radiation for two directions . the positive and negative maxima of the @xmath21 function change places in going from the line center to the wing , which is evidence of a change in the direction of light polarization along the line profile . the positive and negative maxima of the contribution function @xmath23 characterize the predominance of contribution to the absorption of the right - hand and left - hand circularly polarized light , respectively . their contributions are equal at the line center , and so @xmath23 is equal to zero here . in going to the line wing , the value of the @xmath23 maxima varies in accordance with the contribution of the @xmath29 components . one can see in figs 3 and 4 how the stokes profiles @xmath27 , @xmath35 , @xmath36 and their contribution functions calculated for the profile sections @xmath40 , 5 , 10 pm vary when the inclination increases from 0@xmath41 to 90@xmath42 . the function @xmath20 is the most asymmetric for @xmath43 and @xmath44 pm . the contribution of the undisplaced @xmath28 component grows with increasing @xmath32 . this `` smooths '' the @xmath20 contribution function , and its asymmetry diminishes . the maximum values of the @xmath21 function become the greatest at @xmath45 and @xmath46 pm . the greatest positive and negative maxima of the @xmath23 function are observed at @xmath43 , @xmath46 pm and@xmath47 , @xmath48 pm , respectively . ( points ) , 5 pm ( squares ) , 10 pm ( triangles ) . ] the effect of the magnetic field strength on the shape of profiles and contribution functions for the sections @xmath49 , 5 , 10 pm is demonstrated in figs 5 and 6 . the profile shape for @xmath27 varies with increasing field strength . the @xmath20 functions for @xmath31 = 0 , 5 pm shift slightly to deeper layers of the atmosphere . their width increases , and the depression grows for @xmath50 pm with growing @xmath33 , and the values of maxima in the functions @xmath20 grow correspondingly . in the functions @xmath23 , the negative maximum diminishes with increasing @xmath33 at the same distances from the line center , as the @xmath29 group moves away from the line center , and this means that its intensity becomes less at the same @xmath31 . an increase of the field strength favours a `` clearing '' of the right - hand polarized absorption from the left - hand polarized one in the right wing of the line profile . it should be noted that the value of the land factor also affects strongly the shape of the contribution functions . the larger @xmath30 the greater is the splitting of components , and the cases are possible when the components separate entirely , i.e. , a complete splitting occurs , and then the shape of the contribution functions does not differ from a usual symmetric shape . for example , the complete splitting is observed for the line fe i 525.02 nm with @xmath51 which originates in magnetic flux tubes with a longitudinal magnetic field of a strength higher than 0.1 t. so , variations of the contribution functions of magnetically active absorption lines under the effect of a magnetic field correspond to variations of line profiles . the contribution functions of the stokes parameter @xmath27 which refer to steep parts of wings of line profiles become asymmetric due to an increase of the distance between splitted line components . a second maximum may appear for moderately strong lines in which @xmath52 when the magnetic field is longitudinal or is close to longitudinal and its strength is higher than 0.1 t. the contribution functions for the polarization parameters @xmath35 , @xmath53 , and @xmath36 have a complex shape with a positive and a negative maxima . the mean depth of the layer where the processes of selective absorption in a given frequency interval of a spectral line go effectively is called the depth of formation of the absorption line profile section , and it is calculated as a center of gravity or as a weighted mean value : @xmath54 the depression contribution function is used for the weight function @xmath55 . the averaging operation is quite justifiable here ; this follows from @xcite , for example , where it is shown that , if a contribution function positive over the whole height interval is normalized , then the new function @xmath56 thus obtained may be considered as the height distribution of probability density for the given process in the atmosphere . having calculated the center and variance of the distribution , one can use this function for finding not only the depth but the width also of the layer where a significant contribution occurs . the analysis of the contribution functions for the stokes parameters made by us has shown that the contribution function for the parameter @xmath27 can be transformed to the probability density function . the depression function @xmath57 is positive and it is easily normalized by dividing it by the value of relative depression @xmath58 . as a result , we obtain the function @xmath59 where @xmath60 this function characterizes the probability of absorption of the unpolarized and polarized radiation in a given line in a given spectral interval @xmath61 . the depth of formation @xmath62 and the half - width of the effective layer @xmath62 can be calculated now for the parameter @xmath27 by the following expressions : @xmath63 @xmath64 besides that , the so - called averaged depth of formation of the complete line profile can be calculated as a weighted mean quantity : @xmath65 where @xmath66 the relative depression , i.e. , the function @xmath67 is a weight function here ; @xmath68 is the total line depression , or the equivalent width . calculating the quantities @xmath69 and @xmath70 by the expressions given here , we define completely the region of formation of the stokes profile @xmath27 and in that way solve the problem of the depth of formation of a magnetically active absorption line in the presence of a magnetic field . the formulae given above can not be used in this case , since the contribution functions for the stokes parameters @xmath35 , @xmath53 , and @xmath36 can take on both positive and negative values in the height range considered here . in order to find the mean depth of formation of a polarization parameter , it is more convenient to resort to the center of gravity of the plane figure formed by the curve corresponding to the contribution function and by abscissa . for example , let us consider the contribution function @xmath23 for the parameter @xmath36 . the area of the figure @xmath71 is equal to the sum of two areas formed by the positive and negative parts of the @xmath23 curve which lie above and below the abscissa , i.e.,@xmath72 , where @xmath73 @xmath74 is the geometric height at which the function @xmath23 changes its sign . we shall use the expression @xmath75 to determine the depth of formation of the @xmath36 profile . substituting the expression for @xmath71 and transforming to a form convenient for calculations , we get @xmath76 the depths of formation for the parameters @xmath35 and @xmath53 can be found in the same way . thus , to calculate the depths of profile formation for polarization characteristics of magnetically active lines , one can use the formula of weighted mean where the modulus of the depression contribution function of the stokes parameters is used for the weight function . using three lines fe i 525.02 , 523.29 , 643.08 nm as an example ( fig . 7 ) , we illustrated the variations in the depth of formation of the stokes parameters with varying magnetic field strength . we calculated the profiles and the depths of their formation , varying the magnetic field strength @xmath33 from 0 to 0.2 t. figure 8 shows the effect of the field inclination on the depth of formation of the stokes profiles for the line fe i 643.08 nm . when the inclination varies , the azimuth remains equal to zero , and the field strength is 0.1 t. when calculating the depths of formation of the stokes profiles , we adopted the magnetic flux tube model of @xcite and used the geometric scale height @xmath77 and the logarithmic optical depth scale log @xmath78 , as these scales are mostly used in practice . the depths of formation in the figures are given in the geometric scale height . the geometric height zero level @xmath79 corresponds to the optical depth @xmath80 . geometric heights below the zero level take on negative values and above it they are positive . analysis of the results brings us to the following conclusions . as is generally known , the fraunhofer line profiles change their shape in the presence of a magnetic field . it differs from the classic profile shape of a fraunhofer line ( see fig . 7 ) . the depression in the central part of the profile decreases with increasing field strength as compared to the profile calculated without a magnetic field ( a `` nonmagnetic '' profile ) , and in the line wings it increases . the greater the land factor the stronger this effect is . the profiles and the depth of formation of the depression change . it is evident from fig . 7 that , as the magnetic field strength and the land factor are growing , the central part of the profile is formed in more deep layers ( the geometric heights decrease ) , the middle part is formed in more high layers ( the geometric heights increase ) as compared to the `` nonmagnetic '' profile , and far wings are formed in the layers where the far wings of `` nonmagnetic '' profiles are formed . the depth of line formation averaged over the entire profile depends on the manner in which the equivalent width varies . the greater the magnetic broadening of the line the higher in the atmosphere the entire line profile is formed on the average , i.e. , the magnetic field as if `` forces out '' slightly the line upwards . a noticeable growth of the height may be expected for the lines with @xmath81 and @xmath82 pm for @xmath83 t. but calculations show that this growth is several kilometers only . when magnetic fields are measured from the fraunhofer lines , it should be remembered that a steep section of the magnetic line profile can form higher in the atmosphere than the profile center . changes in the magnetic field inclination from @xmath38 to @xmath84 affect little both the profile @xmath27 ( fig . 3 ) and the depths of its formation ( fig . 8) . appreciable changes are observed only at the distances corresponding to the profile half - widths , where the geometric height of formation decreases with increasing inclination . the width of the layer where the main fraction of polarized radiation is absorbed in a certain part of line depends on the value of the line splitting in the magnetic field , @xmath85 . when the splitting grows , the layer width also grows . the complete splitting having been reached , the layer width diminishes . the width is different for different parts of the line . the layer width reaches its maximum value , as a rule , for the middle part of the profile wings . the layer width decreases when the magnetic field inclination varies from @xmath38 to @xmath84 . changes of the layer width that occur under the effect of a magnetic field are as large as tens of kilometers . for example , for a region @xmath86 pm in the line fe i 643.08 nm which is formed in a magnetic flux tube with a field strength of 0 , 0.1 , 0.15 , 0.2 , 0.3 t and @xmath87 , @xmath88 , the maximum half - width of the effective layer reaches 90 , 123 , 126 , 119 , 105 km , respectively . the maximum half - width for inclinations of @xmath38 , @xmath37 , @xmath89 , @xmath84 and @xmath90 t is 123 , 116 , 100 , 92 km , respectively . if the magnetic field is longitudinal , only one polarization parameter @xmath36 exists . the depth of its formation is found to be close to the depth of formation of the @xmath27 profile ( fig . the difference between them reaches several kilometers , @xmath36 being formed slightly higher than @xmath27 . thus , the depths of formation of the stokes profiles @xmath27 and @xmath36 for a region @xmath46 pm of the lines fe i 525.02 , 523.29 , 643.08 nm in a magnetic flux tube with @xmath91 t , @xmath92 , @xmath93 are 187 and 188 , 186 and 201 , 262 and 267 km , respectively . if the magnetic field inclination @xmath32 varies from @xmath38 to @xmath84 at @xmath93 , the parameter @xmath35 appears in addition to @xmath36 , and its profile is formed higher that @xmath27 ( fig . for example , for @xmath48 pm of the line 643.08 nm in a magnetic flux tube with @xmath90 t , @xmath94 , @xmath93 , the @xmath27 , @xmath35 , and @xmath36 profiles are formed at depths of 253 , 258 , 256 km , respectively . if we change the angle @xmath95 , the parameter @xmath53 will appear . figure 9 shows how the depths of formation of @xmath35 and @xmath53 change in these cases . it follows from the results of calculations that the polarization profiles @xmath35 , @xmath53 , and @xmath36 are formed higher than the general relative depression of the line , @xmath96 , but the difference is insignificant on the average . therefore , it may be assumed that all the stokes profiles are formed in the same layers . for calculating the depths of line formation , first of all it is necessary to define physical conditions in the region of formation or to specify a model solar atmosphere . it is well known that even the simplest horizontally homogeneous model atmosphere is defined by a large number of parameters which vary with height . the accuracy of calculations depends therefore completely on the accuracy of simulation of the conditions in the atmospheric regions where the processes of absorption line formation go . the calculated depth of line formation refers directly to the specific model used in the calculations . variations of physical conditions in the atmosphere cause variations of such line parameters as the central intensity , equivalent width , profile shape , and the depth of formation as well . such a notion as the depth of line formation appears to become more indeterminate and becomes meaningless in inhomogeneous model atmospheres . in this case , it is better to use the concept of the effective region of formation . [ cols="^,<,>,>,>,>,>",options="header " , ] to demonstrate the aforesaid , we chose three substantially different models . they are the models of the quiet photosphere ( holmu @xcite ) , a plage area ( wal2 @xcite ) , and a spot ( obst @xcite ) . table 1 gives the parameters of the lines used . the chemical element abundances were taken according to @xcite . the magnetic field is longitudinal with a strength of 0.15 t and the azimuth @xmath93 . the microturbulent velocity is 1 km / s ( macroturbulence was ignored ) , the damping constant is @xmath97 . tables 25 give the results of calculations for the depths of formation of selected lines for each of the models . table 2 contains the calculation data for the quiet photosphere without regard to a magnetic field . the values of the following parameters are given for each line : equivalent width @xmath68 averaged over the entire line profile , depth of formation @xmath70 , optical depth @xmath98 , selected section on the profile @xmath31 , relative line depression @xmath99 , parameters @xmath27 and @xmath36 , depth of formation and half - width of the effective layer of the depression @xmath100 , @xmath101 , @xmath102 . in order to demonstrate variations in the depth of formation over a profile from one model to another , we selected three sections on the profile calculated for the holmu model without a magnetic field : at the line center , at a distance corresponding to the line half - width , and in a wing , where the depression was 1% . further , the depths of formation were calculated in other models for the same sections @xmath31 . an additional point on the profile was calculated in that part of the wing where it became wider . the data given in the tables demonstrate clearly the reaction of lines to the conditions of their formation . as an example , we may point at the lines whose equivalent width more than doubled upon the temperature rise . these are the lines v i 611.16 nm , ti i 612.62 nm , ni i 612.89 nm , fei 524.75 , 525.02 , 615.16 , 617.33 , 630.25 nm . the lines fe ii 523.46 and 636.94 nm react weakly to the temperature rise , but they practically disappear when the temperature drops , in a spot , for example . other conclusions may be also drawn from the data given in the tables , and everyone who is interested in and needs this has such an opportunity . cccccccc + no . & @xmath68 , nm & @xmath70 , km & @xmath103 & @xmath104 , pm & @xmath105 & @xmath106 & @xmath107 , km + + + no . & @xmath68 , nm & @xmath70 , km & @xmath103 & @xmath104 , pm & @xmath105 & @xmath108 & @xmath107 , km + 1 & 2.68 & 166 & -1.11 & 0 & 0.48 & -1.25 0.75 & 187 116 + & & & & 3 & 0.20 & -0.98 0.72 & 145 111 + & & & & 6 & 0.12 & -0.62 0.59 & 91 89 + 2 & 9.66 & 296 & -1.96 & 0 & 0.80 & -2.85 0.69 & 431 104 + & & & & 5 & 0.52 & -1.22 0.71 & 183 110 + & & & & 20 & 0.01 & -0.63 0.53 & 92 81 + 3 & 8.86 & 262 & -1.74 & 0 & 0.78 & -2.59 0.79 & 392 120 + & & & & 6 & 0.28 & -0.62 0.59 & 90 89 + & & & & 17 & 0.01 & -0.34 0.48 & 50 68 + 4 & 6.28 & 315 & -2.08 & 0 & 0.77 & -2.54 0.85 & 384 129 + & & & & 4 & 0.39 & -1.55 0.85 & 233 130 + & & & & 8 & 0.01 & -0.97 0.65 & 143 101 + 5 & 7.59 & 300 & -1.98 & 0 & 0.78 & -2.57 0.74 & 390 111 + & & & & 5 & 0.33 & -1.20 0.74 & 179 115 + & & & & 11 & 0.01 & -0.76 0.56 & 111 86 + 6 & 6.74 & 328 & -2.17 & 0 & 0.79 & -2.69 0.84 & 408 127 + & & & & 4 & 0.46 & -1.60 0.85 & 241 131 + & & & & 8 & 0.01 & -0.96 0.65 & 142 101 + 7 & 3.78 & 177 & -1.18 & 0 & 0.48 & -1.38 0.73 & 208 112 + & & & & 4 & 0.22 & -1.00 0.69 & 148 107 + & & & & 9 & 0.01 & -0.56 0.52 & 82 78 + 8 & 1.08 & 189 & -1.26 & 0 & 0.17 & -1.32 0.78 & 197 120 + & & & & 3 & 0.09 & -1.23 0.76 & 184 118 + & & & & 6 & 0.01 & -1.05 0.72 & 156 112 + 9 & 2.04 & 197 -&1.31 & 0 & 0.30 & -1.40 0.79 & 210 122 + & & & & 3 & 0.17 & -1.27 0.77 & 191 119 + & & & & 7 & 0.0l & -0.97 0.70 & 145 108 + 10 & 2.30 & 201 & -1.34 & 0 & 0.35 & -1.44 0.81 & 217 125 + & & & & 3 & 0.19 & -1.28 0.79 & 192 122 + & & & & 7 & 0.01 & -0.93 0.69 & 138 106 + 11 & 4.83 & 234 & -1.55 & 0 & 0.59 & -1.81 0.79 & 274 121 + & & & & 4 & 0.30 & -1.30 0.77 & 195 119 + & & & & 8 & 0.01 & -0.83 0.62 & 122 96 + 12 & 6.97 & 279 & -1.85 & 0 & 0.68 & -2,33 0.75 & 353 113 + & & & & 5 & 0.35 & -1.29 0.76 & 193 117 + & & & & 11 & 0.01 & -0.74 0.55 & 108 85 + 13 & 9.30 & 267 & -1.77 & 0 & 0.69 & -2.51 0.68 & 380 102 + & & & & 6 & 0.37 & -1.10 0.67 & 163 104 + & & & & 30 & 0.01 & -0.60 0.51 & 88 78 + 14 & 0.45 & 138 & -0.93 & 0 & 0.07 & -0.97 0.70 & 144 108 + & & & & 3 & 0.04 & -0.90 0.69 & 133 106 + & & & & 5 & 0.01 & -0.79 0.65 & 116 100 + 15 & 1.99 & 132 & -0.90 & 0 & 0.28 & -1.00 0.76 & 148 117 + & & & & 3 & 0.17 & -0.85 0.73 & 125 112 + & & & & 7 & 0.01 & -0.55 0.62 & 81 93 + 16 & 11.83 & 358 & -2.37 & 0 & 0.75 & -3.46 0.70 & 522 108 + & & & & 7 & 0.42 & -1.29 0.72 & 193 111 + & & & & 30 & 0.01 & -0.76 0.55 & 111 85 + 17 & 2.72 & 154 & -1.04 & 0 & 0.33 & -1.18 0.70 & 176 109 + & & & & 4 & 0.16 & -0.93 0.67 & 138 104 + & & & & 9 & 0.01 & -0.54 0.52 & 78 77 + cccccccccc + no . & @xmath109 nm & @xmath110 km & @xmath98 & @xmath31,pm & @xmath27 & @xmath111 & @xmath112,km&@xmath113 & @xmath114,km + + no . & @xmath109 nm & @xmath110 km & @xmath98 & @xmath31,pm & @xmath27 & @xmath111 & @xmath115,km&@xmath113 & @xmath114,km + 1 & 2.69 & 166 & -1.11 & 0 & 0.31 & -1.08 0.73 & 161 113 & 0.00 & + & & & & 3 & 0.24 & -1.18 0.75 & 178 116 & 0.21 & 187 115 + & & & & 8 & 0.01 & -0.63 0.59 & 92 89 & 0.01 & 96 92 + 2 & 9.76 & 298 & -1.97 & 0 & 0.77 & -2.40 0.70 & 365 105 & 0.00 & + & & & & 5 & 0.44 & -2.17 0.90 & 329 136 & 0.33 & 336 137 + & & & & 20 & 0.01 & -0.63 0.53 & 92 81 & 0.003 & 92 81 + 3 & 8.88 & 262 & -1.74 & 0 & 0.76 & -2.33 0.79 & 354 119 & 0.00 & + & & & & 6 & 0.34 & -1.21 0.75 & 181 116 & 0.26 & 196 116 + & & & & 17 & 0.01 & -0.34 0.48 & 50 68 & 0.003 & 50 68 + 4 & 6.46 & 312 & -2.06 & 0 & 0.46 & -1.62 0.85 & 244 132 & 0.00 & + & & & & 4 & 0.39 & -2.49 0.87 & 377 132 & 0.38 & 382 129 + & & & & 8 & 0.19 & -1.54 0.85 & 232 131 & 0.19 & 233 131 + & & & & 12 & 0.004 & -0.95 0.63 & 140 99 & 0.002 & 141 100 + 5 & 7.62 & 300 & -1.98 & 0 & 0.38 & -1.24 0.75 & 187 116 & 0.00 & + & & & & 5 & 0.40 & -2.53 0.77 & 384 116 & 0.38 & 387 115 + & & & & 11 & 0.10 & -1.04 0.71 & 155 110 & 0.09 & 158 111 + & & & & 14 & 0.01 & -0.75 0.56 & 111 87 & 0.01 & 111 87 + 6 & 6.78 & 330 & -2.18 & 0 & 0.09 & -1.23 0.78 & 184 121 & 0.00 & + & & & & 4 & 0.38 & -2.40 0.86 & 364 130 & 0.38 & 367 128 + & & & & 8 & 0.37 & -2.27 0.86 & 344 131 & 0.37 & 345 130 + & & & & 12 & 0.02 & -1.09 0.73 & 162 113 & 0.02 & 164 114 + 7 & 3.97 & 174 & -1.17 & 0 & 0.26 & -1.06 0.71 & 158 109 & 0.00 & + & & & & 4 & 0.24 & -1.29 0.73 & 193 113 & 0.22 & 203 111 + & & & & 9 & 0.07 & -0.92 0.68 & 137 105 & 0.07 & 140 106 + & & & & 12 & 0.01 & -0.60 0.54 & 87 82 & 0.01 & 89 84 + 8 & 1.08 & 189 & -1.26 & 0 & 0.07 & -1.21 4.76 & 181 118 & 0.00 & + & & & & 3 & 0.09 & -1.30 0.78 & 195 120 & 0.08 & 199 120 + & & & & 6 & 0.05 & -1.25 0.77 & 187 119 & 0.05 & 188 119 + & & & & 9 & 0.01 & -1.08 0.73 & 161 113 & 0.01 & 162 113 + 9 & 2.08 & 196 & -1.31 & 0 & 0.15 & -1.27 0.77 & 190 119 & 0.00 & + & & & & 3 & 0.15 & -1.36 0.79 & 204 121 & 0.13 & 212 121 + & & & & 7 & 0.06 & -1.24 0.77 & 185 118 & 0.06 & 187 119 + & & & & 10 & 0.01 & -1.03 0.72 & 154 111 & 0.01 & 157 112 + 10 & 2.33 & 200 & -1.33 & 0 & 0.12 & -1.22 0.77 & 182 120 & 0.00 & + & & & & 3 & 0.17 & -1.40 0.80 & 210 124 & 0.15 & 214 124 + & & & & 7 & 0.09 & -1.28 0.79 & 191 122 & 0.09 & 193 121 + & & & & 10 & 0.01 & -1.05 0.73 & 156 113 & 0.01 & 158 114 + 11 & 4.92 & 233 & -1.54 & 0 & 0.19 & -1.19 0.75 & 178 117 & 0.00 & + & & & & 4 & 0.29 & -1.75 0.80 & 265 122 & 0.28 & 270 120 + & & & & 8 & 0.21 & -1.44 0.78 & 217 121 & 0.21 & 219 121 + & & & & 12 & 0.01 & -0.87 0.65 & 128 100 & 0.01 & 131 102 + 12 & 6.99 & 281 & -1.86 & 0 & 0.11 & -1.00 0.70 & 148 108 & 0.00 & + & & & & 5 & 0.33 & -2.16 0.77 & 327 117 & 0.33 & 331 115 + & & & & 11 & 0.23 & -1.45 0.77 & 220 119 & 0.23 & 221 119 + & & & & 16 & 0.01 & -0.75 0.56 & 111 87 & 0.01 & 111 88 + 13 & 9.20 & 270 & -1.79 & 0 & 0.23 & -0.88 0.62 & 129 96 & 0.00 & + & & & & 6 & 0.36 & -2.39 0.77 & 362 116 & 0.33 & 364 115 + & & & & 25 & 0.01 & -0.60 0.51 & 87 78 & 0.005 & 88 78 + 14 & 0.44 & 137 & -0.92 & 0 & 0.02 & -0.85 0.67 & 125 103 & 0.00 & + & & & & 3 & 0.03 & -0.94 0.70 & 140 108 & 0.03 & 145 108 + & & & & 5 & 0.03 & -0.96 0.70 & 142 108 & 0.03 & 144 108 + & & & & 9 & 0.01 & -0.83 0.67 & 123 103 & 0.01 & 124 103 + 15 & 1.99 & 132 & -0.90 & 0 & 0.03 & -0.63 0.66 & 92 99 & 0.00 & + & & & & 3 & 0.08 & -0.85 0.73 & 125 112 & 0.08 & 127 112 + & & & & 7 & 0.13 & -0.97 0.75 & 144 116 & 0.13 & 145 116 + & & & & 12 & 0.01 & -0.59 0.64 & 86 96 & 0.01 & 87 97 + 16 & 12.10 & 360 & -2.38 & 0 & 0.71 & -2.83 0.72 & 428 108 & 0.00 & + & & & & 7 & 0.40 & -2.68 0.96 & 405 145 & 0.31 & 409 146 + & & & & 30 & 0.01 & -0.76 0.55 & 111 85 & 0.002 & 112 85 + 17 & 2.75 & 155 & -1.04 & 0 & 0.02 & -0.59 0.55 & 86 83 & 0.00 & + & & & & 4 & 0.08 & -0.93 0.67 & 137 104 & 0.08 & 141 104 + & & & & 9 & 0.16 & -1.15 0.70 & 172 109 & 0.16 & 173 109 + & & & & 16 & 0.01 & -0.55 0.53 & 80 79 & 0.01 & 81 79 + cccccccccc + no . & @xmath68,nm & @xmath70,km & @xmath98 & @xmath31,pm & @xmath27 & @xmath116 & @xmath115,km&@xmath36 & @xmath114,km + + no . & @xmath68,nm & @xmath70,km & @xmath98 & @xmath31,pm & @xmath27 & @xmath116 & @xmath115,km&@xmath36 & @xmath114,km + 1 & 1.42 & 102 & -0.71 & 0 & 0.16 & -0.69 0.67 & 99 100 & 0.00 & + & & & & 3 & 0.12 & -0.77 0.68 & 111 101 & 0.11 & 121 100 + & & & & 8 & 0.01 & -0.33 0.60 & 47 86 & 0.01 & 52 89 + 2 & 5.71 & 210 & -1.42 & 0 & 0.47 & -1.71 0.53 & 252 90 & 0.00 & + & & & & 5 & 0.27 & -1.53 0.69 & 227 111 & 0.20 & 234 111 + & & & & 17 & 0.01 & -0.27 0.53 & 38 75 & 0.003 & 39 75 + 3 & 5.94 & 320 & -1.98 & 0 & 0.44 & -2.77 0.63 & 461 48 & 0.00 & + & & & & 6 & 0.26 & -1.57 0.90 & 240 157 & 0.19 & 255 158 + & & & & 17 & 0.01 & -0.23 0.57 & 33 82 & 0.002 & 34 83 + 4 & 2.60 & 166 & -1.15 & 0 & 0.14 & -0.93 0.67 & 134 102 & 0.00 & + & & & & 4 & 0.19 & -1.31 0.63 & 190 99 & 0.18 & 193 98 + & & & & 8 & 0.06 & -0.89 0.68 & 128 102 & 0.06 & 129 102 + & & & & 10 & 0.01 & -0.69 0.67 & 98 99 & 0.01 & 101 100 + 5 & 3.98 & 187 & -1.28 & 0 & 0.16 & -0.79 0.66 & 113 99 & 0.00 & + & & & & 5 & 0.23 & -1.59 0.57 & 235 94 & 0.23 & 237 93 + & & & & 11 & 0.04 & -0.62 0.65 & 89 97 & 0.04 & 91 97 + & & & & 12 & 0.01 & -0.39 0.58 & 55 83 & 0.01 & 56 84 + 6 & 2.92 & 176 & -1.21 & 0 & 0.03 & -0.71 0.68 & 102 100 & 0.00 & + & & & & 4 & 0.18 & -1.28 0.63 & 185 99 & 0.18 & 187 98 + & & & & 8 & 0.17 & -1.22 0.64 & 176 99 & 0.17 & 177 99 + & & & & 12 & 0.01 & -0.66 0.67 & 94 98 & 0.01 & 96 99 + 7 & 1.92 & 125 & -0.87 & 0 & 0.12 & -0.79 0.65 & 113 98 & 0.00 & + & & & & 4 & 0.12 & -0.98 0.64 & 141 98 & 0.10 & 151 94 + & & & & 9 & 0.04 & -0.65 0.65 & 94 97 & 0.04 & 97 97 + & & & & 12 & 0.01 & -0.40 0.60 & 56 86 & 0.01 & 60 88 + 8 & 0.32 & 118 & -0.82 & 0 & 0.02 & -0.78 0.67 & 112 101 & 0.00 & + & & & & 3 & 0.03 & -0.84 0.67 & 121 101 & 0.02 & 125 101 + & & & & 6 & 0.02 & -0.81 0.67 & 117 101 & 0.02 & 118 101 + 9 & 0.66 & 122 & -0.85 & 0 & 0.05 & -0.82 0.67 & 118 101 & 0.00 & + & & & & 3 & 0.05 & -0.88 0.67 & 126 101 & 0.04 & 132 100 + & & & & 7 & 0.02 & -0.80 0.67 & 115 101 & 0.02 & 117 101 + 10 & 0.81 & 129 & -0.90 & 0 & 0.04 & -0.82 0.68 & 117 101 & 0.00 & + & & & & 3 & 0.06 & -0.93 0.67 & 134 102 & 0.05 & 138 101 + & & & & 7 & 0.03 & -0.86 0.67 & 124 102 & 0.03 & 124 102 + & & & & 10 & 0.01 & -0.72 0.67 & 103 100 & 0.01 & 105 100 + 11 & 2.01 & 149 & -1.03 & 0 & 0.07 & -0.79 0.67 & 114 100 & 0.00 & + & & & & 4 & 0.13 & -1.15 0.63 & 166 98 & 0.12 & 169 96 + & & & & 8 & 0.08 & -0.97 0.65 & 140 99 & 0.08 & 141 99 + & & & & 12 & 0.004 & -0.53 0.63 & 74 91 & 0.004 & 78 93 + 12 & 3.30 & 183 & -1.26 & 0 & 0.04 & -0.64 0.65 & 92 96 & 0.00 & + & & & & 5 & 0.17 & -1.41 0.58 & 206 94 & 0.17 & 209 93 + & & & & 11 & 0.11 & -1.09 0.63 & 157 97 & 0.12 & 158 96 + & & & & 14 & 0.01 & -0.51 0.62 & 73 90 & 0.01 & 75 91 + 13 & 4.95 & 203 & -1.38 & 0 & 0.11 & -0.66 0.60 & 93 89 & 0.00 & + & & & & 6 & 0.21 & -1.78 0.59 & 265 99 & 0.19 & 267 98 + & & & & 20 & 0.01 & -0.31 0.53 & 44 75 & 0.01 & 45 75 + 14 & 0.19 & 94 & -0.66 & 0 & 0.01 & -0.58 0.66 & 83 97 & 0.00 & + & & & & 3 & 0.01 & -0.68 0.67 & 97 99 & 0.01 & 101 100 + & & & & 5 & 0.02 & -0.69 0.67 & 98 100 & 0.01 & 101 100 + & & & & 7 & 0.01 & -0.64 0.67 & 92 99 & 0.01 & 93 99 + 15 & 1.49 & 160 & -1.08 & 0 & 0.02 & -0.74 0.81 & 109 128 & 0.00 & + & & & & 3 & 0.06 & -1.02 0.82 & 151 133 & 0.06 & 153 133 + & & & & 7 & 0.09 & -1.18 0.82 & 175 134 & 0.09 & 176 134 + & & & & 12 & 0.01 & -0.68 0.81 & 100 125 & 0.01 & 101 126 + 16 & 6.11 & 252 & -1.68 & 0 & 0.39 & -1.90 0.49 & 284 87 & 0.00 & + & & & & 7 & 0.22 & -1.82 0.64 & 273 107 & 0.18 & 278 106 + & & & & 20 & 0.01 & -0.42 0.55 & 59 79 & 0.003 & 61 79 + 17 & 1.35 & 117 & -0.82 & 0 & 0.01 & -0.36 0.58 & 52 83 & 0.00 & + & & & & 4 & 0.04 & -0.71 0.65 & 02 96 & 0.04 & 107 96 + & & & & 9 & 0.08 & -0.93 0.64 & 33 97 & 0.07 & 134 96 + & & & & 14 & 0.01 & -0.47 0.62 & 67 89 & 0.01 & 69 90 + cccccccccc + no . & @xmath68,nm & @xmath70,km&@xmath98 & @xmath31,pm & @xmath27&@xmath116&@xmath115,km&@xmath36 & @xmath114,km + + no . & @xmath68,nm & @xmath70,km & @xmath98 & @xmath31,pm & @xmath27&@xmath116 & @xmath115,km&@xmath36 & @xmath114,km + 1 & 3.30 & 33 & -0.54 & 0 & 0.32 & -0.62 0.67 & 39 40 & 0.00 & + & & & & 3 & 0.25 & -0.71 0.72 & 44 44 & 0.16 & 56 43 + & & & & 8 & 0.04 & -0.04 0.44 & 2 29 & 0.02 & 4 30 + & & & & 14 & 0.01 & 0.02 0.43 & -2 28 & 0.003 & -2 29 + 2 & 16.12 & 64 & -1.08 & 0 & 0.73 & -2.11 0.73 & 125 40 & 0.00 & + & & & & 5 & 0.52 & -1.63 0.98 & 98 56 & 0.21 & 99 58 + & & & & 70 & 0.01 & -0.07 0.51 & 3 33 & 0.001 & 15 55 + 3 & 0.33 & -7 & 0.08 & 0 & 0.04 & 0.07 0.48 & -6 32 & 0.00 & + & & & & 5 & 0.01 & 0.18 0.43 & -13 29 & 0.01 & -11 29 + 4 & 17.15 & 123 & -2.11 & 0 & 0.81 & -2.91 0.83 & 168 46 & 0.00 & + & & & & 4 & 0.50 & -3.88 1.11 & 224 68 & 0.37 & 224 68 + & & & & 8 & 0.50 & -2.24 1.18 & 131 67 & 0.29 & 134 68 + & & & & 60 & 0.01 & -0.35 0.60 & 22 37 & 0.001 & 22 37 + 5 & 33.27 & 93 & -1.59 & 0 & 0.79 & -2.61 0.74 & 152 39 & 0.00 & + & & & & 4 & 0.48 & -4.24 0.80 & 243 50 & 0.39 & 243 51 + & & & & 8 & 0.55 & -3.24 1.17 & 185 65 & 0.32 & 186 65 + & & & & 81 & 0.03 & -0.33 0.55 & 20 35 & 0.004 & 22 34 + 6 & 18.60 & 118 & -2.03 & 0 & 0.56 & -1.39 0.75 & 85 42 & 0.00 & + & & & & 5 & 0.45 & -4.19 0.81 & 242 55 & 0.40 & 242 55 + & & & & 11 & 0.39 & -1.55 0.93 & 93 53 & 0.25 & 99 54 + & & & & 70 & 0.01 & -0.34 0.59 & 21 37 & 0.001 & 22 37 + 7 & 2.54 & 35 & -0.56 & 0 & 0.14 & -0.53 0.63 & 33 39 & 0.00 & + & & & & 4 & 0.14 & -0.72 0.68 & 44 41 & 0.10 & 54 40 + & & & & 9 & 0.05 & -0.33 0.57 & 21 36 & 0.04 & 27 36 + & & & & 16 & 0.01 & -0.04 0.44 & 2 29 & 0.004 & 3 29 + 8 & 9.54 & 108 & -1.84 & 0 & 0.60 & -2.02 0.83 & 120 45 & 0.00 & + & & & & 3 & 0.49 & -2.33 1.14 & 136 63 & 0.25 & 138 65 + & & & & 6 & 0.39 & -2.17 1.00 & 127 55 & 0.29 & 130 55 + & & & & 30 & 0.01 & -0.40 0.57 & 25 35 & 0.002 & 25 35 + 9 & 14.85 & 116 & -1.98 & 0 & 0.75 & -2.87 0.74 & 165 39 & 0.00 & + & & & & 3 & 0.64 & -2.70 1.18 & 155 64 & 0.17 & 155 68 + & & & & 7 & 0.46 & -2.28 1.10 & 133 61 & 0.27 & 135 62 + & & & & 50 & 0.01 & -0.43 0.57 & 27 35 & 0.001 & 28 35 + 10 & 3.29 & 64 & -1.05 & 0 & 0.18 & -0.84 0.79 & 52 47 & 0.00 & + & & & & 3 & 0.21 & -1.23 0.87 & 75 50 & 0.18 & 83 48 + & & & & 7 & 0.13 & -0.99 0.82 & 61 48 & 0.12 & 64 47 + & & & & 14 & 0.01 & -0.24 0.53 & 15 34 & 0.003 & 15 34 + 11 & 7.15 & 80 & -1.32 & 0 & 0.31 & -0.96 0.70 & 59 41 & 0.00 & + & & & & 4 & 0.34 & -1.80 0.93 & 107 53 & 0.26 & 110 53 + & & & & 8 & 0.26 & -1.42 0.82 & 86 47 & 0.22 & 90 46 + & & & & 26 & 0.01 & -0.20 0.51 & 12 33 & 0.003 & 13 33 + 12 & 10.53 & 86 & -1.44 & 0 & 0.25 & -0.70 0.56 & 43 34 & 0.00 & + & & & & 5 & 0.38 & -2.12 1.01 & 124 57 & 0.29 & 126 57 + & & & & 11 & 0.27 & -1.36 0.79 & 83 45 & 0.22 & 87 45 + & & & & 43 & 0.01 & -0.20 0.50 & 12 33 & 0.003 & 13 32 + 13 & 9.61 & 60 & -0.99 & 0 & 0.21 & -0.55 0.46 & 35 29 & 0.00 & + & & & & 6 & 0.33 & -1.63 0.86 & 98 50 & 0.24 & 100 50 + & & & & 47 & 0.01 & -0.07 0.46 & 4 30 & 0.003 & 5 30 + 14 & 0.15 & 23 & -0.37 & 0 & 0.01 & -0.25 0.58 & 15 37 & 0.00 & + & & & & 3 & 0.01 & -0.40 0.66 & 25 41 & 0.10 & 32 42 + & & & & 6 & 0.01 & -0.39 0.65 & 24 41 & 0.01 & 28 41 + 15 & 0.004 & -4 & 0.04 & 0 & 0.0001 & 0.17 0.43 & -12 29 & 0.00 & + & & & & 3 & 0.0003 & 0.05 0.53 & -5 35 & 0.0003 & -4 36 + & & & & 5 & 0.0006 & 0.01 0.56 & -2 37 & 0.0000 & -1 37 + 16 & 23.30 & 93 & -1.57 & 0 & 0.73 & -2.97 0.77 & 172 44 & 0.00 & + & & & & 7 & 0.52 & -2.45 1.21 & 143 68 & 0.23 & 144 69 + & & & & 80 & 0.02 & -0.24 0.51 & 15 33 & 0.001 & 16 33 + 17 & 0.99 & 26 & -0.42 & 0 & 0.02 & -0.98 0.45 & 6 29 & 0.00 & + & & & & 4 & 0.03 & -0.33 0.58 & 21 37 & 0.02 & 27 37 + & & & & 9 & 0.05 & -0.57 0.66 & 35 40 & 0.05 & 38 40 + & & & & 16 & 0.01 & -0.07 0.45 & 4 29 & 0.01 & 5 29 + the depression contribution functions for the stokes parameters of magnetically active absorption lines are of a complex nature in the conditions of the solar atmosphere in the regions where a strong magnetic field is present . the depression functions may be considered as a result of combining the contribution functions of groups of the components into which a line splits in a magnetic field . the function profiles depend to a large extent on the value of splitting @xmath117 and on the intensity of each of the components . the depression functions may be used to determine the depth and width of the effective layer where the stokes profiles are formed . the profiles of the line polarization parameters are formed several kilometers higher than the profile of the parameter which describes the general depression of polarized and unpolarized radiation . the stokes profiles may be considered to form in the same layer due to a small difference between the depths of their formation . magnetically active lines with large values of the land factor which form in the presence of a strong longitudinal magnetic field have a distinctive feature the steep section of the line profile is formed higher than the center of the line profile . the averaged depth of line formation depends on the amount for magnetic broadening . when the latter increases , the whole region of line formation shifts slightly into the layers lying higher . this effect being insignificant , the averaged depth of line formation in a magnetic field does not change . it is not possible to draw up tables of calculated depths of formation of magnetically active lines on account of a strong dependence of the value of depression and the depth of line formation on physical conditions in the medium . the depths of formation should be calculated every time the conditions in the medium change . a. s. gadun and v. a. sheminova , spansat : program for calculating spectral absorption line profiles in stellar atmospheres in the lte approximation [ in russian ] , kiev , 1988 ( institute of theoretical physics as ukr ssr preprint itf-88 - 87p ) . u. grossman - doerth , b. larsson , and s. k. solanki , `` contribution and response functions for stokes line profiles formed in a magnetic field , '' astron . and astrophys . , 1 , pp . 266274 , 1988 . v. a. sheminova , effect of physical conditions in the medium and of atomic constants on the stokes profiles of absorption lines in the solar spectrum [ in russian ] , kiev , 1991 ( institute of theoretical physics as ukr ssr preprint itf-90 - 87p ) . a. a. van ballegooijen , `` contribution functions for zeeman split lines and line formation in photospheric faculae , '' in : measurements of solar vector magnetic fields , m. j. hagyard ( editor ) , nasa conf . 2374 , pp . 322334 , 1985 .

we compute the one - loop divergences of , supergravity and of its reduction to . we study the tensor structure of the counterterms appearing in and and compare these to expressions previously found in the low energy expansion of string theory . the infinities have the primitive yang - mills tree amplitude as a common factor . # 1_#1 _ = 0.1 cm = -2.5 cm = -2.0 cm = 23.2truecm = 16.7truecm 40 pt plus 1000pt minus 100pt # 1 # 1= to [ section ] # 1eq . ( [ # 1 ] ) # 1eq . ( [ # 1 ] ) # 1#2eqs . ( [ # 1 ] ) and ( [ # 2 ] ) # 1#2eqs . ( [ # 1 ] ) and ( [ # 2 ] ) # 1fig . [ # 1 ] # 1#2figs . [ # 1 ] and [ # 2 ] # 1#2figs . [ # 1 ] and [ # 2 ] # 1section [ # 1 ] # 1section [ # 1 ] # 1#2sections [ # 1 ] and [ # 2 ] # 1appendix [ # 1 ] # 1appendix [ # 1 ] # 1#2appendices [ # 1 ] and [ # 2 ] # 1#2appendices [ # 1 ] and [ # 2 ] # 1table [ # 1 ] # 1table [ # 1 ] # 1 = = = by - by - by by 2 -to # 1.#2#1#2 # 1.#2 # 1.#2(#1#2 ) @=11 # 1#2#3#4(#1#2#3#4 ) # 1#2x_#1 # 2 # 1/-0.17 cm # 1 # 1 hep - th/9911158 + ens-99/34 + swat-99 - 241 + november , 1999 + 2.cm * counterterms in type i supergravities * * * 2.cm david c. dunbar , bernard julia , domenico seminara and mario trigiante 0.5 cm _department of physics , university of wales swansea , swansea , uk _ .3 cm _laboratoire de physique thorique , cnrs - ens , paris , france _ ' '' '' width 3.6inresearch supported by tmr grant fmrx - ct96 - 0012research supported by tmr grant fmrx ct96 - 0045 .umr 8549 du cnrs et de lecole normale suprieure .

hep-th9911158

highly extended supergravity theories have long been seen as potential theories of quantum gravity although they were displaced by superstring theories @xcite as favored candidates for a `` final theory '' . the ( quantum ) m theory has 10 dimensional supergravities as low energy limits and might be thought of either as a ( non - perturbative ) quantum version of eleven dimensional supergravity ( including its extended solutions ) or as a non - perturbative completion of superstring theories . the spectacular perturbative ultra - violet finiteness of string theories indicates that strings provide a physical regulator of supergravities in 10 dimensions . in this paper we shall explore the ultra - violet behavior of @xmath13 supergravity and its dimensional descendants from a field theory viewpoint . in this we are attempting to `` work - up '' to the quantum theory of gravity rather than work down from string theory . of the extended supergravities two play a central role . firstly , there is the @xmath14 , @xmath1 maximally extended theory @xcite which reduces to @xmath15 in @xmath16 @xcite . in some ways this theory is the ultimate conventional point - particle field theory . the one - loop amplitude is potentially infinite for @xmath17 although in dimensional regularisation it is only infinite for @xmath2 ( in the dimensional regularisation prescription one - loop amplitudes in odd dimensions are finite and the @xmath0 infinity vanishes onshell ) . we shall work entirely within the dimensional reduction scheme this being the most appropriate to study an anomaly - free supersymmetric theory . at two - loops infinities have been calculated in the amplitudes for @xmath18 @xcite ( including the @xmath14 case . the eleven dimensional counterterm was subsequently evaluated in ref . the expectation is that this theory will be perturbatively infinite in four dimensions @xcite . in this paper we shall examine features of the other interesting extended supergravity . this is the @xmath0 , @xmath1 theory and its dimensional reduction descendants which include the @xmath16 , @xmath19 supergravity theory ( with a specific matter content ) . this supergravity is the gravitational sector of the low energy limit of both type i and heterotic string theories @xcite . for the one - loop amplitudes we may consider the @xmath13 supergravity together with arbitrary matter multiplets , although having less supersymmetry than the @xmath20 theory , it is not clear which is the most fundamental . a web of dualities relates the various string theories and if one introduces for instance the appropriate gauge group so(32 ) in the type i matter sector , this choice cancels the gravitational and gauge anomalies . when compactified to @xmath21 type i and type ii theories give rise to very special and similar symmetry groups . in fact it can be argued that these duality symmetries that have gained importance together with the realisation of the overwhelming necessity of non - perturbative effects are not that different between type i and type ii supergravities . the tantalizing @xmath22 is closely related to the hyperbolic kac - moody algebra obtained from the extrapolation of the chamseddine - sen sequence @xcite namely the hyperbolic extension of @xmath23 . in fact the symmetry groups of type i are fixed point sets under an involution of those of type ii for all dimensions ( at least three ) ; this was mentioned in @xcite and relies on kac s description of the automorphisms of simple lie algebras which can be found for instance in @xcite . clearly , the extended supergravities can only be uv finite if they possess symmetries which we do not fully understand yet , the implications of which will be at the non - perturbative level , or at least will require novel renormalisation techniques . consequently , if these symmetries are operative at all , they should be approximately as powerful for type i and ii theories . and one could infer from this that the structure of divergences should be somehow related in both theories which are two dual perturbative expansions . one should keep in mind though that the relation is nonperturbative . the explicit ( perturbative ) calculation we shall perform is four graviton scattering at one - loop for @xmath24 . this will enable us to evaluate @xmath25 counterterms . for the one - loop amplitudes we may consider the @xmath26 supergravity multiplet with arbitrary matter , we shall call it type i slightly abusively . although the theory will only be free from gravitational anomalies in @xmath0 for special gauge groups , for @xmath27 the gravitational anomalies only manifest themselves in the five or six graviton amplitude . furthermore the contribution is parity odd and finite whereas we shall consider divergences . before calculating the counterterms , we shall calculate the entire amplitudes ( including their finite part ) and at first for a particular choice of external helicities - namely in the case where the external polarisation vectors are forced to be four dimensional . with this simplification it is possible to present the amplitude in a very elegant form as the sum over a few simple integral functions . from these amplitudes we can easily see the presence of ultra - violet infinities in @xmath27 . unlike the situation in four dimensions , there are onshell independent @xmath25 tensors in dimension @xmath28 , so in @xmath2 the counterterms are not determined by single coefficients and it requires a full computation of the amplitude to fix their form . we find a beautiful factorisation of the infinite counterterms in the amplitude into a product of left times right kinematic factors in a manner very reminiscent of the relationships between gravity and yang - mills presented in ref . the factorisation can be understood from a string theory viewpoint but remains more obscure for field theorists . offshell this factorisation is also not obvious at all from examining the field theory counterterms although the counterterms can be manipulated to reflect it . as a simple case , we first study amplitudes where the helicity of the external gravitons are restricted to lie in the four - dimensional space defined by the momenta . in this situation we can calculate the entire amplitude in a fairly simple form by breaking the amplitude into its helicity components . even with the external helicities specified there are different contributions depending upon which supermultiplet is circulating in the internal loop . one may label the possible loop contributions according to the circulating four dimensional supermultiplets . the three contributions we shall distinguish are that from a @xmath15 multiplet , that from a @xmath30 matter multiplet and finally that from a @xmath19 matter multiplet . in terms of these contributions the ( pure ) @xmath19 supergravity one - loop amplitude is @xmath31 also , of more interest to us is the `` @xmath32 '' one - loop amplitude for the dimensional reduction of @xmath33 supergravity to four dimensions which is @xmath34 it is often the case that nonmaximally supersymmetric theories do not remain irreducible upon dimensional reduction . if we consider the @xmath35 ( or type i ) supergravity coupled to @xmath19 super - yang - mills matter the corresponding amplitude is @xmath36 where @xmath37 is the dimension of the gauge group . let us call @xmath38 the combination @xmath39 . we have chosen to organize our amplitudes conveniently as linear combinations of the three supermultiplet contributions . for @xmath0 the three independent supersymmetric contributions are for instance the @xmath15 and the @xmath32 contributions as well as the `` matter '' @xmath19 term . these have in turn reductions below 10 dimensions . if the external polarisations are forced to be four dimensional then the external gravitons have @xmath40 helicity . there are then only three independent `` helicity amplitudes '' , @xmath41 , @xmath42 and @xmath43 . ( we choose a convention where all particles are considered outgoing . ) in any supersymmetric theory , @xmath44 and the only non - zero independent amplitude is @xmath43 . we have calculated our amplitudes using `` string - based rules '' @xcite , ( see appendix a ) and verified the four - dimensional expressions using unitarity techniques . the details are presented in appendix b. the amplitudes are all regulated using dimensional reduction @xcite with parameter @xmath45 . calculating the @xmath15 amplitude gives @xmath46 \left(4\pi \right)^{-d/2 } \times \left ( i^{d}_4(s , t ) + i^{d}_4(s , u ) + i^{d}_4(t , u ) \right ) \equn\label{eqeight}\ ] ] where @xmath47 denotes the @xmath29-dimensional scalar box integral function with ordering of legs @xmath48 , @xmath49 and @xmath50 are the usual mandelstam variables . the @xmath51-point scalar amplitude with external legs @xmath52 defined by @xmath53 is related to the function @xmath54 by @xmath55 in general this is a function of the kinematic variables @xmath56 . often we will express these variables which indicate the ordering of legs . for example @xmath47 has ordering of legs @xmath48 , @xmath57 has ordering of legs @xmath58 etc ... the @xmath15 amplitude was originally evaluated using the low energy limit of string theory by green , schwarz and brink @xcite . unlike the following two amplitudes it is valid for all external polarisations . the remaining two amplitudes are more complicated and have the form @xmath59 \times { ( 4\pi)^{-d/2 } } \times \biggl ( - \frac{1}{s } i_{4}^{d+2}(t , u ) \cr & \null \hskip 1.5 truecm + \frac{(d-4)}{2s } \left ( i^{d+2}_{4}(s , t ) + i^{d+2}_{4}(s , u ) - i^{d+2}_{4}(t , u ) \right ) \biggr)\cr } \equn\label{eqmattersix}\ ] ] @xmath60 \times ( { 4\pi})^{-d/2 } \times \biggl ( -{1 \over 2 s^2 } \left ( i_2^d(t ) + i_2^d(u ) \right ) \cr \null & \hskip -2.0 truecm + { ( d(d+2 ) ) \over 4s^2 } i^{d+4}_4(u , t)+ { ( 2-d)(4-d ) \over 4 s^2 } i^{d+4}_4(s , t)+ { ( 2-d)(4-d ) \over 4 s^2 } i^{d+4}_4(s , u ) \biggr ) \cr } \equn\label{eqmatterfour}\ ] ] note the appearance of the @xmath29-dimensional scalar bubble integral , @xmath61 . the tree amplitude for four - dimensional helicities is @xmath62 where we have expressed the amplitude using a `` spinor helicity '' representation of the polarisation tensors . spinor helicity techniques @xcite were introduced for qcd calculation but by splitting the polarisation tensor @xmath63 they can be applied to gravity calculations also @xcite . these amplitudes are complete . they contain a great deal of information but in particular we can simply extract their ultra - violet infinities . since we are using dimensional regularisation the one - loop integrals are only divergent in even dimensions . for a four point amplitude the expected counterterm is of the @xmath25 type which can only appear for @xmath28 at one - loop . for the @xmath15 amplitude the form of the amplitude is manifestly finite for @xmath64 since scalar box integrals are only divergent for @xmath17 . extracting the divergences for @xmath2 and @xmath0 from the integrals yields @xmath65\times { 1 \over 2 \eps } \cr m^{n=8 , d=10 - 2\eps } & = 0 \cr } \equn\ ] ] the @xmath0 one - loop amplitude is onshell - convergent within dimensional regularisation since the integrals give @xmath66 . a cut - off regularisation would give a non - zero result proportional to @xmath67 which leads to the well known @xcite counterterm @xmath68 . ( @xmath69 will be defined later . ) within dimensional regularisation the two - loop amplitude is the first divergence @xcite . for the remaining two amplitudes we can evaluate the infinities in these expressions quite easily . the expressions are manifestly convergent for @xmath70 for the @xmath30 contribution and @xmath71 for the @xmath19 one . extracting the divergences we find @xmath72 \times - { 1 \over 24 \eps } \cr m^{n=6,matter , d=10 - 2\eps } ( 1 ^ -,2 ^ -,3^+ , 4^+ ) & = \kaps { 1\over ( 4\pi ) ^5 } [ stu m^{tree } ] \times - { s \over 720 \eps } \cr m^{n=4,matter , d=6 - 2\eps } ( 1 ^ -,2 ^ -,3^+ , 4^+ ) & = 0 \cr m^{n=4,matter , d=8 - 2\eps } ( 1 ^ -,2 ^ -,3^+ , 4^+ ) & = \kaps { 1\over ( 4\pi ) ^4 } [ stu m^{tree } ] \times { 1 \over 180 \eps } \cr m^{n=4,matter , d=10 - 2\eps } ( 1 ^ -,2 ^ -,3^+ , 4^+ ) & = \kaps { 1\over ( 4\pi ) ^5 } [ stu m^{tree } ] \times { s \over 3360 \eps } \cr } \equn\ ] ] in both cases the @xmath73 counterterm vanishes . in @xmath73 the expected counterterm is @xmath74 . however there is no supersymmetrisable @xmath74 counterterm and amplitudes contain no infinities . this is exactly the reason why the two - loop infinity vanishes in @xmath16 supergravity . finally , recombining these poles to give the full physical amplitude with four dimensional helicities within type i supergravity we find @xmath75 \times ( 4\pi)^{-d/2 } \biggl ( ( i^{d}_4(s , t ) + i^{d}_4(s , u ) + i^{d}_4(t , u ) ) \cr & -{4 \over s } i_4^{d+2}(t , u ) - { 2(4-d ) \over s } \bigl [ i^{d+2}_{4}(s , t ) + i^{d+2}_{4}(s , u ) - i^{d+2}_{4}(t , u ) ) \bigr ] \cr & + { 2(2-d)(4-d ) \over s^2 } i^{d+4}_4(s , t ) + { 2(2-d)(4-d ) \over s^2 } i^{d+4}_4(s , u ) + { 2d(d+2 ) ) \over s^2 } i^{d+4}_4(u , t ) \cr & -{4 \over s^2 } ( i_2^d(t ) + i_2^d(u ) ) \biggr ) \cr } \equn\ ] ] with divergences @xmath76 \times { 32 \over 45 \eps } \cr m^{n=4*,d=10 } ( 1 ^ -,2 ^ -,3^+ , 4^+ ) & = \kaps { 1\over ( 4\pi ) ^5 } [ stu m^{\rm tree } ] \times { s\over 126 \eps } \cr } \equn\ ] ] if these are coupled to super - yang - mills the divergences become @xmath77 \times { 120 + g \over 180 \eps } \cr m^{n=4*,d=10 } ( 1 ^ -,2 ^ -,3^+ & , 4^+ ) = \kaps { 1\over ( 4\pi ) ^5 } [ stu m^{\rm tree } ] \times { ( 3 g + 56)s\over 10080 \eps } \cr}\equn\ ] ] where @xmath39 . although these amplitudes are indicative of the divergences present they are not sufficient to determine completely the structure of the counterterms in higher dimensions . for example in @xmath2 there are @xmath78 independent @xmath25 tensors ( onshell ) compared to @xmath79 in @xmath16 . when restricting the external helicities and momenta to @xmath16 naturally we loose most of the information . so to determine the exact counterterms we must calculate with arbitrary external helicity . this we shall do for @xmath80 in the following sections . we can also immediately see one interesting fact . for @xmath19 the @xmath0 divergence does not cancel - unlike in the @xmath15 case . this implies that the counterterms must have a different @xmath25 structure . we shall explore this further also . let us conclude the present section with the remark that the strange coefficients appearing in the above formulas will actually be more reasonable looking in the higher dimensional amplitudes and result from the collapse of several invariants onto the same expression in lower dimension . in this section we move on from the complete amplitudes and focus upon their infinity structure . we will relax our restriction to four dimensional helicities and obtain the infinity in the amplitudes in @xmath81 for arbitrary external polarisations . these amplitudes were calculated using the string - based method of ref . @xcite . in this section we examine the @xmath2 infinity structure . the dimensional reduction of @xmath13 to @xmath2 is @xmath82 supergravity @xcite plus matter . in @xmath2 the potential one - loop counterterm is an @xmath25 tensor . this is analogous to the situation in @xmath16 at three loops where a potential @xmath25 term exists for @xmath1 supergravity @xcite . however , four dimensions is rather special because many of the potentially inequivalent @xmath25 tensors become equivalent at low dimensions . in fact , and as alluded to above , from the potential seven tensors onshell ( actually six remain after integration by parts ) only two are inequivalent in four dimensions . of these only one is compatible with supersymmetry - this is the well known bel - robinson tensor @xcite . however for @xmath28 all seven tensors are inequivalent and the structure of @xmath25 tensors is much richer . the supersymmetrisability has been discussed in @xcite . forgetting about supersymmetry we know from @xcite that a general @xmath25 tensor in @xmath2 is @xmath83 where @xmath84 these are onshell the independent tensors ( actually the riemann tensor means the weyl tensor here ) and the combination @xmath85 vanishes ( or rather is a total divergence ) being proportional to the euler form . @xmath86 in order to calculate the appropriate @xmath15 counterterm we evaluate the ( on - shell ) amplitude and we find it factorises in the following way : @xmath87 where @xmath88 where @xmath89 denotes symmetrisation over the four indices @xmath48 . the factorisation is easily understood if one regards @xmath15 supergravity as the low energy limit of string theory , however it is much more obscure from a field theory viewpoint . in fact the tensor @xmath90 appears in string tree and loop amplitudes . ( see ref . @xcite eqs . ( 7.4.42 ) and ( 9.a.19 ) ) . its appearance in many diverse calculations is presumably due to the uniqueness of a tensor compatible with maximal supersymmetry . the counterterm necessary to cancel this infinity is , # 1 1 ( 2 ) ^4 i ( 4)^#1 @xmath91 where @xmath92 - { 1 \over 4 } \left [ t_4 -4t_7 \right ] \equn\ ] ] there is one arbitrary coefficient @xmath93 since the euler term vanishes onshell . choosing @xmath94 we have @xmath95 \equn\ ] ] this combination is precisely , the tensor combination @xmath96 which appears in the derivative expansion of the @xmath97-theory effective action ( @xcite ) . the tensor @xmath69 is defined in @xcite , here we drop its antisymmetric part . the reason why @xmath69 appears in trees as well as loops is connected to the form of the vertex operators and to triality . calculating with the ( formal ) @xmath98 counterterm we find the infinity has the same tensor structure and is @xmath99 finally , and more interestingly , we consider the infinity arising from the @xmath19 _ matter _ multiplet . this also factorises into the form , @xmath100 where @xmath101 we have organised @xmath102 according to the number of @xmath103 . the @xmath89 denotes symmetrising the terms with a single @xmath104 . the counterterms necessary to cancel this are @xmath105 where @xmath106 we can relate this also to specific tensors contracted against @xmath25 . the tensor @xmath69 can be split into two pieces @xmath107 and @xmath108 each having the same symmetry properties as @xmath69 . the tensors @xmath107 and @xmath108 contain 12 and 48 quartic monomials in the @xmath109 s respectively . they are the only two tensors which have the same symmetry properties of @xmath69 itself in eight dimensions @xcite . specifically @xmath110 where 4#1#2#3#4(^#1#3^#2#4-^#1#4^#2#3 ) @xmath111+[k\leftrightarrow l]+[m\leftrightarrow n ] \biggr)\cr } \equn\ ] ] where @xmath112 $ ] denotes antisymmetrisation with respect to @xmath113 and @xmath114 . from these tensors we can define @xmath115 where the @xmath116 denotes the usual contraction of the upper and lower eight indices . we can also express these tensor contractions as traces @xcite @xmath117 in terms of these combinations the @xmath15 counterterm of the type @xmath118 is just @xmath119 the @xmath19 counterterm can also be expressed in terms of @xmath120 and @xmath121 and is @xmath122 this second tensor structure has also played a role in the low energy limit of string theory , although in this case that of heterotic or type i string theory . it has appeared in the ten dimensional effective action as seen by @xcite both as a string tree and one - loop correction . again the string formulation of tree amplitudes and thus of the appropriate invariant tensors gives an understanding of the factorisation of the amplitude which does not have a simple interpretation in field theory . one checks easily that @xmath123 contains a @xmath69 factor . the counterterm is also supersymmetrisable as shown in @xcite where the most general lagrangians of the form @xmath124 were considered ( without vector multiplets ) . considering @xmath32 supergravity coupled to matter we find @xmath125 where @xmath39 . ten dimensions is , of course , the natural home of both @xmath32 supergravity and of superstring theory . however , within dimensional regularisation it is the @xmath2 counterterms that match easily tensors which appear in string theory . in dimensional regularisation possible counterterms in @xmath29 dimensions at @xmath126 loops are @xmath127 where @xmath128 . with a cut - off regulator , string theory being a physical regulator , the equivalent terms are @xmath129 for @xmath0 our counterterms are thus of the form @xmath130 . indices have been suppressed in this expression and the full form is @xmath131 unless the tensor splits as follows @xmath132 it will not be expressible in terms of the set of tensors found in @xmath2 and those arising in string theory . as a matter of fact one of the counterterms we find does not split in this way . recall that in @xmath0 the `` physical '' combinations are @xmath133 the results of calculating the infinities are firstly , @xmath134 as expected . the infinities have two powers more of momentum as compared to @xmath2 . however we still find that all infinities factorise with one factor of @xmath90 . the remaining factor @xmath135 contains the extra two powers of momentum . specifically we calculate @xmath136 where @xmath137 and @xmath138 the infinities can be canceled by specific combinations of @xmath139 counterterms . once more we find local lorentz invariant counterterms to cancel the infinities . as a working hypothesis we assumed factorisability to make this tractable . the strategy is to calculate arbitrary onshell scalars of the form @xmath140 times the previous @xmath141 , to deduce from the resulting three parameter expression for the amplitude the values of these coefficients and finally to replace the two polynomial solutions in the @xmath142 s by the corresponding invariants in the fourth order in the riemann tensor . ( onshell equivalent to the weyl tensor at this order . ) as is explained in detail in appendix c we may choose to express the counterterms in terms of the following set of tensors , @xmath143 for @xmath144 to @xmath145 the derivatives are contracted with each other and these @xmath146 s are related to derivatives acting upon the @xmath147 s of the previous sections . tensors @xmath148 and @xmath149 however have the derivatives contracted into the riemann tensors . of course many tensors of the form @xmath130 vanish onshell since they produce amplitudes of the form @xmath150 in terms of the @xmath146 s the infinities are canceled by the counterterms @xmath151 where @xmath152 it is far from obvious that such counterterms lead to infinities which factorise , however we can manipulate them to do so . in fact it is possible to express both tensors in the form @xmath153 where we have chosen to contract the derivatives into the @xmath154 tensor . the specific tensors are @xmath155 where possible one must antisymmetrise with respect to the pairs of indices @xmath156 etc and symmetrise with respect to pairs of couples of indices @xmath157 . the tensor for @xmath32 and @xmath32 coupled to matter being the appropriate linear combination of these . both tensors define @xmath130 tensor which must be @xmath26 supersymmetrisable . presumably the vanishing of the @xmath158 is because no @xmath159 supersymmetrisable @xmath130 tensor exists . the counterterms for @xmath26 supergravities are given by linear combinations of the various @xmath160 and in fact the infinities are given by @xmath161 since the various infinities factorise in this way . at the level of the counterterms , all type i supergravity divergences contain the factor @xmath69 . in this letter we have determined the one - loop ultra - violet behavior of the dimensional reductions of @xmath13 supergravity . we have examined this by calculations involving physical on - shell ( four - point ) amplitudes . we have found a variety of results ; some in complete agreement with expectations but some not so obvious from the field theory viewpoint . as expected , the amplitudes are one - loop finite for @xmath162 . for @xmath163 there is a richer structure . in @xmath2 we find infinities which correspond to @xmath25 counterterms and we have completely determined the @xmath25 structure . in @xmath16 the structure of @xmath25 terms is fairly simple ; there are only two independent tensors of which one is compatible with supersymmetry - the well known `` bel - robinson '' combination @xcite . for @xmath28 , there are seven independent tensors . for the @xmath164 theory supersymmetry is less restrictive than for @xmath165 and this allows a further @xmath25 counterterm , in fact we do find infinities belonging to a new tensor structure . within dimensional regularisation , @xmath0 counterterms to a four - point infinity are of the form @xmath130 . the vanishing of the @xmath0 , @xmath166 one - loop infinity is presumably due to the non - existence of a supersymmetric combination of @xmath130 counterterms ( which does not vanish onshell . ) for @xmath26 we do however find a @xmath130 counterterm which is non - zero . an interesting feature of the amplitudes ( so far checked up to 1-loop level ) is _ factorisation_. for the @xmath2 counterterms one finds that the infinities in the amplitude factorise as @xmath167 where the @xmath168 are combinations of the external @xmath169 and @xmath52 . similarly the @xmath0 counterterms factorise as @xmath170 . this factorisation is far from manifest when examining the counterterms however . regarding the amplitudes as arising from the low energy limit of string theory , this factorisation is the remnant of the string factorisation into left and right moving amplitudes . however , it should be noted that the string factorisation occurs _ within _ the loop momentum integral whereas the factorisation of infinity occurs in the amplitude . the factorisation is expected whenever the tree corrections generate the same invariants . however the factorisation of the amplitudes is unexpected from a field theory viewpoint and is hinting towards an alternate description of gravity theories as a product of two yang - mills theories - as suggested in @xcite . although such a formalism is natural in string theory it might well exist in a purely field theoretic context . the counterterm structures we find are related to terms appearing in various places in string theory . this is not surprising since string theory provides a physical regularisation of supergravity . also string / m theory shares many symmetries with supergravity theory and hence counterterms / effective actions are subject to the same constraints . in the present work we have calculated purely within a field theory context . the structures we find are presumably inherent to any regulator of supergravity whether a string theory or a more conventional one . acknowledgements . we wish to thank pascal bain , zvi bern , paul howe and david kosower for useful conversations . we have calculated using the string - based techniques , originally developed for yang - mills calculations @xcite and then applied to gravity calculations @xcite . although originating in string theory the same formalism was subsequently shown to arise from a `` world - line '' approach to field theory @xcite . we refer the reader to these papers for the details of the technique , here merely presenting the results for our amplitudes . the techniques provide a formalism for obtaining the feynman parameter integrals for on - shell amplitudes . the initial step is to draw all @xmath171 diagrams , excluding tadpoles . there is no need to include diagrams with a loop isolated on an external leg since these vanish when dimensional regularisation is used . the external legs of these diagrams should be labeled , with diagrams containing all orderings included . the inner lines of trees attached to the loop are labeled according to the rule that as one moves from the outer lines to the inner ones , one labels the inner line with the same label as the most clockwise of the two outer lines attached to it . the contribution from each labeled @xmath51-point @xmath171-like diagram with @xmath172 legs attached to the loop is @xmath173 where the ordering of the loop parameter integrals corresponds to the ordering of the @xmath174 lines attached to the loop , @xmath175 . the @xmath176 are related to ordinary feynman parameters by @xmath177 . @xmath178 is the `` reduced kinematic factor '' , which the string - based rules efficiently yield in a compact form . the lines attached to the loop carry momenta @xmath179 which will be off - shell if there is a tree attached to that line . one obtains @xmath178 by applying substitution rules to an overall kinematic factor , @xmath180 \exp \biggl [ ( k_i\cdot\pol_j - k_j\cdot\pol_i ) \ , \gbd^{ij } - \pol_i\cdot\pol_j\ , \gbdd^{ij } \biggr ] \cr & \hskip 1 cm \times \exp \biggl [ ( k_i\cdot\bar\pol_j - k_j\cdot\bar\pol_i ) \ , \gbdb^{ij } - \bar\pol_i\cdot\bar\pol_j\ , \gbddb^{ij } \biggr ] \exp \biggl[- ( \pol_i\cdot\bar\pol_j + \pol_j\cdot\bar\pol_i ) \ , h_b^{ij } \biggr ] \biggr|_{\rm multi - linear } \cr } \equn\label{masterkin}\ ] ] where the ` multi - linear ' indicates that only terms linear in all @xmath181 and @xmath182 are included . the graviton polarization tensor is reconstructed by taking @xmath183 . from a string theory perspective @xmath184 is the bosonic green function on the string world sheet , @xmath185 and @xmath186 are derivatives of this green function with respect to left - moving variables , and @xmath187 and @xmath188 are derivatives with respect to right - moving ones . ( since a closed string is periodic the variables describing the string world sheet can be split into `` left - moving '' and `` right - moving '' . ) the term @xmath189 is the derivative of the green function with respect to one left mover and one right mover variable . the functions @xmath190 , @xmath191 and @xmath189 are to taken as symmetric in the @xmath113 and @xmath114 indices while @xmath185 is antisymmetric . although the above expression contains much information in string theory , when one takes the infinite string tension limit it should merely be regarded as a function which contains all the information necessary to generate @xmath192 for all graphs . the utility of the string based method partially lies in this compact representation ( which is valid for arbitrary numbers of legs ! ) . the existence of an overall function which reduces to the feynman parameter polynomial for each diagram is one of the most useful features of the string - based rules . slightly different substitutions rules are used depending upon particle type . the technique is particularly simple for supersymmetric theories where the cancellations between particle types circulating are manifest . in general in gravity theories , in a one - loop @xmath51-point amplitude the amplitude is a sum over diagrams with @xmath174 legs attached to the loop where @xmath193 . for these diagrams the integrand is a polynomial of the feynman parameters of degree @xmath194 . the overlying kinematic expression is of the form @xmath195 . the substitutions act upon @xmath196 and @xmath197 independently - each being up to @xmath174 powers of feynman parameters . for a given particle circulating in the loop the substitution is of the form @xmath198 where @xmath199 denotes a part common to all particle types and @xmath200 is a `` cycle '' contribution which depends upon particle type . in general @xmath199 is of degree @xmath174 whereas @xmath201 is typically @xmath202 or less . bosonic / fermionic particles have a @xmath203 term . in a supersymmetric theory we are thus guaranteed that the term @xmath204 cancels and the feynman parameter integral has maximum degree @xmath202 . for extended supersymmetry there can be further cancellations between the @xmath200 terms and for a @xmath51-point amplitude the degree of the polynomial is @xmath205 for @xmath206 . for supersymmetry with @xmath206 the cancellations imply that the only @xmath171 diagrams contributing to a four point amplitude are the three box integrals . ( this is not quite obvious the cancellations occur on @xmath196 only @xcite so if @xmath207 for example the @xmath196 becomes a polynomial of degree @xmath208 . in other words it cancels completely . ) in ref . @xcite the low energy limit of string theory can be taken and the loop contributions from the different particle types disentangled to obtain the contributions due to a single graviton , weyl fermion , scalar etc . here we will be reconstructing the @xmath32 amplitude again . a string theory consists of two sectors - the neveu - schwarz sector ( ns ) and the ramond sector ( r ) . these sectors contribute in the low - energy limit @xmath209 for a type ii superstring which has @xmath15 supergravity as its low energy limit the contributions are of the form @xmath210 adds up to @xmath211 \equn\ ] ] for a superstring with @xmath32 as its low energy limit the contribution is @xmath212 leading to a contribution @xmath213 \equn\ ] ] we use four dimensional helicity to simplify the overall kinematic expression whereas letting the particles circulate in any dimension @xmath214 . this is very similar to the calculations of ref . @xcite although here we let @xmath29 vary from 4 to 10 . the contributions we seek are for the @xmath15 , @xmath30 and @xmath19 matter multiplets . for the amplitudes @xmath43 a choice of spinor helicity basis simplifies the kinematic expression enormously . spinor helicity techniques utilize a representation of the polarisation vectors in terms of spinor products , @xmath215 where @xmath216 is the momentum of the external state and @xmath217 is a `` reference momentum '' satisfying @xmath218 . ( different choices of @xmath217 correspond to different gauge choices for @xmath219 . the advantage of this technique is that the @xmath217 s may be chosen to simplify the combinations @xmath220 and @xmath221 which appear in @xmath178 . for the amplitude @xmath43 choosing @xmath222 then @xmath223 and @xmath224 with this simplification the entire kinematic expression is given by a kinematic factor @xmath225 \equn\ ] ] times a combination of green s functions . this simplifies the calculations considerably . * step 1 * the @xmath226 amplitude in the language of @xcite we want the contribution of the form @xmath227 \equn\ ] ] for the green s functions . each of the combinations @xmath228 and @xmath229 are quadratic in the feynman parameter polynomials however for the combination @xmath230 two powers cancel leaving just a constant . for all three boxes this coefficient is simple as @xmath231 and we find the amplitude is just a sum over scalar box integrals with the overall factor @xmath232 $ ] . * step 2 * the @xmath233 . in the language of @xcite we want the contribution of the form @xmath234 \equn\ ] ] applying the substitution rules we obtain , @xmath235 \times \left ( i_{1234 } [ f_1 ( a_i ) ] + i_{1243 } [ f_2 ( a_i ) ] + i_{1324}[f_3 ( a_i ) ] \right ) \equn\ ] ] where @xmath236 & = i_{1234 } [ a_3 -a_3 ^ 2 + a_1a_3 ] \cr i_{1243 } [ f_2 ( a_i ) ] & = i_{1243 } [ a_1 -a_1 ^ 2 + a_1a_3 -a_1a_2+a_2a_3 ] \cr i_{1324}[f_3 ( a_i ) ] & = i_{1324 } [ a_1 -a_1 ^ 2 -a_1a_3 ] \cr } \equn\label{eqpoly6}\ ] ] note that this is true in any dimension . reducing these integrals to scalar integrals by one s favorite technique ( our is that of ref.@xcite ) we have @xmath237 = { 2 \over s } \times ( 2-d/2 ) i_{1234}^{d+2 } + { 2 \over st } i^{d}_2(t ) \cr i_{1234}^{d } [ 2a_1a_3 ] = { 2s+2t(2-d/2 ) \over su } i_{1234}^{d+2 } -{2 \over su } i^{d}_2(s ) + { 2 \over su } i^{d}_2(t ) \cr } \equn\ ] ] which yields the amplitude @xmath238 as has been observed before , @xcite , the amplitudes with less supersymmetry have the dimensions of the box integrals shifted relative to the maximal supersymmetric case . * step 3 . * the @xmath239 in the language of @xcite we want the contribution of the form @xmath240 \equn\ ] ] for the boxes we get @xmath241 \cr i_{1243}&[(a_1+a_2)(a_3+a_4)(a_3a_1-a_1a_4-a_3a_2 ) + ( a_1a_4+a_3a_2)/4 ] \cr i_{1324 } & [ a_3 ^ 2(-a_4(a_1+a_4)-a_2(a_1+a_2)+a_2a_4 ) -a_3a_2a_4 + a_3(a_2+a_4)/4 ] \cr } \equn\label{eqpoly4}\ ] ] reducing these integrals to scalar integrals we have @xmath242 * step 4 * the @xmath243 adding together the contributions to have @xmath244 the combination is then just @xmath245 \equn\ ] ] as one might deduce directly . at this point we should discuss which string theory is relevant . in the original work on string - based rules the low energy limit of a heterotic string theory was taken to obtain the rules . later it was realised that the same information could be encoded within a formalism where the fermionic sectors were dropped @xcite . one of the challenges in developing this technique was to break the link between the string theory and the field theory limit allowing calculations in more general field theories such as non - supersymmetric qcd to be performed @xcite . in our work we are focusing on @xmath19 supergravity which is the low energy limit of heterotic string theory and as such the string based rules `` recombine '' and simplify . supergravity is of course the field theory limit of two of the fundamental string theories namely the type i theory of open and closed strings and the heterotic string theory consisting of only closed strings . both these string theories contain gauge groups in the low energy limit which we decouple . in the type i theory the decoupling is extremely simple : the gauge particles are in the open string sector so the contribution to a four graviton scattering amplitude from the @xmath32 multiplet will come from the torus and klein bottle amplitudes only . for a heterotic theory one must drop the contributions arising from the gauge particles arising from the bosonic string . in this section we show how the @xmath246graviton amplitude in @xmath16 may be checked by combining unitarity and helicity techniques @xcite . this approach provides an alternative method to obtain part of our results and at same time a strong check of their consistency . let us begin with the @xmath247-channel cut of the four - point amplitude @xmath43 represented pictorially in fig . [ fcute ] . according to the cutkosky rules , it is given by @xmath248 here @xmath249 denotes the usual invariant lorentz phase space and it contains an additional symmetry factor @xmath250 to keep into account that 2 identical particles are going through the cut . since we wish to construct the entire amplitude , we observe that we can replace the phase space integral by the cut of an unrestricted loop momentum integral @xmath251 while eq . ( [ fourptcut ] ) includes only imaginary part , eq . ( [ fourptcut1 ] ) contains both real and imaginary parts . as indicated , eq . ( [ fourptcut1 ] ) holds only for those terms with @xmath252channel branch cut ; terms without an @xmath252channel cut require a separate determination . a very useful feature of this identity is that we are free to use the on - shell conditions @xmath253 to simplify the integrand : in fact only terms containing no cut in this channel would change . the only internal particles , which give a non - vanishing result in the sum in eq . ( [ fourptcut1 ] ) , are gravitons . fermion contributions vanish because their helicity is not flipped by the graviton vertex implying that @xmath254 . the same holds for vector and scalar amplitudes at tree level ( taking particle / antiparticle instead of positive / negative helicity in the latter ) . ( on a more formal ground , this follows from susy and chiral ward identities @xcite . ) thus in our decomposition only the @xmath15 supermultiplet will produce a non - vanishing @xmath252cut , while the @xmath30 and @xmath19 supermultiplets will have only @xmath255cuts . let us focus on the former case . 2truecm 2.2 truein in the standard helicity formalism ( see _ e.g. _ @xcite ) , the tree amplitude has the form @xmath256\times \left [ s\frac{\spa1.2 [ 3 4]}{[1 2 ] \spa3.4}\right]^2 . \equn \label{am1}\ ] ] the equivalence between the two forms of the amplitude can be easily shown by exploiting the identities:@xmath257 \langle b a\rangle=2 ( a , b)$ ] and @xmath258 \langle a_2 a_3\rangle\cdots[a_{n-1 } a_n ] \langle a_n~a_1\rangle=1/2tr[(1 + \gamma_5 ) a_1\cdots a_n]\equiv tr_+(a_1\cdots a_n).$ ] inserting the r.h.s . ( [ am1 ] ) into the sum appearing in eq . ( [ fourptcut1 ] ) , we obtain @xmath259}{[1 2]\langle \ell_1 \ell_2\rangle}s\right]^2\left[\frac{1}{(\ell_1-k_1)^2}+\frac{1}{(\ell_1-k_2)^2}\right]\times \left [ \frac{[3 4]\langle \ell_1 \ell_2\rangle}{\spa3.4 [ \ell_1 \ell_2]}s\right]^2\left[\frac{1}{(\ell_2-k_3)^2}+\frac{1}{(\ell_2-k_4)^2}\right ] } \ ] ] which we can rearrange to be , @xmath260 \left[\frac{1}{(\ell_2-k_3)^2}+\frac{1}{(\ell_2-k_4)^2}\right ] \equn\label{cutaa}\ ] ] where we have factorized out the kinematical factor given by the tree amplitude . substituting this result into eq . ( [ fourptcut1 ] ) and expanding the products , we can interpret each term as coming from a box integral depicted in figure ( [ cuts8a ] ) . 4 truein -.2 cm note that ( a ) is related to ( b ) and ( c ) to ( d ) by redefinition of loop momenta . for example the product of the first two propagator leads to the integral @xmath261 which corresponds to the box integral @xmath262 in fig . ( [ cuts8a ] ) . @xmath263 ( 4\pi)^{-2}\left[i^4_4 ( s , t)+i^4_4(s , u)\right]\right |_{s - cut } , \equn \label{scut}\ ] ] which is manifestly in agreement with the general result given in eq . ( 2.6 ) for the @xmath15 contribution to the amplitude . [ @xmath264 is absent in ( [ scut ] ) because it does not contain any @xmath252cut . its presence can be detected by looking at the @xmath255cuts . ] the @xmath265- and @xmath50-channel cuts require essentially the same calculations since the helicity configuration are the same in both cases . we shall focus on the @xmath265-channel case and deduce the @xmath50-channel result from this by permuting @xmath266 . in this case the possible helicity configurations of the trees are those represented in fig . [ fcutg ] . since helicity is no longer flipped on either tree , the cut receives a non - vanishing contribution from each of the supermultiplet . 2.2 truein -.2 cm the contribution to the @xmath265-cut from states in the self - conjugate supermultiplet with @xmath267 supersymmetries is given by @xmath268 the above sum can be performed by recalling that all amplitudes are related , via supersymmetric ward identities ( see refs . @xcite ) , to the scalar one according to the relations @xmath269 where @xmath270 and @xmath271 and @xmath272 denote the graviton , the gravitino , the vector , the spin @xmath250 and the scalar respectively . using , in fact , the relations ( [ rel18 ] ) we can rearrange eq . ( [ tcut ] ) to be , @xmath273 where @xmath274 with @xmath275 the term @xmath276 is a kinematical factor that encodes the sum over the different particles , while @xmath277 is , in this case , the amplitude for two scalars into two gravitons , whose explicit form can be obtained by combining eq . ( [ rel18 ] ) with eq . ( [ am1 ] ) . if we restrict ourselves to choices with even @xmath278 , we can simplify @xmath279 as follows @xmath280 where we have used the well - known identity @xmath281 . it remains to compute the combination @xmath282 which corresponds to the contribution of the @xmath15 self - conjugate supermultiplet to the @xmath283cut . inserting the explicit expression for the amplitudes and @xmath284 we find @xmath285 which in turn can be rewritten as @xmath286 { t^2 \over 16 ( k_1\cdot \ell_1)(k_2\cdot \ell_1 ) ( k_4\cdot \ell_1 ) ( k_3\cdot \ell_1 ) } } \equn\ ] ] here , as in the case of the @xmath252cut , we have factorized out the tree amplitude and reduced everything to scalar products . therefore , we are led to write the following master - formula that encompasses all the possible contributions to the @xmath283channel coming from a self - conjugate supermultiplet in @xmath16 @xmath287\times \cr & i \int\dlips\ , { t^2 \over 16 ( k_1\cdot \ell_1)(k_2\cdot \ell_1 ) ( k_4\cdot \ell_1 ) ( k_3\cdot \ell_1 ) } \ , \left ( { \tr_+(4 \ell_1 1 2 \ell_1 3 ) \over t\ , s^2}\right)^{4-m } } . \equn \label{gen1}\ ] ] the remarks made after eq . ( [ fourptcut1 ] ) are also valid here . in particular we are free to turn the scalar products appearing in the denominator of eq . ( [ gen1 ] ) into propagators , when it is convenient . the @xmath15 contribution is selected by choosing @xmath288 . in this case , using the same procedure adopted for the @xmath252channel , we can turn eq . ( [ gen1 ] ) into a sum of boxes . we find a very similar result except that @xmath289 . by posing @xmath290 , we are led to compute the @xmath30 contribution . the first step is to expand the trace in the integrand and to simplify the common factors with the denominator @xmath291=\cr & = -\frac{1}{8 } \left [ 2\left(\frac{1}{(\ell_1\cdot k_1)}+ \frac{1}{(\ell_1\cdot k_4)}-\frac{1}{(\ell_1\cdot k_2 ) } -\frac{1}{(\ell_1\cdot k_3)}\right ) -u\left(\frac{1}{(\ell_1\cdot k_1)(\ell_1\cdot k_3)}+\frac{1 } { ( \ell_1\cdot k_2)(\ell_1\cdot k_4)}\right)\right]\cr } \equn \label{cio}\ ] ] by rewriting the denominators in the last line of eq . ( [ cio ] ) as propagators , @xmath292\equn\ ] ] we can finally turn the phase space integral in eq . ( [ gen1 ] ) into the @xmath283cut of an unrestricted loop integral . as in the case of the @xmath252channel , this is achieved through the substitution @xmath293 the final result is a combination of four triangles @xmath294 and two boxes @xmath295 and it is represented pictorially in fig . [ fcuth ] . 6 truein 7 truecm -.2 cm in terms of the integral functions , we have @xmath296(4\pi)^{-2 } \frac{u t}{s^2 } \left ( \frac{2}{u } i^4_3(t)+\frac{2}{t } i^4_3(u ) + i^4_4(t , u ) \right ) \bigg|_{t - cut } \cr } \equn\ ] ] where we have already added the term @xmath297 , whose presence can be detected by looking at the @xmath298channel . then using the identity @xmath299 we can recast the above combination of boxes and triangles integrals as a @xmath300dimensional box @xmath301(4\pi)^{-2 } \frac{1}{s } i^6_4(t , u ) \right |_{t - cut}. \equn\ ] ] this is exactly the result expected from formula ( 2.6 ) . the @xmath19 contribution is more lengthy , but it can be computed along the same lines . we find again agreement with eq . however in this last case some caution is in order : while the @xmath15 and @xmath30 amplitudes were cut - constructible , namely their exact one - loop expression could be deduced using unitarity @xcite , the @xmath19 is not and only the cut expressions are guaranteed to coincide . the aim of the present appendix is to express each factor appearing in the @xmath302 poles of the @xmath2 and @xmath0 amplitudes in terms of lorentz invariants in the product of four gauge field strengths @xmath303 and their derivatives . one of the advantages of this analysis was to suggest a natural ( and correct ) form for the @xmath130 invariants entering the expression of our counterterm in @xmath0 . this would have been otherwise a hard task since the number of independent @xmath130 invariants in @xmath0 dimensions which do not vanish on shell is more than 30 @xcite . at the end we shall briefly comment on a possible superstring interpretation of the observed factorization of the counterterms , which actually extends to the complete 1-loop amplitude as well . it is useful to express the riemann tensor in terms of the symmetrized product of two field strengths : @xmath304}\,;\,\bar{f}_{ab } \equiv k_{\left[a\right.}\bar{\epsilon}_{\left . b\right]}\nonumber\end{aligned}\ ] ] the vector polarizations @xmath305 and @xmath306 being related to the graviton polarization @xmath307 through a traceless symmetrized product : @xmath308 . as shown in sections 3 and 4 , the @xmath2 counterterms we find are of the form @xmath309 and similarly the @xmath0 counterterms have the factorized form @xmath310 ( @xmath311 ) . as we shall show below , each of the factors can be expressed as a suitable invariant in the product of four vector field strengths @xmath303 and their derivatives . in particular , in the @xmath2 case we find that the factors @xmath312 are two independent invariants constructed contracting suitable combinations of the tensors @xmath107 and @xmath108 with four gauge field strengths @xmath303 . + if we define : @xmath313 we can express @xmath90 and @xmath102 as follows : @xmath314 in @xmath0 we find that the factors @xmath135 in the expression of the counterterm are related to invariants involving @xmath140 terms . there are only three independent such terms which do not vanish on shell @xcite which have the following form : @xmath315 in terms of the above invariants , @xmath316 and @xmath317 are expressed as follows : @xmath318 it is useful to express the product @xmath319 in the basis of the invariants @xmath320 ( @xmath321 ) : @xmath322 so that , multiplying by @xmath90 and using eqs . ( a.4 ) we find the expression given in section 4 on the @xmath0 counterterms ( @xmath30 and @xmath19 respectively ) : @xmath323 the expression of @xmath324 in terms of the tensors @xmath325 is related through ( torsion ) bianchi identities to its expression in terms of the @xmath146 . the first two equations in ( a.5 ) expressing @xmath328 in terms of @xmath320 ( @xmath329 ) and ( a.7 ) expressing @xmath324 in terms of the @xmath325 ( @xmath330 ) can be derived directly by writing the three @xmath331 in the following more compact way : @xmath332 and the @xmath333tensor @xmath334 is defined as follows : @xmath335 the observed factorization of the @xmath1 and @xmath166 counterterms is consistent with the interpretation of these theories as low energy limits of suitable closed superstring theories ( heterotic and type ii respectively ) . indeed , as it was shown above , each factor in the expression of the counterterms can be written in terms of invariants within suitable gauge theories . these gauge theories may be thought of as describing the low energy excitations of the open string theories associated with the _ left _ ( l ) and _ right _ ( r ) movers of some closed string . from this point of view the factorization of the counterterms ( and of the whole amplitudes ) would seem quite natural : it is well known that the free fock space of a closed string theory is the tensor product of the two open string fock spaces corresponding to the ( l ) and(r ) movers , moreover , retrieving the supergravity amplitudes from the @xmath336 expansion of closed string amplitudes ( that is using superstring theory as a regulator of the effective supergravity theory ) , up to one loop order the contributions from the two sectors are expected to factorise . however we wish to emphasize that the philosophy underlying our work is not to go from superstring theory `` downwards '' , that is to study supergravity as its effective low energy theory , but on the contrary to move from field theory `` upwards '' and to study its effective action within a qft framework , using dimensional reduction as a regularisation scheme , instead of string theory . from the field theory point of view the observed factorization is a highly non trivial result , consistent with the interpretation of the supergravities with @xmath337 and @xmath338 supercharges as low energy theories of type ii and heterotic superstring theories respectively . in particular the different forms of the counterterms in the maximal and non maximal cases may be related , in this perspective , to different supersymmetry constraints holding on the two sectors of the closed string theory . more specifically our results seem to suggest that in an open string sector the only possible invariants in four @xmath303 consistent with an @xmath1 supersymmetry have to be constructed by saturating the indices of the field strengths with a @xmath69 tensor . this condition would restrict the possible invariants ( non vanishing on shell ) in four @xmath303 and at most two derivatives , to just one possibility , namely @xmath339 . indeed if we interpret the maximal supergravity as the low energy limit of type ii superstring , which has an @xmath1 supersymmetry on both the ( l ) and ( r ) sectors , according to the above positions the only on shell non vanishing counterterm is the one found in @xmath2 ( with no derivatives ) , which is indeed proportional to @xmath340 . as far as the theory with @xmath338 supercharges is concerned , it may be interpreted as the low energy effective theory of the heterotic superstring , which has an @xmath1 supersymmetry on one sector ( l ) and an @xmath341 on the other ( r ) . again , consistently with our results and the above considerations , we can associate the @xmath90 factor of the corresponding counterterms in @xmath2 and @xmath0 with the sector ( l ) constrained by supersymmetry . b. julia , proc . ams - siam summer seminar on applications of group theory in physics and mathematical physics , chicago 1982 , eds . m. flato , p. sally & g. zuckerman , lectures in applied mathematics , vol . 21 ( 1985 ) p.335 . de causmaecker , r. gastmans , w. troost and t.t . wu , phys . * b105*:215 ( 1981 ) , nucl . phys . * b206*:53 ( 1982 ) ; + r. kleiss and w.j . stirling , nucl . * b262*:235 ( 1985 ) ; + j.f . gunion and z. kunszt , phys . * b161*:333 ( 1985 ) ; + z. xu , d .- h . zhang and l. chang , nucl . * b291*:392 ( 1987 ) ; + m.l . mangano and s.j . parke , phys . * 200 * ( 1991 ) 301 . z. bern , l. dixon , d.c . dunbar and d.a . kosower , nucl . * b425*:217 ( 1994 ) [ hep - ph/9403226 ] ; nucl . * b435*:59 ( 1995 ) [ hep - ph/9409265 ] ; [ hep - ph/9706477 ] ; nucl . * 39bc * ( 1995 ) 146 [ hep - ph/9409214 ] . grisaru , h.n . pendleton and p. van nieuwenhuizen , phys . rev . * d15 * : 996 ( 1977 ) ; + m.t . grisaru and h.n . pendleton , nucl . * b124*:81 ( 1977 ) ; + s.j . parke and t. taylor , phys . * b157*:81 ( 1985 ) .

we present a catalogue of h band spectra for 85 stars of approximately solar abundance observed at a resolving power of 3000 with the kpno mayall 4 m fts . the atlas covers spectral types o7m5 and luminosity classes i - v as defined on the mk system . we identify both atomic and molecular indices and line ratios which are temperature and luminosity sensitive allowing spectral classification to be carried out in the h band . the line ratios permit spectral classification in the presence of continuum excess emission , which is commonly found in pre main sequence and evolved stars . we demonstrate that with spectra of obtained at it is possible to derive spectral types within subclasses for late type stars . these data are available electronically through the astronomical data center in addition to being served on the world wide web . # 1#2#3#4#5#6#7 to = -0.2 truein

astro-ph9808206

with the recent development of large format infrared array detectors , high quality photometric surveys are routinely conducted at wavelengths between 12.5 @xmath3 m . soon the completion of the 2 micron all sky survey ( 2mass ; skrutskie 1997 ) and denis ( epchtein et al . 1997 ) will provide comprehensive catalogues of near infrared sources with detection limits sensitive to a wide variety of stellar and non stellar objects . infrared spectra will be required for appropriate identification of many of these sources , and for further study of their astrophysical properties . the pioneering study of johnson and mendez ( 1970 ) was the first to explore the spectra of a large sample of normal stars in the near infrared . however many years passed before improvements in instrumentation made possible similar observations of large numbers of targets of astrophysical interest . the majority of the work done in near infrared spectroscopy to date has been focused on the k band , in large part because intrinsically cool or heavily obscured objects are typically brighter at k band than in the j or h bands . in 1986 , kleinmann and hall ( 1986 ; kh86 ) provided the first comprehensive medium resolution atlas ( @xmath4 ) of stellar spectra in the k - band , covering all luminosity classes , but restricted to spectral types between f8-m7 . more recently , wallace and hinkle ( 1997 ; wh97 ) have extended the kh96 k band atlas , using the same fts spectrograph on the kpno 4 m with @xmath4 , but including stellar spectra spanning spectral types o - m and luminosity classes i - v . they also summarize the considerable body of work directed toward k band spectroscopy in the last decade . while in many situations , the k band will be the wavelength selection of choice for spectroscopic studies of highly obscured or very cool objects , the presence of circumstellar dust ( @xmath5 ; pollack et al . 1994 ) often results in significant excess continuum emission longward of @xmath6 m . this continuum excess is commonly found in two important classes of objects : young stars with circumstellar disks ( e.g. meyer , calvet , & hillenbrand , 1997 ) and evolved stars with extensive envelopes from mass - loss ( e.g. le bertre 1997 ) . near infrared excess due to warm dust can also complicate spectroscopic studies of composite stellar systems aimed at discerning the stellar populations of other galaxies ( e.g. schinnerer et al . continuum excess longward of @xmath6 m will weaken or even render invisible the photospheric features in the k band , while the photosphere will dominate at shorter wavelengths . in such a situation , near infrared spectra shortward of 2 @xmath3 m will be required to see the stellar photosphere too obscured to be detected optically . to date , there has been relatively little work in the h band ( 1.551.75 @xmath3 m ) . recent publications include : i ) observations of 40 g , k , and m stars of luminosity class i and iii at @xmath7 by origlia , moorwood , and oliva ( 1993 ) ; ii ) the library of 56 spectra o m of luminosity class i , ii , and v at @xmath8 ( lancon & rocca volmerange , 1992 ) ; iii ) a library of 37 stars of luminosity classes i , iii , v at @xmath9 ( dallier , boisson , & joly 1996 ) over a limited portion of the h band ; and iv ) a study of 9 ob stars at @xmath10 ( blum et al . 1997 ) . here we present an h band spectral atlas for 85 stars of nearly solar abundance with spectral types on the mk classification system ranging from 07-m5 and luminosity classes i - v . these @xmath4 spectra were collected with the same fts at the kpno 4 m as the k band atlases of kh86 and wh97 . in section 2 , we describe the sample selection and in section 3 we describe the observations and calibration of the data . in section 4 we discuss the dependence of the spectral features on temperature and luminosity and suggest a two dimensional classification appropriate for late - type stars . in section 5 we discuss near ir spectral classification with regard to wavelength range / spectral resolution , and conclude with a summary of our results . in our sample selection , we chose optically visible stars which had previously been identified on the temperature and luminosity scales of the revised mk system ( keenan 1987 ) . the majority of the stars were drawn from the following fundamental lists : i ) morgan , abt , and tapscott ( 1978 ) for 29 stars o6g0 ; ii ) keenan and mcneil ( 1989 ) for 45 stars g0k5 ; and iii ) kirkpatrick , henry , and mccarthy ( 1991 ) for 5 late type dwarfs k5m3 . we supplemented these primary standards with an additional 5 secondary standards from the compilation of jaschek , conde , and de sierra ( 1964 ) and one late type dwarf classified by henry , kirkpatrick , and simons ( 1994 ) . in order to cover as complete a range of stellar temperature and luminosity as possible , we defined a two - dimensional grid with 26 bins of spectral type and three bins of luminosity class . our temperature grid is binned @xmath11 more coarsely in spectral subclass than the revised mk system , so that we typically sample only every other mk subclass . the three luminosity bins are divided into supergiants ( i ii ) , giants ( iii ) , and subgiants / dwarf stars ( iv v ) . a full sampling of this grid would have resulted in 78 distinct temperature / luminosity pairs . our atlas includes a total of 85 sources with 53 of the bins filled . grid coverage was finer among the later spectral types , where for stars go and later , we filled 26 of the 27 bins ( 9 spectral types @xmath12 3 luminosity classes ) . in contrast , for stars earlier than go , only 27 of the 51 bins were covered ( 17 spectral types @xmath12 3 luminosity classes ) . the 85 individual stars in our h band survey are listed in table [ sample ] along with relevent stellar properties taken from the bright star catalogue ( hoffleit & jaschek 1982 ) . additional restrictions on the sample selection included : i ) v@xmath13 @xmath14 250 km s@xmath15 with exception of hr3982 ( b7 v ) ; ii ) near - solar metallicity ( avoiding those mk standards which exhibit spectral peculiarities due to enhanced or deficient metal abundance ) ; and iii ) no visual companions within the beam ( separations @xmath16 ) . our program was begun in advance of the k band fts atlas of wh97 , and their sources are drawn in large measure from our sample . we note in table [ sample ] the stars for which k band spectra can be found in the wh97 digital atlas . table [ temp ] and figure [ fig1 ] provide additional insight into the temperature and luminosity coverage of our sample . in table [ temp ] , we list each of the spectral type and luminosity bins we have `` filled '' . for each bin in which there is at least one spectrum , we give the corresponding effective temperature . for most stars , we adopted the temperature scale of tokunaga ( 1996 ) , except for giants earlier than g0 where we adopted schmidt kaler ( 1982 ) . figure [ fig1 ] provides a schematic illustrating the temperature and luminosity coverage for the 85 stars in our sample . in this illustration we have applied the same main sequence bolometric corrections to both dwarf ( 27 ) and subgiant ( 11 ) stars ; as such they are indistinguishable in this diagram . observations of our 85 sample stars were obtained at the mayall 4 m telescope at kitt peak national observatory during four separate observing runs from 19931994 ( table [ ftslog ] ) . we used the fourier transform spectrometer ( fts ) dual output interferometer ( hall et al . the fts was ideal for this program for several reasons . first , the wavelength coverage of the fts is limited only by the bandpass of the blocking filters , independent of the spectral resolution . this gave us complete coverage in the j and h bands which would have been difficult to obtain with available grating spectrographs . for example , our h band spectral range is a factor of two greater than the spectra of dallier et al . secondly , the spectral resolution is fixed by the path difference scanned with the interferometer so we were able to chose the highest resolution possible and achieve s / n in excess of 75 for the majority of our sources . finally , because of the novel background subtraction algorithm of the 4 m fts described below , we were able to observe the brightest stars in our sample ( @xmath17 ) during good daytime conditions ( typically mornings ) . combining daytime observations with targeted nighttime observations of key faint sources , the fts provided a uniform set of high quality spectra for a large sample of spectral standards . our observing program included simultaneous spectral coverage in both the j band and the h band . however , the j band data presented difficulties which made it expeditious to focus our initial effort on the h band . the primary problems with the j - band spectra were ; i ) the inherent difficulty in data reduction due to rapid temporal variations in telluric water vapor absorption ; and ii ) and the relative paucity of strong features which would allow spectral classification over the full range of stellar temperature and luminosity . we defer discussion of the j band spectra to a future contribution . spectra were collected simultaneously in the j and h bands with the use of a dichroic beamsplitter to separate the wavelengths longward and shortward of 1.5 @xmath3 m . each star was centered within an input aperture of 3.8 arcsec while sky background was measured through an identical aperture 50.0 arcsec away . the interferogram was scanned at a rate of 1 khz as the path difference was varied continuously from 0.00.75 cm providing an unapodized resolution of 0.8 cm@xmath15 . data were obtained as separate scan pairs , with the path difference varied first in one direction and then the other . a forward backward scan pair was treated as an `` observation '' and observations were repeated in beam switching mode ( a b b a ) . because the sky background from each aperture produces an interferogram shifted in phase by 180@xmath19 at each set of detectors , source spectra are background subtracted in fourier space as they are collected . this permits observations of bright stars to be obtained during good daytime conditions . these beam switched observations were repeated and scans were averaged until adequate signal to noise ratio ( snr ) was achieved . the interferograms were transformed at kitt peak national observatory yielding spectra in units of relative flux versus wavenumber ( @xmath20 in cm@xmath15 ) . the transformed spectra were converted into fits format images and all data reduction was performed using the iraf software package . the spectra were then convolved with a gaussian filter of half width @xmath21 cm@xmath15 . this procedure , commonly referred to as apodization , eliminates `` ringing '' observed in the fts spectra due to the finite scan path of the interferometer . the resulting apodized resolution ( rayleigh criterion ) was @xmath22 cm@xmath15 giving a mean resolving power of @xmath4 in the h band . at this stage the j and h band spectra were separated for ease of reduction . the slope of the continuum was normalized to 1.0 using a four segment spline fitting function . care was taken to keep the residuals from this fit to within 1 % . next we corrected the spectra for telluric absorption features present in the spectra which varied with zenith angle . we attempted to construct an opacity map for the earth s atmosphere by dividing normalized spectra obtained of the a0 star standards at different airmass . because of the simplicity of the a0 star spectra , showing primarily hydrogen lines in absorption , it was relatively easy to monitor the degree to which this procedure was successful . in dividing two normalized spectra of the same star taken at different airmass , all stellar photospheric absorption features should directly cancel , leaving only those absorption features due to the earth s atmosphere . if we assume that the opacity of the telluric absorption is directly proportional to airmass we derive : @xmath23\ ] ] where @xmath24 is the atmospheric opacity , @xmath25 is the low airmass value , and @xmath26 is the high airmass value . a typical opacity map derived in this way for the h band is shown in figure [ fig2 ] . several of the features in this map identified with known constituents of the earth s atmosphere such as water vapor , methane , and carbon dioxide , are denoted in figure [ fig2 ] . again if the atmospheric opacity varies linearly with airmass we can simply scale the opacity for each star so that @xmath27 . using this technique we corrected the spectra to zero airmass ; @xmath28 we used the highest signal to noise a0 standard star spectra ( @xmath29 ) with the largest @xmath30 to define the opacity . we found some residual telluric absorptions , possibly due to water vapor which do not vary strictly with airmass . such features severely complicate the reduction of the j band spectra . finally , the forward and backward scans of each star were averaged and residuals of the differenced spectra were calculated in order to evaluate the average snr . the observations were obtained with the goal of achieving snr of 75 or greater . in most cases this was achieved with the highest quality spectra reaching values of several hundreds . the average snr for each stellar spectrum is included in table [ ftshewsup ] table [ ftshewdw ] below . representative h band spectra are shown in figures [ fig3] [ fig6 ] for luminosity classes i ii , iii , iv , and v , with prominent atomic and molecular features identified . line identifications were made for the strongest lines from comparison with the solar photospheric and umbral near infrared atlases ( livingston & wallace 1991 ( lw91 ) ; wallace & livingston 1992 ( wl92 ) ) . however , at our moderate spectral resolution many features are blended and we found the model atmosphere calculations of oliva , moorwood , and origlia ( 1993 ) to be useful in identifying the dominant contributors to a blend in late type stars . visual inspection of the features in figures [ fig3] [ fig6 ] reveals that @xmath4 h band spectra contain sufficent temperature and luminosity sensitive features to enable spectral classes to be distinguished . beginning with the early type stars , the dominant spectral features are hei 5882 cm@xmath15 ( 1.700 @xmath3 m ) and the brackett series of hydrogen from lines 4 - 10 ( 1.736 @xmath3 m ) to 4 - 16 ( 1.556 @xmath3 m ) . the he i line exceeds the strength of the brackett lines in the very earliest stars ( 06 to b0 ) , with a maximum equivalent width of @xmath31 cm@xmath15 ( hr1903 ; b0 ia ) , and recedes to undetectable levels ( @xmath32 cm@xmath15 ) by spectral type b8 . from the late b to early f stars , the brackett series dominates the spectrum , after which lines of neutral atomic metals begin to take prominence . the strongest metallic lines include mgi , sii , cai , ali , and fei , which increase in strength toward the k stars . finally molecular features of oh and co dominate the spectra of the latest type stars from k5m5 . the most striking luminosity - sensitive feature is the second overtone co bandhead @xmath33 $ ] at 6177 cm@xmath15 ( 1.619 @xmath3 m ) , which is found in the spectra of the k and m stars . this feature is signficantly stronger in stars of lower surface gravity at equivalent spectral type . to further enable spectral classification in the h band , we have identified a set of 9 features which are prominent in stars of spectral type a - m . these include a relatively isolated brackett line ( h4 - 11 ) , 5 neutral metals , and 3 molecular bands . in table [ ftshband ] , we define 9 narrow band indices with bandpasses ranging from 10 to 50 cm@xmath15 , which include each of these features . the variable widths of the bandpasses were selected to minimize line blending , contamination from residual telluric absorption , and sensitivity to radial velocity shifts . table [ ftshband ] also identifies the wavenumber of the dominant contributor and the lower state energy level , the central wavenumber and passband of the index , and additional species which may contribute to the index strength . the equivalent widths of these 9 indices were evaluated from the normalized spectra of our 85 survey stars , and are tabulated in tables [ ftshewsup ] to [ ftshewdw ] in units of cm@xmath15 . \times 10 ^ 8 $ ] ] uncertainties in these equivalent widths depend on the snr of the spectrum in question and the bandpass / strength of the feature . errors range from @xmath34 cm@xmath15 exceeding this upper limit in very few cases . multiple observations of several sources are listed for comparison . the temperature and luminosity dependence for four representative indices is illustrated in figure [ fig7 ] . the 411 brackett line ( hi5950 ) behaves as expected , with a rapid rise to a maximum ( at a peak equivalent width of @xmath35 3 cm@xmath15 ) as @xmath36 approaches 10000 k , and a slower decline toward higher temperatures . the behavior of the index is similar in both the dwarfs and the giants , although the luminosity class i / ii sources show a larger scatter , presumably due to intrinsic variability ( e.g. kaufer et al . the general behavior of the neutral atomic features is illustrated by the mg6345 index . in luminosity classes iv v , this index reaches a maximum strength between 5000 - 6000 k , with a peak equivalent width of @xmath35 2.5 cm@xmath15 . in contrast , the maximum strength of this index in the lower surface gravity objects ( also @xmath35 2.5 cm@xmath15 ) is found in the coolest stars in our sample , monotonically decreasing toward higher temperatures ; as expected given the behavior of ionization state as a function of surface gravity . the two sii indices exhibit similar behavior , but the ali index , ( not shown ) turns over at much lower temperatures in our dwarf stars because of its lower ionization potential . note that we have chosen not to form an index based on the strongest sii line at 6292 cm@xmath15 ( 1.5892 @xmath3 m ) because it is coincident with the 414 brackett line of hi at 6297 cm@xmath15 ( 1.5881 @xmath3 m ) . the behavior of the molecular features is illustrated for both the second overtone @xmath37co ( 6,3 ) and the oh ( @xmath38 ) indices . both indices exhibit a similar behavior with temperature and luminosity , becoming detectable around @xmath36 = 5000 , with a strength in the giants approximately twice that in the dwarfs . similar behavior was noted by kh86 in the first overtone co features in the k band . in dwarf stars the second overtone co index reaches a maximum before m5 , and displays a turnover toward the coolest stars . this may be due in part to features of cai and fei which contaminate the index for intermediate spectral types ( f5k3 ) . ali et al . ( 1995 ) find that the relationship between t@xmath39 and equivalent widths of the first overtone co bandheads flatten out between 35005000 k in dwarf stars . from high resolution ( @xmath40 ) fts spectra , wallace & hinkle ( 1996 ) observed that the 2 @xmath3 m continuum in late type dwarf stars is suppressed by numerous water vapor features which are blended at intermediate to low resolution . predicting the equivalent widths of features where the apparent continuum is subject to temperature and luminosity effects is not straight forward . in contrast , both the co and oh indices continue to rise at the coolest temperatures for stars of higher luminosity . however , the magnitude ( and temperature ) of the maximum in the dwarf stars differs between the co and oh indices , which we use in the next section to define a two - dimensional classification scheme for late type stars . we note that the oh index begins to include a contribution from the stark broadened 4 - 11 brackett line at 5949 cm@xmath15 creating a secondary maxiumum in the strength of this index around 10,000 k in the dwarf stars . while the temperature and luminosity dependence of the atomic features is readily understood through application of the saha and boltzman equations governing the population of the ionization states and energy levels respectively , the explanation behind the behavior of the molecular features is more subtle . two possibilities for the factor of two enhancement in the molecular bands in the giants over the dwarfs have been explored in the literature . one attributes the luminosity dependence in the molecular features to differing microturbulence in the atmospheres of dwarfs and giants . the expectation is that larger microturbulence in the lower surface gravity giants effectively broadens the opacity of the feature over a larger frequency interval in these saturated features , thereby enhancing the equivalent width ( mcwilliams & lambert 1984 ) . another possible contributor is the differing depth of the line formation region in the dwarfs versus the giants , which is fixed by the h@xmath41 opacity . as described by gray ( 1992 ) higher surface gravity results in a higher electron pressure ( and thus h@xmath42 column density ) . this brings the co line formation region closer to the stellar surface reducing @xmath43 according to the following proportionality : @xmath44 in any case , this luminosity dependence of the band strength gives an excellent empirical discriminant between giants and dwarfs , which we exploit below to develop a two dimensional spectral index . to discern surface gravity effects between the super giants and giants or between sub giants and dwarf stars requires more careful study . a detailed examination of line strengths as a function of surface gravity at a fixed temperature reveal the expected trends . however , this behavior does not reveal itself in the coarse analysis afforded by our narrow band indices . while the temperature and luminosity sensitivities outlined above can provide good spectral classification in many instances , discriminants which do not rely on absolute line strength are required when a star is subject to near - infrared continuum veiling . in this case line ratio diagnostics are to be preferred , since absorption features will appear shallower in the presence of continuum excess but line ratios will be preserved as long as the excess is not strongly wavelength dependent . we have identified one diagnostic based on line ratios which can be used to evaluate both temperature and luminosity for stars from k3-m5 in the presence of continuum veiling . this two dimensional spectral index is defined as : @xmath45}{ew[mg6345 ] } \ vs. \ \frac{ew[co6018+co6170]}{ew[mg6345]}\ ] ] where ew is the equivalent width in cm@xmath15 for the indices identified in table [ ftshband ] and listed in tables [ ftshewsup] [ ftshewdw ] . the temperature and luminosity dependence of this diagnostic is illustrated in figure [ fig8 ] . in this diagnostic , the ratio of the oh5920 to mg6345 indices is temperature sensitive , with distinct temperature dependences for dwarfs and giants . specifically we find @xmath46}{ew[mg6345]}\ ] ] and @xmath47}{ew[mg6345]}\ ] ] the comparison of this temperature sensitive ratio with the sum of the two @xmath37co indices , also normalized to mg6345 , then provides an excellent means of identifying both the temperature and luminosity class of late type stars . formal errors in the equivalent width suggest that spectral types can be evaluated to within @xmath2 subclasses ( @xmath48 300 k ) from k3m5 using spectra with @xmath1 based on these indices alone . spectral classification in the near - infrared will become increasingly important in the next decade , as the 2mass and denis near - infrared sky surveys reveal unprecented numbers of stars which are optically invisible . because the 12.5 @xmath3 m region is on the rayleigh jeans tail of most stellar seds , it is not an ideal wavelength regime to pursue spectral classification . yet , there are sufficient features in both the h and the k bands to allow most stellar photospheres to be classified . for heavily reddened sources , the k band will be the wavelength of choice . however , continuum emission from circumstellar dust with temperatures less than 2000 k can heavily veil stellar photospheres at wavelengths greater than 2.0 @xmath3 m . in this case , shorter wavelength spectra are required in order to identify the underlying star . the early - type stars are probably the most challenging for spectral classification in the near infrared . we find that a rough classification from 07-b8 can be made in the h band by the relative strengths of hei 5882 cm@xmath15 and the brackett series ( see also blum et al . hanson , conti , and rieke ( 1995 ) have established a classification scheme in the k band for o b stars relying on lines of helium as well as higher ionization species obtained at @xmath49 . for stars a through early k , the h band may be superior to the k band in providing a large number of intermediate ionization potential species with strong features such as mgi , sii , and fei in addition to the numerous brackett series features . stars k3m5 are probably best classified in the k band ( kh86 ; ali et al . 1995 ; wh97 ) using atomic features of mgi , cai , and nai as well as the first overtone co bandheads observed at @xmath50 . however we have found that these stars also have strong temperature and luminosity sensitive features in the h band such as mgi , ali , oh , and the second overtone co bandheads . the very latest type stars ( @xmath51 ) have very strong , broad , molecular features which can be identified at resolutions as low as @xmath52 . kirkpatrick et al . 1993 ( see also jones et al . 1996 ) have classified stars in the i or the j band employing features due to vo , tio , and feh . in addition , broad water vapor bands observed throughout the 12.5 @xmath3 m region ( jones et al . 1995 ) are an important opacity source in the atmospheres of the coolest stars as well as brown dwarfs ( allard & hauschildt 1995 ) . while the i or j bands are probably the best spectral regions to classify extremely cool stars ( as they lie on the wien side of the planck function for these objects ) more heavily obscured objects can still be profitably observed at low resolution in the j and h bands ( e.g. nicmos on hst ) or in the k band ( wilking , greene , & meyer 1998 ) in search of these water vapor absorptions . the h band spectral atlas we have presented is comprised of moderate resolution spectra with @xmath53 . in contrast , most spectral classification is typically carried out with @xmath54 . the strongest and broadest features in the h band are the co(6 - 3 ) bands and the brackett lines . these features could be identified with much lower spectral resolution than our survey , at @xmath55 . the most crowded region in the h band spectra is that in the vicinity of the hi line at 5948.50 cm@xmath15 ( 1.68110 @xmath3 m ) , the ali triplet at 59645980 cm@xmath15 ( 1.6771.672 @xmath3 m ) , and the sii line at 5993.29 cm@xmath15 ( 1.66853 @xmath3 m ) . in order to properly separate these important features from each other , a resolving power of @xmath50 is required . at this resolution , one can also obtain measurements of the hei line at 5882 cm@xmath15 ( 1.700 @xmath3 m ) . at @xmath56 one can resolve individual components of the ali triplet , the mg i doublet , and the co bandheads , as well as the stark broadened brackett lines in the early type dwarf stars ( table [ ftshband ] ) . an additional issue in the near infrared is the significant contribution to shot - noise from air - glow lines . in the h band air glow from oh is sufficently bright and variable that they compromise @xmath57 h band spectral classification for very faint sources . spectral resolution as high as @xmath58 will be required to resolve the bulk of these air glow features and to obtain adequate snr spectra of faint objects . in summary , we present an h band spectral atlas at a resolving power of @xmath56 that spans a wide range in stellar temperature ( o7-m5 ) and luminosity class ( i - v ) . this spectral region contains a number of temperature and/or luminosity sensitive atomic and molecular features which will allow spectral classification to be carried out in the h - band . as an example of the efficacy of this spectral range for distinguishing stellar spectral types , we define a set of narrow band indices which , with @xmath59 , permit classification of late type stars on the mk system within @xmath2 subclasses . it appears however , that for most applications obtaining h band spectra at @xmath50 will be sufficient for classification . the final reduced averaged spectra as well as the difference of the forward and backward scan pairs ( see section 3 for description of the reduction procedure ) are available through the astronomical data center ( adc ) for each observation listed in this paper . the adc can be contacted directly : i ) by post at astronomical data center , nasa goddard space flight center , code 631 , greenbelt , md 20771 ; or ii ) by telephone at ( 301 ) 2868310 ; or iii ) by fax at ( 301 ) 2861771 ; or iv ) via the internet at http://adc.gsfc.nasa.gov . the data are in fits format with pertinent header information included for each image . these fits format files , useful plotting routines , and other relevent information are also available on the world wide web at http://donald.phast.umass.edu . the raw fts data are also available directly from noao ( contact khh for details ) . we would like to thank lori allen , ed chang , lynne hillenbrand , susan kleinmann , michael skrutskie , and lloyd wallace for helpful discussions . special thanks to john carpenter for assisting in the initial compilation of the standard star lists , and to karen strom and stephen friedman for their assistance in making the data available electronically . antonella romano provided assistance in preparing the tables and figures for publication . support for mrm during the final stages of this work was provided by nasa through hubble fellowship grant # hf01098.0197a awarded by the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , for nasa under contract nas 526555 . se acknowledges support from the national science foundation s faculty award for women program . this work was supported in part through a grant from the national science foundation ( # ast9114863 ) to ses . schmidt kaler , t.h . 1982 , physical parameters of stars , landolt - bornstein new series , vol . 2b , astronomy and astrophysics , stars and star clusters , ed . k. shaifers & h. h. voigt , ( springer verlag : new york ) lllllll & & & & & & + 1903 & * 46 @xmath60 ori & 2.4 & b0 & ia & 26sb & 87 + 1203 & 44 @xmath61 per & 3.4 & b1 & ib & 20sb & 59 + 2827 & 31 @xmath62 cma & 2.5 & b5 & ia & 41v & 45 + 1713 & * 19 @xmath63 ori & 0.1 & b8 & ia & 21sb & 33 + 3975 & * 30 @xmath62 leo & 3.3 & a0 & ib & 3v & 20 + 7924 & * 50 @xmath64 cyg & 1.0 & a2 & ia & -5sbo & 21 + 1865 & 11 @xmath64 lep & 2.0 & f0 & ib & 24 & 13 + 1017 & * 33 @xmath64 per & 0.9 & f5 & ib & -2v & 18 + 7796 & * 37 @xmath65 cyg & 1.1 & f8 & ib & -8v & 20 + 8232 & * 22 @xmath63 aqr & 1.5 & g0 & ib & 7v ? & 18 + 7479 & 5 @xmath64 sge & 2.4 & g1 & ii & 2v ? & 0 + 7063 & @xmath63 sct & 2.2 & g4 & iia & -22sb10 & 10 + 8752 & * - & 3.6 & g4v & @xmath66 & -58v ? & 35 + 7314 & * 21 @xmath67 lyr & 2.1 & k0 & + ii & -31v & @xmath1419 + 6713 & 93 her & 2.4 & k0.5 & iib & -24 & @xmath1417 + 8465 & * 21 @xmath61 cep & 1.1 & k1.5 & ib & -18sb & @xmath1417 + 6498 & 49 @xmath20 oph & 1.9 & k2 & ii & -27 & @xmath1419 + 603 & 57 @xmath68 and & -0.5 & k3 & -iib & -12sb & @xmath1417 + 8089 & * 63 cyg & 1.5 & k4 & ib - iia & -26v & + 8079 & * 62 @xmath69 cyg & 0.5 & k4.5 & ib - ii & -20sb & @xmath1417 + 2061 & 58 @xmath64 ori & -2 . ? & m1 - 2 & ia - iab & 21sb & + 1155 & * & 0.5 & m2 & + iiab & -3v & + 921 & 25 @xmath70 per & -1.7 & m4 & ii & 28 & + 7009 & * & 0.6 & m4.5-m5 & + ii & -19 & + 6406 & * 64 @xmath71 her & -2.4 & m5 & ib - ii & -33v & 21 + 1899 & 44 @xmath72 ori & 3.5 & o9 & iii & 22sb2o & 130 + 1552 & 3 @xmath73 ori & 4.1 & b2+b2 & iii & 23sbo & 40 + 5291 & * 11 @xmath64 dra & 3.5 & a0 & iii & -13sbo & 18 + 403 & * 37 @xmath74 cas & 2.3 & a5 & iii - iv & 7sb & 113 + 1412 & * 78 @xmath75 tau & 2.9 & a7 & iii & 40sb1o & 78 + 4031 & * 36 @xmath61 leo & 2.8 & f0 & iii & -16sb & 84 + 21 & * 11 @xmath63 cas & 1.6 & f2 & iii iv & 12sb & 70 + 5017 & * 20 cvn & 3.9 & f3 & iii & 8v ? & 17 + 2706 & 48 gem & 5.0 & f5 & iii - iv & 13v & 74 + 8905 & * 68 @xmath76 peg & 3.3 & f8 & iii & -11 & 79 + 4883 & * 31 com & 3.0 & g0 & iii & -1v ? & 77 + 4716 & 5 cvn & 2.8 & g6 & iii & -12sb & @xmath1417 + 7328 & 1 @xmath77 cyg & 1.6 & g9 & iii & -29sb & @xmath1417 + 7949 & 53 @xmath60 cyg & 0.2 & k0 & -iii & -11sb ? & @xmath1417 + 8317 & * 11 cep & 2.3 & k1 & iii & -37 & @xmath1417 + 6299 & * 27 @xmath77 oph & 0.8 & k2 & iii & -56v & @xmath1417 + 165 & 31 @xmath74 and & 0.5 & k3 & iii & -7sb1o & @xmath1417 + 6705 & * 33 @xmath65 dra & -1.2 & k5 & iii & -28 & @xmath1417 + 152 & & 1.7 & k5-m0 & iii & -33v & @xmath1417 + 4517 & * 3 @xmath78 vir & 0.3 & m1 & iiiab & 51v ? & + 6242 & & 0.4 & m4 & + iii - iiia & -7v ? & + 7886 & * & -0.6 & m6 & iii & -66v ? & + lllllll & & & & & & + 6588 & 85 @xmath72 her & 4.3 & b3 & iv & -20sb1o & 11 + 4033 & * 33 @xmath79 uma & 3.3 & a2 & iv & 18v & 48 + 1351 & 57 tau & 4.9 & f0 & iv & 42sb1 ? & 109 + 5235 & 8 @xmath62 boo & 1.5 & g0 & iv & 0sb1o & 13 + 5409 & 105 @xmath80 vir & 2.8 & g2 & iv & -10sb & 0 + 6623 & 86 @xmath3 her & 1.4 & g5 & iv & -16v & 20 + 995 & 59 ari & 3.9 & g6 & iv & 0v & + 7602 & 60 @xmath63 aql & 1.7 & g8 & iv & -40v & @xmath14 16 + 7957 & * 3 @xmath62 cep & 1.2 & k0 & iv & -87 & @xmath1417 + 5901 & 11 @xmath77 crb & 2.5 & k1 & iva & -24 & @xmath1417 + 6014 & & 3.6 & k1.5 & iv & -4v & + 2456 & * 15 mon & 5.5 & o7 & v(e ) & 33sb & 63 + 5191 & * 85 @xmath62 uma & 2.4 & b3 & v & -11sb ? & 205 + 3982 & * 32 @xmath64 leo & 1.6 & b7 & v & 6sb & 329 + 7001 & * @xmath64 lyr & 0.0 & a0 & v & -14v & 15 + 2491 & * 9 @xmath64 cma & -1.5 & a1 & vm & -8sbo & 13 + 4534 & * 94 @xmath63 leo & 2.0 & a3 & v & 0v & 121 + 4357 & * 68 @xmath74 leo & 2.3 & a4 & v & -20v & 181 + 4931 & 78 uma & 4.1 & f2 & v & -10v ? & 92 + 1279 & & 5.1 & f3 & v & 36sb1 ? & 25 + 2943 & * 10 @xmath64 cmi & -0.6 & f5 & iv - v & -3sbo & 6 + 1538 & 59 eri & 2.3 & f6 & v & 35 & + 4375 & * 53 @xmath69 uma & 3.0 & g0 & v & -16sb1o & 1 + 4983 & 43 @xmath63 com & 3.1 & f9.5 & v & 6sb ? & 6 + 483 & * & 3.7 & g1.5 & v & 4v ? & 2 + 4374 & 53 @xmath69 uma & 3.5 & g0 & v & -16sb1o & 3 + 5072 & 70 vir & 3.6 & g4 & v & 5v & 1 + 4496 & * 61 uma & 3.8 & g8 & v & -5v & @xmath1417 + 7462 & * 61 @xmath20 dra & 3.0 & k0 & v & 27v & @xmath1417 + 1084 & * 18 @xmath60 eri & 1.6 & k2 & v & 15v ? & @xmath1417 + 8832 & * & 3.2 & k3 & v & -18v & + & gl570a & 3.0 & k4 & v & - & - + 8085 & * 61 cyg & 2.4 & k5 & v & -64v & @xmath1417 + 8086 & 61 cyg & 3.1 & k7 & v & -64v ? & = @xmath1425 + & gl338a & 4.5 & m0 & v & - & - + & gl526 & 4.5 & m1.5 & v & - & - + & * gl411 & 3.6 & m2 & v & - & - + & * gl725a & 4.7 & m3 & v & - & - + rlll & & & + o6-o8 & & 37000 & 38000 + o9 & 32500 & 32000 & 33200 + o9.5 & & & 31450 + b0 & 26000 & 29000 & 29700 + b1 & 20700 & 24000 & 25600 + b2 & 17800 & 20300 & 22300 + b3 & 15600 & 17100 & 19000 + b4 & 13900 & & 17200 + b5 & 13400 & 15000 & 15400 + b6 & 12700 & 14100 & 14100 + b7 & 12000 & 13200 & 13000 + b8 & 11200 & 12400 & 11800 + b9 & 10500 & 11000 & 10700 + a0 & 9730 & 10100 & 9480 + a1 & 9230 & 9480 & + a2 & 9080 & 9000 & 8810 + a5 & 8510 & 8100 & 8160 + a7 & & 7650 & 7930 + f0 & 7700 & 7150 & 7020 + f2 & 7170 & 6870 & 6750 + f5 & 6640 & 6470 & 6530 + f7 & & & 6240 + f8 & 6100 & 6150 & + g0 & 5510 & 5910 & 5930 + g2 & & & 5830 + g3 & 4980 & & + g4 & & 5190 & 5740 + g6 & & 5050 & 5620 + g8 & 4590 & 4960 & + k0 & 4420 & 4810 & 5240 + k1 & 4330 & 4610 & + k2 & 4260 & 4500 & 5010 + k3 & 4130 & 4320 & + k4 & & 4080 & 4560 + k5 & 3850 & 3980 & 4340 + k7 & & & 4040 + m0 & 3650 & 3820 & 3800 + m1 & 3550 & 3780 & 3680 + m2 & 3450 & 3710 & 3530 + m3 & 3200 & 3630 & 3380 + m4 & 2980 & 3560 & 3180 + m5 & & 3420 & 3030 + m6 & & 3250 & 2850 + lllllll & & & & & & + mgi(4s4p ) & 5.39 & 5843.41 & 1.71133 & 5844 & 10 & co , fe , ni , oh + oh(@xmath38 ) & 0.76 & 5920 : & 1.689 & 5920 & 20 & c , co , fe , ni + hi(4 - 11 ) & 12.75 & 5948.50 & 1.68110 & 5950 & 20 & co , fe , ni , si + ali(4p4d tr ) & 4.09 & 5963.76 & 1.67679 & 5972.5 & 25 & co , fe , ni , oh + & & 5968.31 & 1.67552 & & & + & & 5979.60 & 1.67235 & & & + sii(4p3d ) & 5.98 & 5993.29 & 1.66853 & 5993 & 10 & co , fe , ni , oh + @xmath37co(8,5)bh & 1.55 & 6018 & 1.662 & 6017.5 & 15 & fe , oh , s + @xmath37co(6,3)bh & 1.05 & 6177 & 1.619 & 6170 & 50 & ca , fe , ni , oh , si + sii(4p5s ) & 5.98 & 6263.92 & 1.59644 & 6264 & 10 & fe , mg , ni , oh + mgi(4s4p tr ) & 5.93 & 6341.10 & 1.57701 & 6345 & 20 & cn , co , fe , h@xmath81o , ni , oh + & & 6347.88 & 1.57533 & & & + & & 6351.22 & 1.57450 & & & + llllllllllll & & & & & & & & & & & + hr1903 & 26000 & 203 & -0.03 & 0.09 & 0.75 & 0.17 & -0.02 & 0.11 & 0.27 & 0.11 & 0.47 + hr1903 & 26000 & 167 & 0.01 & 0.07 & 0.73 & 0.11 & -0.00 & 0.09 & 0.19 & 0.12 & 0.32 + hr1203 & 20700 & 160 & -0.01 & 0.12 & 0.80 & 0.18 & 0.01 & 0.11 & 0.23 & 0.11 & 0.56 + hr2827 & 13400 & 045 & -0.05 & -0.02 & 1.06 & -0.04 & 0.20 & -0.04 & 0.01 & -0.05 & 0.08 + hr1713 & 11200 & 218 & 0.04 & 0.06 & 1.27 & 0.23 & 0.05 & 0.09 & 0.19 & 0.17 & 0.50 + hr3975 & 9730 & 068 & -0.08 & 0.11 & 2.12 & 0.23 & 0.12 & -0.09 & 0.33 & 0.20 & 0.58 + hr7924 & 9080 & 225 & -0.05 & 0.00 & 1.40 & 0.20 & -0.02 & 0.03 & -0.14 & 0.08 & 0.08 + hr1865 & 7700 & 196 & -0.01 & 0.22 & 2.21 & 0.32 & 0.14 & -0.13 & 0.30 & 0.25 & 0.52 + hr1017 & 6640 & 248 & 0.08 & 0.25 & 1.88 & 0.43 & 0.06 & -0.02 & 0.53 & 0.41 & 1.05 + hr7796 & 6100 & 324 & 0.10 & 0.29 & 1.58 & 0.69 & 0.14 & 0.03 & 0.79 & 0.60 & 0.94 + hr8232 & 5510 & 290 & 0.09 & 0.37 & 1.28 & 0.67 & 0.17 & 0.03 & 1.18 & 0.67 & 1.19 + hr8752 & 5510 & 084 & -0.13 & 0.34 & 1.16 & 0.14 & 0.07 & 0.02 & 0.16 & 0.16 & 0.41 + hr7479 & 5333 & 096 & 0.15 & 0.13 & 1.13 & 0.70 & 0.10 & 0.00 & 0.92 & 0.41 & 0.69 + hr7479 & 5333 & 109 & 0.10 & 0.22 & 1.07 & 0.49 & 0.11 & 0.02 & 0.76 & 0.54 & 1.04 + hr7479 & 5333 & 182 & 0.13 & 0.12 & 1.08 & 0.56 & 0.05 & 0.07 & 0.94 & 0.49 & 1.06 + hr7479 & 5333 & 093 & 0.06 & -0.02 & 1.20 & 0.53 & 0.03 & 0.14 & 1.05 & 0.54 & 0.93 + hr7063 & 4902 & 260 & 0.17 & 0.12 & 0.94 & 0.84 & 0.14 & 0.25 & 1.66 & 0.76 & 1.06 + hr7314 & 4420 & 193 & 0.23 & 0.15 & 0.84 & 0.88 & 0.15 & 0.46 & 2.25 & 1.00 & 1.60 + hr7314 & 4420 & 257 & 0.15 & 0.17 & 0.67 & 0.59 & 0.10 & 0.37 & 1.91 & 0.86 & 1.36 + hr6713 & 4375 & 265 & 0.22 & 0.19 & 0.81 & 0.80 & 0.16 & 0.40 & 1.86 & 0.79 & 1.25 + hr8465 & 4295 & 308 & 0.13 & 0.28 & 0.68 & 0.70 & 0.19 & 0.63 & 2.35 & 1.02 & 1.79 + hr6498 & 4260 & 247 & 0.24 & 0.30 & 0.84 & 1.00 & 0.21 & 0.83 & 2.75 & 1.11 & 1.84 + hr603 & 4130 & 453 & 0.24 & 0.34 & 0.69 & 0.61 & 0.19 & 0.49 & 2.09 & 0.87 & 1.46 + hr8089 & 3990 & 202 & 0.21 & 0.45 & 0.50 & 0.67 & 0.14 & 0.68 & 2.26 & 0.94 & 1.59 + hr8079 & 3920 & 445 & 0.24 & 0.67 & 0.62 & 0.80 & 0.24 & 1.05 & 2.70 & 1.05 & 1.72 + hr2061 & 3550 & 327 & 0.54 & 1.08 & 0.36 & 1.39 & 0.23 & 1.46 & 3.92 & 0.89 & 2.22 + hr1155 & 3450 & 237 & 0.46 & 1.31 & 0.58 & 1.12 & 0.38 & 1.37 & 3.35 & 1.12 & 2.03 + hr921 & 2980 & 225 & 0.77 & 1.41 & 0.29 & 1.27 & 0.37 & 1.10 & 4.08 & 0.67 & 1.52 + hr7009 & 2925 & 292 & 0.63 & 1.43 & 0.79 & 1.55 & 0.54 & 1.79 & 4.68 & 1.26 & 2.21 + hr6406 & 2800 & 319 & 0.59 & 1.41 & 0.80 & 1.66 & 0.54 & 1.85 & 4.98 & 1.31 & 2.31 + llllllllllll & & & & & & & & & & & + hr1899 & 32000 & 164 & -0.03 & 0.17 & 0.54 & 0.15 & -0.04 & 0.03 & 0.18 & 0.10 & 0.49 + hr1899 & 32000 & 182 & -0.04 & 0.20 & 0.62 & 0.11 & 0.05 & -0.03 & 0.07 & -0.00 & 0.07 + hr1552 & 20300 & 115 & -0.07 & 0.13 & 1.29 & 0.31 & 0.10 & -0.09 & 0.12 & -0.01 & 0.09 + hr5291 & 10100 & 159 & -0.07 & 0.19 & 2.65 & 0.78 & 0.02 & 0.05 & 0.28 & 0.11 & 0.41 + hr403 & 8100 & 202 & -0.02 & 0.39 & 2.79 & 0.77 & 0.09 & -0.13 & 0.25 & 0.08 & 0.54 + hr1412 & 7650 & 186 & -0.02 & 0.43 & 2.70 & 0.71 & 0.05 & -0.10 & 0.60 & 0.25 & 0.91 + hr4031 & 7150 & 193 & 0.02 & 0.29 & 2.30 & 0.54 & 0.12 & -0.09 & 0.25 & 0.13 & 0.48 + hr21 & 6870 & 245 & 0.03 & 0.32 & 1.99 & 0.55 & 0.04 & -0.11 & 0.20 & 0.22 & 0.82 + hr21 & 6870 & 179 & 0.01 & 0.25 & 2.21 & 0.63 & 0.13 & -0.14 & 0.46 & 0.24 & 0.69 + hr5017 & 6700 & 174 & 0.04 & 0.28 & 2.58 & 0.83 & 0.07 & -0.03 & 0.43 & 0.36 & 0.96 + hr2706 & 6470 & 086 & 0.01 & 0.39 & 2.08 & 0.47 & 0.16 & -0.10 & 0.13 & 0.20 & 0.64 + hr8905 & 6270 & 059 & 0.06 & 0.13 & 1.06 & 0.45 & -0.01 & -0.16 & 0.30 & 0.28 & 1.02 + hr4883 & 5910 & 280 & 0.15 & 0.20 & 1.07 & 0.52 & 0.09 & 0.04 & 0.90 & 0.43 & 1.02 + hr4716 & 5050 & 133 & 0.17 & 0.10 & 0.97 & 0.47 & 0.13 & 0.07 & 1.29 & 0.54 & 0.92 + hr7328 & 4885 & 281 & 0.19 & 0.20 & 0.78 & 0.57 & 0.11 & 0.11 & 1.33 & 0.67 & 1.35 + hr7949 & 4810 & 369 & 0.21 & 0.24 & 0.77 & 0.59 & 0.13 & 0.18 & 1.39 & 0.67 & 1.38 + hr8317 & 4710 & 191 & 0.29 & 0.14 & 0.63 & 0.58 & 0.08 & 0.26 & 1.79 & 0.83 & 1.64 + hr6299 & 4500 & 532 & 0.26 & 0.09 & 0.77 & 0.81 & 0.10 & 0.44 & 1.92 & 0.76 & 1.40 + hr165 & 4320 & 266 & 0.34 & 0.24 & 0.64 & 0.58 & 0.18 & 0.41 & 1.98 & 0.89 & 1.45 + hr6705 & 3990 & 298 & 0.39 & 0.60 & 0.77 & 1.12 & 0.26 & 0.96 & 3.05 & 1.12 & 1.75 + hr6705 & 3990 & 694 & 0.41 & 0.61 & 0.76 & 1.07 & 0.25 & 0.89 & 2.87 & 1.06 & 1.66 + hr152 & 3956 & 270 & 0.37 & 0.86 & 0.58 & 0.92 & 0.34 & 0.87 & 2.59 & 0.78 & 1.40 + hr4517 & 3780 & 673 & 0.66 & 1.06 & 0.33 & 1.07 & 0.21 & 0.89 & 3.14 & 0.70 & 1.93 + hr6242 & 3560 & 458 & 0.52 & 1.12 & 0.57 & 1.18 & 0.39 & 1.45 & 3.93 & 1.16 & 1.96 + hr7886 & 3250 & 574 & 0.61 & 1.54 & 0.89 & 1.71 & 0.64 & 2.25 & 5.69 & 1.41 & 2.42 + llllllllllll & & & & & & & & & & & + hr6588 & 19000 & 162 & -0.03 & 0.10 & 1.65 & 0.48 & -0.03 & 0.02 & 0.17 & 0.07 & 0.40 + hr4033 & 8810 & 146 & -0.00 & 0.33 & 2.62 & 0.73 & 0.06 & 0.02 & 0.27 & 0.13 & 0.35 + hr1351 & 7020 & 094 & -0.06 & 0.40 & 2.18 & 0.62 & 0.19 & -0.19 & -0.01 & 0.10 & 0.24 + hr5235 & 5930 & 341 & 0.16 & 0.12 & 1.21 & 0.83 & 0.11 & 0.03 & 1.30 & 0.64 & 1.14 + hr5235 & 5930 & 263 & 0.17 & 0.26 & 1.26 & 0.68 & 0.18 & -0.01 & 0.97 & 0.50 & 1.03 + hr5409 & 5830 & 158 & 0.16 & 0.16 & 1.02 & 0.51 & 0.12 & -0.03 & 0.91 & 0.44 & 0.95 + hr6623 & 5680 & 211 & 0.33 & 0.12 & 0.87 & 0.87 & 0.09 & 0.09 & 1.21 & 0.69 & 1.38 + hr995 & 5620 & 143 & 0.22 & 0.08 & 0.80 & 0.61 & 0.23 & 0.03 & 1.09 & 0.48 & 1.14 + hr7602 & 5430 & 056 & 0.46 & 0.15 & 0.71 & 0.86 & 0.01 & 0.17 & 1.40 & 0.57 & 1.49 + hr7957 & 5240 & 189 & 0.22 & 0.09 & 0.68 & 0.70 & 0.07 & 0.15 & 1.04 & 0.46 & 0.99 + hr7957 & 5240 & 192 & 0.09 & 0.18 & 0.56 & 0.52 & -0.01 & 0.09 & 0.75 & 0.40 & 1.17 + hr5901 & 5125 & 395 & 0.37 & 0.13 & 0.55 & 0.62 & 0.08 & 0.21 & 1.61 & 0.74 & 1.60 + hr5901 & 5125 & 153 & 0.37 & 0.09 & 0.75 & 0.93 & 0.15 & 0.30 & 1.54 & 0.71 & 1.13 + hr6014 & 5068 & 072 & 0.31 & -0.12 & 0.39 & 0.59 & 0.14 & 0.05 & 1.10 & 0.70 & 1.40 + hr6014 & 5068 & 133 & 0.42 & 0.11 & 0.58 & 0.55 & 0.18 & 0.09 & 1.48 & 0.65 & 1.24 + llllllllllll & & & & & & & & & & & + hr2456 & 38000 & 053 & -0.03 & 0.36 & 0.39 & 0.26 & 0.16 & 0.12 & -0.32 & 0.11 & 0.31 + hr2456 & 38000 & 073 & -0.12 & 0.18 & 0.48 & 0.10 & 0.06 & -0.15 & 0.15 & 0.03 & 0.10 + hr5191 & 19000 & 282 & -0.05 & 0.19 & 1.71 & 0.57 & 0.08 & -0.07 & 0.22 & 0.07 & 0.24 + hr3982 & 13000 & 241 & -0.08 & 0.13 & 2.22 & 0.47 & 0.04 & -0.09 & 0.28 & 0.03 & 0.15 + hr3982 & 13000 & 241 & -0.07 & 0.09 & 2.14 & 0.51 & 0.06 & -0.09 & 0.32 & 0.12 & 0.26 + hr7001 & 9480 & 678 & -0.05 & 0.36 & 2.83 & 1.06 & 0.03 & -0.03 & 0.41 & 0.15 & 0.52 + hr7001 & 9480 & 146 & -0.10 & 0.38 & 2.81 & 1.02 & 0.10 & -0.05 & 0.43 & 0.17 & 0.41 + hr2491 & 9145 & 111 & -0.07 & 0.36 & 2.71 & 0.83 & 0.16 & -0.14 & 0.37 & -0.08 & 0.10 + hr2491 & 9145 & 148 & -0.14 & 0.40 & 2.67 & 0.93 & 0.12 & -0.13 & 0.39 & 0.05 & 0.20 + hr4534 & 8593 & 205 & -0.04 & 0.45 & 2.67 & 1.01 & 0.11 & -0.12 & 0.30 & 0.07 & 0.34 + hr4357 & 8377 & 192 & -0.05 & 0.40 & 2.83 & 0.96 & 0.09 & -0.12 & 0.24 & 0.06 & 0.38 + hr4931 & 6750 & 119 & 0.04 & 0.24 & 1.47 & 0.62 & -0.01 & -0.06 & 0.09 & 0.09 & 0.86 + hr1279 & 6677 & 080 & 0.10 & 0.32 & 1.62 & 0.67 & 0.15 & -0.09 & 0.26 & 0.12 & 0.64 + hr2943 & 6530 & 327 & 0.04 & 0.33 & 1.60 & 0.57 & 0.15 & -0.14 & 0.54 & 0.27 & 0.74 + hr1538 & 6385 & 281 & 0.06 & 0.34 & 1.26 & 0.53 & 0.18 & -0.11 & 0.51 & 0.22 & 0.90 + hr4375 & 6085 & 254 & 0.26 & 0.15 & 0.63 & 0.56 & 0.15 & -0.06 & 0.68 & 0.37 & 1.09 + hr4983 & 5930 & 123 & 0.21 & 0.16 & 0.75 & 0.66 & 0.09 & 0.04 & 0.71 & 0.45 & 1.20 + hr4983 & 5930 & 173 & 0.19 & 0.21 & 0.90 & 0.52 & 0.19 & -0.06 & 0.87 & 0.47 & 0.96 + hr483 & 5855 & 059 & 0.24 & 0.02 & 0.68 & 0.29 & 0.04 & -0.17 & 0.18 & 0.37 & 1.09 + hr483 & 5855 & 079 & 0.24 & 0.19 & 0.90 & 0.46 & 0.19 & -0.11 & 0.67 & 0.48 & 1.30 + hr4374 & 5830 & 250 & 0.28 & 0.20 & 0.62 & 0.63 & 0.10 & -0.01 & 0.90 & 0.47 & 1.35 + hr5072 & 5740 & 197 & 0.26 & 0.15 & 0.74 & 0.53 & 0.16 & -0.04 & 0.99 & 0.53 & 1.28 + hr4496 & 5430 & 105 & 0.35 & 0.17 & 0.59 & 0.59 & 0.21 & -0.08 & 1.12 & 0.61 & 1.45 + hr4496 & 5430 & 090 & 0.31 & 0.16 & 0.39 & 0.50 & 0.13 & -0.04 & 0.69 & 0.43 & 1.08 + hr7462 & 5240 & 131 & 0.52 & 0.04 & 0.38 & 0.72 & 0.09 & 0.09 & 1.20 & 0.62 & 1.89 + hr7462 & 5240 & 095 & 0.51 & 0.07 & 0.37 & 0.69 & 0.13 & 0.06 & 1.16 & 0.72 & 2.05 + hr1084 & 5010 & 223 & 0.61 & 0.19 & 0.44 & 0.74 & 0.24 & -0.01 & 1.59 & 0.76 & 2.00 + hr8832 & 4785 & 111 & 0.88 & 0.14 & 0.29 & 1.06 & 0.17 & 0.18 & 1.88 & 0.90 & 2.50 + gl570a & 4560 & 120 & 0.90 & 0.22 & 0.37 & 1.13 & 0.28 & 0.04 & 2.02 & 0.90 & 2.29 + hr8085 & 4340 & 125 & 0.65 & 0.18 & 0.01 & 1.15 & 0.11 & 0.15 & 1.59 & 0.60 & 2.38 + hr8086 & 4040 & 170 & 0.72 & 0.39 & 0.06 & 1.46 & 0.14 & 0.11 & 1.78 & 0.50 & 2.07 + hr8086 & 4040 & 181 & 0.72 & 0.32 & 0.07 & 1.48 & 0.15 & 0.12 & 1.86 & 0.52 & 2.18 + gl338a & 3800 & 072 & 0.98 & 0.37 & 0.08 & 1.36 & 0.13 & 0.02 & 1.51 & 0.34 & 1.80 + gl526 & 3605 & 068 & 0.62 & 0.42 & -0.13 & 1.32 & 0.16 & 0.10 & 1.04 & 0.04 & 0.92 + gl411 & 3530 & 202 & 0.48 & 0.53 & -0.04 & 1.29 & 0.16 & 0.01 & 1.17 & 0.10 & 1.19 + gl411 & 3530 & 189 & 0.47 & 0.55 & -0.07 & 1.29 & 0.18 & 0.06 & 1.21 & 0.12 & 1.23 + gl725a & 3380 & 083 & 0.52 & 0.57 & -0.10 & 1.15 & 0.06 & 0.03 & 0.93 & -0.00 & 0.99 + gl725a & 3380 & 107 & 0.54 & 0.56 & -0.12 & 1.25 & 0.02 & 0.08 & 1.20 & -0.02 & 1.09 +

we study the geometric change of chow cohomology classes in projective toric varieties under the weil - mcmullen dual of the intersection product with a lefschetz element . based on this , we introduce toric chordality , a generalization of graph chordality to higher skeleta of simplicial complexes with a coordinatization over characteristic 0 , leading us to a far - reaching generalization of kalai s work on applications of rigidity of frameworks to polytope theory . in contrast to `` homological '' chordality , the notion that is usually studied as a higher - dimensional analogue of graph chordality , we will show that toric chordality has several advantageous properties and applications . most strikingly , we will see that toric chordality allows us to introduce a higher version of dirac s propagation principle . aside from the propagation theorem , we also study the interplay with the geometric properties of the simplicial chain complex of the underlying simplicial complex , culminating in a quantified version of the stanley murai nevo generalized lower bound theorem . finally , we apply our technique to give a simple proof of the generalized lower bound theorem in polytope theory and prove the balanced generalized lower bound conjecture of klee and novik .

1503.06640

a notion at the very core of graph theory , chordality is a statement about the geometry and complexity of cycles in a graph , stating in essence that a cycle is decomposable in the most economic way imaginable . the relation of graph chordality to commutative algebra in particular has motivated many to attempt a generalization of graph chordality to higher dimensions , often using a homological or combinatorial perspective ( we surveyed and summarized this perspective in @xcite ) . however , homological notions , as we shall argue here , are not adequate to capture graph chordality in simplicial complexes , especially not in relation to commutative algebra , and the apparent failure of several important results of graph chordality for homological chordality is a flaw of the approach rather than the problem . as we will see , higher chordality has a better life if we add more information , and let it live in the toric variety ( for simplicial polytopes at least ) , and is a natural symptom of the hard lefschetz theorem , reflecting geometric changes ( as measured by the size of the support ) brought about by multiplication with a lefschetz element . philosophically , we argue that to describe the geometric behavior of simplicial chain complexes , it is better if the boundary map is close to injective . toric geometry allows us to control the boundary map by removing redundant chains in its kernel by factoring out the torus action . to understand this for general simplicial complexes , we use stress spaces @xcite , a construction going back to equilibrium problems in mechanics and elastic introduced by varignon @xcite . it turned out much later that they form a useful model for the intersection theory of a toric variety , in particular the fact that it allows us to describe the support of a chow cohomology class straightforwardly and combinatorially @xcite . the key observation inspiring us was made by kalai @xcite ( compare also ( * ? ? ? * section 2.4.10 ) ) , using a synthetic approach to relate chordality of the graph of a simplicial polytope to minimal rigidity . this beautiful observation deserves a full understanding , especially so because it gives a useful picture of simplicial polytopes with vanishing of primitive second betti numbers ( with respect to the action of a lefschetz element in the cohomology ring of the associated toric variety ) . two natural questions that arise in this context are to extend the observation to all primitive betti numbers and , far more challenging , quantify it beyond the extreme case of vanishing primitive betti numbers . the first problem has been addressed in the celebrated generalized lower bound theorem of murai and nevo @xcite , though without clarifying the beautiful relation to the geometry of homology cycles kalai exhibited . additionally , their proof relies rather heavily on earlier work of green which seems hard , if not impossible , to quantify satisfyingly due to the reliance on generic initial ideals that destroy much of the combinatorial properties of the simplicial complex . the first goal is therefore to understand , reprove and generalize kalai s theorem in a proper context . a secondary objective is to provide a direct and , most importantly , quantifiable proof of the generalized lower bound theorem of murai and nevo @xcite , thereby solving both problems at once , using the newly introduced method of toric chordality . we will also apply our technique to the balanced generalized lower bound conjecture of klee and novik @xcite . more conceptually , this investigation is in line with combinatorial - geometric approaches to intersection theory that has seen remarkable successes disconnecting it from the algebraic varieties it classically lives on ( compare @xcite ) . toric chordality is way to study models of intersection theory in a rather precise geometric way without going to the toric variety . in particular , the present research also has some application to conjectures detailing the interplay between toric algebraic geometry and approximation theory @xcite . * acknowledgements * we thank satoshi murai and eran nevo for extraordinarily useful comments and inspiring conversations concerning their work on the generalized lower bound theorem , and robert macpherson for pointing out some historical facts . we also thank the referee for a thorough read of this paper . before we turn to defining chordality within models of intersection theory , it is useful to recall the homological approach to higher chordality @xcite . while it is an immediate and naive generalization of graph chordality to higher dimensions , we will find it rather unsatisfactory . nevertheless , it is useful to keep in mind , as it does not depend on a geometric realization of the simplicial complex , and it is easily understandable in terms of simplicial homology . let @xmath0 be a simplicial complex . we will consider the * collection of @xmath1-faces * @xmath2 , and the simplicial complex of faces of dimension @xmath3 , the * @xmath1-skeleton * @xmath4 of @xmath5 . a * @xmath1-clique * is a simplicial complex of dimension @xmath1 that contains all possible faces of dimension @xmath3 on its vertex set , and we say that a simplicial complex has a * complete @xmath1-skeleton * if its @xmath1-skeleton is a @xmath1-clique . with this , one can associate to any simplicial complex its * @xmath1-clique complex * , defined as @xmath6 the * support * of a simplicial @xmath1-chain ( cf . @xcite for basics of simplicial homology ) is the simplicial complex generated by the @xmath1-faces on which the chain is non - zero . a graph @xmath7 is * chordal * if every simple cycle of length @xmath8 has a chord , i.e. , an edge connecting non - adjacent vertices of the cycle . the homological statement this translates to is explaining that a graph is chordal if and only if every cycle @xmath9 can be written as a sum of @xmath10-cycles of length @xmath11 that support no vertices that are not already vertices of @xmath9 . said in yet another way , graph chordality is equivalent to the condition that for any @xmath10-cycle @xmath9 in @xmath12 , there exists a @xmath13-chain @xmath14 in @xmath0 with @xmath15 and @xmath16 . this leads us to basic notions of homological chordality . for simplicity , we restrict to homology with real coefficients here . we say that a @xmath17-chain @xmath18 is a * resolution * of a @xmath1-cycle @xmath19 if @xmath20 and @xmath15 . we say that a simplicial complex @xmath0 in which every @xmath1-cycle admits a resolution is * ( resolution ) @xmath1-chordal*. the * @xmath1-leray property * encodes the property that a simplicial complex is @xmath21-chordal for all @xmath22 . a fundamental property of graph chordality , which goes back at least as far as to dirac , is that graph chordality implies the leray property immediately once the trivial obstruction vanishes . we call this the * propagation property * of graph chordality . recall that a * nonface * of a simplicial complex @xmath5 is , naturally , a simplex on groundset @xmath23 that is not a face of @xmath5 . a * minimal nonface * , or * missing face * , of @xmath5 is an inclusion minimal nonface of @xmath5 . equivalently , a simplex @xmath24 is a missing face of @xmath5 iff @xmath25 , but @xmath26 . if @xmath0 is a simplicial complex such that @xmath27 is a chordal graph . @xmath0 has no missing faces of dimension @xmath28 . then @xmath0 is @xmath10-leray . one natural test of strength for a generalization of chordality is whether it satisfies a propagation principle , and homological chordality falls short of this aspect in dimension @xmath28 . [ thm : prp_exterior ] let @xmath0 denote any ( abstract ) simplicial complex without missing faces of dimension @xmath29 . the following are equivalent : @xmath0 is resolution @xmath21-chordal for @xmath30 $ ] . @xmath0 is @xmath1-leray , i.e. , it is resolution @xmath21-chordal for @xmath22 . however , for every @xmath31 , there is a simplicial complex @xmath32 that is resolution @xmath21-chordal for @xmath33 $ ] , has no missing face of dimension @xmath34 , but is not resolution @xmath35-chordal . we take this as a motivation to come up with a better notion of higher chordality that can better control the geometry of cycles . there are several important models for the intersection theory of a complete toric variety @xcite , but two shall be of special importance here : the intersection ring generated by divisors and chow cohomology . their combinatorial cousins are stanley - reisner rings and minkowski weights ( or mcmullen lee stress spaces ) , respectively . because toric chordality will be natural for geometric simplicial complexes beyond polytopes we will be working with these models rather than within toric geometry . if @xmath0 is an abstract simplicial complex on groundset @xmath36:=\{1,\cdots , n\}$ ] , let @xmath37 denote the nonface ideal in @xmath38 $ ] ( cf . @xcite ) , where @xmath38={\mathbb{r}}[x_1,\cdots , x_n]$ ] . let @xmath39:={\mathbb{r}}[{\mathbf{x}}]/i_\delta$ ] denote the stanley reisner ring of @xmath0 . a collection of linear forms @xmath40 in the polynomial ring @xmath38 $ ] is a * partial linear system of parameters * if @xmath41/\theta { \mathbb{r}}[\delta]=\dim { \mathbb{r}}[\delta]-\ell\ ] ] for @xmath42 the krull dimension . if @xmath43 = \dim \delta + 1 $ ] , then @xmath44 is simply * linear system of parameters*. this model was used by stanley to relate combinatorial problems for simplicial polytopes to the hard lefschetz theorem for quasismooth projective toric varieties ( see below ) . rather than working with the intersection ring directly , it shall be more useful to adopt a dual perspective that in the toric setting goes back to weil @xcite where the support of a class is more readily accessible : dual to the stanley - reisner ring in characteristic @xmath45 we consider the stress spaces of @xmath0 ( cf . @xcite ) : the dual action of @xmath38 $ ] acting on itself by multiplication is the action of the polynomial ring on itself by partial differentials ( where naturally every variable in a polynomial @xmath46 is replaced with a corresponding partial differential @xmath47 ) . in details , we define the * stress space * of @xmath0 as @xmath48\rightarrow { \mathbb{r}}[{\mathbf{x}}]\right]\ ] ] and by minding also the linear system of parameters @xmath49.\ ] ] the elements of this space are called stresses ; the spaces themselves are very familiar in toric geometry and coincide simply with the top homology groups of the ishida complex @xcite . note that @xmath50 and the stanley reisner ring @xmath39/\theta { \mathbb{r}}[\delta]$ ] are isomorphic as graded vector spaces . note further that the dual @xmath51 of a linear form @xmath52 acts on the linear stress space , and therefore defines an action @xmath53\times \widetilde{{{\mathrm{s}}}}(\delta;\theta^\vee)\rightarrow \widetilde{{{\mathrm{s}}}}(\delta;\theta^\vee).\ ] ] we refine these definitions in several ways : a * relative simplicial complex * @xmath54 is a pair of simplicial complexes @xmath55 with @xmath56 . if @xmath57 is a relative simplicial complex , it is not hard to modify this definition to obtain the correct picture : we simply define the * relative stress spaces * by @xmath58 $ } \big/ \raisebox{-.3em}{$\widetilde{{{\mathrm{s}}}}[\gamma]$}}}\ ] ] ( where @xmath59/\widetilde{{{\mathrm{s}}}}[\gamma]$ ] denotes a quotient of vector spaces ) and @xmath60\ ] ] and analogously for the relative stanley reisner module of @xmath61 . observe furthermore that @xmath44 induces a map @xmath62 by associating to the vertices of @xmath0 the coordinates @xmath63 , where @xmath64 . hence , as is standard to do when considering stress spaces , we identify a pair @xmath65 with a * geometric simplicial complex * , i.e. a simplicial complex with a map of the vertices to @xmath66 . the differentials given by @xmath67 are therefore @xmath68 , where @xmath69 is the gradient . conversely , the canonical stress spaces and reduced stanley reisner rings , respectively , of a geometric simplicial complex are those given by the linear system of parameters given by the geometric realization , so that we usually leave out the linear system of parameters when denoting the stress space of a geometric simplicial complex . a geometric simplicial complex in @xmath70 is * proper * if the image of every @xmath1-face , @xmath71 , linearly spans a subspace of dimension @xmath72 . a sequence of linear forms is a ( partial ) linear system of parameters if the associated coordinatization is proper . [ cnvt ] the stress space of a _ geometric simplicial complex _ is considered with respect to its natural system of parameters induced by the coordinates . we will make this clear by simply writing @xmath73 for @xmath50 , where @xmath44 is the linear system induced by the vertex coordinates of @xmath0 . finally , we say a sequence of linear forms @xmath44 of size @xmath74 is * regular * ( up to degree @xmath1 ) if , for every truncation @xmath75 , @xmath76 , we have a surjection @xmath77 for every @xmath21 ( at most @xmath1 ) . the * depth * of a simplicial complex is the length of the longest regular sequence that its stress space admits ; this is essentially independent of the sequence of linear forms chosen as long as it is a ( partial ) linear system of parameters . * cohen macaulay * singles out the abstract simplicial complexes for which depth equals krull dimension . the following results are central to combinatorial commutative algebra , and relate topological and geometric features of the simplicial complex to the algebraic features of the ring . recall that the * star * and * link * of a face @xmath24 in @xmath5 are the subcomplexes @xmath78 _ the hochster hibi theorem , cf . _ a simplicial complex @xmath0 of dimension @xmath79 is of depth @xmath80 if and only if its @xmath81-skeleton @xmath82 is cohen macaulay if and only if for every face @xmath24 of @xmath0 , the reduced homology ( with real coefficients ) of @xmath83 vanishes below dimension @xmath84 . _ the hard lefschetz theorem , cf . _ if @xmath85 is a simplicial @xmath86-polytope , @xmath87 and @xmath88 denotes the canonical differential on @xmath38 $ ] ( corresponding to the anticanonical divisor ) associated to the homogenizing embedding @xmath89 , then @xmath90 is an isomorphism for every @xmath91 . we shall use @xmath92 to denote the canonical differential on vertex set @xmath93 , i.e. @xmath94 . for the results that follow , we always think of every simplicial complex as coming with an explicit coordinatization in a vector space over @xmath95 . recall that for us , it is in general not required that this realization be an embedding of the simplicial complex ; rather , we usually require some properness with respect to the coordinates of vertices . we close with a useful notion related to simplicial stresses : the space of squarefree coefficients of @xmath96 is also called the space of * minkowski weights * @xcite , denoted by @xmath97 . consider a relative proper simplicial @xmath86-complex @xmath61 in @xmath70 . then @xmath98 the map restricting a stress to its squarefree terms , is injective ( and therefore an isomorphism ) . moreover , @xmath99 is the space of weights on @xmath100-faces that satisfy the * minkowski balancing condition * : for every @xmath101-face @xmath102 of @xmath61 , @xmath103 i.e. the sum on the left - hand side lies in the linear span of @xmath102 . here , we use @xmath96 to denote the stress space of the geometric simplicial complex @xmath61 as convened upon in convention [ cnvt ] . in this situation , we therefore extend the action of the stanley reisner ring on the stress spaces to minkowski weights , compare also @xcite for a direct definition of this action . to define toric chordality , it is useful to consider first a situation where it stands out as an extreme case . we use @xmath104 to denote the nonnegative part of a real function . combinatorially , we think of a stress in a proper simplicial complex as the underlying * support * , i.e. the simplicial complex generated by faces whose coefficients are non - zero , and therefore speak of faces of a stress in a simplicial complex . [ thm : stress_quant ] we consider a simplicial polytope @xmath85 of dimension @xmath86 , and its boundary complex @xmath0 . let @xmath105 , and let @xmath106 be any integer . there exists a set @xmath107 of at most @xmath108 vertices such that , for every @xmath109 , we have @xmath110 the cokernel of @xmath111 is of dimension @xmath112 . the second part is a simple conclusion of the hard lefschetz theorem ; for the first fact , we observe an auxiliary lemma . the * boundary of a star * is @xmath113 and the * open star * of @xmath24 in @xmath0 is the relative complex @xmath114 [ lem : quant ] in the situation of theorem [ thm : stress_quant ] , the map @xmath115 is an injection for all but at most @xmath116 vertices of @xmath0 . let us recall two simple facts : _ mcmullen s integral formula for the @xmath117-vector , cf . * prop.4.10 ) ) . _ for any proper geometric simplicial @xmath81-sphere @xmath0 in @xmath70 , and any integer @xmath1 , we have @xmath118 _ cone lemma i ( ( * ? ? ? * cor.1.5 ) & ( * ? ? ? * thm.7 ) ) . _ for any vertex @xmath119 , @xmath0 , a geometric simplicial complex in @xmath120 , any integer @xmath1 , and @xmath121 the orthogonal projection of @xmath122 to the orthogonal complement of the vector @xmath123 , we have an isomorphism @xmath124 mcmullen and swartz state the integral formula only for the @xmath125-vector ( defined in this context by @xmath126 ) , but the form for the @xmath117-vector given here follows directly from the underlying identity for hilbert functions implicit in swartz proof . for an appropriate combinatorial definition of @xmath127 , this fact in particular extends to all simplicial complexes . we need also a second version of the cone lemma that corresponds to a pullback in chow rings . in the situation of the first cone lemma and @xmath0 in @xmath128 we have a natural isomorphism @xmath129 the map @xmath130 is clearly an isomorphism . let @xmath131 denote an orthogonal basis of @xmath132 . then the preimage of @xmath133 vanishes under the elements of @xmath131 by injectivity of @xmath134 , so that @xmath135 is surjective ; injectivity follows as the map is obtained as restriction of an injective map . it suffices to show that @xmath136 for all but at most @xmath137 vertices . but @xmath138 for all @xmath123 by the hard lefschetz theorem , so the claim follows by mcmullen s integral formula . claim ( b ) is clear . it remains to discuss claim ( a ) . by lemma [ lem : quant ] , the map is an isomorphism except for a small set @xmath139 of vertices @xmath123 . in particular , if @xmath140 is a stress , and we consider the restriction @xmath141 of @xmath142 to the open star of @xmath123 in @xmath0 , then @xmath143 unless @xmath144 . the extreme cases of theorem [ thm : stress_quant ] lead us to consider toric chordality : consider a geometric simplicial complex @xmath0 , and a linear differential @xmath145 . we say that @xmath146 is * toric @xmath1-chordal * for some integer @xmath1 if @xmath145 induces a surjection @xmath147 and an injection @xmath148 we say the pair @xmath146 is * weakly toric @xmath1-chordal * if @xmath149 is an injection for all vertices @xmath123 of @xmath0 . it is useful to remark that the three conditions for toric chordality do not come without interplay . we have @xmath150 , so injection is implied by . if @xmath0 is a proper geometric simplicial complex in @xmath70 , then toric @xmath86-chordality implies resolution @xmath81-chordality . indeed , in this situation and following the action @xmath39\acts { { \mathrm{m}}}(\delta)$ ] , we have a natural isomorphism @xmath151 compare also ishida @xcite and equivalently tay whiteley @xcite . indeed , following ishida , stress spaces are but special simplicial chains and cycles with real coefficients . we observe as above : [ lem : quant2 ] if @xmath146 is weakly toric @xmath1-chordal for some @xmath106 , then for every @xmath142 in @xmath152 , we have @xmath153 argue as in the proof of theorem [ thm : stress_quant ] . it follows in particular that the map @xmath154 is injective . [ cor : quant2 ] weak toric chordality propagates , i.e. a weakly toric @xmath1-chordal complex , @xmath106 , is also weakly toric @xmath74-chordal for every @xmath155 . let us justify the notion of toric chordality by reconsidering theorem [ thm : stress_quant ] : consider the boundary of a simplicial @xmath86-polytope @xmath0 . if @xmath156 ( for some @xmath106 ) , then @xmath0 is toric @xmath1-chordal . this follows at once from the lefschetz theorem . the main motivation for toric chordality , however , is that it satisfies the propagation principle ; clearly in the form of theorem [ thm : prp_exterior ] , but even after it is satisfied only in a single degree . [ thm : toric_propagation ] assume that , , for some @xmath106 , a geometric properly embedded simplicial complex @xmath0 is toric @xmath1-chordal w.r.t . a linear differential @xmath145 and that @xmath0 has no missing faces of dimension @xmath72 . then @xmath0 is toric @xmath17-chordal . consider a @xmath17-stress @xmath142 in @xmath0 , and its restriction @xmath157 to @xmath158 for some @xmath159 . then @xmath160 is , by assumption , in the image of @xmath161 . we proceed to show that there exists a relative @xmath162-stress @xmath163 in @xmath158 that maps to @xmath164 under @xmath134 . to this end we show that @xmath165 is supported in @xmath166 . since there are no missing faces of cardinality @xmath167 , a @xmath1-face @xmath168 is in @xmath166 if and only if @xmath169 is in @xmath166 . hence , it suffices to show that for every @xmath170 , the stress @xmath171 is supported in the star of @xmath123 . consider the commutative diagram @xmath172{r}{\ \upomega\ } & { { \mathrm{s}}}_{k-1}(\delta)&\\ { { \mathrm{s}}}_{k+1}(\operatorname{st}_v^\circ \delta ) \arrow{r}{\ \upomega\ } \arrow{ur}{\updelta_v } & { { \mathrm{s}}}_{k}(\operatorname{st}_v^\circ \delta ) \arrow{ur}{\updelta_v } & & \end{tikzcd}\ ] ] the top horizontal map @xmath145 is an injection , and the bottom map is a surjection when restricted to the image of the restriction map @xmath173 hence , the inverse of @xmath174 under @xmath175 maps to @xmath176 under @xmath134 . hence @xmath176 is a stress supported in the star of @xmath123 , which implies that @xmath177 is supported in the star of @xmath123 . via the cone lemma , we obtain as desired a preimage for @xmath157 under @xmath145 in @xmath178 . finally , to obtain the surjection @xmath179 let @xmath142 denote a @xmath17-stress in @xmath0 , and let @xmath157 denote is restriction to the relative stress in @xmath158 . by the previous argument , we construct relative stresses @xmath163 that map to @xmath157 under @xmath145 . now , by the cone lemma and weak toric chordality , @xmath145 is injective as a map from @xmath180 to @xmath181 for every edge @xmath182 of @xmath0 . hence , @xmath163 and @xmath183 , for vertices @xmath123 , @xmath184 of @xmath0 , coincide in @xmath185 . hence , there exists a stress @xmath186 that restricts to @xmath163 on @xmath187 : it is given by assigning the monomial @xmath188 , @xmath189 , the coefficient of @xmath188 in @xmath163 . it follows that @xmath146 is toric @xmath17-chordal . [ cor : vanishstresshvecor ] consider a proper geometric simplicial complex in @xmath70 , and assume @xmath0 has no missing faces in dimension @xmath29 for some @xmath106 . if @xmath146 is toric @xmath1-chordal , then @xmath190=0\ ] ] for all @xmath22 . the following fact is also a simple consequence of theorem [ thm : toric_propagation ] . [ cor : propagation_sr ] let @xmath0 denote an abstract simplicial complex , @xmath145 a linear differential , and @xmath67 a coordinatization of @xmath0 in a real vector space . assume that , for some @xmath106 , with respect to the coordinatization and the differential @xmath145 , @xmath0 is toric @xmath1-chordal , and @xmath0 has no missing faces of dimension @xmath29 , and that @xmath191 is regular up to degree @xmath1 on @xmath73 . then @xmath192 is a regular system of parameters , and @xmath0 is cohen macaulay . this fact is very easy to see when adopting the algebraic perspective : it uses the simple fact that a graded module @xmath193 over @xmath194 $ ] such that @xmath195 is injective for every linear coordinate change @xmath196 is free . the proof only needs one easy observation ; to simplify notation , let us just look at the case @xmath197 . how would we show that the multiplication by @xmath198 defines an injective map on @xmath193 ? simply because we can exploit the linear transformation @xmath199 , and use the assumption coupled with the fact that @xmath200 dirac s cut theorem is a second influential theorem on chordal graphs , cf . @xcite ; acting in the opposite direction of the propagation theorem , and is a cornerstone of results in chordal graph theory . let @xmath201 denote a graph , and let @xmath202 denote its induced intersection ( i.e. their intersection is an induced subgraph of @xmath7 ) . assume that we have a surjections @xmath203 for @xmath204 , and that @xmath7 is chordal . then @xmath205 is connected . if , moreover , a cycle @xmath206 has preimages @xmath207 so that @xmath208 for @xmath209 , then @xmath9 has a resolution in @xmath205 ( in the sense of resolution chordality ) . again , a version of this theorem also holds in the toric case : let @xmath210 denote a geometric simplicial complex in @xmath211 , and let @xmath212 denote their induced intersection ( i.e. @xmath213 is an induced subcomplex of @xmath0 ) . assume that , for two differentials @xmath214 , and some @xmath106 , we have surjections @xmath215 for @xmath204 , and @xmath216 , \upomega)$ ] is toric @xmath1-chordal . then @xmath217 surjective . if , moreover , a stress @xmath218 has preimages @xmath219 so that @xmath220 for @xmath209 , then @xmath221 can be chosen so that @xmath222 . here , @xmath223\ ] ] can naturally be seen as the stress space of a geometric simplicial complex in @xmath70 , justifying the second condition . for a @xmath100-stress @xmath142 in @xmath224 , consider its preimages @xmath225 , @xmath226 under @xmath227 . then @xmath228 is a stress in @xmath229 ( as @xmath230 ) , so that @xmath231 , restricted to @xmath232 , is a preimage @xmath233 of @xmath234 under @xmath145 . but then @xmath235 is a stress in @xmath236 , and @xmath237 as desired . the second part of the theorem follows if we consider the support of @xmath238 . so far , toric chordality is only an interesting concept , without any clear connection to homological chordality studied classically . let us try to remedy this by translating theorem [ thm : stress_quant ] to a statement about simplicial homology . the results of this section refine observations by kalai @xcite from graph cycles to general homology cycles ; see also @xcite for a generalization of his ideas for geometrically restricted homology cycles . we use toric chordality instead . [ lem : torreschord ] let @xmath0 in @xmath70 be any proper geometric simplicial complex . assume that , and some @xmath106 , @xmath239 for every @xmath240 of cardinality @xmath241 , @xmath242= \alpha$ ] and that @xmath146 is weakly toric @xmath1-chordal . then @xmath243 for every @xmath244 . in particular , if @xmath146 is toric @xmath1-chordal and ( a ) holds then @xmath0 is resolution @xmath100-chordal , and has in particular no missing @xmath1-faces . here @xmath245 is the subcomplex of @xmath0 induced by a vertex set @xmath244 , and @xmath246 denotes reduced betti numbers with real coefficients . [ cor : torreschord ] let @xmath0 be the boundary of any simplicial @xmath86-polytope , and let @xmath247 . then @xmath248 for @xmath244 . theorem [ cor : torreschord ] alone can also be proven rather elegantly using the stratification of the toric variety by torus orbits , which can be thought of as a diagram of spaces built over the intersection poset of @xmath85 . studying the associated resolution of the ishida complex gives the desired fact easily ( compare also @xcite ) . lemma [ lem : torreschord ] can be proven this way as well with just a little more work . the situation of @xmath249 has an advantage , as alexander duality and the dehn sommerville relations allow us to make a statement for @xmath91 as well . let @xmath0 be the boundary of any simplicial @xmath86-polytope , and any @xmath91 , @xmath250 this corollary alone can also be shown using the mayer vietoris resolution of the ech complex of local stress spaces as in @xcite . compare also @xcite for an application to a conjecture of kalai . to prove the desired inequality , we provide iteratively an embedding of homology classes associated to certain critical morse loci into the to the space of coprimitive stresses @xmath251 . we argue by induction on @xmath252 , where we note that the case @xmath253 is done already : * induction start : * to start with , assumption ( c ) gives us an injection @xmath254:{\widetilde}{h}_{d-1}(\delta_{w})\ { \ensuremath { \lhook\joinrel\relbar\joinrel\rightarrow}}\ { { \mathrm{s}}}_{d}(\delta)/\upomega{{\mathrm{s}}}_{d+1}(\delta)\ ] ] from homology classes to coprimitive stresses , see isomorphism . * induction step : * the key observation for the induction is that , since @xmath255 following assumption ( a ) , homology classes of @xmath256 correspond to relative classes in @xmath257 . to exploit this , we use morse theory , following the construction of morse homology : remove the vertices of @xmath0 one by one , exploring the complex as a morse function does , until we arrive at the desired vertex set @xmath93 . we use the morse function to build an injection @xmath258:{\widetilde}{h}_{k-1}(\delta_{w})\ { \ensuremath { \lhook\joinrel\relbar\joinrel\rightarrow}}\ { { \mathrm{s}}}_{k}(\delta)/\upomega{{\mathrm{s}}}_{k+1}(\delta).\ ] ] if , in some stage of removal , the map @xmath259 induced by inclusion is _ not _ a injection , then we call @xmath123 a _ critical point_. given the sequence @xmath260 we see that the boundaries of relative classes in @xmath261 generate the kernel of @xmath262 . let @xmath263 denote a subspace of @xmath264 so that @xmath265 is exact . by the cone lemmas , we can think of @xmath266 as the nonrelative simplicial complex @xmath267 in @xmath268 , so that we can apply the induction assumption to @xmath267 and obtain an embedding @xmath269 : { \widetilde}{h}_{k}(\delta_{x},\delta_{x\setminus \{v\}})\ { \ensuremath { \lhook\joinrel\relbar\joinrel\rightarrow}}\ { { \mathrm{s}}}_{k+1}(\operatorname{st}_v^\circ\delta_x)/\upomega{{\mathrm{s}}}_{k+2}(\operatorname{st}_v^\circ\delta_x).\ ] ] we compose this with the partial differential @xmath270 and obtain a map to @xmath271 via the cone lemma , obtaining an embedding @xmath272\ } \ { { \mathrm{s}}}_{k}(\operatorname{st}_v\delta_x)/\upomega{{\mathrm{s}}}_{k+1}(\operatorname{st}_v\delta_x).\ ] ] since , by assumption , elements of the kernel of @xmath262 generate homology classes of dimension @xmath273 in @xmath274 , the image of this embedding lies in @xmath275 , so that we can see the previous map as an embedding @xmath272\ } \ { { \mathrm{s}}}_{k}(\delta_x)/\upomega{{\mathrm{s}}}_{k+1}(\delta_x).\ ] ] we now proceed iteratively , and remove vertex after vertex from @xmath276 . the combination of the maps constructed in each step gives a map @xmath277 : \bigoplus_{j=1,\cdots , n } \mathcal{a}_{x_j , v_j,\delta , k } \ \longrightarrow\ { { \mathrm{s}}}_{k}(\delta)/\upomega{{\mathrm{s}}}_{k+1}(\delta)\ ] ] where @xmath278 is a sequence of vertices in @xmath276 with @xmath279 and @xmath280 . since every homology class in dimension @xmath273 has a unique representation as linear combination of elements in @xmath281 ( it was chosen to be a basis of the generated homology at the critical points of the morse function , after all ) , this induces a map @xmath282 to see that this map is an injection , assume the contrary . since the map is locally ( i.e. in every construction step when removing a vertex ) an injection , there must be @xmath283 so that @xmath284)(\gamma_j ) \in { { { \mathrm{s}}}_{k}(\delta)}\ = \ \overline{\gamma}\ \in\ \upomega{{\mathrm{s}}}_{k+1}(\delta).\ ] ] let @xmath285 denote the smallest @xmath286 such that @xmath287 . but then @xmath288 is a chain supported in @xmath289 by weak toric chordality . hence @xmath290 is a boundary in @xmath291 , in contradiction with the construction . note that the proof of lemma [ lem : torreschord ] was very wasteful , only recording the birth , not the death , of homology classes . in particular , it is immediate that the inequalities above are not sharp , especially when the primitive betti numbers are far from minimal . it would be interesting to understand whether the inequalities , or the connection between stress spaces and ordinary homology in general , could reveal some interesting results on higher connectivity of polytope graphs and skeleta . in the same vein , the perspective via toric geometry could also be interesting to study higher - dimensional expansion ( cf . @xcite ) , in particular in connection with hodge theory of toric varieties . indeed , we saw here that toric chordality is an effective approach to study what is essentially the case of minimal connectivity , while higher expansion focuses on the case of maximum connectivity , so that it seems only natural to think that both can be studied in the same setting . for a dual result , we consider the setting in which chordality fails at a stable set of vertices . [ thm : torreschor ] let @xmath0 in @xmath70 be any proper geometric simplicial complex . assume that , for some @xmath106 , @xmath239 for every @xmath240 of cardinality @xmath241 , @xmath292 is an injection for all but a set @xmath139 of vertices @xmath123 of @xmath0 , and that @xmath293 is a surjection . then @xmath294 for every subset @xmath295 . the proof is analogous to the one of lemma [ lem : torreschord ] , this time taking care that none of the `` bad '' vertices of @xmath139 get removed in the iterative removal of vertices going from @xmath276 . let @xmath296 be such a removed vertex , going from a vertex set @xmath297 containing @xmath123 and @xmath139 to the vertex set @xmath298 . following the proof of lemma [ lem : torreschord ] above , a supposed homology @xmath100-class @xmath142 generated in the attaching step corresponds injectively to a stress in @xmath299 . but such a stress is in the image of @xmath145 in @xmath0 ; so that @xmath142 is a boundary in @xmath300 ; a contradiction . we formulate a useful consequence . [ cor : betti ] let @xmath0 be the boundary of any simplicial @xmath86-polytope , and let @xmath301 for some @xmath106 . then there exists a subset @xmath302 , @xmath303 such that @xmath294 for @xmath244 . in particular , nontrivial homology classes in @xmath304 are not only boundaries in @xmath0 , but one can find them to be boundaries of chains with at most @xmath305 additional vertices . we again apply alexander duality to obtain a dual result for @xmath306 . in the situation of theorem [ cor : betti ] , we have @xmath307 for @xmath244 . we now give a new and simple proof of the generalized lower bound theorem , for simplicial polytopes and for lefschetz spheres , which includes a new characterization in terms of toric chordality ( similar to the one of kalai for the special case @xmath308 , cf . @xcite ) . [ thm : toricglbcpolytopes ] let @xmath85 denote any simplicial @xmath86-polytope . then the following are equivalent : @xmath309 for some @xmath310 . @xmath311 is toric @xmath1-chordal . @xmath85 admits a @xmath1-stacked triangulation , i.e. a triangulation without interior faces of dimension @xmath312 . \(c ) @xmath313 ( a ) is easy @xcite , ( a ) @xmath313 ( b ) follows from theorem [ thm : stress_quant ] ( where @xmath314 ) . finally , for ( b ) @xmath313(c ) , we use the fact that @xmath315 is cohen macaulay by corollary [ cor : propagation_sr ] . the rest of the proof is as in @xcite : by the work of mcmullen @xcite , @xmath316 is a geometric subcomplex of @xmath70 . by corollary [ cor : propagation_sr ] , @xmath316 is also cohen macaulay , and thus triangulates @xmath85 . [ rem : glbt ] virtually the same result holds for all proper geometric simplicial rational homology @xmath81-spheres @xmath0 in @xmath70 with the * weak lefschetz property * ( cf . @xcite ) , i.e. for which there exists a linear differential @xmath145 such that @xmath317 is , for every @xmath1 , injective or surjective . the only thing that needs amendment in theorem [ thm : toricglbcpolytopes ] is to weaken ( c ) to say that @xmath318 is merely a @xmath1-stacked triangulated ball in homology . only the implication ( b ) @xmath319 ( c ) needs elaboration , but it follows with lemma [ lem : torreschord ] : to see this , let @xmath320 and @xmath321 , where @xmath123 is any new vertex ( and naturally @xmath322 and @xmath297 are identified along the common subcomplex @xmath0 ) , which we realize in @xmath323 using generic coordinates . it follows with lemma [ lem : torreschord ] , or even only isomorphism , that @xmath297 is acyclic and @xmath324 is cohen macaulay , so that the stress space in degree @xmath325 is generated by a single element @xmath326 . now , as a cone over a rational homology sphere , @xmath327 has a perfect poincar pairing with respect to its fundamental class @xmath328 in degree @xmath86 @xcite , i.e. every stress of @xmath329 is a derivative of @xmath330 . it follows with the cone lemma that every stress of @xmath327 is a derivative of @xmath331 . but every stress of @xmath297 is supported in @xmath0 by toric chordality and propagation , so that @xmath324 has perfect poincar pairing and is therefore a rational homology sphere . we conclude that @xmath297 , as the deletion of @xmath123 from @xmath324 , is a rational homology ball . let us turn our attention to balanced polytopes . a simplicial @xmath81-complex @xmath0 is called * balanced * if there exists a `` coloring map '' @xmath332=\{1,\cdots , d\}\ ] ] that is an injection on every facet of @xmath0 . if @xmath333 is a subset of @xmath334 $ ] , we also use @xmath335 to denote the restriction of @xmath0 to a specific color set . the smallest balanced @xmath81-sphere is , quite clearly , the boundary of the crosspolytope . klee and novik @xcite conjectured a far - reaching generalized lower bound conjecture for balanced polytopes and weak lefschetz spheres . the following result of juhnke - kubitzke and murai @xcite resolves the first part of this conjecture . the proof is based on a clever induction and heavy use of the weak lefschetz property . [ lem : js ] let @xmath0 denote the boundary of a balanced simplicial @xmath86-polytope , and let @xmath333 denote any subset of @xmath334 $ ] . let @xmath336 denote a generic projection of @xmath337 to an @xmath338-dimensional linear subspace . then , for every @xmath339 , we have a surjection @xmath340 for some generic @xmath145 . as we will see , we could choose @xmath341 in this lemma , though this is not necessary . for purposes of self - containedness , we give a simple proof based on partition of unity . in its general form , it is stated as follows if @xmath0 is cohen macaulay and proper , these facts are essentially equivalent by the following `` partitioning lemma '' : [ lem : partyyyyy ] consider a proper geometric cohen macaulay @xmath81-complex @xmath0 in @xmath70 . then , for every @xmath71 , we have a surjection @xmath342 for a lemma this simple to state , it is a little tricky to prove , but it will not be needed here in this strength : we will content ourselves with a simpler version . let us start by giving an elementary version of lemma [ lem : partyyyyy ] . a simplicial complex is * pure * if all its facets are of the same dimension . a pure @xmath81-dimensional simplicial complex @xmath0 is * shellable * if there is a total order on the facets of @xmath0 such that the intersection of every facet with the union of all subsequent facets is pure @xmath343-dimensional simplicial complex , or empty . [ lem : partyyyyy2 ] consider a proper geometric shellable @xmath81-complex @xmath0 in @xmath70 . then , for every @xmath71 , we have a surjection @xmath342 we refer to @xcite for an introduction to shellability , and mention only that boundaries of simplicial polytopes are always shellable @xcite and that shellable complexes are always cohen macaulay @xcite . the proof is exceedingly simple : consider a shelling step @xmath0 to @xmath344 , consisting of the removal of a single facet @xmath24 from @xmath0 . let @xmath102 denote the minimal face of @xmath0 not in @xmath344 . it is well - known and easy to see that @xmath345 is an isomorphism in every degree except for @xmath346 , where it is a injection with a one - dimensional cokernel . any stress in @xmath347 not in @xmath348 therefore is a @xmath1-stress that contains @xmath102 in its support . let @xmath123 denote any vertex of @xmath0 not in @xmath24 , which exists unless @xmath349 . the same then applies for @xmath350 choose a stress @xmath351 representing the cokernel of this map . then every stress in @xmath347 is supported in @xmath348 modulo a scalar multiple of @xmath142 , so that the theorem follows by induction . to prove lemma [ lem : js ] , we use the same idea : [ lem : partey ] consider a shellable balanced @xmath81-complex @xmath0 with a proper embedding into @xmath211 . then we have a surjection @xmath352}).\ ] ] the proof is the same as in the case of lemma [ lem : partyyyyy ] ; we just have to observe that the shelling of @xmath0 restricts to a shelling of @xmath353}$ ] and observe the changes in stresses in every step . this extends to cohen macaulay complexes , see @xcite . it follows that every stress can be partitioned into stresses supported in links of vertices of any chosen color . if we assume now that @xmath0 is the boundary of a simplicial balanced polytope , then we in turn apply the hard lefschetz theorem to every link . the claim of lemma [ lem : js ] , and thereby the first part of the klee novik balanced generalized lower bound conjecture , follows immediately for @xmath354 . iterating the argument proves lemma [ lem : js ] in full generality . the second part of the klee novik conjecture concerned a conjecture for the case of equality , which asserts the existence of a certain nice triangulation for @xmath0 , that can be constructed in three steps : consider the boundary @xmath0 of a balanced @xmath86-polytope on color set @xmath334 $ ] , and let @xmath1 denote a non - negative integer . the construction consists of three steps consider any pair of vertices @xmath355 in @xmath0 of the same color , called * antipodes*. let @xmath356 where @xmath357 denotes the subcomplex of @xmath358 induced by faces of dimension @xmath359 , and @xmath360 denotes the suspension with apices @xmath361 . if the star @xmath362 has no @xmath100-stress , proceed no further . otherwise , let @xmath363 denote a vertex created anew ( and given color @xmath364 ) , the * center of the antipode * , and let @xmath365 we attach @xmath366 to @xmath0 using the natural embedding @xmath367 . this is repeated with every antipode in @xmath0 . consider now @xmath368 , @xmath369 two antipodes of @xmath0 in the resulting complex . if @xmath370 , or ( equivalently ) @xmath371 , then we now identify @xmath372 . repeat this with every pair of antipodes . observe that @xmath373 is a canonical balanced triangulation of the cell complex defined by klee novik @xcite in proposal of their balanced lower bound conjecture . whenever this construction is used in the context of a geometric simplicial complex , the centers of antipodes are put at the origin . for a color @xmath14 , we use @xmath375 to denote the sum of partial differentials over vertices with color @xmath14 . [ lem : balanced_prop ] let @xmath0 be the boundary of a balanced simplicial @xmath86-polytope , and assume that , for some @xmath376 , and for all @xmath377 $ ] , @xmath378 , the surjection of equation is also an injection . moreover , if @xmath155 , we have an exact sequence @xmath382})\ \longrightarrow\ { { \mathrm{s}}}_{\ell } ( { \mathcal{t } } ) \xrightarrow{\ \updelta_{d+1}\ } { { \mathrm{s}}}_{\ell-1}({\mathcal{t}})\ \longrightarrow\ 0.\ ] ] the polytopality assumption can be weakened : similar to lemma [ lem : js ] , the lemma also holds in the generality of weak lefschetz spheres ; one reason we do not adopt this level of generality here is that we want to remain as self - contained as possible , and rely on shellings heavily . that said , everything needed is a generalization of lemma [ lem : partey ] for balanced cohen macaulay complexes ( see @xcite ) . lemma [ lem : balanced_prop ] is essentially the balanced analogue of theorem [ thm : toric_propagation ] ; we do only not call it such to express a mild dissatisfaction with it . indeed , as we shall see in the proof , it is critically used that our complex @xmath0 is a shellable sphere ( or at least @xmath13-cohen macaulay ) already , which is not needed in the original theorem . indeed , only the final part of this proof could be justifiably called a balanced propagation theorem , albeit requiring somewhat special assumptions . * secondly * , note that @xmath0 has no missing @xmath21-faces , @xmath383 ( that is has different colors under @xmath384 if @xmath385 ) . this follows as in lemma [ lem : torreschord ] , but let us give a more direct argument : let us first assume that @xmath386 . assume there is a missing @xmath21-face , let @xmath9 be the @xmath387-cycle defined by its boundary . consider furthermore @xmath333 a colorset of cardinality @xmath388 that contains all but one of the colors of @xmath9 . consider finally a generic projection @xmath389 of @xmath0 to @xmath390 and @xmath123 a vertex of color @xmath391 of @xmath9 . let @xmath392 . we have a natural embedding @xmath393 , inducing an embedding of stress spaces @xmath394 we therefore obtain a commutative square @xmath395{r}{\ \upomega\ } & { { \mathrm{s}}}_{i-1}({\widetilde}{\delta}_s ) \\ { { \mathrm{s}}}_i(\operatorname{lk}_v { \widetilde}{\delta}_{s ' } ) \arrow[hook , two heads]{r}{\ \upomega\ } \arrow[hookrightarrow]{u } & { { \mathrm{s}}}_{i-1}(\operatorname{lk}_v { \widetilde}{\delta}_{s ' } ) \arrow[hookrightarrow]{u } \end{tikzcd}\ ] ] where the horizontal maps are isomorphisms . since , in a generic embedding , every face and every complete cycle is contained in the support of some nontrivial stress , it follows that @xmath396 is the boundary of a @xmath387-face in @xmath122 if and only if it bounds such a face in @xmath397 . if @xmath398 , @xmath399 , consider @xmath400 , and consider a face @xmath24 of cardinality @xmath286 in @xmath0 . then , following what we proved above , @xmath401 has no missing faces of dimension @xmath402 , so that @xmath0 has no missing face of dimension @xmath21 containing @xmath24 . as a * third step * , we progress to prove that the map @xmath403 is surjective ; we will prove this first for @xmath404 , and then use a balanced version of toric propagation to extend this to all @xmath74 . we need a lemma of the partitioning type . for simplicity , we call a pair @xmath407 * @xmath21-rigid * if @xmath408 is a surjection . we say it is * minimally @xmath21-rigid * if the map is an isomorphism . we say that it is * trivially @xmath21-rigid * if @xmath409 is trivial . [ lem : partitioning ] let @xmath123 denote any vertex of @xmath410 of color @xmath411 , let @xmath412 , and @xmath413 denote a generic projection of @xmath410 to @xmath414 . then @xmath415 is an isomorphism for @xmath416 . indeed ; this map is clearly an injection by assumption . moreover , for @xmath417 of cardinality at most one , we obtain surjectivity by lemma @xmath420 . consider now some multiindex @xmath421 and a stress @xmath142 in @xmath422 . by lemma [ lem : js ] , we have that @xmath142 has a preimage in @xmath423 , @xmath424 . since @xmath145 is injective , the preimages must coincide , and we conclude that the preimage is supported in @xmath422 . * surjectivity : * we follow kalai s argument for the chordality of polytopes with @xmath426 , and more detailedly the generalization by adiprasito nevo samper ( * ? ? ? * lemma a.1 ) . abbreviate @xmath427 , i.e. the restriction of @xmath413 to colorset @xmath333 . consider the cover of @xmath428 by @xmath429 , where @xmath123 stands for vertices of @xmath0 of color @xmath14 . in particular , the generalized mayer - vietoris principle induces a resolution of the ishida complex for the stress space of @xmath428 given by its coordinates and @xmath145 . now , to understand the total complex of the resulting double complex , we can make two simple observations computing the homology of the total complex can be done in an easy way for instance by using first the fact that the resolution induced by the cover is exact , so that we are only left with the ishida complex for @xmath428 given by its coordinates and @xmath145 easily ; its top homology group ( i.e. in degree @xmath1 ) is therefore isomorphic to @xmath430.\ ] ] on the other hand , the degree @xmath100-column of the double complex contains as a direct summand the resolution of the ishida complex for @xmath428 with respect to its coordinates only : @xmath431 whose entries vanish under the differential of the ishida complex ; the first homology of this complex is therefore injects into the top homology of the total complex . to see how this implies surjectivity , note that following brugesser mani @xcite , @xmath413 can be shelled in such a way that removes the star of @xmath123 first . it follows with lemma [ lem : partey ] that we have a surjection @xmath432 so that we write @xmath433 as a sum of @xmath100-stresses supported in @xmath434 vertices @xmath435 of color @xmath14 . now , @xmath436 is surjective . at the same time @xmath437 is surjective . hence , @xmath433 is in the image of two different @xmath1-stresses unless the decomposition of @xmath433 above is already supported in the link of @xmath123 , see figure [ fig : surj ] unless it defines a homology class in @xmath438 of the chain complex . the former , however , is excluded by the injectivity of @xmath439 . * injectivity : * the kernel of the partitioning map is spanned by @xmath100-stresses in @xmath440 for @xmath441 of color @xmath14 ; hence , we simply have to show that @xmath442 is trivially @xmath1-rigid . but we already observed that the complexes @xmath443 are @xmath1-rigid for every choice of @xmath441 of color @xmath14 ( this is isomorphism for two - element sets ) . moreover @xmath444 is minimally @xmath1-rigid ( this is isomorphism for one - element sets ) . in particular , @xmath445 are @xmath1-rigid . but then their intersection @xmath442 must be trivially @xmath1-rigid by the mayer vietoris formula for the ishida complex . we start by proving the first claim for @xmath451 . for every colorset @xmath452 $ ] of cardinality @xmath406 , @xmath453 , @xmath411 , @xmath454 and a generic projection @xmath455 of @xmath456 to @xmath457 , the previous lemma gives a surjection @xmath458 for every vertex @xmath123 of @xmath410 of color @xmath411 . by surjectivity of @xmath459 for @xmath460 , ( or even just perfectness of the poincar pairing in @xmath0 ) , we have in particular also @xmath461 for all @xmath451 . since the complexes @xmath462 and @xmath463 coincide up to their @xmath101-skeleton , we obtain a surjection @xmath464 for all @xmath451 . by partition of unity , we have a surjection @xmath465 so that it follows then that we have a surjection @xmath466 so the first claim of the lemma follows . for the characterization of the kernel , consider @xmath467 denote a generic projection of @xmath468 to @xmath414 , so that the map @xmath469 has kernel @xmath470 , so corollary [ cor : part ] for @xmath425 follows . * finally * , we get to balanced toric propagation . it follows at once as in the proof of propagation of weak toric chordality ( corollary [ cor : quant2 ] ) that holds for @xmath472 . we wish to prove that map is a surjection for all @xmath472 as well ; assume it is proven for @xmath473 . it is sufficient to show that @xmath482 is in the image of @xmath134 for every @xmath184 of color in @xmath334 $ ] . to this end , we consider in analogy to the toric propagation the commutative diagram @xmath483 where the top map has kernel @xmath484})$ ] . hence , @xmath485 gives the desired preimage of @xmath486 as in the proof of the classical propagation theorem . we therefore obtained a preimage of @xmath157 under @xmath446 in @xmath480 . since the kernel of the map is @xmath45-dimensional for @xmath489 because @xmath490\ = \ { { \mathrm{s}}}_{d+1}({\mathcal{t}}_{[d]})\ = \ 0.\ ] ] we observe with isomorphism that @xmath456 is acyclic . the same holds for links of faces on colorset @xmath334 $ ] up to dimension @xmath252 by the cone lemma and as in lemma [ lem : quant2 ] and theorem [ thm : stress_quant ] . let now @xmath491 , which we realize with generic coordinates in @xmath323 using a new vertex @xmath123 . by the observation we just made and isomorphism , for every face @xmath24 of @xmath481}$ ] of dimension at most @xmath252 , @xmath492 is generated by a single element @xmath493 . by partition of unity ( lemma [ lem : partey ] ) , we have @xmath494 and we conclude that every stress of degree at least @xmath495 is a derivative of @xmath496 . moreover , since @xmath497}^{(\le d - k)}$ ] , we have a surjection @xmath498 for all @xmath499 . but @xmath0 is a rational homology sphere @xcite , so every stress of degree at most @xmath500 in @xmath324 is a derivative of @xmath496 . hence we have a perfect poincar pairing in @xmath324 , so that @xmath324 is a ( rational ) homology sphere . removing @xmath123 yields a ( rational ) homology ball @xmath456 , the first claim follows . finally , the link of every center admits a fix - point free @xmath374 that is induced as the diagonal embedding of the @xmath374-action on antipodes of a given color ; as every individual action was transitive on the colorset , so is the diagonal embedding . hence the link of every center of an antipode must be a crosspolytope .

time - resolved , multi - epoch , _ hubble space telescope _ ultraviolet eclipse spectrophotometry is presented for the magnetic cataclysmic variables ( mcvs ) v1309 ori , mn hya , v2301 oph , and v1432 aql . separation of the eclipse light curves into specific wavebands allows the multiple emission components to be distinguished . accretion - heated photospheric spots are detected in v1309 ori and v2301 oph , indicating covering factors and.008 , and blackbody temperatures k and,000 k , respectively . the cyclotron - emitting shock on mn hya is detected in the optical / near - uv and found to occupy no more than of the stellar surface . emission from the accretion stream is a prominent component in v1309 ori and v2301 oph , and the protracted eclipse ingress of uv emission lines in the latter object indicates that the stream penetrates the white dwarf magnetosphere to a height of before it is funneled onto the magnetic poles . the uv emission - line spectrum of v1309 ori ( and to a lesser extent , mn hya ) is unusual among mcvs , with the relative strength of in v1309 orimatched only by the asynchronous system by cam . the prominence of suggests that an overabundance of nitrogen may be the most likely explanation . three epochs of observation of the asynchronous v1432 aql cover about one - third of a 50-day lap cycle between the white dwarf spin and binary orbit . the eclipse light curves vary enormously from epoch to epoch and as a function of waveband . the dereddened uv spectrum is extremely bright compared with the optical , and the overall spectral energy distribution coupled with the duration of eclipse ingress indicate that the dominant source of energy is a hot ( k ) white dwarf . this explanation also accounts for the modest brightness variation out of eclipse and the weakness of optical circular polarization due to cyclotron emission . undiminished line emission through eclipse indicates that the eclipse itself is caused by a dense portion of the accretion stream , not the secondary star . the temperature of the white dwarf in v1432 aql greatly exceeds that in any other am her system except the recent nova v1500 cyg . this , combined with its current asynchronous nature and rapid timescale for relocking , suggests that v1432 aql underwent a nova eruption in the past yr . the reversed sense of asynchronism , with the primary star currently _ spinning up _ toward synchronism , is not necessarily at odds with this scenario , if the rotation of the magnetic white dwarf can couple effectively to the ejecta during the dense wind phase of the eruption .

astro-ph0010011

at least 11 of the nearly 5 dozen known am her variables are true eclipsing binaries where light curves through eclipse can be used to analyze the structure and relative importance of emission regions on the magnetic white dwarf and in the accretion column that connects it to the mass - losing companion . the eclipsing systems are found with magnetic field strengths @xmath8 mg ( mg @xmath9 g ) , span the orbital period range @xmath10 hr , and include one asynchronous system . thus , there is reasonable assurance that an understanding gained from the eclipsing systems can be applied to the class as a whole ( see cropper 1990 for a general review ) . because the magnetic field funnels accreting gas onto impact regions as small as @xmath11 , high time resolution , @xmath12 s , plus an adequate photon rate are required to resolve the structures . the wavelength region of choice is the ultraviolet , where the white dwarf and heated impact region are prominent ( @xmath13 k and @xmath14 k , respectively ) , line and continuum emission from the funnel gas is available , and the contribution from the late - type secondary star is minimized . the first two papers in this series presented _ hst _ time - resolved spectroscopy of dp leo ( stockman et al . 1994 , hereafter paper i ) and uz for ( stockman & schmidt 1996 , hereafter paper ii ) ; the latter study also applied the method to st lmi , where the accretion footpoint undergoes `` self - eclipse '' when it rotates behind the white dwarf limb . dp leo was observed in a state of very low accretion , nevertheless the eclipse light curves revealed the contribution of an accretion - heated spot ( @xmath15 k ) against the surrounding white dwarf photosphere ( @xmath16 k ) . the fractional surface area of the near - uv spot , @xmath17 , was found to be very similar to that of the optical cyclotron source in a high state . observations of uz for showed that the accretion stream itself can be a dominant uv source in an active accretion state , owing to both strong resonance line emission and free - bound continua . the temperature of the hot spot was measured at @xmath18 k and found to occupy at least @xmath19 of the white dwarf surface area . attempts to model the line emission led to a two - phase composition for the gas stream , consisting of low filling - factor , dense clumps threading a more rarefied flow . many of the observed properties of uz for in the uv are shared by the well - studied hu aqr ( e.g. , schwope 1998 ) , and similar conclusions have been reached . in this paper we present faint object spectrograph ( fos ) spectroscopy of the final 4 eclipsing magnetic cataclysmic variables ( mcvs ) observed in this program . all four are relatively bright when actively accreting , @xmath20 , and together display an interesting variety of properties : v1309 ori is the longest - period am her system known , with an eclipse persisting for more than 40 min . given the rather modest magnetic field of @xmath21 mg , spin - orbit locking in this binary challenges current theories of the magnetic interaction mechanism . in the high state , mn hya is a classic 2-pole accretion system whose strong optical polarization enables detailed models . the weakest magnetic field measured on any am her system is found on v2301 oph , a characteristic which has frustrated polarimetrists but motivated a number of photometric and spectroscopic analyses . aql is the only one of the 4 for which an eclipse by the secondary star is in doubt ; it is also the sole asynchronous eclipsing system , made all the more unusual because the spin period appears to _ exceed _ the orbital period by 0.3% . these and other properties of the systems are summarized together with basic references to previous work in table 1 . recently , several sophisticated techniques used to study the emission of disk cvs have found their way to magnetic systems . these include doppler tomography ( e.g. , schwope et al . 1999 ) and maximum entropy methods ( mem ) for modeling eclipse light curves in broad bands ( harrop - allin , hakala , & cropper 1999b ; harrop - allin et al . 1999a ) as well as of full - orbit phase - resolved emission - line spectroscopy ( sohl & wynn 1999 ; vrielmann & schwope 1999 ; kube , gnsicke , & beuermann 2000 ) . several of these approaches utilize genetic algorithms ( e.g. , hakala 1995 ) to efficiently approach global optimization of the many model variables . each of the methods has its strengths and limitations . doppler tomography maps the velocity distribution of emitting regions , not their locations , and the technique is insensitive to motions out of the orbital plane ( see , e.g. , schwope et al . 1999 ) . mem and genetic optimization codes were developed in an attempt to deal with the tremendous ambiguity of generalized model - fitting approaches , where the number of emitting sources may exceed the number of independent data points . the various tools are in many ways complementary and together represent the state - of - the - art in analytical techniques for time - resolved studies of accretion binaries . we have chosen to apply a more qualitative approach to interpret the eclipse light curves of these mcvs for several reasons . first , the photon rate in the near - uv from _ hst_+fos can not compete with what is now commonplace from modern instrumentation and large ground - based telescopes working in the visible ( @xmath1@xmath22 m ) . this , plus the low spectral resolution of fos effectively rule out quantitative doppler studies of the emission lines . time allocation considerations on _ hst _ limited the observing program on each source to windows just encompassing eclipse , and for some targets scheduling constraints precluded obtaining both ingress and egress at the same epoch . with such limitations , the data presented here are amenable to identifying the location and relative importance of principle sources of line and continuum emission , but not for unraveling their detailed structure or kinematics . time - resolved uv spectra of v1309 ori , mn hya , v2301 oph , and v1432 aql were obtained with _ hst _ at several epochs between 1996 aug . 3 and 1996 dec . 24 ( cycle 6 , post - costar ) , as recorded in table 2 . setup of the fos was as follows : the g160l grating and 0.86 square c-1 aperture pair to provide a spectral coverage of @xmath23 , 2@xmath24 substepping for a resolution of @xmath17 ; rapid readout mode with a time resolution of 0.8201 s per spectrum ; and 5@xmath24 overstepping to eliminate gaps caused by dead diodes . data calibration was carried out as in papers i and ii , using the _ calfos _ reduction pipeline . observations of v1309 ori and v1432 aql were obtained at several epochs each : the former to adequately cover the long eclipse and the latter to sample the range of eclipse behavior shown by this asynchronous system . the eclipse interval of v2301 oph was missed in the initial observation due to a spacecraft scheduling error , and the object was revisited a month later . even then , egress was not adequately covered . mn hya was found in a very faint state in the first pointing but had brightened by an order of magnitude upon a second visit 70 days later . the low accretion state of mn hya in mid - oct . 1996 was confirmed by a contemporaneous ground - based spectrum that showed a visual magnitude @xmath25 and no obvious emission lines . this and a number of other supporting optical observations were obtained over the time period of the spacecraft observations . in the cases of v1309 ori , mn hya , and v2301 oph , the blue channel spectrograph of the mmt provided flux - calibrated spectra covering the range @xmath26 . in addition , a full orbital cycle of unfiltered photometry and circular polarimetry with a time resolution of @xmath145 s was acquired during an active phase of mn hya on 1996 jan . 27 ( 9 months prior to the initial _ hst _ visit ) with the octopol polarimeter attached to the 2.3 m bok telescope of steward observatory . a @xmath27 eclipse light curve of the same object with 128 ms time samples was obtained on 1996 mar . 10 using a programmed - readout mode of the ccd in the mmt red channel spectrograph ( schmidt , weymann , & foltz 1989 ) . for this observation , the disperser was replaced by a plane mirror and both the target and a nearby comparison star were located in the spectrograph slit . the light curve of the latter object enabled a quality eclipse profile of mn hya despite the presence of light clouds . finally , occasional spectra and circular spectropolarimetry spanning @xmath1@xmath28 were obtained with the ccd spectropolarimeter ( schmidt , stockman , & smith 1992 ) attached to the steward observatory 2.3 m telescope . the ground - based observations are also summarized in table 2 . light curves were prepared from the time - resolved uv spectroscopy in 4 wavelength bandpasses in a manner similar to the procedure described in papers i and ii . the wavebands extracted from the dispersed spectrum are : a near - uv interval ( nuv ; @xmath29 ) ; the far - uv continuum ( fuv ; @xmath30 excluding @xmath51397 ) ; and the emission lines @xmath31 + @xmath32 , constructed by summing a pair of continuum - subtracted bands covering @xmath33 and @xmath34 . in addition , a `` white - light '' curve was derived from the zero - order undispersed image . this bandpass has an effective width of @xmath11900 for these sources and central wavelength of @xmath13400 . the light curves for v1309 ori , mn hya , v2301 oph , and v1432 aql are shown as montages plotted vs. orbital phase in figures @xmath35 , respectively . an approximate absolute calibration of the 3 uv channels in units of flux per detected count per 0.82 s sample is : + : 2.0@xmath36 ergs @xmath37 s@xmath38 ; fuv : 9.3@xmath39 ergs @xmath37 s@xmath38 @xmath38 ; nuv : 7.4@xmath40 ergs @xmath37 s@xmath38 @xmath38 . note that the two continuum bands are characterized per unit wavelength . neither v1309 ori nor v2301 oph displayed discernible variations in eclipse shape over the various epochs , in the former despite a span of more than 2 months . for these objects , mean light curves are shown . mn hya was too faint in the initial visit to provide useful data so only the second data set is displayed . it should be noted that the quoted accuracies of all ephemerides ( table 1 ) are sufficient to phase the data presented here to an accuracy of @xmath41 for all but v1309 ori , where the uncertainty is @xmath42 . the mean spectrum of each of the 4 objects as it emerged from the processing pipeline showed evidence for the 2200 silicate depression due to interstellar absorption . using the extinction curve from cardelli , clayton , & mathis ( 1989 ) and @xmath43 , corrections were applied to each spectrum with an eye toward producing as smooth a continuum as possible in the region @xmath1@xmath44 . the results , quoted in table 1 , indicate visual extinctions in the range @xmath45 mag , with uncertainties of @xmath46 mag . for typical system parameters , eclipse by the secondary of a point on the white dwarf has a maximum duration of @xmath47 ( @xmath1@xmath48 min total length ) , with a weak dependence on binary period . eclipse transition across the white dwarf disk itself requires a phase interval @xmath49 ( at least 30 s ) , with the inequality reflecting the dependence on primary star mass , orbital period ( greater for shorter @xmath50 ) , and inclination ( greater for smaller @xmath51 ) . previous studies ( e.g. uz for , paper ii ) have found that the accretion column is a strong source of not only line emission , but also optical - uv continuum . because the gas stream approaches the primary in a curved , prograde path ( see , e.g. , lubow & shu 1975 ) and dynamically couples to the magnetic field en route , covering of the accretion column can commence prior to the white dwarf eclipse and continue long after . moreover , depending on the colatitude of the impact region on the star , the stream may arc out of the orbital plane and undergo only a partial eclipse . at the accretion shock , optical - ir cyclotron emission typically originates over a region of the stellar surface with @xmath52 . a somewhat larger portion of the disk @xmath53 is heated by low - level accretion or the euv - bright base of the accretion column to a characteristic temperature @xmath54 ev . for a circular profile , eclipse of this hot spot will produce an abrupt drop in the fuv flux over an interval 1 s @xmath55 s , depending on the observed waveband . the smallest spots are not resolvable at our sampling frequency ( and measured uv count rates ) . therefore , the baseline behavior that we expect through eclipse is : 1 ) initially a gradual extinction of the emission - line flux , beginning with the gas stream near the secondary and proceeding along the flow ; 2 ) a brief decline in the white light and nuv continuum bands that is well - resolved at the fos sampling rate , corresponding to the limb of the secondary beginning to pass over the white dwarf disk ; 3 ) an abrupt , possibly unresolved , drop in the fuv and to a lesser extent , the nuv bands , as the accretion - heated spot on the white dwarf is eclipsed . this transition will very nearly coincide with a similar extinction of the cyclotron flux in the white light channel . continuing through the light curve , we find 4 ) the conclusion of the resolved white dwarf eclipse ; 5 ) continued obscuration of the prograde portion of the gas stream ; 6 ) a period of nearly constant minimum light lasting up to several minutes ; and 7 ) @xmath56 11 ) an egress sequence that is a vertically - mirrored repeat of steps 1 ) @xmath56 5 ) , in order , as each component is progressively exposed . barring changes in viewing perspective or in the beaming factor of an emission component , amplitudes of corresponding ingress / egress transitions will be identical . the rosetta stone for this basic behavior is uz for ( paper ii ) . v1309 ori has the longest orbital period of the known am her binaries at 7.98 hr . the uv ( figure 5 ) and optical spectra ( garnavich et al . 1994 ; shafter et al . 1995 ) are dominated by continuum and line emission from the accretion stream . buckley & shafter ( 1995 ) report the detection of weak , variable circular polarization in a white - light optical bandpass . spectropolarimetric observations of the system obtained by us in 1994 nov . ( not shown ) confirm a low level of polarization for the system ( @xmath57 ) as well as the existence of cyclotron emission harmonics centered near @xmath58 in both total flux and polarization . identifying these with harmonic numbers @xmath59 , respectively , we substantiate previous estimates of the primary magnetic field strength near 61 mg ( garnavich et al . 1994 ; shafter et al . 1995 ; harrop - allin et al . the polarization appears to vary synchronously on the orbital period and is probably strongly diluted by light from the accretion stream . it is remarkable that such a wide binary can maintain synchronism given the implied accretion rate and modest magnetic field ( see also frank , lasota , & chanmugam 1995 and harrop - allin et al . 1997 ) . nevertheless , the 1996 uv light curves shown in figure 1 are so similar to published visible - light profiles from @xmath60 ( e.g. shafter et al . 1995 ) , and so unlike the variable light curves exhibited by asynchronous systems ( e.g. , 3.5 ) , that we conclude the spin and orbital motions are indeed locked . interestingly , v1309 ori displays an extremely peculiar x - ray behavior ( walter , scott , & adams 1995 ; de martino et al . 1998 ) , accreting in bursts as short as a few seconds duration whose local @xmath61 values fall in the domain of buried accretion shocks ( beuermann & woelk 1996 ) , and which cool primarily in the soft x - rays ( e.g. , szkody 1999 ) . this extreme flickering does not appear in the uv light curves recorded by fos . the v1309 ori eclipse light curves shown in figure 1 and at an expanded scale in figure 6 confirm the dominance of the accretion stream in the uv . apart from inflections in the continuum bands at the beginnings of ingress and egress , all four channels display protracted transitions ( @xmath62 ) . the lack of sharp transitions in the white - light continuum places an upper bound on the cyclotron emission in this band ( @xmath635% ) . the inflections themselves are of approximately the correct duration ( @xmath64 ) to ascribe to an eclipse of the white dwarf , but this explanation requires a hot photosphere , @xmath65 k. the interpretation is also complicated by the simultaneous egress of the line - emission region at @xmath66 as is evidenced by the emission - line channel . on the other hand , a sharp drop in the fuv channel at @xmath67 is strong evidence for the presence of an accretion - heated spot on the white dwarf . the data are sufficiently noisy that we show this light curve in figure 6 after smoothing by a running mean of width 10 samples ( 8 s ) . the ingress time for the spot as measured from the two contacts @xmath68 indicated in the figure is @xmath69 s , implying that the transition is unresolved in the smoothed data . we therefore can set only an upper limit to the azimuthal extension of the spot as @xmath70@xmath71 cm . the implied covering factor for a circular spot on a 0.6 @xmath72 white dwarf is @xmath73 , possibly larger if the star is over - massive . the reduction in fuv flux corresponding to this drop is @xmath12.2 counts sample@xmath38 , or a dereddened flux of @xmath74 ergs @xmath37 s@xmath38 @xmath38 at an effective wavelength of @xmath11400 . for a distance @xmath75 pc , the corresponding blackbody temperature is @xmath2 k ( @xmath76 ev ) . for v1309 ori , removal of the 2200 feature was accomplished with a minor amount of absorption , @xmath77 mag , despite a distance of at least 500 pc . the dereddened spectra are shown in figure 5 and characterized by the line strength measurements compiled in table 3 . out of eclipse , v1309 ori shows the high excitation line @xmath51370 as well as extraordinarily strong @xmath51240 . the lines persist weakly through eclipse ( @xmath78 ) , attesting to a small , continuously - exposed portion of the funnel or corona on the secondary star . the remarkable strength of @xmath79 has already been noted by szkody & silber ( 1996 ) from an _ iue _ spectrum . the ratio @xmath79/@xmath80 is exceeded among known mcvs only by by cam at one epoch ( bonnet - bidaud & mouchet 1987 ; shrader et al . 1988 ) ; indeed among mcvs in general this ratio rarely approaches unity . with @xmath130 ev separating the ionization stages , the very large ratio implies at minimum a hot ionizing continuum and high density conditions . however , more detailed analysis by bonnet - bidaud & mouchet found that the line ratios for by cam were so extreme that a nitrogen enhancement was indicated . proposed enrichment mechanisms included accretion from a companion whose hydrogen - rich layers had already been stripped , or mixing of normal stellar material with nuclear - processed matter : either nova ejecta or possibly surface material of the white dwarf . for stripping of an initially main - sequence secondary , the orbital period of v1309 ori would require that well over a solar mass of material has been shed by the secondary over the lifetime of the binary a very unlikely scenario . the fact that the / , / line ratios place v1309 ori precisely atop the tight grouping of 4 of the 5 measurements for by cam ( bonnet - bidaud & mouchet 1987 ) would seem to argue against a mechanism that involves mixing of unprocessed and processed material either from a previous nova eruption or from within the secondary star . nevertheless , a nitrogen abundance anomaly would seem to be the best explanation since @xmath51718 is also overly strong in both systems relative to the other atomic species ( figure 5 and table 3 ) . the interpretation of mn hya is relatively straightforward . except for the eclipse , circular polarization varies smoothly from @xmath5620% to + 10% through the 3.39 hr period ( buckley et al . our orbit of polarimetry displayed in figure 7 exhibits this behavior at a time resolution of 45 s ( @xmath81 ) . the presence of both senses of polarization during the orbital cycle indicates a 2-pole accretion system . ramsay & wheatley ( 1998 ; cf . buckley et al . 1998b ) have proposed a model in which a dominant , positively - polarized pole is hidden behind the stellar limb except during the interval @xmath82 . a weaker , negatively - polarized pole is visible between @xmath83 and responsible for the strong polarization around the time of eclipse . though a portion of our optical light curve is complicated by the passage of clouds , the polarization curve is unaffected and consistent with this interpretation . furthermore , the @xmath84-wave in circular polarization in the interval @xmath85 can be explained in this model as the successive disappearance of the negative pole and appearance of the positive pole over the white dwarf limb , with the double sign change due to viewing each of the slightly elevated shocks `` from below '' ( ref . the polarization undershoot in vv pup ; liebert & stockman 1979 ) . a flux minimum at this time is consistent with the grazing viewing aspect of both accretion regions . our single epoch of fos spectroscopy of mn hya presented in figure 2 shows sharp , well - defined transitions in the white - light and nuv bands , indicating a compact cyclotron - emitting shock . very little flux is detected in the two high - frequency channels . the light curves closely resemble optical eclipse profiles measured by buckley et al . ( 1998a ) but are quite unlike the x - ray light curve ( buckley et al . 1998b ) , where a broadened eclipse and hardening of the spectrum indicates absorption by the intervening accretion stream . in figure 8 the fos white - light and nuv curves are compared to an eclipse profile obtained on 1996 mar . 10 with the mmt using a ccd continuously clocked at a rate of 7.8 rows s@xmath38 . in the latter curve , the effective time resolution is set by the fwhm of the seeing profile at @xmath86 s , similar to the fos data . ingress and egress of the bright spot are measured from the optical light curves as @xmath87 s , and @xmath88 s , i.e. essentially unresolved . the implied linear dimension on the white dwarf is @xmath63@xmath89 cm ( @xmath90 ) . a non - zero flux and concave profile through minimum light is common to at least the white - light and nuv channels . the uv spectrum from the eclipse interval , @xmath91 , shown in figure 9 , exhibits strong emission lines superposed on a weak , rather red , continuum . clearly , a small portion of the stream remains exposed even at mid - eclipse . the extinction correction for mn hya was based on the out - of - eclipse spectrum and amounts to only @xmath92 mag . it is important to note that , like v1309 ori , mn hya is characterized by an unusually strong @xmath79 line , with the flux ratio @xmath79/ @xmath93 . once again , @xmath51718 appears to be present . when gauged according to the / , / ratios , mn hya falls in the gap between the extreme objects by cam and v1309 ori , and the remaining 9 mcvs available to bonnet - bidaud & mouchet ( 1987 ) , all of which have @xmath79/ @xmath94 , @xmath79/ @xmath95 . to that latter group we can add uz for and st lmi ( paper ii ) , and v2301 oph and v1432 aql ( szkody & silber 1996 ; friedrich , staubert , & la dous 1996a ; this study table 3 ) . thus , the emission - line spectra of by cam , v1309 ori , and to a lesser extent , mn hya , remain anomalous . with the lowest mean surface field yet measured for any am her system ( @xmath96 mg ) , v2301 oph is probably synchronized only because it is also very compact ( @xmath97 hr ) . the suggestion that the binary may house a small accretion disk ( silber et al . 1994 ) has not found support in subsequent studies , all of which indicate a system accreting solely via a stream ( barwig , ritter , & brnbantner 1994 ; ferrario et al . 1995 ; hessman et al . 1997 ; simi et al . 1998 ; steiman - cameron & imamura 1999 ) . with such a weak field , circular polarization has eluded detection , and thus far there is no indication of the orientation of the white dwarf within the binary . the uv light curves shown in figure 3 record one clean ingress but only the latter portion of egress prior to the center of minimum light , as has been found in other studies . ] . in the expanded plot shown as figure 10 , ingress in the optical and nuv is seen to be composed of a well - resolved ledge beginning at @xmath98 and a final decline to minimum light at @xmath99 . from nearly coincident optical and x - ray photometry , steiman - cameron & imamura ( 1999 ) showed that the x - ray eclipse is complete , with ingress occurring during the optical ledge and x - ray egress near @xmath100 . the totality of eclipse at short wavelengths is confirmed by our nuv , fuv , and light curves . in addition , the dereddened ( @xmath101 mag ) eclipse spectrum shown in figure 11 exhibits neither significant continuum nor line emission . the optical decline that extends from @xmath102 to 0.993 implies a source much larger than the white dwarf , and was interpreted by steiman - cameron & imamura ( 1999 ) to be the extinguishing of the prograde portion of the accretion stream . the idea is verified by the similarity to our light curve over this phase interval . we return to this point below . in figure 10 we identify contact points @xmath103 and @xmath104 with ingress and egress of the ledge in the nuv , and measure a time interval @xmath105 s or @xmath106 . this is significantly shorter than the @xmath150 s duration measured in a broad optical bandpass by steiman - cameron & imamura ( 1999 ) , but is still comparable in size to the disk of a white dwarf ( @xmath1@xmath107 cm ) , and we follow their assignment as such . within this interval , points @xmath108 and @xmath109 mark the contacts of a precipitous drop in the fuv flux which is strong evidence for an accretion - heated spot on the stellar surface . the ingress interval of @xmath110 s implies a linear distance of @xmath1@xmath111 cm , or a covering factor for a circular spot of @xmath112 . the amplitude of the drop , @xmath14.4 counts sample@xmath38 , corresponds to a decline in ( dereddened ) flux of @xmath113 ergs @xmath37 s@xmath38 @xmath38 . for @xmath114 pc ( silber et al . 1994 ) , the spot brightness temperature is @xmath115 k. out of eclipse , the _ hst _ light curves covering @xmath116 show a nearly constant flux level in the nuv and white - light bands , a slow decline in the fuv , and steady rise in line emission . these features can be explained by a geometry in which the dominant pole is receding toward the stellar limb during the eclipse interval , causing a reduction in fuv flux due to foreshortening , but a rise in the line emission as the roughly radial , optically - thick funnel presents an increasing cross - section to our view . the implication is that the orientation of the magnetic axis of the white dwarf within the binary is more closely orthogonal to than it is aligned with the stellar line of centers . the protracted ingress of the accretion stream , which extends nearly to phase zero or some 140 s beyond eclipse of the white dwarf , is a valuable diagnostic of the extent that the accretion stream overshoots in azimuth the white dwarf . such a path is of course expected for gas traveling in a purely ballistic trajectory ( lubow & shu 1975 ) . the simultaneous and rather abrupt fall to zero light in all bands indicates that the final ingress truly defines an `` edge '' to the gas , as opposed to a gradual thinning out ( cf . the eclipse profiles of v1309 ori in figures 1 and 6 ) . the implied extension of the emitting gas above the white dwarf is @xmath117 cm , or @xmath19 lubow & shu characterize stream paths for gas unaffected by magnetic fields according to the point of closest approach from the accreting star , @xmath118 , and the radius of the disk that would ensue , @xmath119 . for a non - disk system , we would expect that the maximum projected altitude of the stream above the white dwarf at the time of eclipse would lie intermediate between these characteristic heights . using parameters estimated for v2301 oph(@xmath120 , @xmath121 ; barwig et al . 1994 ) , we find @xmath122 cm and @xmath123 cm , which indeed bracket the measured extent above . it should be noted that because of the dynamical nature of the stream / magnetosphere interaction , this height might change with time , e.g. with varying accretion rate ( cf . the variable , @xmath124 height measured by steiman - cameron & imamura 1999 ) . variations in the shape of stream egress in the interval @xmath125 can be appreciated by an intercomparison of the curves of steiman - cameron & imamura ( 1999 ) and to the white - curve in figure 3 . asynchronous am her systems have proven to be interesting puzzles . the prototype , v1500 cyg ( stockman , schmidt , & lamb 1988 ) , was unusually straightforward because it had already accumulated a rich photometric record due to its 1975 nova outburst ( e.g. , patterson 1979 ) and the system accretes continuously onto both magnetic poles . among the subsequent 3 discoveries5840=euvej2115@xmath56586 ] , it has been more common to find periods that are sidebands and/or harmonics of the underlying @xmath126 and @xmath127 due to pole - switching and migration of the accretion footpoints through the lap period ( see ferrario & wickramasinghe 1999 and mouchet , bonnet - bidaud , & de martino 1999 for recent discussions ) . the bright and highly - polarized by cam , for example , required study over a period of several years , including an intense photometric campaign , to clearly isolate the spin period of the white dwarf ( silber et al . 1992 , 1997 ) . the presence of an eclipsing system among the known asynchronous mcvs is a particular bonus because successive light curves portray an advancing configuration between the stream and stellar magnetosphere . an extensive observational program could in principle track not only the accretion footpoint(s ) across the white dwarf but also the gas trajectory which gives rise to those variations . such information would be invaluable in forming a general understanding of the processes which couple the accreting gas to the magnetic field . v1432 aql thus occupies a position of unique importance for all mcvs . because it is eclipsing , the binary period of v1432 aql is unequivocal at 3.365 hr . however , controversy immediately arises over the origin of the eclipses , and they have been more commonly labeled `` dips '' to reflect this uncertainty . watson et al . ( 1995 ) noted phase jitter as well as an offset between the optical / x - ray dips and the velocity zero - crossing of the emission lines in proposing that the dips are due to obscuration by a high - density accretion stream . an extensive photometric campaign by patterson et al . ( 1995 ) led to the identification of a superposed periodicity at 3.375 hr , taken to be the spin period of the white dwarf and implying a lap period of 50 d. however , the secondary star was preferred as the eclipsing object . several ensuing studies ( e.g. , friedrich et al . 1996b ; staubert et al . 1995 ; geckeler & staubert 1997 ) have supported and extended the duality of periods , but have returned to an eclipse by the stream ( but see geckeler & staubert 1999 ) . recently , even the period identification was called into question after another variation at @xmath1@xmath128 the orbital period was noted in archival x - ray data ( mukai 1998 ) . the defining measurement detection of phase - modulated circular polarization consists only of unpublished data ( from buckley , quoted by watson 1995 ) which appears to support the period assignment of patterson et al . ( 1995 ) . four epochs of fos spectroscopy were attempted of v1432 aql ; unfortunately the final observation was lost due to an on - board tape recorder failure . overall , the data span @xmath1@xmath128 of a 50 d lap cycle . as shown in figure 4 , dip ingress and egress were each recorded on only one occasion . the light curves vary enormously among themselves and are quite unlike those of the other eclipsing systems . at all epochs a downward inflection is apparent near @xmath129 , at approximately the phase of optical ingress ( e.g. , watson et al . 1995 ) . however , the duration of this transition varies with epoch and fos waveband from @xmath130 ( 360 s ) for all continuum bands on 19 aug . 1996 to @xmath64 ( 60 s ) for the fuv approximately 2 weeks later . the dips approach totality only in the fuv , and even here significant signal remains at minimum light for all but the 3 sep . observation . the phase of optical egress , @xmath66 , is barely covered on 29 aug . , but there is no indication of a sharp inflection in any of the spacecraft bands at this time ; indeed the data are consistent with a smooth rise in count rate from the brief minimum around @xmath131 . by contrast , the emission - line channel shows no evidence of dips at all , with only a gradual , ill - defined fading through the observation of 19 aug . the data from this epoch were also searched for spectral shifts in the centroid and shape of @xmath51549 during the `` ingress '' period @xmath132 , such as might occur from the eclipse of a disk . no such variations were detected to a limit of about one diode ( @xmath1600 km s@xmath38 ) . the top panel of figure 12 presents the pipeline - processed spectrum of v1432 aqlaveraged over all epochs . the plot basically resembles the single _ iue _ spectrum obtained by friedrich et al . ( 1996a ) . however , two new features are evident in the much higher quality fos data : a prominent 2200 depression , and the presence of ly@xmath133 in absorption . the interstellar absorption feature can be erased cleanly for a visual extinction @xmath134 mag . eclipse ( @xmath135 ) and post - eclipse ( @xmath136 ) spectra corrected for this effect are shown in the bottom panel of figure 12 , both taken from the run on 29 aug . 1996 , and line strengths computed from the mean dereddened spectrum are listed in table 3 . a total extinction of @xmath137 mag is by far the largest found among the objects studied here , but in good agreement with a value of 0.6 mag derived from the burstein & heiles ( 1982 ) maps for @xmath138 , @xmath139 . the implied neutral hydrogen column is @xmath140 @xmath37 , and the corresponding equivalent width ( ew ) of damped ly@xmath133 is @xmath125 ( wolfe et al . 1986 ) . thus , the interstellar medium is an important contributor to the deep absorption feature we detect . we have added a sequence of panels to the light curves in figure 4 depicting the ew of the ly@xmath133 line . in these plots , ew @xmath141 corresponds to an absorption line , ew @xmath142 indicates emission , and the measurements have been smoothed with a running - mean of width 40 s ( @xmath143 ) . this index measures the net strength of the feature , since these low resolution data do not resolve the absorption component from any emission line due to the gas stream . note that the feature shows roughly constant ew except for phase intervals when the fuv flux is near zero i.e. @xmath144 during the deep dips of 29 aug . and 3 sep . the dereddened out - of - eclipse spectrum in figure 12 exhibits a continuum that rises smoothly to the blue , somewhat flatter than rayleigh - jeans . when compared to blackbodies , a temperature of @xmath145 k is found to reproduce the uv slope and not exceed the dereddened optical continuum of the only published spectrum of v1432 aql that is provided in absolute units ( patterson et al . 1995 ) . to compute the size of the object with this effective temperature , the distance estimate of @xmath1230 pc ( watson et al . 1995 ) must be reduced by 16% to @xmath1200 pc to account for the implied 0.3 mag of @xmath146-band extinction , where detection of the m4 v secondary was made . for a circular object , the uv - optical source is then estimated to have a radius @xmath147 cm . if white dwarf model atmospheres are used instead ( bergeron 1994 ) , a somewhat lower temperature , @xmath148 k , and larger size , @xmath149 cm , ensue . the model atmospheres also predict a ly@xmath133 absorption feature with an ew of 30 or more which would blend with the interstellar component at this resolution . the similarity of the derived source size to the disk of a white dwarf is very attractive , and at face value suggests a low - mass star ( @xmath150 ) . however , there is significant latitude in the combined uncertainties of the extinction correction and distance estimate . moreover , the white dwarf need not supply the entire observed continuum in either the uv or optical ( note the residual flux level in the eclipse spectrum of figure 12 ) . thus , a smaller stellar size ( larger mass ) is possible . a somewhat smaller size , @xmath151 cm , is also indicated by the 40 s duration of the most abrupt fuv eclipse ingress on 3 sep . 1996 ( figure 4 ) . the corresponding luminosity for emission by an entire white dwarf with @xmath152 k is @xmath153 , of which @xmath12% is intercepted by the secondary star . this illumination amounts to at most 50% of the @xmath154 nuclear luminosity of an m4 dwarf ( cf . v1500 cyg schmidt & stockman 1995 ) , so the brightness modulation at the orbital period due to reprocessing would be expected to be modest . indeed , the average light curve of patterson et al . ( 1995 ) shows a @xmath10.5 mag variation away from the dip feature , peaking near @xmath155 , as is appropriate for an origin on the inner hemisphere of the companion ( see the discussion of this point by patterson et al . ) . a contribution from the white dwarf and illuminated secondary is also consistent with the weak level of optical circular polarization measured for v1432 aql ( @xmath156% , friedrich et al . 1996b ; watson 1995 ) . because a hot white dwarf can account for the basic characteristics of the uv - optical spectral energy distribution and is consistent with the mean optical light curve and polarization , we adopt this explanation throughout the remainder of the paper . with the observed energy distribution from the uv through the optical dominated by the hot white dwarf , we now investigate the origin of the photometric dips . discussion on this topic has centered around the regularity of the features , their depth in various wavelength regions , the duration of ingress / egress , and their phasing relative to other indicators of orbital motion . secondary star proponents point out the very predictable dip ephemeris and phasing relative to the emission - line radial velocities ( patterson et al . gas stream advocates , on the other hand , note an energy dependence of eclipse depth in x - rays and the ingress / egress intervals , which exceed those expected for the sharp - edged eclipse of a white dwarf ( watson et al . 1995 ; friedrich et al . 1996b ; geckeler & staubert 1997 ) . new information on this question is provided by the fos spectroscopy . first , the dips in figure 4 are seen to be highly variable in both _ shape _ and _ depth _ from epoch to epoch . note in particular the absence of a defined ingress on 19 aug . , the lack of an expected egress on 30 aug . , and the variable depth at minimum light . such behavior is natural for absorption due to a dense , ill - defined gas stream but at odds with a stellar eclipse . this is particularly true for the fuv channel , which from our prior conclusions is dominated by the hot white dwarf . indeed , based on our results for uz for ( paper ii ) and the other systems studied here , it is difficult to imagine any source of appreciable light in the @xmath1@xmath157 region that would not be located on or very near the white dwarf . second , the uv emission lines are essentially _ undiminished _ through the absorption dips . this fact has already been noted for the optical lines by watson et al . ( 1995 ) and contested by patterson et al . ( 1995 ; see their figures 12 and 13 ) . the issue revolves around the various line components and their places of origin . in the case of the high - ionization lines of and , the lack of eclipse features in figure 4 is unequivocal . we also point out that if the white dwarf were being eclipsed by a secondary which fills its roche lobe , the observed duration of the dip ( 820 s or @xmath158 ) requires an inclination @xmath159 . in this case , the eclipse track crosses behind the secondary disk @xmath1@xmath160 of the way from the center . the remainder of the disk would be expected to obscure line - emitting gas in the lower accretion streams(s ) out to @xmath17 @xmath4 above the orbital plane for a period of up to @xmath1800 s. no such reduction in emission - line light is evident . we therefore conclude that the absorption dips in v1432 aql are due to the intervention of a dense portion of the accretion flow . in the context of emission by a white dwarf disk and eclipse by a stream , the curious `` choppy '' appearance to the light curves _ at all orbital phases _ ( watson et al . 1995 ; patterson et al . 1995 ) is suggestive of persistent obscuration by gas clumps , such as might occur in a circulating accretion `` ring '' . a similar structure is inferred to exist in the asynchronous mcv v1500 cyg from its ever - present sinusoidal polarization curve that implies 2-pole accretion ( schmidt , liebert , & stockman 1995 ) . we also note that the lack of an identifiable eclipse by the secondary implies that the inclination of v1432 aqlis @xmath161 . a very rough estimate can also be gained from the observed radial velocity half - amplitude of the narrow emission lines ( 160 km s@xmath38 ; patterson et al . 1995 ) , which for reasonable white dwarf and secondary star masses ( @xmath10.6 and 0.3 @xmath72 , respectively ) suggests @xmath162 . the link between the nova event and asynchronism was established with the discovery that the 1975 nova system v1500 cyg is an mcv ( stockman et al . 1988 ) . in support of this position , patterson et al . ( 1995 ) has pointed out the vanishing likelihood that all 4 asynchronous systems would be synchronizing for the first time , given that the spin - orbit period difference is universally @xmath11% and that , when measured , locking timescales are all in the range @xmath163 yr ( schmidt et al . 1995 ; mason et al . 1995 ; geckeler & staubert 1997 ) . the likely alternative is that most , if not all , of the asynchronous am her systems were previously phase - locked and that the magnetic connections were broken in recent nova events . only in the case of v1500 cyg is the post - nova classification certain . the object has an extremely rich observational history post-1975 , the white dwarf temperature has been measured at @xmath164 k and decreasing , and the object lies at the center of an expanding debris shell . none of the other asynchronous systems have been associated with a nebula , despite searches . the likelihood that the aforementioned abundance anomalies of by cam ( 3.2 . ) are indicative of mixing with nova - processed material is reduced now that the synchronized systems v1309 ori and mn hya have been found to show elevated line strengths . a nova event in the recent past of the 8 hr - period v1309 ori appears particularly improbable , as it should be especially easy to disrupt and slow to resynchronize . excluding v1500 cyg , the temperature of @xmath135,000 k that we have measured for the white dwarf in v1432 aql is far and away the hottest yet found on an am her system ( sion 1999 ; gnsicke 1999 ) . when all cvs are considered , white dwarfs this hot have been detected _ only _ in dwarf nova and nova - like classes ( sion 1999 ) , both of which are thought to be eruptive . it is therefore tempting to classify v1432 aql as the second post - nova am her system . using the cooling calculations of prialnik ( 1986 ) as a guide , the current stellar luminosity of @xmath165 for v1432 aql vs. @xmath15 @xmath166 for v1500 cyg ( schmidt et al . 1995 ) suggests that the interval since the last eruption of v1432 aql is in the range @xmath167 yr . at a distance @xmath168 pc and an expansion velocity @xmath169 km s@xmath38 , the shell could now be several arcminutes across . because the surface brightness of a nebular remnant scales as @xmath170 ( cohen 1988 ) , direct imaging of it might be very difficult ( see also wade 1990 ) . we note that the faintness and lack of a hot blackbody component in the near - uv spectrum of by cam ( bonnet - bidaud & mouchet 1987 ; shrader et al . 1988 ) suggest that , if it is also a post - nova system , it is considerably older still and thus detection of a shell would be hopeless . the fly in the ointment for a post - nova explanation of v1432 aql is the fact that in this case @xmath171 . the mechanism outlined by stockman et al . ( 1988 ) for decoupling v1500 cyg involved the transfer of a small amount of orbital angular momentum to the bloated white dwarf during the common - envelope phase of the nova . when the photosphere contracted back within the binary and eventually to a normal stellar radius , the two motions could be distinguished observationally , and @xmath172 . to achieve the v1432 aqlconfiguration from an outburst in an initially synchronized binary requires either decreasing the orbital period of the binary ( such that the secondary spirals inward ) or increasing the spin period of the primary . shara et al . ( 1986 ) has shown that the net effects of frictional angular momentum loss on a secondary orbiting within the envelope results in a decrease in the binary dimension @xmath173 only for slow novae with very low - mass secondaries ( @xmath174 ) . of course , for v1432 aql , it is not sufficient merely that @xmath175 , but that the resulting decrease in @xmath127 exceed the decrease in @xmath126 due to spin - up of the envelope ( the v1500 cyg effect ) . thus , an explanation of the sense of asynchronism of v1432 aql purely in terms of dynamical effects is not likely . more promising is rotational braking of the white dwarf during the mass - loss phase via the strong magnetic field ( the magnetic propeller effect ) . this problem has been presented in many ways for single stars as well as members of binary systems . we choose the convenient formulation of kawaler ( 1988 ) , which can be expressed in terms of the fractional change in rotational angular momentum @xmath176 here , @xmath177 is the total mass lost in the eruption , @xmath178 ( radial ) is the alfvn radius out to which corotation is maintained for a radial magnetic field line distribution , @xmath179 is an index that denotes the departure from a radial field geometry , and we have approximated the moment of inertia of the white dwarf by that of a uniform - density sphere . a value @xmath180 denotes radial field lines , while @xmath181 is appropriate for a dipole . the expression for @xmath178 provided by kawaler ( 1988 ) can be rewritten using parameters appropriate for a 0.6 @xmath72 white dwarf as @xmath182 with @xmath183 being the surface magnetic field in gauss and @xmath184 in gm s@xmath38 . assuming @xmath185 g and continuous mass loss totalling @xmath186 over a wind phase of @xmath187 s ( macdonald , fujimoto , & truran 1985 ) , we find a decoupling radius for radial field lines @xmath188 , well inside the orbit of the secondary . the total angular momentum lost is then @xmath189 , with the two limits signifying dipolar and radial geometries , respectively . the consistency with the 0.3% period difference currently observed for v1432 aql is encouraging . the above derivation is admittedly crude . it serves , however , to illustrate the possible importance of magnetic braking during the outburst . in view of this result , it might be better to view v1500 cyg as an example , rather than a template , of a nova eruption in an mcv , and given that the net outcome is the result of a competition between at least two effects with magnitudes of a few percent each , a system @xmath171 might also be expected . in fact , with @xmath178 being most sensitive to the magnetic field strength and duration of the wind phase ( @xmath190 ) , a very fast nova like v1500 cyg could experience spin - up , while a higher - field , slow nova might be spun down . time - resolved _ hst _ uv spectroscopy of four eclipsing mcvs has enabled us to identify : * accretion - heated photospheric spots in v1309 ori and v2301 oph with covering factors @xmath0 and @xmath10.008 , and blackbody temperatures @xmath2 k and @xmath190,000 k , respectively . * a cyclotron shock on mn hya with @xmath191 . * prominent emission from the accretion stream in v1309 ori and v2301 oph . a protracted emission - line ingress in the latter object indicates the stream penetrates the magnetosphere to a height of @xmath19 @xmath4 before it is funneled onto the magnetic poles . * unusually strong @xmath51240 in v1309 ori and , to a lesser extent , mn hya , that is suggestive of an overabundance of nitrogen in the gas stream . * a spectral energy distribution for v1432 aql that strongly suggests the dominant source of energy is a hot ( @xmath6 k ) white dwarf . the undiminished line emission through eclipse indicates that the eclipse is caused by a dense portion of the accretion stream , not the secondary star . the very hot white dwarf , combined with the current asynchronous nature and rapid timescale for relocking , suggests that v1432 aql underwent a nova eruption in the past @xmath1@xmath7 yr . a reversed sense of asynchronism , with the primary star currently spinning up toward synchronism , is not at odds with this history , if the white dwarf rotation can magnetically couple to the ejecta during the dense wind phase of the eruption . time - tagged uv spectrophotometry is an important tool for understanding the sizes and relative contributions of the accretion stream , hot spot , and white dwarf photospheres in mcvs . both the space telescope imaging spectrograph ( stis ) and the cosmic origins spectrograph ( cos ) offer sensitivities an order of magnitude greater than the fos . unfortunately , with the limited aperture of _ hst _ , it will be difficult to resolve objects smaller than 10% of the white dwarf radius . future uv telescopes , with apertures exceeding 4 m and high efficiency , energy - resolving detectors will be ideal facilities for discerning details of structures as small as the accretion shock itself . special thanks go to carolyn pointek for assistance with the initial data analysis and to r. wagner for spearheading the development of high - speed ccd photometry at the mmt and for lending assistance at the telescope . bergeron kindly provided da model atmosphere spectra . support was provided through nasa grant go-6498 from the space telescope science institute , which is operated by the association of universities for research in astronomy , inc . , under nasa contract nas 5 - 26555 . ground - based studies of magnetic stars and stellar systems are funded by nsf grant ast 97 - 30792 to gds . de martino , d. , barcaroli , r. , matt , g. , mouchet , m. , belloni , t. ; beuermann , k. , chiappetti , l. , done , c. , gaensicke , b.t . , la franca , f. , & mukai , k. 1998 , , 332 , 904 ferrario , l. , & wickramasinghe , d.t . 1999 , , 309 517 ferrario , l. , wickramasinghe , d. , bailey , j. , & buckley , d. 1995 , , 273 , 17 friedrich , s. , staubert , r. , & la dous , c. 1996a , , 315 , 411 friedrich , s. , staubert , r. , lamer , g. , knig , m. , geckeler , r. , bssgen , m. , kollatschny , w. , streicher , r. , james , s.d . , & sood , r.k . 1996b , , 306 , 860 gnsicke , b.t . 1999 , in asp conf . 157 , annapolis workshop on magnetic cataclysmic variables , ed . c. hellier & k. mukai ( san francisco : asp ) , 261 hakala , p.j . 1995 , , 296 , 164 harrop - allin , m.k . , cropper , m. , hakala , p.j . , hellier , c. , & ramseyer , t. 1999a , , 308 , 807 harrop - allin , m.k . , cropper , m. , potter , s.b . , dhillon , v.s . , & howell , s.b . 1997 , , 288 , 1033 mason , p.a . , andronov , i.l , kolesnikov , s.v . , pavlenko , e.p . , & shakovskoy , n.m . 1995 , in asp conf . 85 , cape workshop on magnetic cataclysmic variables , ed . buckley & b. warner , ( san francisco : asp ) , 496 schwope , a.d . , schwarz , r. , staude , a. , heerlein , c. , horne , k. , & steeghs , d. 1999 , in asp conf . 157 , annapolis workshop on magnetic cataclysmic variables , ed . c. hellier & k. mukai ( san francisco : asp ) , 71 silber , a.d . , szkody , p. , hoard , d.w . , hammergren , m. , morgan , j. , fierce , e. , olsen , k. , mason , p.a . , rolleston , r. , ruotsalainen , r. , pavlenko , e.p . , shakhovskoy , n.m . , shugarov , s. , andronov , i.l . , kolesnikov , s.v . , naylor , t. , & schmidt , e. 1997 , , 290 , 25 simi , d. , barwig , h. , bobinger , a. , mantel , k .- h . , & wolf , s. 1998 , , 329 , 115 staubert , r. , friedrich , s. , lamer , g. , knig , m. , & geckeler , r. 1995 , in asp conf . 85 , cape workshop on magnetic cataclysmic variables , ed . buckley & b. warner , ( san francisco : asp ) , 264 stockman , h.s . , & schmidt , g.d . 1996 , , 468 , 883 ( paper ii ) stockman , h.s . , schmidt , g.d . , & lamb , d.q . 1988 , , 332 , 282 stockman , h.s . , schmidt , g.d . , liebert , j. , & holberg , j.b . 1994 , , 430 , 323 ( paper i ) szkody , p. 1999 , in asp conf . 157 , annapolis workshop on magnetic cataclysmic variables , ed . c. hellier & k. mukai ( san francisco : asp ) , 195 szkody , p. , & silber , a. 1996 , , 112 , 289 vrielmann , s. , & schwope , a.d . 1999 , in asp conf . 157 , annapolis workshop on magnetic cataclysmic variables , ed . c. hellier & k. mukai ( san francisco : asp ) , 93 walter , f.m . , scott , j.w . & adams , n.r . 1995 , , 440 , 834 watson , m.g . 1995 , in asp conf . 85 , cape workshop on magnetic cataclysmic variables , ed . buckley & b. warner , ( san francisco : asp ) , 259 watson , m.g . , rosen , s.r . , odonoghue , d. , buckley , d.a.h . , warner , b. , hellier , c. , ramseyer , t. , done , c. , & madejski , g. 1995 , , 273 , 681 wolfe , a.m. , turnshek , d.a . , smith , m.g . , & cohen , r.d . 1986 , , 61 , 249 lccccccl v1309 ori & 7.98271 & @xmath192500(@xmath1921500 ? ) & @xmath193 & @xmath194 & @xmath190 & 61 & 3,5,7,10 + mn hya & 3.38985 & @xmath195 & @xmath196 & @xmath130 & @xmath10 & 20 : 42 : & 2,9,13 + v2301 oph & 1.88280 & @xmath197 & @xmath198 & @xmath199 & @xmath194 & 7 & 1,4,11 + v1432 aql & 3.36564 & @xmath200 & @xmath201 & @xmath158 ? & asynchronous & @xmath194 & 6,8,12,13 + llccl v1309 ori & 1996 aug 4 & @xmath202 & os & @xmath203 + ( rx j0515.6 + 0105 ) & 1996 aug 12 & @xmath204 & us & + & 1996 aug 22 & @xmath205 & us & + & 1996 aug 22 & @xmath206 & us & + & 1996 aug 24 & @xmath207 & us & + & 1996 aug 24 & @xmath208 & us & + & 1996 aug 26 & @xmath209 & us & + & 1996 aug 26 & @xmath210 & us & + & 1996 sep 16 & @xmath211 & us & + & 1996 sep 27 & @xmath212 & us & + & 1996 oct 1 & @xmath213 & us & + & 1996 oct 1 & @xmath214 & us & + & 1996 oct 25 & @xmath215 & us & + & 1996 oct 26 & @xmath216 & us & + & 1996 dec 3 & @xmath217 & os & @xmath218 + + mn hya & 1996 jan 27 & @xmath219 & p , ph & @xmath220 to @xmath221 + ( rx j0929.1@xmath562404 ) & 1996 mar 10 & @xmath222 & ph & 128 ms time samples + & 1996 oct 14 & @xmath223 & us & extremely faint + & 1996 oct 14 & @xmath224 & us & extremely faint + & 1996 oct 15 & @xmath225 & os & @xmath25 + & 1996 dec 4 & @xmath226 & os & @xmath227 + & 1996 dec 24 & @xmath228 & us & + & 1996 dec 24 & @xmath229 & us & + + v2301 oph & 1996 aug 3 & @xmath230 & us & + ( 1h 1752 + 081 ) & 1996 aug 3 & @xmath231 & us & + & 1996 aug 4 & @xmath232 & os & @xmath233 + & 1996 aug 30 & @xmath234 & us & + & 1996 aug 30 & @xmath235 & us & + + v1432 aql & 1996 aug 19 & @xmath236 & us & + ( rx j1940.1@xmath561025 ) & 1996 aug 29 & @xmath237 & us & + & 1996 aug 29 & @xmath238 & us & + & 1996 sep 3 & @xmath239 & us & + & 1996 sep 9 & @xmath240 & us & fgs / tape recorder failure + & 1996 sep 9 & @xmath241 & us & fgs / tape recorder failure + lcccc @xmath51176 & @xmath194 & @xmath194 & 8.2 & 25.8 + ly@xmath133 @xmath51216 & 2.6 & 3.9 & 8.3 & absorption + @xmath51240 & 17.4 & 2.3 & 13.3 & 41.8 + @xmath51298 & 1.2 & 0.35 & 4.1 & 4.5 + @xmath51335 & 0.32 & @xmath194 & 3.8 & 4.3 + @xmath51371 & 0.93 & 0.14 : & @xmath194 & 5.1 + @xmath51397 & 5.4 & 1.3 & 13.3 & 23.8 + @xmath51549 & 2.4 & 0.76 & 29.4 & 83.4 + @xmath51640 & 4.5 & 0.94 & 7.2 & 24.6 + @xmath51718 & 1.6 & 0.09 : & @xmath194 & 1.7 + @xmath51857 & 1.2 & 0.25 & 3.8 & 2.9 + @xmath52297 & 0.29 : & @xmath194 & 0.87 & 2.4 + @xmath52383 & 0.18 : & @xmath194 & @xmath194 & 1.1 + plus @xmath32 emission lines ; _ fuv : _ the @xmath242 continuum ; _ nuv : _ the @xmath243 continuum ; _ white light : _ the zero - order image , a broad passband with @xmath244 . horizontal line segments in the bottom panel indicate phase intervals of the individual time series . rate units are detected photon counts per 0.8201 s sample . , title="fig : " ] 7.truein ) . the width of the latter window is indicated by the gap between opposing arrows . eclipse contacts @xmath68 in the fuv channel indicate ingress of a hot accretion spot which is unresolved in the smoothed data.,title="fig : " ] 7.3truein in the nuv channel , suggesting a structure approximately the size of the white dwarf , and @xmath245 in the fuv , which is evidence for an accretion - heated spot on the white dwarf of linear dimension @xmath1@xmath111 cm and @xmath112 . the extended decline to minimum light in all bands corresponds to the eclipse of the portion of the accretion stream which overshoots in azimuth the white dwarf.,title="fig : " ] 7.3truein

the two - nucleon density distributions in states with isospin , spin=1 and projection=0 and are studied inh ,he ,li ando . the equidensity surfaces for=0 distributions are found to be toroidal in shape , while those of= have dumbbell shapes at large density . the dumbbell shapes are generated by rotating tori . the toroidal shapes indicate that the tensor correlations have near maximal strength at fm in all these nuclei . they provide new insights and simple explanations of the structure and electromagnetic form factors of the deuteron , the quasi - deuteron model , and the , and=2 (-wave ) components inhe ,he andli . the toroidal distribution has a maximum - density diameter of fm and a half - maximum density thickness of.9 fm . many realistic models of nuclear forces predict these values , which are supported by the observed electromagnetic form factors of the deuteron , and also predicted by classical skyrme effective lagrangians , related to qcd in the limit of infinite colors . due to the rather small size of this structure , it could have a revealing relation to certain aspects of qcd . experiments to probe this structure and its effects in nuclei are suggested . pair distribution functions in other channels are also discussed ; those in have anisotropies expected from one - pion exchange interactions . the tensor correlations in states are found to deplete the number of pairs in nuclei and cause a reduction in nuclear binding energies via many - body effects . = 10000

nucl-th9603035

nuclear structure has been discussed mostly in the context of the liquid drop and shell models . these models have been extremely useful in explaining many observed nuclear properties . however , they are based on macroscopic concepts , and do not address the simplest nuclei , hydrogen and helium . furthermore , recent @xmath22 experiments @xcite have indicated that in heavier near - closed - shell nuclei , less than @xmath23 of the nucleons are in the single - particle orbitals that would be fully occupied in the simple shell model . to obtain a more microscopic description of nuclear structure , we may regard the nucleus as a collection of interacting nucleons described by the hamiltonian @xcite @xmath24 the interactions @xmath25 and @xmath26 are not exactly known , but @xmath25 is well constrained by the available scattering data , and binding energies and theoretical considerations place important constraints on @xmath26 . the structure of the ground - state wave function @xmath27 at small interparticle distances is influenced by the repulsive core and tensor parts of @xmath25 . most realistic models of @xmath25 contain these components and , for example , the reid @xcite , paris @xcite , urbana @xcite and new argonne @xmath28 @xcite models seem to predict similar structures . the three - nucleon interaction @xmath26 is much weaker than the @xmath25 . due to a large cancellation between kinetic and two - body interaction energies , it has a significant effect on nuclear binding energies @xcite but its effect on the structure of @xmath27 is much less than that of the better known @xmath25 . due to the strong spin - isospin dependence of @xmath25 and @xmath26 it is difficult to solve the schrdinger equation with the hamiltonian ( 1.1 ) . only recently it has been possible to obtain accurate solutions for @xmath29 nuclei @xcite with the green s function monte carlo ( gfmc ) method . accurate variational wave functions @xmath30 , which contain less than @xmath31 admixture of excited states are known for @xmath32 = 3 and 4 nuclei @xcite . the available @xmath30 for @xmath6li @xcite and @xmath7o @xcite are certainly not as accurate as those for @xmath32 = 3 and 4 , nevertheless they presumably contain most of the important structure of the @xmath27 . in this paper we examine the short - range structure of @xmath4h , @xmath5he , @xmath6li and @xmath7o by calculating the two - nucleon density distributions in states with isospin @xmath33 , spin @xmath1 , and spin projection @xmath2 . since the deuteron has only two nucleons , its one- and two - body density distributions are trivially related . variational wave functions @xmath30 and monte carlo methods are used for @xmath34 . the two - nucleon distributions in the @xmath20 states have a strong dependence on the spin projection @xmath2 . the equidensity surfaces , spanning the top three quarters of the density range in @xmath2=0 states , have toroidal shape . these tori are produced by the joint action of the repulsive core and tensor interactions . in contrast the equidensity shapes in the @xmath2=@xmath31 have dumbbell shapes , which have been studied earlier in the deuteron @xcite . a brief description of the two - nucleon interaction in @xmath20 states is given in section ii , and the density distribution of the deuteron is discussed in detail in section iii , where we show that the dumbbell - shaped distributions in @xmath2=@xmath31 states are produced by rotating tori . commonly used models of @xmath25 predict that the maximum density torus has a diameter of @xmath171 fm , and the half - maximum density torus has a thickness of @xmath170.9 fm . in section iii we relate these dimensions of the toroidal distribution to the observed electromagnetic form factors of the deuteron . the structures are rather dense ; current models predict the maximum one - body density of the torus inside the deuteron to be @xmath35 @xmath36 , i.e. , approximately twice nuclear matter density . the two - nucleon @xmath20 , @xmath37 density distributions in @xmath5he , @xmath6li and @xmath7o are compared with those of the deuteron in section iv . the distributions for @xmath8 fm differ only by a single scale factor . they indicate that in the @xmath0 state , the tensor correlations have near maximal strength for @xmath8 fm in all these nuclei . the scale factor is identified as the levinger - bethe quasi - deuteron number , and its value is compared to the ratios of total photon ( @xmath38 = 80 to 120 mev ) and pion ( @xmath39115 mev ) absorption cross sections . in order to study the nature of many - body structures induced by these compact two - body structures we study the @xmath9 , @xmath10 and @xmath11 overlaps with the @xmath30 of @xmath14he , @xmath15he and @xmath16li in section v. these depend strongly on the spin projection @xmath40 of the deuteron and indicate the presence of anisotropic structures in all these nuclei . experiments to probe these structures are suggested . pair distribution functions in other @xmath41 states are discussed in section vi . those in @xmath42 , @xmath43 states are anisotropic as expected from the pion - exchange tensor force . we also find that the number of @xmath44 pairs in a nucleus is reduced due to many - body effects involving the strong @xmath45 tensor correlations . this reduction gives a significant contribution to the saturation of nuclear binding energies . the skyrme field theory @xcite , related to qcd in the limit of large number of colors @xmath46 , has predicted toroidal shapes for the deuteron @xcite in the classical limit . density distributions of the ground states with 3 to 6 baryons have also been calculated @xcite in this limit . in the last section , vii , we summarize our results , obtained with conventional nuclear many - body theory , and indicate their relation to those of the skyrme field theory in the classical limit , and of the constituent quark model . nuclear forces are not yet quantitatively understood from qcd . however , many realistic models have been constructed by fitting the available two - nucleon scattering data . the shape of the short - range structures in the @xmath20 state appears to be relatively model - independent . the interaction @xmath47 in the @xmath20 state in reid , urbana and argonne models can be expressed as @xmath48 while a @xmath49 operator is used instead of @xmath50 in the paris potential . the structures are formed mostly by the static part of the interaction : @xmath51 it is instructive to study the expectation values of @xmath52 in eigenstates of the position operator @xmath53 with spin projections @xmath2=0 and @xmath2=@xmath31 . these depend upon @xmath54 and @xmath55 , the polar angle of @xmath53 with respect to the spin - quantization axis @xmath56 , and are given by @xmath57 the @xmath2=0 expectation value of @xmath52 has the largest variation with @xmath55 as illustrated in fig . [ potfig ] . the static potential has a repulsive core ; outside the core it is very attractive for @xmath55=@xmath58 and repulsive for @xmath55=0 and @xmath59 . therefore , in this state the @xmath60 pairs form a toroidal density distribution in the @xmath61 plane . the potential in the @xmath2=@xmath31 states is attractive for @xmath55=0 and @xmath59 , and equal to that for @xmath2=0 , @xmath55=@xmath58 , while it is repulsive for @xmath55=@xmath58 and half way between the @xmath2=0 , @xmath55=0 and @xmath58 potentials . thus the @xmath2=@xmath31 potential has two distinct minima separated by a barrier , and therefore the density distributions have a dumbbell shape in this state . at @xmath621.5 fm , the @xmath52 is dominated by the one - pion - exchange potential , while at smaller @xmath54 it has a significant model dependence . much of this model dependence is cancelled by the differences in the momentum - dependent terms in the models . in particular , the deuteron wave functions calculated from these potentials have much smaller model dependence . these are commonly written as @xmath63 where @xmath64 and @xmath65 are the @xmath1- and @xmath13-state radial wave functions and @xmath66 are the spin - angle functions . the @xmath67 and @xmath68 calculated from the different potential models are shown in fig . [ r0r2fig ] . the short - range structures are related to the @xmath67 and @xmath68 functions , and therefore we expect them to be fairly model - independent . however , the `` full '' bonn potential @xcite offers an exception . the @xmath67 and @xmath68 predicted by this one - boson - exchange model of the @xmath69 interaction are similar to other predictions at larger @xmath54 , but they have an additional sharp structure with a range of @xmath170.2 fm . we will not consider the possibility of such an additional structure in this work . the short - range structure of the deuteron is most obvious in its density distribution @xmath70 which depends upon the projection @xmath40 of the total deuteron angular momentum , the distance @xmath71 from the deuteron center of mass , and the polar angle @xmath55 of @xmath72 ; it is independent of the azimuthal angle @xmath73 . note that the interparticle distance @xmath74 , and the standard normalizations , @xmath75 & = & 1 , \\ \int\mbox{d}^3{\bf r}^{\prime}\ \rho_d^{m_d}({\bf r}^{\prime } ) & = & 2,\end{aligned}\ ] ] are used in this work . the @xmath76 is given by @xmath77 , where the factor 16 comes from the difference in normalizations ( 3.1 ) and ( 3.2 ) . a simple calculation using eq . ( 2.5 ) yields @xmath78 \>\>\ > , \\ \rho_d^{\pm 1}({\bf r}^{\prime})&=&\frac{4}{\pi}\left [ c_0(2r^{\prime})+c_2(2r^{\prime})\mbox{p}_2(\cos\theta ) \right ] \>\>\>,\end{aligned}\ ] ] with @xmath79 the interesting structure of these density distributions is shown in figs . [ figdeutxz]-[surffig ] . [ figdeutxz ] shows @xmath80 along @xmath81 and @xmath82 directions , noting that @xmath83 the above densities are the largest while @xmath84 is the smallest as expected from the properties of @xmath52 discussed in sec . the small value of the ratio @xmath85 indicates that the deuteron has near maximal tensor correlation at distance @xmath86 fm or equivalently at @xmath8 fm . this ratio is @xmath170 for maximal tensor correlations . figures [ rhod0 ] and [ rhod1 ] show the density distributions predicted by the argonne @xmath28 model in the @xmath87 plane . the maximum value of @xmath88 is fairly model - independent ( fig . [ figdeutxz ] ) and large ( @xmath170.35 @xmath36 ) . the maxima of @xmath89 ( fig . [ rhod0 ] ) form a ring with a diameter of @xmath171 fm , denoted by @xmath90 , in the @xmath91 plane , while the @xmath92 has two equal maxima on the @xmath93-axis ( fig . [ rhod1 ] ) at @xmath93=@xmath94 . the three - dimensional distributions @xmath76 can be obtained by rotating the distributions shown in figs . [ rhod0 ] and [ rhod1 ] about the @xmath93-axis . they are represented by equidensity surfaces shown in fig . [ surffig ] for @xmath88=0.24 and 0.08 @xmath36 ; all four sections are drawn to the same scale . the maximum value of @xmath95 is @xmath170.05 @xmath36 ( fig . [ figdeutxz ] ) . therefore the equidensity surfaces for @xmath89 having @xmath960.05 @xmath36 can not intersect the @xmath93-axis , and thus have toroidal shapes shown in figs . [ surffig]b and [ surffig]d . the central hole in these tori is due to the repulsive core in @xmath52 , and their angular confinement is due to the tensor force . in absence of the tensor force , @xmath97 , the @xmath98 , and the equidensity surfaces are concentric spheres . the maximum value of @xmath99 is @xmath170.19 @xmath36 as can be seen from fig . [ figdeutxz ] . therefore the equidensity surfaces of @xmath92 for @xmath1000.19 @xmath36 can not cross the @xmath55=@xmath58 plane ; they have two disconnected parts forming a dumbbell as shown in fig . [ surffig]a . at smaller values of @xmath88 we also obtain two equidensity @xmath92 surfaces ( fig . [ surffig]c ) , consisting of an inner surface due to the repulsive core enclosed by an outer . at very small @xmath88 ( @xmath1010.05 @xmath36 ) the equidensity surfaces of @xmath89 also have disconnected inner and outer parts , neither close to spherical in shape . the toroidal shape of the @xmath40=0 equidensity surfaces is more compact and persists down to smaller @xmath88 , or equivalently to larger values of @xmath71 , as can be seen from figs . [ figdeutxz]-[surffig ] . in the classical skyrmion field theory only this shape is obtained for the distribution of baryon density in the ground state for two baryons @xcite . the deuteron can be considered to be more deformed in the @xmath40=0 state . for example , the expectation values of the quadrupole operator @xmath102 obey the relation @xmath103 it is rather simple to obtain the @xmath104 from the @xmath105 . we rotate the @xmath105 about the @xmath106-axis by an angle of @xmath58 so that the toroidal ring is in @xmath107 plane with @xmath108 as the symmetry axis . this places the deuteron in the superposition of @xmath40=@xmath31 states . the @xmath40=@xmath31 states are obtained by spinning the rotated toroid about the @xmath93-axis , and the @xmath104 is just the average value of the density of the spinning toroid , i.e. , @xmath109 the @xmath12=1 , @xmath110=@xmath31 states of the harmonic oscillator , given by @xmath111=e@xmath112i@xmath113 ) , are obtained in the same way from the @xmath111=e@xmath114 , @xmath12=1 , @xmath110=0 state . therefore it is tempting to consider the toroidal shape of @xmath89 as the basic shape of the deuteron . the expectation value of the current operator is zero in the @xmath40=0 state , therefore one may regard that as the `` static '' state of the deuteron . note that the toroidal shapes can not be obtained by rotating the dumbbell by @xmath58 about the @xmath106-axis and spinning it about the @xmath93-axis . this gives @xmath115 which is not true . the dumbbell- or cigar - shaped density distribution of the deuteron in the @xmath40=@xmath31 state has been studied earlier @xcite . unfortunately the toroidal distribution of the @xmath40=0 state was not studied , and its similarity with the predictions of skyrmion field theory was not noticed . the deuteron electromagnetic structure functions @xmath116 and @xmath117 , and the tensor polarization @xmath118 in elastic electron - deuteron scattering have been extensively studied experimentally @xcite and theoretically @xcite . they are usually calculated from the @xmath1- and @xmath13-wave functions @xmath67 and @xmath68 obtained from realistic interactions , by including in the nuclear electromagnetic current , in addition to the dominant impulse approximation ( ia ) operators , relativistic corrections and two - body meson - exchange contributions @xcite . more recently , calculations of these observables based on quasipotential reductions of the bethe - salpeter equation and one - boson - exchange interaction models , constrained to fit nucleon - nucleon data , have also been carried out @xcite . the theoretical predictions for the structure functions based on both the nonrelativistic and relativistic approaches are in good agreement with data . our interest here is not to improve upon the present theoretical predictions , but to relate the values of the minima and maxima of @xmath118 and @xmath117 to the size of the toroidal structure in the deuteron . the @xmath119-values of these extrema may be shifted by @xmath1710@xmath120 by corrections to the impulse approximation used in the following pedagogical analysis . the charge form factors , defined as @xmath121 are shown in fig . [ deutlong ] . at large @xmath119 the @xmath122 gets most of its contribution from the two peaks of @xmath123 ( fig . [ rhod1 ] ) at @xmath93=@xmath94 . the fourier transform of the sum of two @xmath124-functions at @xmath93=@xmath94 is given by @xmath125 with zeros at @xmath126=@xmath59 , 3@xmath59 , @xmath127 . these zeros are due to the cancellation of the contribution from the two peaks , and they persist even when the peaks have a finite width . the first two minima of @xmath128 , obtained from the argonne @xmath28 @xmath123 , occur at @xmath129=3.6 and @xmath130=12.6 @xmath131 . the effective values @xmath132 estimated from the minima @xmath133 , using @xmath132=@xmath134 , are 0.87 and 0.75 fm for @xmath135=1 , 2 . these values are smaller than the diameter @xmath90=1 fm because the dumbbell ends ( fig . [ surffig]a ) are not spherical . nevertheless the minima of @xmath136 seem to be primarily determined by the diameter @xmath90 of the maximum - density torus . the fourier transforms of a disc of thickness @xmath137 , with @xmath138 perpendicular to the disc , are proportional to @xmath139 irrespective of the shape of the disc . these have zeros at @xmath140=2@xmath59 , 4@xmath59 , @xmath127 , which may be used to obtain the thickness @xmath137 . the first two zeros of @xmath141 at @xmath119=9.2 and 19.5 @xmath131 ( fig . [ deutlong ] ) give values 0.68 and 0.64 fm for the effective thickness of the torus . the maximum thickness along the @xmath142-axis of the calculated equidensity surface at half maximum density is 0.88 fm . the @xmath118 form factor of the deuteron has small magnetic contributions which depend upon the electron scattering angle @xmath55 . the extrema of @xmath143 are not significantly affected by this magnetic contribution as can be seen from fig . [ deutt20 ] , and ignoring it we obtain a rather simple equation : @xmath144 its minima occur when @xmath128=0 , while the maxima have @xmath145=0 . these minima and maxima correspond to those values of @xmath119 at which the recoiling deuteron has only @xmath40=0 or @xmath40=@xmath31 , respectively . the first minimum of @xmath143 is experimentally known to occur at @[email protected] @xmath131 in agreement with the value @xmath1471 fm predicted by realistic potentials . the first maximum of @xmath143 has not yet been experimentally located ; it provides a measure of the thickness @xmath137 . in magnetic elastic scattering the deuteron spin projection @xmath40 in the @xmath138 direction changes by @xmath31 since the photon has @xmath148=@xmath31 . thus the magnetic form factor @xmath149 is a transition form factor . it has convection current and spin - flip terms @xcite of which the latter is dominant . the @xmath149 calculated with and without the convection current term are not too different ( fig . [ deutmq ] ) . the spin flip part of @xmath149 can be obtained from the transition density @xmath150 : @xmath151 \mbox{p}_2(\cos\theta)\right\}.\end{aligned}\ ] ] this transition density is shown in fig . [ tranden ] ; it is dominated by the toroidal @xmath105 , and its effective thickness along @xmath93 axis , obtained from the zeros of @xmath149 ( fig . [ deutmq ] ) , is @xmath170.85 fm . the minimum of @xmath149 is observed @xcite at @xmath152 @xmath131 , supporting the theoretical estimates of @xmath137 . the deuteron wave function in momentum space is defined as @xmath153 with @xmath154 the momentum distributions @xmath155 , given by @xmath156 , are then easily obtained as @xmath157 \>\>\ > , \\ { \tilde { \rho}}_d^{\pm 1}({\bf k})&=&\frac{1}{4 \pi}\left [ { \tilde { c}}_0(k)+{\tilde { c}}_2(k)\mbox{p}_2(\cos\theta_k ) \right ] \>\>\>,\end{aligned}\ ] ] where @xmath158 is the angle between @xmath159 and @xmath56-axis , and the @xmath160 are defined as in eqs . ( 3.5)-(3.6 ) with @xmath161 replaced by @xmath162 . note that the @xmath163 are normalized such that @xmath164 the momentum distributions @xmath165 and @xmath166 for @xmath167 and @xmath168 are shown in fig . [ rhodk ] . note that @xmath169 the zeros of @xmath170 and @xmath171 occur at @xmath172 @xmath131 and @xmath173 @xmath131 , respectively , and are related to the spatial dimensions of the torus . in naive estimates these minima occur at @xmath174 and @xmath175 , respectively . thus measuring the positions of the zeros in these momentum distributions would provide an independent estimate of the spatial dimensions of the toroidal structure in the deuteron . this information would be complementary to that yielded by elastic form factors measurements . the momentum distributions @xmath176 could in principle be measured by @xmath22 scattering on polarized deuterons . in the one - photon exchange approximation the @xmath177 cross section , in the laboratory frame , is generally expressed as @xmath178 where @xmath40 is the target spin projection , @xmath179 and @xmath180 are the energy and solid angle of the final electron , and @xmath181 is the solid angle of the ejected proton . the coefficients @xmath182 are defined in terms of the electron variables , while the structure functions @xmath183 involve matrix elements @xmath184 of the charge ( @xmath185 ) and current ( @xmath186 ) operators between the initial deuteron and final @xmath187 states . the neutron momentum is @xmath188=@xmath189 , @xmath138 is the momentum transfer , @xmath190=@xmath188 is the missing momentum , and @xmath191 and @xmath192 are the proton and neutron spin projections , respectively @xcite . the cross section for unpolarized deuterons , @xmath193 has been measured up to @xmath194500 mev / c , and there is good agreement between theory and experiment @xcite . in plane - wave - impulse - approximation ( pwia ) , obtained by neglecting interaction effects in the final @xmath187 states as well as relativistic corrections and two - body terms in the charge and current operators , the @xmath40-dependent cross section is proportional to @xmath195 using , for example , tensor polarized deuterium , it should be possible to measure the difference between @xmath196 and @xmath197 , and therefore empirically determine the positions of the minima in these momentum distributions . clearly , such an analysis is justified if the pwia is valid . this assumption has been tested by carrying out the full and pwia calculations of the @xmath198 in parallel kinematics with @xmath119 fixed at 500 mev / c , @xmath199 in the range 290 - 390 mev , and the electron scattering angle @xmath200 . the results , shown in fig . [ 2heep ] , indicate that , while fsi , two - body current , and relativistic corrections are not entirely negligible , at least in the kinematical region which has been studied here , their effect is small compared to the difference between the cross sections for @xmath40=0 and @xmath31 . we therefore conclude that the results of such an experiment could be used to empirically study the diameter and thickness of the torus . one might argue that this information could be more easily obtained from elastic form factors measurements , as discussed above . however , it should be realized that , in contrast to the @xmath201 data , the double - coincidence data would allow us to ascertain to what extent this toroidal structure is due to nucleonic degrees of freedom . the two - nucleon density - distributions in @xmath33 , @xmath1 , @xmath202 , @xmath2 two - nucleon states are defined as @xmath203 here @xmath204 denotes the ground state of the nucleus with total angular momentum @xmath205 and projection @xmath206 , and @xmath207 projects out the specific two - nucleon state with @xmath208 . for @xmath209 nuclei , the wave functions used in this study are symmetric under exchange of neutrons and protons ; hence @xmath210 is independent of @xmath202 . for @xmath14he , we have @xmath211 while for larger @xmath212 nuclei , the @xmath202 dependence is nontrivial . in the following we discuss @xmath213 , the average over @xmath202 of @xmath214 . the @xmath213 is normalized such that @xmath215 which is the number of pairs in the nucleus . it is a function of @xmath54 and @xmath55 independent of the azimuthal angle @xmath73 . it can be verified from eqs.(3.3 - 3.6 ) and ( 4.1 ) that in the deuteron @xmath216 note that the spin - dependent two - body density on the left ( @xmath217 ) is an average over projections @xmath40 in the deuteron , while the polarized one - body density on the right ( @xmath218 ) has been summed over spins . the @xmath213 in @xmath5he , @xmath6li , and @xmath7o have been calculated from variational wave functions using monte carlo techniques . for the a @xmath101 7 nuclei , these wave functions minimize the expectation value of a hamiltonian consisting of the argonne @xmath28 two - nucleon and urbana ix three - nucleon potentials @xcite ( for @xmath32 = 6,7 the minimization is constrained by the experimental rms radii ) ; a detailed description of the form of the wave functions is given in refs . the wave function for @xmath7o was obtained from the variationally best wave function by slightly increasing the radius of the single - particle part of the wave function so as to reproduce the experimental rms radius of @xmath7o . the details of the @xmath7o wave function will be published elsewhere @xcite . a cluster - expansion including up to four - body clusters with monte carlo integration @xcite was used to compute the two - body densities in @xmath7o . to reduce statistical fluctuations in the calculation , we write @xmath219 and directly compute the @xmath220 as @xmath221 where @xmath222 represents the coordinates @xmath223 . because of the average over the total spin of the nucleus , the @xmath224 are zero for @xmath225 , and @xmath226 for the deuteron , the @xmath227 are related to the @xmath228 of eqs.(3.3 - 3.6 ) by @xmath229 ; @xmath230 . the shapes of @xmath231 are very similar at @xmath232 2 fm in all the nuclei considered . in order to study the evolution of @xmath233 with @xmath32 we divided the @xmath234 by the ratio @xmath235 defined as @xmath236 the densities so normalized are compared in fig . [ rhoa ] , and the values of @xmath235 are listed in table i. fig . [ rhoa ] shows @xmath237 , @xmath238 , and @xmath239 for @xmath4h , @xmath15he , and @xmath7o . note that @xmath240 = @xmath241 by virtue of equations ( 4.4 ) , ( 4.6 ) and ( 4.7 ) in all nuclei . after normalization by @xmath235 , the various densities for @xmath14he lie between those of @xmath4h and @xmath15he , while those for @xmath6li are in between the @xmath15he and @xmath7o results . it is obvious from fig . [ rhoa ] that the equidensity surfaces of the two - body density @xmath242 are very similar to those of the deuteron density shown in figures [ figdeutxz]-[surffig ] at @xmath243 2 fm ( @xmath244 1 fm ) . at @xmath243 2 fm the ratio @xmath245 is very small , indicating that the tensor correlations have near maximal strength in all the nuclei considered . in @xmath7o the @xmath233 becomes approximately independent of @xmath2 only for @xmath246 fm . bethe and levinger suggested in 1950 @xcite that at small distances the relative @xmath18 = 0,1 neutron - proton wave function in a nucleus is likely to be similar to that in the deuteron . we find that this is a good approximation . the expectation value of any short - ranged two - body operator that is large only in the @xmath20 state scales as @xmath235 . in table i we list values of the ratios of the calculated expectation values of the one - pion exchange part of the argonne @xmath28 potential , the observed low - energy ( 118 mev for @xmath14he @xcite and @xmath15he @xcite , and 115 mev for @xmath7o @xcite ) pion absorption cross sections and the average value of the observed photon absorption cross sections in the range @xmath38 = 80 to 120 mev . all these processes are dominated by the @xmath20 pairs , and seem to scale as @xmath235 . while comparing these ratios in detail it should be realized that @xmath247 in nuclei has a relatively small contribution from @xmath248 states , absent in the deuteron , which makes @xmath249 slightly larger than @xmath235 . the @xmath250 = 21.3 mev for the argonne @xmath28 model , and it accounts for most of the deuteron potential energy , @xmath251 = 22.1 mev . also in larger nuclei , the @xmath252 gives a large fraction of the total two - body interaction energy @xcite . direct comparison of the ratio of pion absorption cross sections with @xmath235 may not be strictly valid . the scattering and absorption of pions by spectator nucleons , absent in the deuteron , is expected to reduce the ratio @xmath253 , while three - body and higher absorption mechanisms , also absent in the deuteron , will increase it . after correcting for final state interactions of the two outgoing protons , the two - body ( @xmath254 ) part is estimated to account for @xmath255 of the total absorption cross section for 115 mev @xmath256 by @xmath7o @xcite . in @xmath14he @xmath257 of the 118 mev @xmath256 absorption cross section has three - body character @xcite , however a part of this @xmath258 must be due to initial and final state interactions . results of mainz experiments @xcite on @xmath259li and @xmath7o are used to calculate the average value of @xmath260 in the energy interval @xmath38 = 80 to 120 mev . the @xmath261 averaged over the same energy interval is @xmath262 mb @xcite . the only available data for @xmath14he in this energy range are from the experiments done in the 1960 s @xcite and 1970 s @xcite . the average cross section of the two - body photodisintegration of @xmath14he , in the energy range 80 - 120 mev , is @xmath170.03 mb @xcite , and that for the three - body process is @xmath170.10 mb @xcite , giving total cross section of @xmath170.13 mb . the average cross section for total absorption of photons by @xmath15he in the same energy range is crudely estimated from fig . [ potfig ] in ref . @xcite to be @xmath170.3 mb . the total number of pairs with given @xmath18 in nuclei can be computed as @xmath263 the values for @xmath20 and the corresponding naive independent - particle model values are also shown in table i. we see that the correlations induced by the potentials do not significantly change the @xmath264 from their independent - particle ( ip ) values ; however , as will be discussed later , this is not true for @xmath33=1 pairs . for few - body nuclei , @xmath235 is significantly larger than @xmath264 , however , in @xmath7o @xmath264 has a large contribution from pairs with large @xmath54 and @xmath235 is smaller than @xmath264 . the calculated value of @xmath235 for @xmath7o is much smaller than levinger s estimate @xmath265 @xcite . the @xmath266 has its half - maximum value at @xmath2671.8 fm ( fig . [ rhoa ] ) . if we identify the region with @xmath268 fm as the `` quasi - deuteron '' , then the probability that the @xmath60 pair in a deuteron is in the quasi - deuteron region is @xmath170.25 , and the number of quasi - deuterons in a nucleus is @xmath269 . in the past , however , @xmath235 itself has been interpreted as the number of quasi - deuterons in the nucleus . the strong spin - dependent anisotropies of the two - nucleon densities suggest that three - nucleon and higher distribution functions in nuclei could also be anisotropic . a general study of these higher distributions is beyond the scope of this work ; however the two - cluster distribution functions @xmath270 in @xmath14he , @xmath271 in @xmath15he , and @xmath272 in @xmath16li are simple to study with the monte carlo method @xcite . they provide some information on the higher distribution functions , and may be relatively accessible by @xmath273 and @xmath274 experiments . the two - cluster overlap function can be written generally as @xmath275 where @xmath276 is the is the relative coordinate between the centers of mass of the two clusters and @xmath277 is an antisymmetrization operator for the two - cluster state . the @xmath278 radial functions can be evaluated from @xmath279^{\dagger } y^*_{lm_l}(\hat{r}_{ab } ) \frac{\delta(r - r_{ab})}{r^2_{ab } } \psi^{m_j}({\bf r } ) \ , \end{aligned}\ ] ] where @xmath280 represents the coordinates of particles in cluster @xmath281 . we note that in pwia the @xmath282 cross section is proportional to the momentum distribution @xmath283 obtained from the fourier transform of the overlap function @xmath284 . in the present work , the integrations have been made with monte carlo techniques akin to ref.@xcite , but with some improvements . configurations are sampled with the weight function @xmath285 containing the full variational wave function . in ref.@xcite only a single term in the antisymmetric product in eq . ( 5.2 ) is calculated . the efficiency of the monte carlo sampling has been improved by evaluating all possible partitions of the nucleus into clusters @xmath286 and @xmath287 at each configuration @xmath222 . we also use a much larger sample of configurations than in the previous calculations . we can define a two - cluster wave function , in analogy with the deuteron wave function of eq . ( 2.5 ) , using the radial overlap functions @xmath288 where @xmath289 and @xmath290 denote spin states @xmath291 and @xmath292 of @xmath286 and @xmath287 . for the cases @xmath293 = @xmath9 , @xmath10 , and @xmath11 there are both s- and d - wave states in the two - cluster wave function . in these cases the well known @xmath294 parameter can be defined by means of the @xmath295 and @xmath296 radial functions @xcite : @xmath297 although in the present paper we emphasize the short - range structure of nuclei , it is also interesting to study the asymptotic behavior of the overlap radial functions . of particular interest is the asymptotic d / s ratio @xmath298 , where @xmath299 and @xmath300 are the asymptotic normalization constants of @xmath295 and @xmath296 respectively : @xmath301 here @xmath302 is the spherical hankel function of first kind and @xmath303 is the wave number associated with the separation energy of the nucleus into clusters @xmath286 and @xmath287 . we must point out that the present variational method , as well as the gfmc method , determine the wave functions by energy minimization , to which long - range configurations contribute very little . therefore these methods are not very sensitive to the asymptotic part of the wave functions , and consequently our values for @xmath304 should be considered only as estimates . the two - cluster density distribution for a given set of spin projections is defined as : @xmath305 in each of the cases studied here , it exhibits spin - dependent spatial anisotropies which are easily understood in terms of the toroidal or dumbbell structure of the polarized deuteron . the density is enhanced in the direction corresponding to the most efficient or compact placement of the deuteron with the remaining cluster , and reduced in those directions that would lead to very extended structures . finally , we are also interested in the total normalizations @xmath306 of the s- and d - wave two - cluster distributions : @xmath307 these quantities can be related to spectroscopic factors and give the total s- and d - state fractions . all the results presented here are obtained from the argonne @xmath28 two - nucleon and urbana model ix three - nucleon interactions . the calculated @xmath308 and @xmath309 are shown in fig . [ r02dp ] ; the @xmath309 is negative and smaller in magnitude than the @xmath68 in deuteron ( fig . [ r0r2fig ] ) . the @xmath310 value obtained with these radial functions is @xmath311 fm@xmath4 , a little smaller than experimental estimates , ranging from @xmath313 to @xmath314 , obtained through dwba analysis of ( @xmath90,@xmath14he ) transfer reactions @xcite . in fig . [ r02dp ] we also show our asymptotic fit to the s- and d - waves . the result is @xmath315 , somewhat smaller than the faddeev result @xcite , @xmath316 . experimental estimates , also obtained through dwba analysis of ( @xmath90,@xmath14he ) transfer reactions , range from @xmath317 to @xmath318 @xcite . the total normalizations are @xmath319 and @xmath320 . the sum , 1.33 , can be interpreted as the number of deuterons in @xmath14he @xcite . it is less than 1.49 ( table i ) , the number of @xmath45 pairs , because the pairs are not always in the deuteron state . it is also smaller than @xmath321 inferred from short range distribution functions ( fig . [ rhoa ] and table i ) . this is probably because @xmath14he is more compact than the deuteron . the @xmath322 and @xmath323 are shown in figs . [ 3hemd0 ] and [ 3hemd1 ] . when @xmath324 , the @xmath325 proton is preferentially along the @xmath142-axis ; in contrast , when @xmath326 , the @xmath327 proton is more likely to be in the @xmath61 plane . in the first density distribution the s- and d- wave amplitudes interfere constructively , to enhance the probability of finding the proton along the @xmath142-axis , whereas in the latter the interference is constructive on the transverse @xmath61-plane . consequently @xmath67 and @xmath68 have opposite signs and @xmath310 and @xmath328 are both negative . the spin - dependent @xmath329 anisotropies are favored by both tensor and central forces , and lead to more compact three - body states . the momentum distribution of @xmath9 clusters in @xmath330 @xmath14he is shown in fig . [ 3hemdk ] for @xmath40 , @xmath191 = 0 , @xmath331 and 1 , @xmath332 for momenta parallel and transverse to the @xmath142-axis . in pwia the @xmath333 cross - section is directly related to these momentum distributions . a large spin dependence of the missing - momentum distribution for protons ejected in parallel kinematics is predicted for @xmath138 parallel to @xmath56 . the minimum for the momentum distribution along @xmath142-axis for @xmath325 occurs at @xmath171.4 @xmath131 , while that for @xmath327 is at @xmath172.4 @xmath131 . thus the spin asymmetry , @xmath334 , of the protons ejected from polarized @xmath14he changes from @xmath17 1 to + 1 as the missing momentum varies from @xmath171.4 to 2.4 @xmath131 in parallel kinematics and pwia . the @xmath9 momentum distribution in unpolarized @xmath14he has been studied at saclay @xcite up to @xmath172.5 @xmath131 . the observed distribution is generally smaller than the pwia prediction @xcite indicating attenuation due to fsi . however , a part of the fsi attenuation will cancel in the asymmetry , and moreover , it is now possible to perform continuum faddeev calculations @xcite including fsi . the calculated @xmath335 and @xmath336 are shown in fig . [ r02dd ] . the @xmath337 value obtained with these radial functions is @xmath338 fm@xmath4 . in fig . [ r02dd ] we also show our asymptotic fit to the s- and d - waves . the result is @xmath339 . the integrals of these functions yield @xmath340 and @xmath341 . the number of deuterons present is greater than twice the sum of these quantities , 2.0 , when one allows for the additional presence of @xmath342 configurations . the @xmath343 and @xmath344 are large in @xmath15he and their anisotropies , induced by the tensor interaction and the shapes of deuterons , are shown in fig . [ 4hedd ] . the @xmath343 is largest when @xmath345 is along the @xmath142-axis , i.e. , when the deuterons are in the toroidal shape and have a common axis . it is smallest when @xmath346 is transverse ( two tori side by side ) and equal to that for @xmath344 with @xmath346 parallel to @xmath56 ( two dumbbells in a line ) . the latter distribution is of intermediate strength when @xmath346 is transverse ( two dumbbells side by side ) . again in the first ( second ) density distribution the s- and d - wave amplitudes interfere constructively ( destructively ) along the @xmath56-axis . therefore @xmath67 and @xmath68 have opposite signs and @xmath337 and @xmath347 are both negative . the momentum distributions are also anisotropic ( fig . [ 4heddk ] ) . in particular the @xmath348 has a dip at @xmath3491.7 @xmath131 that is absent in the @xmath350 . it may be possible to study these with @xmath273 reactions . the unpolarized @xmath15he@xmath351 reaction has been studied at nikhef @xcite . the observed cross sections are much smaller than estimates using the @xmath10 momentum distribution and either pwia or dwia . the calculated @xmath352 and @xmath353 are shown in fig . [ r02ad ] . the @xmath352 and @xmath353 both exhibit nodes at short distances and have opposite signs almost everywhere . this nodal structure has been predicted in @xmath354 and @xmath355 cluster models , but not always with the correct relative sign @xcite . the asymptotic behavior is correlated with the quadrupole moment @xmath356 ; obtaining the experimental value of @xmath357 fm@xmath4 has been a long - standing problem in @xmath355 cluster models . the variational wave function used here gives @xmath358 fm@xmath4 , i.e. , the correct sign but far too large in magnitude . small changes in the long range part of the @xmath16li wave function have effects of order 1 fm@xmath4 on the quadrupole moment . thus the values of the asymptotic properties , @xmath359=0.29 fm@xmath4 and @xmath360=0.07@xmath361 , obtained with this wave function may not be very accurate . the total normalizations are @xmath362 and @xmath363 . the resulting spectroscopic factor , 0.84 , is in good agreement with the value of 0.85 obtained in radiative capture experiments @xcite . the two - cluster densities @xmath364 , multiplied by @xmath365 , are shown in fig . [ 6liad ] . they have two peaks ; the smaller inner peak at @xmath366 fm is almost spherically symmetric , while the larger peak at @xmath367 fm is anisotropic . in particular @xmath368 is much larger than @xmath369 for @xmath370 fm . in the former configuration the @xmath54 is along the axis of the torus , while in the latter it is transverse . this anisotropy is also a consequence of the toroidal shape of the deuteron in the @xmath40=0 state . in this section we discuss the properties in nuclei of pairs of nucleons with @xmath371 , @xmath372 , and @xmath373 . like the @xmath20 channel discussed in the previous sections , the @xmath19 channel also has a tensor potential , but it has the opposite sign of that for @xmath20 . therefore the role of @xmath2 is reversed compared to that in @xmath0 states ; @xmath374 pairs have maximum density along the @xmath142-axis , while @xmath375 pairs have maximum density in the @xmath61 plane as can be seen in fig . [ ts11fig ] , which shows @xmath376 for @xmath15he , @xmath16li , and @xmath7o . the curves for @xmath377 , @xmath378 are between the two sets of curves shown in the figure ; to reduce clutter they are not shown . the curves for @xmath15he and @xmath16li have been renormalized by the factors @xmath379 to have the same peak height as for @xmath7o ; these factors are shown in table ii . we see that the shapes of the @xmath19 density profiles are quite different in the different nuclei and that the anisotropy decreases as the number of nucleons increases . the analog of the deuteron in the @xmath21 channel is the @xmath380s@xmath381 virtual bound state ( vbs ) . for the argonne @xmath28 potential , this is a pole on the second energy sheet at @xmath382 mev or @xmath383 @xmath131 . although the wave function is not normalizable , it has a local peak which we scale to compare to the unpolarized deuteron in fig . [ ts10fig ] . we see that it peaks at a slightly larger radius and is broader . the figure also shows the @xmath384 of @xmath15he , @xmath16li , and @xmath7o ; the curve for @xmath14he is between those of @xmath15he and @xmath16li , while the curve for @xmath259li is very close to that of @xmath16li . again the curves have been normalized to the peak height of the @xmath7o density . we see that the short - range shapes of the @xmath385 in nuclei are well reproduced by the vbs density . finally , fig . [ ts00fig ] shows the densities for the @xmath371 channel , again normalized to @xmath7o . as is the case for @xmath19 , there is no common shape . table ii also shows the number of pairs , @xmath386 , in these @xmath18 channels and the corresponding ip values . as is the case for @xmath20 ( table i ) , the number of pairs increases more rapidly with a than does @xmath387 , because of the increasing proportion of pairs with large separation . using the projection operators @xmath388 and @xmath389 for @xmath0 and 1 pairs we find that the total number of @xmath0 and 1 pairs in a nucleus depends only on its mass number @xmath32 and isospin @xmath390 : @xmath391 , \\ n_{1,0}^a+n_{1,1}^a & = & \frac{1}{8}\left[3a^2 - 6a+4t_a(t_a+1)\right].\end{aligned}\ ] ] the above relations are obeyed by @xmath392 obtained from either the @xmath393 or correlated wave functions , since , in the present study , both are eigenstates of @xmath390 . if the total spin , @xmath394 were to be a good quantum number we would have similar relations , @xmath395 , \\ n_{0,1}^a+n_{1,1}^a & = & \frac{1}{8}\left[3a^2 - 6a+4s_a(s_a+1)\right],\end{aligned}\ ] ] for the total number of pairs with spin 0 and 1 . they are obeyed by the @xmath392 calculated for the @xmath393 states which have @xmath396 and 0 for @xmath4h , @xmath14he , @xmath15he , @xmath16li , @xmath259li and @xmath7o respectively . however , @xmath397 is not a good quantum number ; tensor correlations admix states with larger @xmath397 in the ground state . these reduce the @xmath398 and increase the @xmath399 by the same amount due to eq . ( 6.2 ) . in @xmath14he ( @xmath15he ) the @xmath399 is given by 1.5@xmath400 ( 3@xmath400 ) , where @xmath400 is the fraction of @xmath401 , @xmath402 ( @xmath403 ) state in the nuclear ground state . the interaction in the @xmath21 state is much more attractive than that in the @xmath19 state . hence the depletion of @xmath21 pairs by tensor correlations reduces the binding energy of nuclei significantly . for example , in @xmath15he the @xmath21 interaction gives @xmath404 mev per pair , while the @xmath19 interaction gives only 0.8 mev per pair . thus the conversion of 0.47 @xmath33=1 pairs from @xmath1=0 to @xmath1=1 state raises the energy of @xmath15he by @xmath405 mev . it should be stressed that this is a `` many - body '' effect absent in the two - body cluster approximation of either brueckner or variational methods . the tensor interaction between nucleons @xmath135 and @xmath406 can flip their spins and convert pairs @xmath407 and/or @xmath408 from @xmath1=0 to @xmath1=1 . the main conclusions of this study of nuclear structure , as predicted by realistic models of nuclear forces , are : \(i ) the static part of the two - nucleon potential in @xmath409 state has a large angular dependence due to the tensor interaction dominated by one - pion exchange . at @xmath410 fm the difference between this potential at @xmath55 = @xmath58 and 0 is @xmath411 mev in most models ( fig . [ potfig ] ) . it confines @xmath409 pairs to the small @xmath55 region producing toroidal distributions . the central hole in these tori is due to the repulsive core in @xmath69 interaction . the maximum density in the tori is large , due to which the peak one - body density in deuterium exceeds 0.3 @xmath36 in most models . \(iii ) the diameter of the maximum density torus , and the thickness of the half - maximum density torus are predicted to be @xmath413 and 0.9 fm , respectively ; these values are supported by the observed elastic electron - deuteron scattering . \(v ) the pair distribution functions in @xmath20 and 1,0 states in different nuclei , can be scaled to lie on universal surfaces for @xmath414 fm . these universal surfaces are predicted by the density distributions of the deuteron and the @xmath415 virtual bound state . the scaling factor @xmath235 for the @xmath20 densities provides a rigorous definition of the levinger - bethe quasi - deuteron number of the nucleus . the calculated values of @xmath235 are significantly different from estimates based on independent - particle models , and in qualitative agreement with photon and pion absorption data . \(vi ) the many - body distribution functions are also predicted to be anisotropic . in particular the anisotropies of the @xmath329 , @xmath416 , and @xmath417 distributions in @xmath14he , @xmath15he and @xmath16li are strongly influenced by the toroidal structure of the deuteron . \(vii ) tensor correlations convert @xmath33=1 pairs of nucleons from @xmath1=0 to @xmath1=1 state . this many - body effect reduces the binding energies of nuclei . it does not appear as if many - body effects reduce the magnitude of tensor correlations for the range of nuclei studied here : @xmath4h to @xmath7o . due to the small size of this toroidal structure it may be worthwhile to attempt to understand it from the more basic quark degrees of freedom . within the constituent quark model @xcite it requires a solution of the six - quark problem with a suitably chosen hamiltonian . many attempts have been made ( see refs . @xcite for example ) to calculate the nucleon - nucleon interaction from approximate solutions of the six - quark hamiltonian . a direct coupling of the pions to the quarks is used to obtain the tensor part of the interaction . the toroidal structure is presumably very sensitive to this coupling and to the tensor part of the quark - quark interaction in the framework of the constituent quark model . as is well known , toroidal structure for the ground state of the deuteron was predicted many years ago @xcite using classical skyrme field theory @xcite related to qcd in the @xmath46 limit . in the classical limit one obtains a toroidal shape of @xmath418 fm in size and a binding energy of @xmath419 mev . from fig . [ potfig ] it is obvious that in the classical limit realistic models of nuclear forces would also give a deuteron binding energy in the 100 to 200 mev range . there have been attempts to include quantum corrections to this theory . a recent calculation @xcite obtains an energy of 6.18 mev for the deuteron in this model . ground states of the classical skyrme field with baryon numbers 3 to 6 have also been studied @xcite . the baryon equidensity surfaces of these classical solutions are highly anisotropic . however , the nucleon equidensity surfaces of the @xmath420 and @xmath421 @xmath14he and @xmath15he must be spherically symmetric , thus a direct comparison is not possible . nevertheless the anisotropic @xmath329 and @xmath416 distributions in @xmath14he and @xmath15he may have some relation to the baryon density distributions in the skyrme model . the authors wish to thank brian pudliner for many interesting discussions . wishes to thank a. m. eir for useful discussions on the two - cluster overlap functions . r.b.w . wishes to thank dieter kurath for useful comments on constructing wave functions for @xmath6li . the calculations were made possible by grants of computer time from the mathematics and computer science division of argonne national laboratory , the pittsburgh supercomputing center , the cornell theory center and the national energy research supercomputer center . the work of j.l.f , v.r.p . and a.a . has been partially supported by u.s . national science foundation via grant no . phy94 - 21309 , that of s.c.p . and r.b.w . by the u.s . department of energy , nuclear physics division under contract no . e-31 - 109-eng-38 , that of a.a . by universidade de lisboa , junta de investigao cientfica e tecnolgica under contract no . pbic / c / cen/1108/92 , and that of r.s . by the u.s . department of energy . electronic address : [email protected] electronic address : [email protected] electronic address : [email protected] electronic address : [email protected] electronic address : [email protected] electronic address : [email protected] r. g. arnold _ et al_. , phys . lett . * 35 * , 776 ( 1975 ) . g. g. simon , ch . schmitt , and v. h. walther , nucl . * a364 * , 285 ( 1981 ) . r. cramer _ et al_. , z. phys . c * 29 * , 513 ( 1985 ) . s. platchkov _ et al_. , nucl . phys . * a510 * , 740 ( 1990 ) . s. auffret _ et al_. , phys . lett . * 54 * , 649 ( 1985 ) . r. g. arnold _ et al_. , phys . rev . * 58 * , 1723 ( 1987 ) . m. e. schulze _ et al_. , phys . * 52 * , 597 ( 1984 ) . i. the _ et al_. , phys . rev . * 67 * , 173 ( 1991 ) . v. f. dmitriev _ et al_. , phys . * 157b * , 143 ( 1985 ) . r. gilman _ et al_. , phys . lett . * 65 * , 1733 ( 1990 ) . w. fabian and h. arenhvel , nucl * a258 * , 461 ( 1976 ) . e. hummel and j. a. tjon , phys . lett . * 63 * , 1788 ( 1989 ) . r. schiavilla and d. o. riska , phys . c * 43 * , 437 ( 1991 ) . j. w. van orden , n. devine , and f. gross , phys . lett . * 75 * , 4369 ( 1995 ) . cdddddd @xmath422nucleus & @xmath235 & @xmath423 & @xmath424 & @xmath425 & + & & & & & ip & @xmath30 + @xmath14he & 2.0 & 2.1 & 2.4(1 ) & @xmath172 & 1.5 & 1.49 + @xmath15he & 4.7 & 5.1 & 4.3(6 ) & @xmath174 & 3 . & 2.99 + @xmath16li & 6.3 & 6.3 & & & 5.5 & 5.46 + @xmath259li & 7.2 & 7.8 & & 6.5(5 ) & 6.75 & 6.73 + @xmath7o & 18.8 & 22 & 17(3 ) & 16(3 ) & 30 . & 30.1 + cddddddddd @xmath422nucleus & @xmath426 & & @xmath427 & & @xmath379 & + & & ip & & & ip & & & ip & + @xmath14he & 0.087 & 1.5 & 1.35 & 0.0016 & 0 . & 0.01 & 0.012 & 0 . & 0.14 + @xmath15he & 0.22 & 3 . & 2.5 & 0.0085 & 0 . & 0.01 & 0.060 & 0 . & 0.47 + @xmath16li & 0.24 & 4.5 & 4.0 & 0.061 & 0.5 & 0.52 & 0.104 & 4.5 & 4.96 + @xmath259li & 0.37 & 6.75 & 6.1 & 0.118 & 0.75 & 0.77 & 0.18 & 6.75 & 7.41 + @xmath7o & 1 . & 30 . & 28.5 & 1 . & 6 . & 6.05 & 1 . & 54 . & 55.5 +

the recent results on higgs boson searches from lhc experiments provide significant guidance in exploring the minimal supersymmetric ( susy ) standard model ( mssm ) higgs sector . if we accept the existence of a sm - like higgs boson in the mass window of 123 gev gev as indicated by the observed events , there are two distinct mass regions ( in ) left in the mssm higgs sector : ( a ) the lighter cp - even higgs boson being sm - like and the non - sm - like higgs bosons all heavy and nearly degenerate above 300 gev ( an extended decoupling region ) ; ( b ) the heavier cp - even higgs boson being sm - like and the neutral non - sm - like higgs bosons all nearly degenerate around 100 gev ( a small non - decoupling region ) . on the other hand , due to the strong correlation between the higgs decays to and to predicted in the mssm , the apparent absence of a final state signal is in direct conflict with the peak . if we consider the channel on its own , the absence of the signal would imply that the sm - like higgs boson has reduced coupling to , and that the other non - sm - like higgs bosons should not be too heavy and do not decouple . if both the excess and the absence of a signal continue , new physics beyond the mssm will be required . a similar correlation exists between the and channels : a reduced channel would force the channel to be larger . future searches for the sm - like higgs boson at the lhc will provide critical tests for the mssm prediction . we also study the signals predicted for the non - sm - like higgs bosons and emphasize the potential importance of the electroweak processes , which are independent of the susy parameters except for their masses . in addition , there may be sizable contributions from and in the low - mass non - decoupling region , which may serve to discriminate the model parameters . we allow variations of the relevant susy parameters in a broad range and demonstrate the correlations and constraints on these parameters and associated susy particles .

1203.3207

the outstanding performance of the lhc experiments has led the field of high energy physics into unprecedented territory in the energy and luminosity frontier . major discoveries at the tera - scale are highly anticipated . one of the primary motivations for lhc experiments is the exploration for the mechanism of electroweak symmetry breaking . among the many possibilities for new physics beyond the standard model ( sm ) , supersymmetry ( susy ) provides a natural framework for electroweak symmetry breaking . although the signals for susy are still elusive at the lhc , significant progress has been made in the search for the higgs boson . recently , the atlas and cms collaborations have reported their updated searches for the sm higgs boson @xcite . continuously extending the previous lep2 mass bound for a sm higgs ( 114.4 gev ) @xcite , the lhc search has reached an impressively wide coverage for the mass parameters . the main conclusions are * a sm - like higgs boson was excluded at @xmath11 c.l . in the mass range of @xmath12 gev , in @xmath13 gev @xmath14 gev @xcite and in @xmath15 gev@xmath16 gev @xcite , thus leaving a @xmath11 c.l . mass window @xmath17 * an excess of events above the background expectation was observed in the final state of @xmath1 , at 126 gev with 2.5@xmath18 by the atlas collaboration @xcite and at 125 gev with 2.8@xmath18 by the cms collaboration @xcite , thus giving a tantalizing hint for a higgs boson in the mass range @xmath19 * no significant excess of events above the sm backgrounds was observed in the final states of @xmath20 , however , a small excess has been seen in @xmath21 @xcite . although inconclusive with the current data , each one of the statements above has significant impact on our understanding of electroweak symmetry breaking and thus guides us for the next step of the higgs search . in this paper , we study the consequences of the above findings on the higgs sector within the framework of the minimal supersymmetric standard model ( mssm ) @xcite . we first recollect the existing constraints from all the current bounds of the direct searches from lep2 @xcite , the tevatron @xcite and the lhc @xcite . if we accept the existence of a cp - even higgs boson in the mass range of eq . ( [ eq : mh1 ] ) as observed in the @xmath1 mode , we then find very interesting features for the mssm higgs sector and some other relevant susy parameters . there are two distinctive scenarios , both of which incorporate a sm - like higgs boson . * decoupling " regime with @xmath22 gev @xcite : the light cp - even higgs @xmath23 is in the mass range of eq . ( [ eq : mh1 ] ) and sm - like . the non - sm - like higgs bosons ( heavy cp - even state @xmath24 , cp - odd state @xmath25 and the charged state @xmath26 ) are all heavy and nearly degenerate , with masses around @xmath27 . * non - decoupling " regime with @xmath28 around @xmath29 gev : the heavy cp - even higgs @xmath24 is in the mass range of eq . ( [ eq : mh1 ] ) and sm - like , while the light cp - even higgs @xmath23 is non - sm - like . the masses of the light cp - even higgs and the cp - odd higgs are nearly degenerate while the charged higgs is nearly degenerate with @xmath30 @xcite . each of these two cases predicts unique signatures to establish the nature of the mssm at the lhc . while the current searches continue to improve in the future runs and the standard electroweak production processes @xmath31 are still available , we would like to point out the potential importance of the electroweak processes @xmath32 which are via pure gauge interaction and independent of the susy parameters except for their masses . in addition , there may be sizable contributions from @xmath33 in the low - mass non - decoupling region , which may be used to distinguish the model - parameters . on the other hand , due to the strong positive correlation between the higgs decays to @xmath3 and to @xmath4 predicted in the mssm , the observed @xmath4 signal and the apparent absence of the @xmath3 final state signal near the peak would be mutually exclusive to each other . namely , the suppression of the @xmath3 channel would automatically reduce the @xmath4 channel , in direct conflict with the observed @xmath4 excess . we also found another interesting inverse correlation between the higgs decays to @xmath3 and to @xmath34 . in this case , the suppression to the @xmath3 channel would automatically force the @xmath34 channel to be bigger . if the deficit in the @xmath3 channel persists and the result is strengthened for an extended mass range in the future run at the lhc , it would imply that the sm - like higgs boson has reduced couplings to @xmath35 , rendering it less sm - like . consequently , the other non - sm - like higgs bosons can not be deeply into the decoupling regime , and thus can not be too heavy , typically below 350 gev , making them more accessible at the lhc . moreover , if the excess in the @xmath4 channel and the absence of an excess in the @xmath3 channel continue to be strengthened at the lhc , new physics beyond the mssm will be required could help to enhance the branching fraction of the @xmath36 channel . ] . in the current study , we wish to focus on the essentials of the higgs sector in the mssm and to minimize the effects from other susy sectors @xcite . nevertheless , a few other susy parameters , the higgs mixing @xmath37 , the stop mixing @xmath38 and the stop soft susy masses @xmath39 and @xmath40 , play crucial roles in the higgs sector . we explore the effects of the higgs searches on those susy parameters by scanning them in a wide range and we find clear correlations and thus predictions on them . the rest of the paper is organized as follows . in sec . [ sec : mssmhiggs ] , we give a brief introduction to the mssm higgs sector , focusing on the mass corrections as well as the coupling structures that are relevant for our discussion below . in sec . [ sec : parameterregion ] , we discuss our broad scanning of the relevant mssm parameters by imposing the existing constraints of the direct searches from lep2 , the tevatron and the lhc . we obtain the surviving regions for the higgs mass and the other parameters . with the further improvement expected at the lhc with 8 tev and 14 tev , we discuss the consequence of the sm - like higgs boson searches on the mssm higgs sector in sec . [ sec : future ] . in light of the current direct search , we present the dominant production and decay channels as well as the characteristic channels for the non - sm higgs bosons in sec . [ sec : lhcsearch ] to test the mssm in the future runs . we conclude in sec . [ sec : conclude ] . unlike in the standard model where the higgs mass is a free parameter in the theory , in the mssm with two higgs doublets , the masses of the five physical higgs bosons ( two cp - even higgs bosons @xmath41 and @xmath42 , one cp - odd state @xmath43 and a pair of charged higgs @xmath44 ) at tree level and the mixing angle of the cp - even higgs bosons @xmath45 , can be expressed in terms of two parameters @xcite , conventionally chosen as the mass of @xmath46 and the ratio of the two vacuum expectation values ( @xmath47 ) : @xmath48 we will call the cp - even higgs boson that couples to @xmath49 more strongly the `` standard model - like '' higgs as we discuss it s properties further in the next section . for a low - mass @xmath50 , or a high mass @xmath51 , the higgs boson masses can be approximated by @xmath52 because of the large yukawa coupling of the top quark and the possible large mixing of the left - right top squark , the cp - even higgs boson masses receive significant radiative corrections . for nearly degenerate soft susy breaking parameters in the stop sector : @xmath53 , the correction to the mass of the sm - like higgs can be approximately expressed as is sm - like , this expression also applies to the correction of @xmath54 . ] @xcite @xmath55 + \ldots , \label{eq : deltamh}\end{aligned}\ ] ] where the mixing in the stop sector is given by @xmath56 for @xmath57 , the corrections to the higgs mass from the stop sector is minimized , this is the so - called @xmath58 " scenario @xcite , where the radiative contributions could give rise to a higgs mass as high as 117 gev including a dominant two - loop corrections for a stop mass up to about 2 tev . for @xmath59 , the second term in eq . ( [ eq : deltamh ] ) is maximized , leading to the so - called @xmath60 " scenario @xcite , where a maximum higgs mass of about 127 gev can be reached in such a scenario . to obtain a relatively large correction to the light cp - even higgs mass , relatively heavy stop masses ( at least for one of the stops ) as well as large lr mixing in the stop sector is needed . when two - loop corrections of the oder of @xmath61 are included , there is an asymmetric contribution to the higgs mass from the @xmath62 term , where postitive @xmath62 gives a few gev larger correction compared to the negative @xmath62 case . note that there are uncertainties of a few gev coming from higher loop orders , as well as from the uncertainties in @xmath63 , @xmath64 , etc .. for detailed calculations and results on the higgs mass corrections in the mssm , see refs . @xcite . another important aspect is the couplings of the higgs bosons to the sm particles @xcite . the couplings to gauge bosons behave like @xmath65 where @xmath66 is the weak coupling . either @xmath41 or @xmath42 can be sm - like when it has a stronger coupling to @xmath3 and @xmath67 . in the `` decoupling limit '' @xmath68 , @xmath69 . then @xmath41 is light and sm - like , while all the other higgs bosons are heavy , nearly degenerate , and the @xmath42 coupling to @xmath70 is highly suppressed . in the non - decoupling region @xmath71 , @xmath72 . then @xmath42 is sm - like , while all the other neutral higgs bosons are lighter , nearly degenerate , and the @xmath41 coupling to @xmath3 and @xmath67 are highly suppressed . note that the couplings of the pair of higgs bosons @xmath73 to a gauge boson are of pure gauge coupling strength and are independent of the model parameters . the tree - level couplings of the higgs bosons to the sm fermions scale as @xmath74,\ \quad h^{0}u\bar u : m_{u } [ \sin(\beta-\alpha ) + \cot\beta \cos(\beta-\alpha ) ] , \\ \nonumber & & h^{0 } d\bar d : m_{d } [ \cos(\beta-\alpha ) + \tan\beta \sin(\beta-\alpha)],\ \quad h^{0 } u \bar u : m_{u } [ \cos(\beta-\alpha ) - \cot\beta \sin(\beta-\alpha ) ] , \\ % \nonumber & & a^{0}d \bar d : m_{d}\tan\beta\ \gamma_{5},\quad a^{0}u\bar u : m_{u}\cot\beta\ \gamma_{5 } , \quad h^{\pm}d \bar{u } : m_{d } \tan\beta\ p_{r } + \ m_{u } \cot\beta\ p_{l } , \end{aligned}\ ] ] where @xmath75 are the left- and right - projection operators . in the decoupling limit , these result in the branching fractions for the leading channels , @xmath76 in the non - decoupling limit , the couplings of @xmath24 to the sm fermions become sm - like , while the above branching fraction relations still approximately hold for @xmath23 , @xmath25 and @xmath26 , except that the top quark channel would not be kinematically open . radiative corrections can change the above relations @xcite , in particular for the channels involving @xmath77 and @xmath78 . both the mixing in the higgs sector , as well as the top and bottom yukawa couplings could receive relatively large loop corrections in certain regions of the mssm parameter space . in particular , a large positive mssm correction to @xmath79 , defined as @xcite @xmath80 where @xmath81 is the bottom yukawa coupling , leads to a suppression in @xmath82 decay , resulting in an enhancement in @xmath83 and @xmath67 . for more discussion on this , see sec . [ sec : discussion ] . we wish to examine the theoretical parameter space of the mssm higgs sector as generally as possible . to do so , we study the 6-dimensional parameter space in the ranges @xmath84 the lower limit of @xmath85 is chosen based on the lep2 higgs search exclusion @xcite , while the upper limit takes into account the perturbativity of the bottom yukawa coupling . we limit @xmath28 within 500 gev since it already reaches the decoupling region . a higher value for @xmath28 simply pushes up the nearly degenerate masses for @xmath24 , @xmath25 and @xmath26 while it does not affect the phenomenology of the light cp - even higgs @xmath23 . the ranges of @xmath86 and @xmath37 are motivated by the naturalness consideration , as well as the current collider search limits for susy particles . the range of @xmath62 is chosen to cover both the limiting scenarios of @xmath58 and @xmath60 as mentioned below eq . ( [ eq : at ] ) . it turns out that @xmath38 is of critical importance . it dictates the mixing of the stop sector . in turn , it has significant effects on the radiative corrections to the higgs mass , mixing in the cp - even higgs sector , @xmath87 via stop loops , as well as a contribution to @xmath88 through a chargino - stop loop . indeed , we find that changing the sign of @xmath38 could lead to potentially distinctive results . the effects of the other susy parameters on the higgs sector phenomenology is small . therefore we take the simplified approach in our analyses to decouple their effects by setting the other susy soft mass scales to be 3 tev . some notable effects in special cases will be discussed in sec . [ sec : discussion ] . we used feynhiggs 2.8.6 @xcite to calculate the mass spectrum and other susy parameters , as well as the higgs decay widths and branching fractions and dominant higgs production cross sections . we used higgsbound 3.6.1beta @xcite to check the exclusion constraints from lep2 @xcite , the tevatron @xcite and the lhc @xcite . in practice , we generated a large monte carlo sample to scan over the multiple dimensional parameter region and test against the experimental constraints . for the following presentation , the allowed points ( or regions ) in the plots are indicative of consistent theoretical solutions satisfying experimental constraints , but are not meant to span the complete space of possible solutions . for the light cp even @xmath23 ( green circles ) , and the heavy cp even @xmath24 ( red crosses ) , and the charged @xmath44 ( black pluses ) , scanned over the parameter ranges given in eq . ( [ eq : para ] ) , for ( a ) satisfying the lep2 bounds , and ( b ) further including the bounds from the tevatron and the lhc . , title="fig:",width=302 ] for the light cp even @xmath23 ( green circles ) , and the heavy cp even @xmath24 ( red crosses ) , and the charged @xmath44 ( black pluses ) , scanned over the parameter ranges given in eq . ( [ eq : para ] ) , for ( a ) satisfying the lep2 bounds , and ( b ) further including the bounds from the tevatron and the lhc . , title="fig:",width=302 ] ( b ) for the light cp even @xmath23 ( green circles ) , the heavy cp even @xmath24 ( red crosses ) , including the bounds from the lep2 , the tevatron and the lhc for ( a ) @xmath89 and ( b ) @xmath90 . the horizontal lines mark the mass range of eq . ( [ eq : mh1 ] ) . , title="fig:",width=302 ] ( b ) for the light cp even @xmath23 ( green circles ) , the heavy cp even @xmath24 ( red crosses ) , including the bounds from the lep2 , the tevatron and the lhc for ( a ) @xmath89 and ( b ) @xmath90 . the horizontal lines mark the mass range of eq . ( [ eq : mh1 ] ) . , title="fig:",width=302 ] we first reexamine the higgs boson masses for @xmath23 , @xmath24 , @xmath26 subject to various current constraints from the direct searches . in fig . [ fig : mass](a ) , we present the scanning output which satisfies the lep2 @xcite bounds . the band widths reflect the scanning of the other susy parameters . the lep2 bound sharply cuts off the allowed masses at a little above @xmath91 gev near the kinematic limit for @xmath92 or @xmath93 . figure [ fig : mass](b ) further includes the tevatron @xcite and the most recent lhc bounds c.l . bounds . ] @xcite with the search for the light higgs boson and @xmath94 . we find that although many of the points that passed lep2 are no longer allowed , the result is qualitatively the same in terms of the @xmath95 coverage . figure [ fig : masscloseup](a ) and ( b ) show the allowed mass values of the cp - even higgs bosons in the close - up region as in fig . [ fig : mass](b ) with @xmath89 and @xmath90 . the horizontal lines mark the mass range of eq . ( [ eq : mh1 ] ) . we see the subtle difference between the signs of @xmath38 , for which @xmath90 yields more accessible solutions especially for a heavier @xmath96 due to two - loop radiative corrections . versus @xmath2 for ( a ) the @xmath97 final state with @xmath41 ( green circles ) and @xmath42 ( red crosses ) , ( b ) the @xmath4 final state , and the branching fraction correlation ( @xmath98 ) for ( c ) @xmath99 versus @xmath100 and for ( d ) @xmath101 versus @xmath100 . all the lep2 and hadron collider direct search bounds are imposed . the black dots in all the panels represent those satisfying the narrower higgs mass window in eq . ( [ eq : mh1 ] ) . the light blue triangles are those satisfying the cross section requirement eq . ( [ eq : sigma ] ) . other parameters are scanned over the range in eq . ( [ eq : para ] ) with @xmath102 . , title="fig:",width=302 ] versus @xmath2 for ( a ) the @xmath97 final state with @xmath41 ( green circles ) and @xmath42 ( red crosses ) , ( b ) the @xmath4 final state , and the branching fraction correlation ( @xmath98 ) for ( c ) @xmath99 versus @xmath100 and for ( d ) @xmath101 versus @xmath100 . all the lep2 and hadron collider direct search bounds are imposed . the black dots in all the panels represent those satisfying the narrower higgs mass window in eq . ( [ eq : mh1 ] ) . the light blue triangles are those satisfying the cross section requirement eq . ( [ eq : sigma ] ) . other parameters are scanned over the range in eq . ( [ eq : para ] ) with @xmath102 . , title="fig:",width=302 ] versus @xmath2 for ( a ) the @xmath97 final state with @xmath41 ( green circles ) and @xmath42 ( red crosses ) , ( b ) the @xmath4 final state , and the branching fraction correlation ( @xmath98 ) for ( c ) @xmath99 versus @xmath100 and for ( d ) @xmath101 versus @xmath100 . all the lep2 and hadron collider direct search bounds are imposed . the black dots in all the panels represent those satisfying the narrower higgs mass window in eq . ( [ eq : mh1 ] ) . the light blue triangles are those satisfying the cross section requirement eq . ( [ eq : sigma ] ) . other parameters are scanned over the range in eq . ( [ eq : para ] ) with @xmath102 . , title="fig:",width=302 ] versus @xmath2 for ( a ) the @xmath97 final state with @xmath41 ( green circles ) and @xmath42 ( red crosses ) , ( b ) the @xmath4 final state , and the branching fraction correlation ( @xmath98 ) for ( c ) @xmath99 versus @xmath100 and for ( d ) @xmath101 versus @xmath100 . all the lep2 and hadron collider direct search bounds are imposed . the black dots in all the panels represent those satisfying the narrower higgs mass window in eq . ( [ eq : mh1 ] ) . the light blue triangles are those satisfying the cross section requirement eq . ( [ eq : sigma ] ) . other parameters are scanned over the range in eq . ( [ eq : para ] ) with @xmath102 . , title="fig:",width=302 ] versus @xmath96 for ( a ) @xmath3 and ( b ) @xmath34 final state , with the lep2 and hadron collider direct search bounds except the latest bounds on @xmath6 and @xmath7 channel from atlas searches @xcite . the upper solid curve in each panel is from the current @xmath11 c.l . bound @xcite . the two lower curves indicate the estimated improvements at a 8 tev and 14 tev lhc ( see sec . [ sec : future ] ) . the vertical bands indicate the narrow mass window in eq . ( [ eq : mh1 ] ) . other parameters are scanned over the range in eq . ( [ eq : para ] ) with @xmath102 . , title="fig:",width=302 ] versus @xmath96 for ( a ) @xmath3 and ( b ) @xmath34 final state , with the lep2 and hadron collider direct search bounds except the latest bounds on @xmath6 and @xmath7 channel from atlas searches @xcite . the upper solid curve in each panel is from the current @xmath11 c.l . bound @xcite . the two lower curves indicate the estimated improvements at a 8 tev and 14 tev lhc ( see sec . [ sec : future ] ) . the vertical bands indicate the narrow mass window in eq . ( [ eq : mh1 ] ) . other parameters are scanned over the range in eq . ( [ eq : para ] ) with @xmath102 . , title="fig:",width=302 ] we next calculate the cp - even susy higgs production cross section for the channels @xmath103 let us consider the cp - even higgs bosons @xmath41 and @xmath24 after passing both the lep2 and the hadron collider bounds . figure [ fig : csr ] presents the ratios of the mssm cross sections to the sm values versus @xmath2 for ( a ) @xmath3 and ( b ) @xmath4 final states , with green circles for @xmath41 and red crosses for @xmath42 . the result for the @xmath67 channel is very similar to the @xmath3 channel due to the su(2 ) symmetry . for a sm - like higgs boson , the higgs-@xmath104 coupling is the main source for both the @xmath104 and @xmath1 decay channels . in the sm , the ratio at @xmath105 gev is fixed as br@xmath106br@xmath107 . in the mssm even with our broad parameter scan , there is a strong correlation . this is shown in fig . [ fig : csr](c ) for br@xmath108 versus br@xmath109 . we see an empirical linear relation @xmath110 the smaller - than - unity prefactor is due to some level of cancellation in the loops of @xmath111 . in fig . [ fig : csr](d ) , we show another correlation for the channels of @xmath34 and @xmath3 . the sm prediction is at a value br@xmath106br@xmath112 at 125 gev . it is interesting to note that they are `` anticorrelated '' . thus a consistency check of the predicted correlations as shown in fig . [ fig : csr ] could provide crucial information regarding the underlying theory . in fig . [ fig : csrmh ] , we show the cross section ratios @xmath113 versus @xmath96 for ( a ) @xmath3 and ( b ) @xmath34 final states , with the lep2 and hadron collider direct search bounds except the latest bounds on @xmath6 and @xmath7 channel from atlas searches @xcite . the solid curve in each panel is from the current @xmath11 c.l . bound @xcite . the vertical bands indicate the narrow mass window in eq . ( [ eq : mh1 ] ) . we see that the recent atlas bounds from those channels alone are not strong enough to have a direct impact on the existing bounds , leaving solutions with a factor of 1.5 larger than the sm predictions . given the tantalizing hint for the @xmath4 events near 125 gev , we take an important step to assume the existence of a cp - even higgs boson @xmath114 the mass window requirement in eq . ( [ eq : mass ] ) yields a very selective parameter region as indicated by the black dots in the panels in fig . [ fig : csr ] . the simultaneous requirement of the sizable cross section for the @xmath4 mode forces @xmath2 into two distinct and separate regions , as seen from the light blue triangles above the dashed horizontal line in fig . [ fig : csr ] ( b ) . the bulk region of the allowed parameter space is pushed to heavy @xmath2 ( roughly @xmath115 gev ) , the `` decoupling region '' with the light cp - even higgs being sm - like . there is , however , a small region at lower @xmath2 that survives in the `` non - decoupling '' region ( roughly @xmath116 ) with the heavy cp - even higgs being sm - like @xcite . the non - decoupling region , which satisfies both the mass and the cross section requirement as in eqs . ( [ eq : mass ] ) and ( [ eq : sigma ] ) , occurs mainly for @xmath90 . this is because a suppression of @xmath117 is needed in order for @xmath118 to be above 0.8 of the sm value . such a suppression could be due to a large positive radiative correction to the bottom yukawa , @xmath79 ( as defined in eq . ( [ eq : deltamb ] ) ) , as well as a small @xmath119 , where @xmath120 is the cp - even higgs mixing parameter @xmath45 with radiative corrections . both could be realized in the positive @xmath62 case , where @xmath79 is always positive , and @xmath119 could be as small as zero , while keeping @xmath62 large enough to satisfy the mass region in eq . ( [ eq : mass ] ) . for negative @xmath62 , due to the cancellation between the sbottom - gluino loop ( @xmath121 ) and the stop - higgsino loop ( @xmath122 ) , a small @xmath123 is preferred to obtain a positive @xmath79 . in addition , to get a small value for @xmath119 also requires a relatively small @xmath123 . the radiative correction to the higgs mass , however , is small for such a small value of @xmath123 , leading to a strong tension between the higgs mass requirement in eq . ( [ eq : mass ] ) and the cross section requirement in eq . ( [ eq : sigma ] ) for @xmath89 . we summarize these two distinctive regions as @xmath124 the non - decoupling region is of great interest both in terms of the theoretical implication and the lhc searches . versus @xmath2 for ( a ) @xmath125 and for ( b ) @xmath102 , respectively . the region with purple diamonds satisfies all the lep2 , tevatron and the lhc direct search constraints . the black dots represent those in the narrow mass window in eq . ( [ eq : mass ] ) . the light blue triangles are those satisfying the cross section requirement eq . ( [ eq : sigma ] ) . other parameters are scanned over the range in eq . ( [ eq : para ] ) . , title="fig:",width=302 ] versus @xmath2 for ( a ) @xmath125 and for ( b ) @xmath102 , respectively . the region with purple diamonds satisfies all the lep2 , tevatron and the lhc direct search constraints . the black dots represent those in the narrow mass window in eq . ( [ eq : mass ] ) . the light blue triangles are those satisfying the cross section requirement eq . ( [ eq : sigma ] ) . other parameters are scanned over the range in eq . ( [ eq : para ] ) . , title="fig:",width=302 ] it turns out that the above constraints have significant implication for the other susy parameters associated with the higgs sector . we first examine the allowed region of @xmath85 versus @xmath2 . we present the region for @xmath89 in fig . [ fig : tanb](a ) , and for @xmath90 in fig . [ fig : tanb](b ) . not shown in the figures are the regions allowed by lep2 alone , which are uniformly from @xmath126 gev and on . the bounds from the hadron colliders ( purple diamonds ) remove the region of low @xmath28 and high @xmath85 . this is largely due to the searches for @xmath127 @xcite , as well as @xmath128 @xcite . the final requirements for the existence of a sm - like higgs as in eqs . ( [ eq : mass ] ) and ( [ eq : sigma ] ) once again highly limit the parameter space ( black dots and light blue triangles , respectively ) . requiring the existence of a sm - like higgs in the mass range of 123 @xmath0 127 gev results in @xmath129 gev for @xmath89 and @xmath22 gev for @xmath90 . versus @xmath2 . the legends are the same as in fig . [ fig : tanb ] . , title="fig:",width=302 ] versus @xmath2 . the legends are the same as in fig . [ fig : tanb ] . , title="fig:",width=302 ] the higgs mixing parameter @xmath37 plays an important role for radiative corrections to the higgs production and decay channels and we vary it in the range of eq . ( [ eq : para ] ) . we show the impact on this parameter in fig . [ fig : mu ] , where the legends are the same as in fig . [ fig : tanb ] . we note the interesting correlation in the decoupling region for @xmath102 once we impose the cross section requirement as in eq . ( [ eq : sigma ] ) ( regions indicated by light blue triangles ) that a lower value of @xmath2 results in a higher @xmath37 . this is because a smaller @xmath37 leads to a suppressed @xmath131 and is , therefore , disfavored @xcite . versus @xmath2 and ( b ) @xmath38 versus @xmath39 . the legends are the same as in fig . [ fig : tanb ] . , title="fig:",width=302 ] versus @xmath2 and ( b ) @xmath38 versus @xmath39 . the legends are the same as in fig . [ fig : tanb ] . , title="fig:",width=302 ] the next susy parameter relevant to the higgs sector is @xmath38 ( see eq . ( [ eq : at ] ) ) and we vary it in the range of eq . ( [ eq : para ] ) . we show the effect on this parameter in fig . [ fig : at](a ) , with the legends the same as in fig . [ fig : tanb ] . the smaller @xmath133 region is disfavored due to the smallness of @xmath96 , while the large @xmath133 region is removed by demanding sizable @xmath134 cross sections @xcite . such correlation of @xmath62 with @xmath28 is more pronounced for the negative @xmath62 case . similar effects were already observed earlier in @xcite . in fig . [ fig : at](b ) , we present the allowed region in the plane of the soft susy breaking scale @xmath39 and @xmath62 , with the legends the same as in fig . [ fig : tanb ] . the behavior for @xmath40 is very similar . an approximate @xmath135 relation of @xmath136 and/or large @xmath39 , @xmath40 are needed to have a relatively heavy higgs mass in the range of 123 to 127 gev @xcite . imposing the cross section requirement of eq . ( [ eq : sigma ] ) further narrows down the range of @xmath62 . in particular , for the negative @xmath62 case , @xmath62 is typically in the narrow range from @xmath137 to @xmath138 gev , while for the positive @xmath62 case , the allowed region is much broader , from 1000 gev and higher . the difference between positive and negative @xmath62 is mainly due to the difference in the radiative correction to @xmath79 from the stop sector @xcite . versus @xmath2 and ( b ) @xmath37 versus @xmath2 in the non - decoupling region for @xmath90 . the legends are the same as in fig . [ fig : tanb ] . , title="fig:",width=302 ] versus @xmath2 and ( b ) @xmath37 versus @xmath2 in the non - decoupling region for @xmath90 . the legends are the same as in fig . [ fig : tanb ] . , title="fig:",width=302 ] as discussed earlier , the non - decoupling region mainly appears when @xmath90 . in fig . [ fig : non - decoupling ] , we zoom into the non - decoupling region and impose the mass and cross section requirements as in eqs . ( [ eq : mass ] ) and ( [ eq : sigma ] ) . panel ( a ) shows that only a narrow region of @xmath139 can accommodate a sm - like heavy cp - even higgs in the mass range of 123 @xmath0 127 gev @xcite . panel ( b ) shows that a higher value of @xmath37 is preferred for larger @xmath28 after imposing the cross section requirement . this is because a large @xmath37 leads to a larger positive @xmath79 , resulting in a more suppressed @xmath140 and a more enhanced @xmath141 . the surviving region in @xmath62 versus @xmath2 and @xmath62 versus @xmath39 are similar to the decoupling case . in our study , we scanned over the six parameters , @xmath28 , @xmath85 , @xmath37 , @xmath39 , @xmath40 and @xmath62 , which are the parameters most relevant to the higgs sector phenomenology . the other mssm sectors , _ i.e. _ , sbottoms , staus etc . , could also contribute to the higgs sector , radiatively , as we briefly summarize below . most of our discussion applies to the sm - like higgs boson being either @xmath23 in the decoupling region or @xmath24 in the non - decoupling region . versus @xmath142 for ( a ) @xmath89 and ( b ) @xmath90 . the legends are the same as in fig . [ fig : tanb ] . , title="fig:",width=302 ] versus @xmath142 for ( a ) @xmath89 and ( b ) @xmath90 . the legends are the same as in fig . [ fig : tanb ] . , title="fig:",width=302 ] as seen in eq . ( [ eq : deltamh ] ) , the stop sector provides substantial radiative corrections to the higgs mass . large contributions from the stop sector need a relatively large @xmath62 term and at least one of the stop mass parameters ( @xmath39 or @xmath40 ) to be large . in particular , when we restrict the higgs mass to the narrow window of @xmath143 gev , the mass splitting between the two stop mass eigenstates is found to be at least 200 gev ( 300 gev ) for @xmath144 ) . although one of the stops can still be as light as 100 @xmath0 200 gev @xcite , the lighter the stop mass is , the larger the mass split would have to be , as seen from fig . [ fig : stopmass ] . the collider phenomenology of the light stop @xmath145 ( as well as @xmath146 when it is within collider reach ) depends on the stop mixing angle @xmath147 and on the spectrum of gauginos , which is under current investigation @xcite . another way to reach a large positive correction to @xmath148 is to allow extremely heavy stop masses , which we did not explore . even when the stop masses are pushed up to 5@xmath010 tev , we could barely obtain a sm - like higgs boson with a mass of around 125 gev for @xmath149 . such a heavy stop mass would suffer from a severe fine - tuning problem , unless we envision the focus point scenario @xcite . note that there could be negative contributions to the higgs mass from the sbottom and stau sectors when those states are light . the mixing parameter takes the form @xmath150 . sizable corrections could be obtained for large @xmath85 and large @xmath37 , with @xmath151 ( for sbottom contribution ) and @xmath152 ( for stau contribution ) @xcite . observation of the processes @xmath154 is of the utmost importance for the discovery and determination of the properties of the higgs boson . as we discussed above , in the decoupling region ( @xmath155 gev ) @xmath23 is sm - like and all the partial widths @xmath156 and @xmath67 are typically slightly suppressed compared to the sm values , while they are highly correlated in the generic mssm sector . however , there are certain mssm parameter regions where br@xmath157 are not suppressed and even enhanced , and the predicted correlation is modified . given the dominant decay of @xmath158 , a suppression of the @xmath159 coupling leads to the enhancement of the decay branching fractions of all three channels . there are two ways to suppress the @xmath159 coupling , either through the higgs mixing effects in the cp - even higgs sector , or through the suppression of the bottom yukawa coupling via susy radiative corrections . the former is referred to as the small @xmath120 region " in the literature @xcite . when the loop correction from the stop , sbottom , or stau sector to @xmath160 is large and positive , the higgs mixing angle @xmath120 is small , leading to a suppressed @xmath159 coupling , which is proportional to @xmath161 . such a region typically appears for moderate to large @xmath85 , small to moderate @xmath28 , light stop , sbottom , stau masses , as well as large @xmath62 , @xmath162 , @xmath163 and @xmath37 @xcite . the bottom yukawa could also receive large radiative corrections in the mssm , which can either be enhanced or suppressed compared to its tree - level value @xcite . in particular , strong suppression of the bottom yukawa could be achieved for a large and positive value of @xmath164 @xcite . while the partial decay width for @xmath165 ( which are @xmath166 ) are typically suppressed in the mssm compared to the sm values , loop induced decay of @xmath167 , on the other hand , could be enhanced with stop , sbottom , or stau contributions with large left - right mixing and small sparticle masses . for light stop and light sbottom , however , the simultaneous suppression of the production channel @xmath168 results in an overall suppression of @xmath131 . stau , on the other hand , does not lead to the suppression of @xmath168 . for stau mass around 300 gev with large @xmath85 and @xmath163 , an enhancement of @xmath131 as large as a factor of 2 is possible @xcite . as noted above , the stop left - right mixing @xmath38 is of critical importance since it has multiple roles here . first , it affects the correction to the higgs mass with positive @xmath62 and gives a larger correction compared to the case of negative @xmath62 , due to a two - loop contribution with gluino and stops . @xmath62 could also affect the mixing in the cp - even higgs sector , bottom yukawa , higgs coupling to @xmath1 , as well as the production of @xmath168 . third , the sign of @xmath62 also changes the sign of the chargino contribution to @xmath169 , as discussed below . note that similar effects could also occur in the non - decoupling region with @xmath24 being the sm - like higgs . our discussion above is still valid with the substitution of @xmath23 by @xmath24 , @xmath170 by @xmath171 and @xmath161 by @xmath172 . the dominant indirect constraints on a light higgs sector comes from @xmath173 . the current observed value is @xmath174 @xcite and the next - to - next - to - leading order qcd correction gives @xmath175 @xcite . there are two dominant mssm contributions , namely , charged higgs - top loop corrections and chargino - stop loop corrections . while charged higgs loops always give positive contributions , contributions from the chargino loops depend on the signs of @xmath176 , @xmath37 , and @xmath62 @xcite . in particular , the contribution from the higgsino - stop loop that is proportional to the left - right mixing in the stop sector gives a negative contribution for @xmath177 and a positive contribution for @xmath178 . for our choice of @xmath179 , @xmath180 , the rest of the chargino loop contributions typically provides a negative correction to @xmath173 . in the nondecoupling region with small @xmath28 , significant negative contributions from chargino loops are needed to cancel the charged higgs contribution , which typically requires a small @xmath176 . in the decoupling region where the charged higgs contribution is negligible , given that the current sm prediction is lower than the experimental value , a positive @xmath62 is slightly preferred so that the mssm corrections do not make the deviation worse . , ( c ) and ( d ) in the non - decoupling region for @xmath42 . , title="fig:",width=302 ] , ( c ) and ( d ) in the non - decoupling region for @xmath42 . , title="fig:",width=302 ] , ( c ) and ( d ) in the non - decoupling region for @xmath42 . , title="fig:",width=302 ] , ( c ) and ( d ) in the non - decoupling region for @xmath42 . , title="fig:",width=302 ] in anticipation of the successful operation at the energies of 8 and 14 tev , the lhc will deliver a large amount of quality data in the years to come . if a signal for a sm - like higgs boson is confirmed , then the task would be to determine its basic properties to good precision @xcite . on the other hand , if the signal for a sm - like higgs boson continues to be elusive , it would provide further important information about the mssm higgs sector . we first reiterate the production and decay of a sm - like higgs boson in the mssm at the lhc . for the convenience of future discussions , we divide the @xmath2 mass parameter into two regions , @xmath181 in fig . [ fig : cs_h ] , the total cross sections ( left panels ) at 14 tev and decay branching fractions ( right panels ) for the leading channels of the sm - like higgs boson are shown after passing all the constraints , ( a)-(b ) in the decoupling region for the sm - like @xmath41 , and ( c)-(d ) in the non - decoupling region for the sm - like @xmath42 . as before , other parameters in the mssm are scanned over the range in eq . ( [ eq : para ] ) . the leading production channel is via @xmath182 fusion and of a rate at the order of 50 pb @xcite @xmath183 the @xmath184 initial process is known to be small in the sm at the order of 0.6 pb for a 125 gev mass at 14 tev @xcite , but it could be significantly enhanced in certain susy parameter region especially at large @xmath185 @xcite . this is seen in the plot by the large spread in figs . [ fig : cs_h](a ) and ( c ) . the electroweak processes of vector - boson - fusion and higgs - strahlung are the next important sources for the sm - like higgs boson production @xmath186 which are roughly in the range of @xmath187 pb . for those production channels that do not involve heavy quarks , the cross section rates are well predicted as seen from the narrow bands . for the branching fractions , the @xmath188 modes are stable due to the cancellation of a common factor @xmath185 in the ratios , while all other modes result in a large spread . because of the nature of the sm - like higgs boson , in either the decoupling or the non - decoupling region , the cross sections of @xmath41 or @xmath42 behave similarly . versus @xmath2 with the 8 tev improvement of sensitivity at the lhc for ( a ) the @xmath6 channel , and for ( b ) the @xmath4 channel . the legends are the same as in fig . [ fig : csr ] . other parameters are scanned over the range in eq . ( [ eq : para ] ) . , title="fig:",width=302 ] versus @xmath2 with the 8 tev improvement of sensitivity at the lhc for ( a ) the @xmath6 channel , and for ( b ) the @xmath4 channel . the legends are the same as in fig . [ fig : csr ] . other parameters are scanned over the range in eq . ( [ eq : para ] ) . , title="fig:",width=302 ] we now consider improved measurements for the search for the sm - like higgs boson and see the implication for the higgs sector of the mssm . without going through detailed signal and background simulations , we simply assume the future data collection as in table [ tab : improve ] . the signal sensitivity improvements are scaled with @xmath189 where @xmath190 is the total cross section ( we use @xmath191 as an illustration . ) for sm higgs boson production @xcite , and @xmath192 is the integrated luminosity . .statistical improvement factors for the sm - like higgs boson search with @xmath193 at the different energies of the lhc and with different luminosity assumption . [ cols="^,^,^,^",options="header " , ] in light of the powerful results presented by atlas and cms for the sm higgs boson searches at the lhc , along with the data from the lep2 and tevatron , we reexamined the mssm higgs sector for their masses , couplings and other related susy parameters . instead of only presenting benchmark scenarios , we allowed variations of other susy parameters in a broad range . if we accept the existence of a sm - like higgs boson in the mass window of 123 gev@xmath0127 gev as indicated by the observed @xmath1 events , we found that there are two distinctive mass regions left in the mssm higgs sector : ( a ) the lighter cp - even higgs boson being sm - like and the non - sm - like higgs bosons all heavy and nearly degenerate above 300 gev ( an extended decoupling region ) ; ( b ) the heavier cp - even higgs boson being sm - like and the neutral non - sm - like higgs bosons all nearly degenerate around 100 gev ( a small non - decoupling region ) . these features were shown in figs . [ fig : csr](a ) and [ fig : csr](b ) . on the other hand , due to the strong positive correlation between the higgs decays to @xmath3 and to @xmath4 predicted in the mssm , as seen in fig . [ fig : csr](c ) and eq . ( [ eq : r ] ) , the observed @xmath4 signal and the apparent absence of the @xmath3 final state signal near the peak would be mutually exclusive to each other . namely , the suppression to the @xmath3 channel would automatically reduce the @xmath4 channel . in fact , the theoretical expectation for the @xmath4 signal in the mssm relative to that in the sm is even smaller than that for the @xmath3 channel [ @xmath194 eq . ( [ eq : r ] ) ] . to accommodate both the @xmath3 deficit and the @xmath4 enhancement , physics beyond the mssm would be needed . we also found another interesting inverse correlation between the higgs decays to @xmath3 and to @xmath34 , as seen in fig . [ fig : csr](d ) . the suppression to the @xmath3 channel would automatically force the @xmath34 channel to be larger . if the absence of the @xmath3 signal persists and the observation is strengthened for an extended mass range in the future run at the lhc , it would imply that the sm - like higgs boson has reduced couplings to @xmath35 , rendering it less sm - like . although less statistically significant , the lack of the @xmath34 final state signal could also contribute to reach a consistent picture . consequently , the other non - sm - like higgs bosons can not be deeply into the decoupling regime , and thus can not be too heavy , typically below 350 gev , making them more accessible at the lhc . future searches for the sm - like higgs boson at the lhc will provide critical tests for the mssm predictions for those points , as presented in sec . [ sec : future ] . guided by those observations , we studied the signals predicted for the non - sm - like higgs bosons satisfying the current bounds . along with the standard searching processes @xmath195 as shown in fig . [ fig : cs_h](c ) , we emphasize the potential importance of the electroweak processes @xmath8 in fig . [ fig : nondec14tev](e ) , which are independent of the susy parameters except for their masses . in addition , there may be sizable contributions from @xmath9 in the low - mass non - decoupling region , which may serve to discriminate the model parameters . these cross sections can be as large as that of the @xmath196 associated production , which sensitively depends on @xmath85 . the stringent constraints also imply nontrivial correlation and prediction to some other susy parameters relevant to the higgs sector , such as @xmath197 etc . further explorations may lead to predictions for other susy signals for gaugino and stops . over all , the search for the sm higgs boson will prove crucial in understanding the susy higgs sector . we would like to thank s. heinemeyer , t. hahn , c. wagner and h. baer for helpful discussions . the work of n.c . and t.h . was supported in part by pitt pacc , and in part by the lhc - ti under u.s . national science foundation , grant nsf - phy-0705682 . the work of s.s . was supported by the department of energy under grant de - fg02 - 04er-41298 . j. f. gunion , h. e. haber , g. l. kane and s. dawson , front . * 80 * , 1 ( 2000 ) ; j. f. gunion and h. e. haber , nucl . b * 272 * , 1 ( 1986 ) [ erratum - ibid . b * 402 * , 567 ( 1993 ) ] . a. djouadi , phys . rept . * 459 * , 1 ( 2008 ) [ hep - ph/0503173 ] . t. aaltonen _ et al . _ [ cdf collaboration ] , phys . lett . * 103 * , 101803 ( 2009 ) [ arxiv:0907.1269 [ hep - ex ] ] ; v. m. abazov _ et al . _ [ d0 collaboration ] , phys . b * 682 * , 278 ( 2009 ) [ arxiv:0908.1811 [ hep - ex ] ] . cms collaboration , cms - pas - hig-11 - 009 ; g. aad _ et al . _ [ atlas collaboration ] , phys . b * 705 * , 174 ( 2011 ) [ arxiv:1107.5003 [ hep - ex ] ] . cms collaboration on @xmath198 , cms cms - pas - hig-11 - 029 ( 2011 ) . h. e. haber , hep - ph/9505240 . gerard and m. herquet , phys . lett . * 98 * , 251802 ( 2007 ) [ hep - ph/0703051 [ hep - ph ] ] ; e. cervero and j. -m . gerard , arxiv:1202.1973 [ hep - ph ] . m. carena , s. gori , n. r. shah and c. e. m. wagner , arxiv:1112.3336 [ hep - ph ] . some recent papers dealt with the other mssm aspects in light of the recent higgs searches , see , _ e.g. _ h. baer , v. barger and a. mustafayev , arxiv:1112.3017 ; t. moroi and k. nakayama , arxiv:1112.3123 ; o. buchmueller _ et al . _ , arxiv:1112.3564 ; s. akula , b. altunkaynak , d. feldman , p. nath and g. peim , arxiv:1112.3645 ; m. kadastik , k. kannike , a. racioppi and m. raidal , arxiv:1112.3647 ; j. cao , z. heng , d. li and j. m. yang , arxiv:1112.4391 [ hep - ph ] ; p. draper , p. meade , m. reece and d. shih , arxiv:1112.3068 [ hep - ph ] ; n. desai , b. mukhopadhyaya and s. niyogi , arxiv:1202.5190 [ hep - ph ] ; h. baer , v. barger and a. mustafayev , arxiv:1202.4038 [ hep - ph ] . j. l. feng , k. t. matchev and d. sanford , arxiv:1112.3021 . m. s. carena , m. quiros and c. e. m. wagner , nucl . b * 461 * , 407 ( 1996 ) [ arxiv : hep - ph/9508343 ] . m. s. carena , j. r. espinosa , m. quiros and c. e. m. wagner , phys . b * 355 * , 209 ( 1995 ) [ arxiv : hep - ph/9504316 ] . m. s. carena , s. heinemeyer , c. e. m. wagner and g. weiglein , eur . j. c * 26 * , 601 ( 2003 ) [ arxiv : hep - ph/0202167 ] . g. degrassi , s. heinemeyer , w. hollik , p. slavich and g. weiglein , eur . j. c * 28 * , 133 ( 2003 ) [ hep - ph/0212020 ] . s. heinemeyer , w. hollik and g. weiglein , eur . j. c * 9 * , 343 ( 1999 ) [ hep - ph/9812472 ] . m. s. carena , s. mrenna and c. e. m. wagner , phys . d * 62 * , 055008 ( 2000 ) [ arxiv : hep - ph/9907422 ] . m. s. carena , s. mrenna and c. e. m. wagner , phys . d * 60 * , 075010 ( 1999 ) [ arxiv : hep - ph/9808312 ] . l. j. hall , r. rattazzi and u. sarid , phys . d * 50 * , 7048 ( 1994 ) [ arxiv : hep - ph/9306309 ] . r. hempfling , phys . d * 49 * , 6168 ( 1994 ) . m. s. carena , m. olechowski , s. pokorski and c. e. m. wagner , nucl . b * 426 * , 269 ( 1994 ) [ arxiv : hep - ph/9402253 ] . j. a. bagger , k. t. matchev , d. m. pierce and r. j. zhang , phys . * 78 * , 1002 ( 1997 ) [ erratum - ibid . * 78 * , 2497 ( 1997 ) ] [ arxiv : hep - ph/9611229 ] . m. frank , t. hahn , s. heinemeyer , w. hollik , h. rzehak and g. weiglein , jhep * 0702 * , 047 ( 2007 ) [ hep - ph/0611326 ] , and references therein . s. heinemeyer , w. hollik and g. weiglein , comput . commun . * 124 * , 76 ( 2000 ) [ hep - ph/9812320 ] , and references therein . p. bechtle , o. brein , s. heinemeyer , g. weiglein and k. e. williams , comput . commun . * 181 * , 138 ( 2010 ) [ arxiv:0811.4169 [ hep - ph ] ] , and references therein . p. bechtle , o. brein , s. heinemeyer , g. weiglein and k. e. williams , comput . commun . * 182 * , 2605 ( 2011 ) [ arxiv:1102.1898 [ hep - ph ] ] , and references therein . s. heinemeyer , o. stal and g. weiglein , arxiv:1112.3026 . i. low and s. shalgar , jhep * 0904 * , 091 ( 2009 ) [ arxiv:0901.0266 [ hep - ph ] ] . j. cao , z. heng , j. m. yang , y. zhang and j. zhu , arxiv:1202.5821 [ hep - ph ] . d. asner _ [ heavy flavor averaging group ] , arxiv:1010.1589 [ hep - ex ] . m. misiak _ et al . _ , phys . lett . * 98 * , 022002 ( 2007 ) [ arxiv : hep - ph/0609232 ] . m. misiak and m. steinhauser , nucl . b * 764 * , 62 ( 2007 ) [ arxiv : hep - ph/0609241 ] . r. barbieri and g. f. giudice , phys . b * 309 * , 86 ( 1993 ) [ arxiv : hep - ph/9303270 ] . m. duhrssen , s. heinemeyer , h. logan , d. rainwater , g. weiglein and d. zeppenfeld , phys . d * 70 * , 113009 ( 2004 ) [ hep - ph/0406323 ] . s. dittmaier _ et al . _ [ lhc higgs cross section working group collaboration ] , arxiv:1101.0593 [ hep - ph ] . f. maltoni , z. sullivan and s. willenbrock , phys . d * 67 * , 093005 ( 2003 ) [ hep - ph/0301033 ] . j. l. diaz - cruz , h .- j . he , t. m. p. tait and c. p. yuan , phys . lett . * 80 * , 4641 ( 1998 ) [ hep - ph/9802294 ] . a. pukhov , e. boos , m. dubinin , v. edneral , v. ilyin , d. kovalenko , a. kryukov and v. savrin _ et al . _ , hep - ph/9908288 , a. belyaev , n. d. christensen and a. pukhov , arxiv:1207.6082 [ hep - ph ] . s. dawson , s. dittmaier and m. spira , phys . d * 58 * , 115012 ( 1998 ) [ arxiv : hep - ph/9805244 ] . atlas collaboration , atlas - conf-2012 - 014 ( 2012 ) . cao , s. kanemura and c .- p . yuan , phys . d * 69 * , 075008 ( 2004 ) [ hep - ph/0311083 ] . a. belyaev , q .- h . cao , d. nomura , k. tobe and c .- p . yuan , phys . lett . * 100 * , 061801 ( 2008 ) [ hep - ph/0609079 ] . altas collaboration , expected performance of the atlas experiment : detector , trigger and physics , cern - open-2008 - 020 . m. hashemi , s. heinemeyer , r. kinnunen , a. nikitenko and g. weiglein , arxiv:0804.1228 [ hep - ph ] . s. gennai , s. heinemeyer , a. kalinowski , r. kinnunen , s. lehti , a. nikitenko and g. weiglein , eur . j. c * 52 * , 383 ( 2007 ) [ arxiv:0704.0619 [ hep - ph ] ] .

we consider a collinear effective theory of highly energetic quarks with energy , interacting with collinear and soft gluons by integrating out collinear degrees of freedom to subleading order . the collinear effective theory offers a systematic expansion in power series of a small parameter , where is the transverse momentum of a collinear particle . we construct the effective lagrangian to first order in , and discuss its features including additional symmetries such as collinear gauge invariance and reparameterization invariance . heavy - light currents can be matched from the full theory onto the operators in the collinear effective theory at one loop and to order . we obtain heavy - light current operators in the effective theory , calculate their wilson coefficients at this order , and the renormalization group equations for the wilson coefficients are solved . as an application , we calculate the form factors for decays of mesons to light energetic mesons to order and at leading - logarithmic order in .

hep-ph0201197

when a @xmath4 meson decays into light mesons , we can explore different kinematic regions depending on the momenta carried by the light mesons . when a light meson is emitted from a heavy quark with momentum of order @xmath6 , this decay can be successfully described by the heavy quark effective theory ( hqet ) @xcite . the momentum of a heavy quark can be decomposed as @xmath7 , where @xmath8 is the residual momentum of order @xmath6 . the leading contribution to the decay corresponds to the partonic result , and the corrections can be systematically expanded in power series of @xmath9 and @xmath5 . inclusive decays of heavy mesons with large momentum transfer can be treated in the hqet with the operator product expansion @xcite . exclusive decays with heavy - heavy currents and heavy - light currents can also be treated in the context of the hqet @xcite . if a light meson from @xmath4 decays carries a large energy , hqet alone is no longer useful since the large energy of a light meson can be as large as @xmath10 . then an expansion in @xmath9 alone is not appropriate . in this case , however , we can construct a different type of an effective theory by taking the energy @xmath0 of the energetic light quark to infinity . in this limit , nonperturbative effects can also be systematically obtained . in fact , this effective theory is more complicated than the hqet and the naive power counting in @xmath11 should be modified since the system involves several energy scales . another complication arises in decays of a heavy quark with an energetic light quark due a sudakov logarithm since there are both collinear and infrared divergences @xcite . there has been some discussion of summing sudakov logarithms using effective field theories @xcite . such an approach has an advantage over conventional methods since effective theories are valid beyond perturbation theory , and it is straightforward to go beyond the leading approximation by including higher - dimensional operators . the main advantage of using effective theories in this case is that we can reproduce the sudakov logarithm easily without dividing all the kinematic regions @xcite , and the calculation is manifest in the calculational procedure . however , we need an effective theory in which logarithms arising at one loop in the effective theory should match logarithms arising at one loop in qcd for any matching scale @xmath12 in the minimal subtraction scheme . only in this case , these logarithms may be summed using the renormalization group equations . the large - energy effective theory suggested by dugan and grinstein @xcite does not satisfy this criterion since it does not include the effects of collinear gluons properly . recently bauer et al . @xcite have proposed a new effective theory called the `` collinear - soft effective theory '' . if a light quark moves with a large energy , the momentum has three distinct scales . the momentum component in the light cone direction @xmath13 is the largest , of the order of the energy of the quark , @xmath0 . the transverse momentum is smaller than @xmath0 , and the momentum component opposite to the light cone direction is the smallest . in order to disentangle the three scales conveniently , a small parameter @xmath3 is introduced . the largest component has the momentum of order @xmath0 . the transverse component is of order @xmath14 , and the smallest component is of order @xmath15 . between @xmath0 and @xmath14 , we have collinear modes and soft modes for the light quark . here we integrate out all the collinear modes above some scale @xmath12 , and the result is the effective theory consisting of collinear quarks and soft quarks . the effective theory at this stage is called the collinear - soft effective theory , which we will call the `` collinear effective theory '' for brevity . below the scale @xmath14 and above @xmath15 , we integrate out all the collinear modes , and there remain only soft modes in the final soft effective theory . this actually corresponds to the large - energy effective theory suggested by dugan and grinstein @xcite , in which there are only soft modes . in ref . @xcite , they show that at each stage of the effective theories , the infrared behavior of the full theory is correctly reproduced by including the effects of collinear gluons . therefore heavy - light currents in the full theory finally can be matched onto operators in the effective theories , their wilson coefficients are calculable and the renormalization group equation can be solved . if we consider exclusive @xmath4 decays via heavy - light currents in the scheme of effective theories , it is sufficient to consider the collinear effective theory between the scale @xmath0 and @xmath14 and integrate out all the degrees of freedom above some scale @xmath12 . at this scale , we describe a heavy quark in terms of hqet , and treat an energetic light quark in the collinear effective theory . this limit corresponds to @xmath16 with @xmath17 fixed . we can calculate the wilson coefficients of various operators in the effective theory by matching to the full theory and can obtain anomalous dimensions of various operators . in this paper , we extend further the idea of the collinear effective theory and derive the effective lagrangian to subleading order in @xmath3 and renormalize the effective theory at one loop . also we consider the correction to heavy - light currents to order @xmath3 and to leading logarithmic order in @xmath5 . in section [ cet ] , we briefly review the collinear effective theory , and derive the effective lagrangian to order @xmath3 . we also discuss a collinear gauge invariance in the effective theory . in section [ repinvar ] , we discuss reparameterization invariance in the collinear effective theory . the reparameterization invariance ensures that the kinetic energy term is not renormalized to all orders in @xmath5 . it is also useful in deriving high - dimensional operators for heavy - light currents in the collinear effective theory and in obtaining the wilson coefficients and the renormalization behavior of these high - dimensional operators . in section [ match ] , we match heavy - light currents between the full qcd and the collinear effective theory , and consider the effects of radiative corrections at one loop . in section [ rg ] , we compute the anomalous dimensions of various heavy - to - light operators to order @xmath3 at one loop , and solve the renormalization group equation for the wilson coefficients in the collinear effective theory . in section [ form ] , we consider form factors of heavy - light currents for the vector and the axial vector currents to order @xmath3 . in section [ conc ] , we present a conclusion and perspectives of the collinear effective theory . in appendix , we present an explit calculation to show that the effective lagrangian at order @xmath3 is not renormalized at one loop . we construct an effective theory which describes the dynamics of energetic light quarks . a detailed derivation of the effective theory at leading order in @xmath3 is described in refs . @xcite , and we will briefly review the idea . then we construct the effective theory to order @xmath3 . let us consider a reference frame in which a light quark carries a large energy @xmath0 . if we neglect the quark mass , the only large parameter in this system is the energy @xmath0 itself . since we are interested in decays of heavy mesons to energetic light hadrons , we can conveniently choose a reference frame as the rest frame of a heavy meson , in which the energy of light hadrons is indeed large in the heavy quark limit . in this reference frame , light particles lie close on the light - cone direction @xmath13 , and we describe their dynamics using the light - cone variables @xmath18 , where @xmath19 , and @xmath20 . we choose the axis such that @xmath21 , @xmath22 with @xmath23 . for the energetic quark , there are three distinct energy scales , with @xmath24 being large , while @xmath2 and @xmath25 are small . if we take a small parameter as @xmath26 , we can write @xmath27 since @xmath28 . therefore we have three distinct energy scales @xmath0 , @xmath14 and @xmath15 , making the effective theory more complicated than the hqet . it is similar to the case of nonrelativistic qcd ( nrqcd ) for quarkonium states , in which there are also three distinct scales @xmath29 , @xmath30 and @xmath31 , where @xmath29 is the heavy quark mass and @xmath32 is the typical velocity of a quark inside a quarkonium @xcite . the collinear quark can emit either a soft gluon with momentum @xmath33 or a collinear gluon with @xmath34 to the large momentum direction and can still be on its mass shell . due to the infrared sensitivity with collinear loop momentum , the effective theory is more complicated , and the relevant scales must be treated separately to obtain a consistent power counting method . in the collinear effective theory , the power counting in @xmath11 is troublesome , but the expansion in the small parameter @xmath3 offers a consistent power counting and there is no mixing of operators with different powers of @xmath3 . this will be discussed in detail in section v. the lagrangian in the collinear effective theory can be obtained from the full qcd lagrangian at tree level by expanding it in powers of @xmath3 . the full qcd lagrangian for massless quarks and gluons is given by @xmath35 where the covariant derivative is @xmath36 , and @xmath37 is the gluon field strength tensor . we remove large momenta from the lagrangian , similar to the method employed in the hqet . the quark momentum is split as @xmath38 the large part of the quark momentum @xmath39 and @xmath2 , denoted by @xmath40 , will be removed by defining a new field as @xmath41 a label @xmath42 in @xmath43 refers to only the components @xmath44 and @xmath2 . the derivative @xmath45 on the field @xmath43 gives @xmath46 contributions . now we introduce projection operators which project out large components @xmath47 and small components @xmath48 in the direction @xmath13 as @xmath49 the fields @xmath47 , @xmath48 satisfy @xmath50 and @xmath51 we can eliminate the small component @xmath48 at tree level by using the equation of motion @xmath52 and the lagrangian can be written in terms of @xmath47 . it is convenient to separate the collinear and soft parts in gluon modes as @xmath53 in the covariant derivative @xmath54 , such that the covariant derivative involves only soft gluons . the typical scale for the collinear gluons is @xmath55 , while the typical scale for the soft gluons is @xmath56 . since the collinear gluon carries a large momentum @xmath57 , derivatives on this field can yield order @xmath58 and @xmath59 contributions . to make this explicit , we extract the large momentum part containing @xmath60 by redefining the field @xmath61 . then the lagrangian can be written as @xmath62 \frac{\fmslash{\overline{n}}}{2 } \xi_{n , p}. \label{lag1}\end{aligned}\ ] ] here the summation over the labels @xmath40 and @xmath63 and the phase factors for each collinear field are suppressed . from now on , in order to simplify the notation further , we suppress the label momenta for the collinear fields when there can be no confusion . it should be understood that , when @xmath64 and @xmath65 appear , the summation on the label momentum @xmath42 , @xmath66 , the large phases , and the conservation of the label momenta are implied . the method to insert all the summations , the phases , and the label momenta are nicely summarized in ref . @xcite . in order to obtain the effective lagrangian , we expand eq . ( [ lag1 ] ) in powers of @xmath3 . in the power counting of the fields in @xmath3 , we follow the procedure of moving all the dependence on @xmath3 into the interaction terms to make the kinetic terms of order @xmath58 . this is done by assigning a @xmath3 scaling to the fields in the effective theory , as given in table i @xcite . .[lambdacount]power counting for the effective theory fields . [ cols="^,^,^,^,^,^,^",options="header " , ] bauer and stewart @xcite suggested a closed form to include the effects of collinear gluons to all orders . we define an operator @xmath67 which acts on products of effective theory fields . when acting on collinear fields , @xmath67 gives the sum of large momentum labels on fields minus the sum of large momentum labels on conjugate fields . then , for any function @xmath68 , we have @xmath69 the operator @xmath67 has mass dimension 1 , but power counting dimension @xmath58 . the conjugate operator @xmath70 acts only to its left and gives the sum of large momenta on conjugate fields minus the sum of large momenta on fields . let us consider gauge symmetries of the effective theory . since there are several gluon modes , there are possible @xmath71 color gauge transformations for each mode . we consider gauge symmetries that have support over collinear momenta . the collinear effective theory is invariant under a collinear nonabelian gauge transformation of the form @xmath72 $ ] . a set of these collinear gauge transformations is a subset of all the gauge transformations , which satisfies @xmath73 . it is useful to decompose this collinear transformation into a sum over the collinear momenta @xmath74 where @xmath75 . when we expand the gauge transformation , we obtain simple transformation rules for collinear fermions and gluons . the transformation for collinear fermions and gluons are given by @xmath76 . \label{cotran}\ ] ] here @xmath77 produces a sum of momenta of order @xmath3 , and the last term produces a momentum of order @xmath78 . and the soft modes transform as @xmath79 under a collinear gauge transformation . let us define a function @xmath80 of @xmath81 such that @xmath82 is invariant under the transformation in eq . ( [ cotran ] ) . the operators @xmath80 and @xmath83 are defined as @xmath84 , \ \ w^{\dagger}=\bigl [ \exp \bigl(g\overline{n } \cdot a_n^ * \frac{1}{\overline{\mathcal{p}}^{\dagger } } \bigr ) \bigr ] , \label{ws}\ ] ] which satisfy @xmath85 . in the expansion of the exponential , the @xmath86 acts to the right on all gluon fields in the square bracket . under a collinear gauge transformation , @xmath80 transforms as @xcite @xmath87 which makes @xmath82 invariant under a collinear gauge transformation . when we expand the exponential in @xmath80 , we have an infinite series of collinear gluons . but all of them are of order @xmath58 , and should be included . the operator @xmath88 of order @xmath3 transforms as @xmath89 under a collinear gauge transformation . with these transformation properties , we can write the lagrangian @xmath90 in a closed form including an infinite number of collinear gluons as @xmath91 where @xmath92 is the lagrangian at order @xmath93 . the expression in eq . ( [ eftlag ] ) is manifestly invariant under a collinear gauge transformation , and we use the fact that for any function @xmath68 , @xmath94 . the feynman rules for the propagator of a collinear quark and the interaction vertices from @xmath95 are shown in fig . [ feynman ] . here @xmath96 is defined as @xmath97 there are other interaction vertices such as the one with two collinear quark fields and two gluons , and those with triple gluons . we omit them here since they do not contribute to one - loop corrections to order @xmath3 in dimensional regularization . for a heavy quark , we employ hqet for the heavy quark field @xmath98 . the effective lagrangian for hqet is given by @xmath99 the covariant derivative in eq . ( [ hqet ] ) contains only soft gluons because the heavy quark field does not couple to collinear gluons . according to the power counting in table [ lambdacount ] , the corrections in @xmath9 in the hqet lagrangian are suppressed by @xmath78 compared to the leading lagrangian , and we will not consider them here . when we decompose a quantity into a large part and a small part , the decomposition is not unique . we can always shift the large part such that a change in the small part compensates this change to make the total quantity unchanged . the physics should be invariant under such a change . the invariance under this shift is called the reparameterization invariance . in hqet , there is a reparameterization invariance @xcite . it means that the decomposition of the heavy quark momentum @xmath100 into @xmath101 and the residual momentum @xmath8 is not unique . typically @xmath8 is of the order of @xmath6 , which is much smaller than @xmath10 . a small change in the four velocity of the order of @xmath102 can be compensated by a change in the residual momentum . the physics of heavy quarks should be invariant under different decomposition of momenta . a consequence of this reparameterization invariance is that the kinetic energy term in hqet is not renormalized to all orders . besides , we can obtain higher - dimensional operators for heavy - light currents using the reparameterization invariance . and we can easily obtain the wilson coefficients and the anomalous dimensions of higher - dimensional operators without any explicit calculation . a similar reparameterization invariance occurs in the collinear effective theory . the energetic light quark momentum @xmath42 is given by @xmath103 from now on , we will consider a small change of order @xmath3 , neglecting terms of order @xmath78 , which can be included in a straightforward way . as in hqet , the decomposition of @xmath42 into @xmath104 , @xmath2 is not unique . a small change in @xmath13 of order @xmath3 can be compensated by a change in @xmath105 , @xmath106 where @xmath107 is of order @xmath3 . and the physics for collinear quarks should be invariant under different decompositions of momenta . since @xmath104 satisfies @xmath108 , the parameter @xmath107 must satisfy @xmath109 , neglecting terms of order @xmath110 . the light quark spinor @xmath64 must also change to preserve the constraint @xmath111 . consequently , if @xmath64 changes as @xmath112 , @xmath113 satisfies @xmath114 to first order in @xmath115 , one finds @xmath116 therefore a suitable choice for the change in @xmath64 is @xmath117 the lagrangian in eq . ( [ eftlag ] ) must be invariant under the combined changes @xmath118 where the prefactor @xmath119 causes a shift @xmath120 . in order to prove the reparameterization invariance , it is convenient to write the lagrangian @xmath121 as @xmath122 where we included @xmath123 which does not affect the lagrangian , but the addition makes the lagrangian manifestly invariant under a collinear gauge transformation . the change of the lagrangian is given by @xmath124 \frac{\fmslash{\overline{n}}}{2 } \xi_n . \label{lchange}\end{aligned}\ ] ] the change @xmath125 vanishes , which can be easily seen when we disregard gauge fields . then the first line in eq . ( [ lchange ] ) exactly cancels the second line . therefore we have proved that the lagrangian is reparameterization invariant under a shift of order @xmath3 . as a result , the kinetic energy terms appearing both in @xmath95 and @xmath126 are not renormalized . the explicit calculation to show that the kinetic energy term at order @xmath3 is not renormalized at one loop is given in appendix . we can make a stronger statement by combining the reparameterization invariance and the collinear gauge invariance of the collinear effective theory . in the lagrangian @xmath126 at order @xmath3 , the kinetic energy part is given by @xmath127 which is not renormalized due to the reparameterization invariance . however , in order to make this part collinear gauge invariant , @xmath128 should be replaced by @xmath129 . there is no constraint from the collinear gauge invariance on whether we should replace the derivative operator with a covariant derivative including a soft gluon . however , if we require the invariance under ultrasoft gauge transformations @xcite , the derivative operator should be replaced by the covariant derivative . therefore the extension of the kinetic energy term which is invariant under the collinear and the ultrasoft gauge transformation is given by @xmath130 this is not renormalized to all orders in @xmath5 due to the reparameterization invariance and the gauge invariance . and the remaining part in @xmath126 is not renormalized at one loop , hence the whole lagrangian @xmath126 is not renormalized at leading logarithmic accuracy . we can fix the form of some corrections at order @xmath3 from the operators at @xmath58 using the reparameterization invariance . for example , the vector current @xmath131 in the full theory is written as @xmath132 in the collinear effective theory to order @xmath3 . the collinear gauge - invariant form of this operator is given by @xmath133 where the second term is an operator for heavy - light currents at order @xmath3 in the effective theory . we consider the matching of heavy - light currents of the form @xmath134 , where @xmath135 denotes @xmath136 or @xmath137 . below the scale @xmath44 , the hadronic current is matched onto currents in the collinear effective theory and the hqet . this introduces a new set of wilson coefficients . we will match the current operators in the full theory with the current operators in the collinear effective theory and the hqet in a single step neglecting the sum of logarithms of order @xmath138 , which is quite small since @xmath139 . the vector - current operator @xmath140 in the full theory can be matched to the effective theory as @xmath141 the operators @xmath142 are the operators at leading order in @xmath3 , and there are three such operators , which are given as @xmath143 similarly , @xmath144 are a complete set of operators at order @xmath3 there are four such operators and a convenient basis for these operators is given by @xmath145 the operators in eqs . ( [ leading ] ) , and ( [ qlambda ] ) are written in such a way that they are manifestly invariant under a collinear gauge transformation . we also include the nonlocal operators @xmath146 arising from an insertion of the order @xmath3 correction to the effective lagrangian into matrix elements of the leading - order currents , which are defined as @xmath147 our goal is to calculate the wilson coefficients @xmath148 , @xmath149 and @xmath150 in the leading logarithmic approximation . the wilson coefficients are defined by requiring that matrix elements of the vector current in the full theory are the same , to any order in @xmath3 , as matrix elements calculated in the effective theory . before we proceed to explicit calculation , note that there are nontrivial relations between the coefficients @xmath151 and @xmath152 imposed by the reparameterization invariance . this is because operators of order @xmath3 acting on a collinear quark field must always appear in certain combinations with operators of order @xmath58 . in our case , there is a unique way in which the operators @xmath153 can be combined with @xmath154 in a reparameterization invariant way , that is , @xmath155 this implies that , to all orders in perturbation theory , @xmath156 and the coefficients @xmath157 have been calculated at leading logarithmic order in ref . this is our new result and it imposes an important constraint on the theory , which must be obeyed by an explicit calculation . the operator product expansion of the axial vector current @xmath158 can be simply obtained from eq . ( [ vectorc ] ) by replacing @xmath159 if we perform the calculation using the dimensional regularization with modified minimal subtraction ( @xmath160 ) and the ndr scheme with anticommuting @xmath161 . we can rewrite the axial current as @xmath162 . the @xmath161 matrix acting on the massless quark @xmath66 becomes @xmath163 depending on the chirality of the quark . chirality is conserved by the qcd interactions , so the calculation of matching conditions proceeds just as in the vector current case , except that @xmath164 is replaced everywhere by @xmath165 . at the end of the calculation , the @xmath161 is moved back next to @xmath98 , producing a compensating minus sign for @xmath166 , but neither for @xmath167 nor for @xmath168 . thus , for axial vector currents , all the coefficients are the same in magnitude , and only @xmath169 , @xmath170 , and @xmath171 do not change sign , while all the remaining coefficients change sign . bauer et al . @xcite have explicitly showed that the collinear effective theory , indeed , reproduces the infrared behavior of the full theory by including the effects of collinear gluons . once we know that the effective theory reproduces the long - distance physics of the full theory , the matching procedure is independent of any long - distance physics such as infrared singularities , nonperturbative effects and the choice of external states . thus there is a freedom in choosing the external states and the infrared regularization scheme . we find it most convenient to perform the matching of qcd onto the collinear and the heavy quark effective theory using on - shell external quark states and dimensional regularization for both the ultraviolet and infrared divergences encountered in calculating loop diagrams . this scheme has the great advantage that all loop diagrams in the effective theory vanish , since there is no mass scale other than the renormalization scale @xmath12 . it means that matrix elements in the effective theory are given by their tree - level expressions . we assign momentum such that the incoming heavy quark has momentum @xmath172 ( with @xmath173 ) , while the outgoing light energetic quark carries momentum @xmath174 ( with @xmath175 ) . the matrix elements of operators can be written as @xmath176 where @xmath177 and @xmath178 are on - shell spinors for a massless , energetic quark field @xmath64 in the collinear effective theory , and a heavy quark field @xmath98 in the hqet , respectively . they satisfy @xmath179 and @xmath180 . we compute , in the full theory , the vector current matrix element between on - shell quark states at one - loop order in order to do the matching . the relations of the heavy quark spinors and the light quark spinors between qcd and the effective theory are given by @xmath181 the correction to the heavy quark field , which involves @xmath182 , is suppressed by @xmath78 , and it is discarded in our matching at order @xmath3 . we match the coefficients at one loop by employing the dimensional regularization in @xmath183 dimensions . in the full theory , there is no ultraviolet divergence due to current conservation . the residue at the physical mass pole in the propagator is infrared in nature , and it should be added to the vertex correction . the residue at the physical mass pole for the heavy quark in the @xmath160 scheme at order @xmath5 is given by @xcite @xmath184 and in the hqet , the residue at order @xmath5 is given as @xmath185 the residue for the light quark at order @xmath5 in the collinear effective theory is the same as the residue in the full theory , and it is given as @xmath186 since the residues for the light quarks are the same , they cancel each other when we match both theories . the matrix element of the vector current between free quark states with the residues of the external quarks in the full theory can be expressed in terms of the matrix elements in the collinear and the heavy quark effective theory as @xmath187 \bigr\ } \langle j_1^{\mu } + o_1^{\mu}\rangle \nonumber \\ & + & \frac{\alpha_s c_f}{4\pi } \bigl [ \frac{2}{1-x } + \frac{2x}{(1-x)^2 } \ln x \bigr ] \langle j_2^{\mu } + o_2^{\mu } \rangle \nonumber \\ & + & \frac{\alpha_s c_f}{4\pi } \bigl [ -\frac{x}{1-x } + \frac{x(1 - 2x)}{(1-x)^2 } \ln x \bigr ] \langle j_3^{\mu } + o_3^{\mu } + 2 o_4^{\mu } \rangle , \label{matching}\end{aligned}\ ] ] where @xmath188 and @xmath189 is the dilogarithmic function . here we have confirmed the consequence of the reparameterization invariance at one loop explicitly . the infrared behavior of the full qcd is reproduced in the collinear effective theory , and the infrared divergences in both theories cancel in matching . the wilson coefficients @xmath190 for @xmath154 at the renormalization scale @xmath12 are given by @xmath191 , \nonumber \\ c_2 ( \mu ) & = & \frac{\alpha_s c_f}{4\pi } \bigl [ \frac{2}{1-x } + \frac{2x}{(1-x)^2 } \ln x \bigr ] , \nonumber \\ c_3 ( \mu ) & = & \frac{\alpha_s c_f}{4\pi } \bigl [ -\frac{x}{1-x } + \frac{x(1 - 2 x)}{(1-x)^2 } \ln x \bigr ] , \label{wilson}\end{aligned}\ ] ] and the coefficients @xmath192 are given as @xmath193 this relation is expected from the reparameterization invariance , and the operators @xmath153 have the same anomalous dimension as those of the leading operators @xmath154 . the explicit calculation that the operators @xmath194 to @xmath195 have the same ultraviolet behavior as their corresponding leading - operators is shown in section [ rg ] . the coefficients @xmath196 are given by the product of those for @xmath154 and @xmath126 , and they are given by @xmath197 the fact that the wilson coefficients @xmath196 are the same as @xmath190 is because the effective lagrangian @xmath126 at order @xmath3 is not renormalized at leading logarithmic order . the perturbative expansion of the wilson coefficients contains large logarithms of the type @xmath198^n$ ] , which should be summed to all orders . we employ the renormalization group to improve one - loop results . the reason why we choose @xmath3 as the small parameter is because various operators with different orders of @xmath3 do not mix in this power counting . if we choose to expand in powers of @xmath11 , when we renormalize operators , a factor @xmath0 in the numerator could be induced from loop calculations . this is expected since the propagator of a collinear quark explicitly involves @xmath0 in the @xmath11 expansion . therefore higher - dimensional operators in @xmath11 can mix with those operators with one less power of @xmath10 or @xmath0 , and a power counting in @xmath11 is inappropriate . however , if we expand the effective lagrangian in powers of @xmath3 , such mixing never occurs , and we can do the power counting in @xmath3 consistently . in general , the coefficients of the operators with the same power of @xmath3 mix into themselves and satisfy a renormalization group equation of the form @xmath199 since eq . ( [ rgeq ] ) is homogeneous , we can reproduce the exponentiation of sudakov logarithm . the renormalization of the operators @xmath154 at order @xmath58 was performed in ref . the counterterm for the operators @xmath154 in the effective theory using the feynman gauge is given by @xmath200.\ ] ] this counterterm is the same for all @xmath154 , and is independent of the dirac structure of the operators since the propagators and the vertices in the collinear effective theory do not alter the dirac structure of the operators . furthermore there is no operator mixing . the anomalous dimensions are given by @xmath201 where @xmath202 here we have used @xmath203 . this gives the anomalous dimension @xmath204 the divergence in eq . ( [ anomd ] ) is cancelled , and solving the renormalization group equation eq . ( [ rgeq ] ) , we obtain @xmath205 where @xmath206 , and @xmath207 are the wilson coefficients at @xmath208 , as given in eq . ( [ wilson ] ) . at order @xmath3 , we need to renormalize the operators @xmath153 . let us first consider the renormalization of @xmath209 to @xmath210 . the feynman rules for the vertex from these operators with a collinear gluon are given in fig . [ feyn1 ] . the feynman diagrams to renormalize the operators @xmath211 ( @xmath212 ) at order @xmath5 are shown in fig . [ renlam1 ] . since the loop calculation does not alter the dirac structure , we can treat the renormalization of these operators in the same way for all the three operators . the feynman diagrams in fig . [ renlam1 ] give the amplitude @xmath213.\ ] ] note that there is no mixing for the operators @xmath153 . if we add the residues from the propagators of a heavy quark and a collinear quark , we have the counterterm @xmath214 , \ \ ( i=1,2,3 ) , \label{counter}\ ] ] which is identical to the counterterm for the leading operators @xmath215 . we can do the same calculation for the operator @xmath216 and it turns out that the operator @xmath216 has the same dependence on @xmath107 as @xmath217 . and the counterterm is also given by eq . ( [ counter ] ) . therefore the operators @xmath153 ( @xmath218 ) have the same anomalous dimensions as the leading operators . this is the explicit proof of the reparameterization invariance at one loop and order @xmath3 . for the time - ordered products @xmath146 , the anomalous dimensions are the same as those of @xmath219 because the lagrangian @xmath126 in defining @xmath146 is not renormalized at one loop . here we see that the reparameterization invariance and the gauge invariance influence the structure of the theory . furthermore , since there is no mixing , the perturbative corrections to heavy - light currents take a simple form to order @xmath3 . as an application of the collinear effective theory , we can consider the form factors for @xmath4 mesons into light mesons . we consider the kinematic region in which the energy @xmath0 of the light quark is large , @xmath220 which equivalently means that the momentum transfer squared through the weak current is small @xmath221 . in this case , the off - shellness of the light quark is @xmath222 , where @xmath223 , thus @xmath224 . therefore our formulation to order @xmath3 gives the correction to the form factors at order @xmath225 . for simplicity , we will consider the form factors for the vector and the axial vector currents . the form factors for @xmath226 decays into light pseudoscalar and vector mesons from the vector current @xmath227 , and the axial vector current @xmath228 are defined as @xmath229 + f_0 ( q^2 ) \frac{m^2 -m_p^2}{q^2 } q^{\mu } , \nonumber \\ \langle v ( p^{\prime } , \epsilon^*)| v^{\mu } |\overline{b } ( p ) \rangle & = & \frac{2v(q^2)}{m+m_v } i\epsilon^{\mu\nu \alpha \beta } \epsilon^*_{\nu } p^{\prime}_{\alpha } p_{\beta } , \nonumber \\ \langle v ( p^{\prime } , \epsilon^*)| a^{\mu } |\overline{b } ( p ) \rangle & = & 2m_v a_0 ( q^2 ) \frac{\epsilon^ * \cdot q}{q^2 } q^{\mu } + ( m+m_v ) a_1 ( q^2 ) \bigl [ \epsilon^{*\mu } -\frac{\epsilon^ * \cdot q}{q^2 } q^{\mu } \bigr ] \nonumber \\ & & -a_2 ( q^2 ) \frac{\epsilon^ * \cdot q}{m+m_v } \bigl[p^{\mu } + p^{\prime \mu } -\frac{m^2-m_v^2}{q^2 } q^{\mu } \bigr ] , \label{parform}\end{aligned}\ ] ] where @xmath230 , @xmath231 ( @xmath232 ) is the mass of the pseudoscalar ( vector ) meson , @xmath233 is the polarization vector of the vector meson , and @xmath234 is the mass of a @xmath226 meson . we use the sign convention @xmath235 . we can calculate these form factors systematically in powers of @xmath3 in the collinear effective theory . the matrix elements in the full theory are matched to the matrix elements in the collinear effective theory using eq . ( [ vectorc ] ) . however , here we do not include interactions where a collinear gluon is exchanged with the spectator quarks inside a @xmath4 meson . in ref . @xcite it was argued that these spectator effects could be of the same order in @xmath3 and @xmath9 as the soft contributions , but they are suppressed by a power of @xmath236 . they are therefore just as important as the one - loop corrections to the matching coefficients such as @xmath157 . here we apply the collinear effective theory to the soft contributions only . it means that the effective theory applies to light mesons produced in an asymmetric configuration , in which a single quark from the @xmath237 decay carries almost all the momentum . if we consider this process as light - cone dominated , this is not a typical configuration . a typical configuration is for both quark and antiquark have nearly equal momentum . and spectator interactions can play an important role in this configuration . in heavy - to - heavy transitions such as @xmath238 in the heavy quark limit , the interactions of a heavy quark with the soft degrees of freedom around the heavy quark do not change even when there is a transition . on the contrary , in heavy - to - light transitions , the soft degrees of freedom around the heavy quark experience an abrupt change . if an energetic quark and the soft degrees of freedom move somehow elastically with almost the same velocity , we can safely consider the interaction of an energetic quark with the soft degrees of freedom in terms of the collinear effective theory . this corresponds to the soft contribution to form factors . if only an energetic quark is pushed to the light - cone direction , the soft degrees of freedom around the heavy quark should arrange themselves to follow the energetic quark to form light mesons . in this process , hard gluons should be exchanged between the energetic quark and the previous soft degrees of freedom in the heavy quark . this corresponds to the hard spectator interaction . this hard spectator interaction should be considered separately , and we leave the hard spectator contributions for future study . a convenient way to evaluate hadronic matrix elements in the effective theory is to associate the spin wave function @xmath239 with the eigenstates of the effective lagrangian , where @xmath234 is the mass of the meson . the form factors in the effective theory can be written as @xmath240 , \ \ ( l = p , v)\ ] ] where @xmath135 denotes a dirac structure , and @xmath241 are the spin wave functions associated with a light meson and a @xmath226 meson respectively . the normalization factor @xmath242 appearing in @xmath243 is absorbed in @xmath244 . the function @xmath245 contains the long - distance dynamics , and it is independent of the dirac structure @xmath135 in the current . the most general form for @xmath244 is given by @xmath246 but due to the properties of the projection operators in @xmath247 and @xmath248 , not all of them are independent . for @xmath249 , there is one independent term , and for @xmath250 , there are two independent terms . charles et al . @xcite have shown that there are only three independent matrix elements in heavy - to - light transitions by employing the hqet and the large - energy effective theory to obtain the leading result in @xmath11 . however this is not sufficient to describe heavy - to - light decays because interactions with collinear gluons should be included . though the argument is different , there are also three independent matrix elements in the collinear effective theory . at order @xmath3 , we have the form factors of the form @xmath251,\ ] ] where @xmath252 contains the long - distance dynamics and they are independent of the dirac structure @xmath135 of the current . since the operator is proportional to @xmath105 , the only allowed vector component for @xmath253 is @xmath96 . therefore the most general form for @xmath253 is given by @xmath254.\ ] ] as in the case of @xmath255 , all the terms are not independent due to the projection operators in @xmath247 , and @xmath248 . for @xmath249 , there is only one independent term , and for @xmath250 , there are two independent terms . similarly , the matrix elements of the time - ordered products @xmath256 can be written as @xmath257,\ ] ] and the most general form for @xmath258 is written as @xmath259,\ ] ] because @xmath126 is of order @xmath3 and it typically depends on @xmath105 . here also we have one independent term for @xmath249 , and two independent terms for @xmath250 . in summary , we can write the parameters describing the long - distance physics as @xmath260 note that the convention for the longitudinal form factor @xmath261 is the same as that of ref . @xcite , and is related to the corresponding form factor @xmath262 defined in ref . @xcite by @xmath263 . the matrix elements of all the operators can expressed in terms of these nonperturbative parameters . at order @xmath58 , the matrix elements for pseudoscalar bosons are given by @xmath264 using the above relations , we can determine the heavy - to - light form factors at leading order in @xmath3 and @xmath5 . @xmath265 where @xmath266 , @xmath267 . from the results in section iv , we can include the perturbative corrections , which change the relation between form factors . we find that , at leading order in @xmath3 and at leading logarithmic order in @xmath5 , @xmath268 , \ \ \frac{f_0}{x } = \xi_p ( e ) \bigl [ c_1 + \bigl(1-\frac{x}{2 } \bigr ) c_2 + c_3\bigr ] , \nonumber \\ \frac{v}{1+\hat{m}_v } & = & c_1 \xi_{\perp } ( e ) , \ \ \frac{2\hat{m}_v}{x } a_0 = \xi_{\parallel}(e ) \bigl [ c_1 + \bigl(1-\frac{x}{2 } \bigr ) c_2 + c_3 \bigr ] , \nonumber \\ \frac{1+\hat{m}_v}{x } a_1 & = & c_1 \xi_{\perp } ( e ) , \ \ \frac{a_2}{1+\hat{m}_v } = c_1 \xi_{\perp } ( e ) -\bigl ( c_1 + \frac{x}{2 } c_2 + c_3 \bigr ) \xi_{\parallel } ( e).\end{aligned}\ ] ] these results are the same as those derived by bauer et al . @xcite , though our basis is different from theirs . in ref . @xcite , beneke and feldmann have calculated the soft contribution to the form factors using the large - energy effective theory . as we have stressed , the matching to the full theory is impossible in this case . however , they judiciously absorbed the infrared divergences into the nonperturbative parameters such as @xmath269 , @xmath270 or @xmath261 by observing the dirac structure of the matrix elements . in the process , the nonperturbative parameters are defined at each order in @xmath5 . since we can match the collinear effective theory to the full theory , we can check their calculations . we find that their perturbative corrections in eqs . ( 30 ) , ( 32 ) and ( 33 ) in ref . @xcite are correct when we compare them with the exact results in the collinear effective theory . now we include the nonperturbative corrections at order @xmath3 , along with the perturbative correction . at order @xmath3 , the matrix elements of @xmath153 for pseudoscalar mesons are given as @xmath271 finally , for the time - ordered products , we have @xmath272 and the time - ordered products involving the heavy - light currents with @xmath161 vanish . for the matrix elements of the time - ordered products for vector mesons , we find @xmath273 combining all these form factors , we obtain in the collinear effective theory @xmath274 \nonumber \\ & & + 2e v^{\mu } \bigl [ c_2 \xi_p + \frac{1}{2e } \bigl ( a_p ( 2b_1 + b_2 ) + b_p a_2 \bigr ) \bigr ] , \nonumber \\ \langle v|v^{\mu } |\overline{b}\rangle & = & 2e i \epsilon^{\mu \nu \alpha \beta } \epsilon^*_{\nu } n_{\alpha } v_{\beta } \bigl [ c_1 \xi_{\perp } + \frac{1}{2e } \bigl ( b_4 a_{v1}+a_1 b_{v1 } \bigr ) \bigr ] , \nonumber \\ \langle v|a^{\mu } |\overline{b}\rangle & = & 2e \epsilon^{*\mu } \bigl [ c_1 \xi_{\perp } + \frac{1}{2e } \bigl ( b_4 a_{v1}+ a_1 b_{v1}\bigr ) \bigr ] \nonumber \\ & - & 2e(\epsilon^ * \cdot v ) n^{\mu } \bigl[c_1 \xi_{\perp } - ( c_1 + c_3 ) \xi_{\parallel } \nonumber \\ & & + \frac{1}{2e } \bigl ( ( b_1 -b_3 ) a_{v2 } + b_4 a_{v1 } + a_1 b_{v1 } -(a_1+a_3 ) b_{v2 } \bigr ) \bigr ] \nonumber \\ & + & 2e ( \epsilon^ * \cdot v ) v^{\mu } \bigl [ c_2 \xi_{\parallel } + \frac{1}{2e } \bigr ( ( 2b_1+b_2 ) a_{v2}+ a_2 b_{v2}\bigr ) \bigr ] . \end{aligned}\ ] ] from these relations , we can obtain the form factors to order @xmath3 and to leading - logarithmic order in @xmath5 as @xmath275 \bigl [ \xi_p + \frac{1}{2e } ( a_p + b_p ) \bigr]-(2-x ) c_1 \frac{a_p}{2e } , \nonumber \\ \frac{f_0}{x } & = & \bigl [ c_1 + \bigl(1-\frac{x}{2 } \bigr ) c_2 + c_3 \bigr ] \bigl [ \xi_p + \frac{1}{2e } ( a_p + b_p)\bigr ] -xc_1 \frac{a_p}{2e},\nonumber \\ \frac{2\hat{m}_v}{x } a_0 & = & \bigl [ c_1 + \bigl(1- \frac{x}{2 } \bigr ) c_2 + c_3 \bigr ] \bigl [ \xi_{\parallel } + \frac{1}{2e } ( a_{v2 } + b_{v2})\bigr ] -xc_1 \frac{a_{v2}}{2e } , \nonumber \\ \frac{1+\hat{m}_v}{x } a_1 & = & \frac{v}{1+\hat{m}_v}= c_1 \bigl(\xi_{\perp } + \frac{b_{v1}}{2e } \bigr ) + c_3 \frac{a_{v1}}{e } , \nonumber \\ \frac{a_2}{1+\hat{m}_v } & = & c_1 \bigl ( \xi_{\perp } + \frac{b_{v1}}{2e } \bigr ) + c_3 \frac{a_{v1}}{e } - \bigl [ c_1 + \frac{x}{2 } c_2 + c_3 \bigr ] \bigl [ \xi_{\parallel } + \frac{1}{2e } ( a_{v2 } + b_{v2})\bigr ] \nonumber \\ & & + ( 2-x ) c_1 \frac{a_{v2}}{2e}. \label{formlam } \end{aligned}\ ] ] here we keep @xmath276 explicitly even though @xmath277 in our power counting . it is because meson masses are inserted in the definition of form factors in eq . ( [ parform ] ) without regard to the power counting in the collinear effective theory . however , we neglect the terms proportional to the mass squared of the light meson compared to @xmath278 . and we use the relations among the wilson coefficients to express the result in terms of @xmath190 only . at leading order in @xmath3 , there are three unknown nonperturbative parameters @xmath279 , @xmath280 , and @xmath281 . these are dimensionless functions . while the isgur - wise function in hqet is normalized to one at maximal momentum transfer due to the heavy quark symmetry , there is no constraint in the normalization of these unknown parameters @xcite . at order @xmath3 , there are six additional nonperturbative parameters : @xmath282 , @xmath283 , @xmath284 , @xmath285 , @xmath286 , and @xmath287 . in our convention , all these parameters have mass dimension , for example , @xmath288 where the last relation comes from the kinematics . the remaining five unknown parameters are of the same order in @xmath3 . therefore eq . ( [ formlam ] ) is our result for the form factors to order @xmath225 . there are interesting relations among the form factors in the effective theory . at zeroth order in @xmath3 and @xmath5 , those relations are given by @xmath289 these relations are modified at order @xmath3 and at leading - logarithmic order in @xmath5 as @xmath290 , \nonumber \\ \frac{v}{1+\hat{m}_v } & = & \frac{1+\hat{m}_v}{x } a_1,\nonumber \\ \frac{2\hat{m}_v}{x } a_0&=&\frac{1+\hat{m}_v}{x } a_1 -(1-\hat{m}_v ) a_2 \nonumber \\ & & + ( 1-x ) c_2 \bigl [ \xi_{\parallel } + \frac{1}{2e}(a_{v2 } + b_{v2 } ) \bigr ] + ( 1-x)c_1 \frac{a_{v2}}{e}. \label{formdiff}\end{aligned}\ ] ] note that the second relation in eq . ( [ formlead ] ) still holds to order @xmath3 and at leading - logarithmic order in @xmath5 . and the tree - level results hold only in the limit @xmath291 . we have shown that heavy meson decays in which light mesons are emitted with large energy can be consistently described by the collinear effective theory combined with the hqet . and we can obtain a systematic expansion of the effective lagrangian in powers of @xmath3 . heavy - light currents can also be expanded consistently in powers of @xmath3 , and the wilson coefficients of various operators in the effective theory can be computed by matching the effective theory to the full theory . it is crucial to note that the collinear effective theory reproduces the infrared behavior of the full theory by including the effects of collinear gluons . there is a reparameterization invariance in the collinear effective theory , in which a slight change of the light - cone direction @xmath13 can be compensated by a change of @xmath2 to make the physics invariant under this transformation . if we also require that the theory be invariant under collinear gauge transformations , we can prove that the effective lagrangian @xmath126 at order @xmath3 is not renormalized . this reparameterization invariance is also useful in deriving the operators of order @xmath3 from the operators of order @xmath58 . the wilson coefficients and the anomalous dimensions can be obtained from the operators which are related by the reparameterization invariance . the reparameterization invariance and the collinear gauge invariance put a serious constraint in the structure of heavy - light currents in the collinear effective theory . the development of the collinear effective theory casts a renewed view on heavy quark decays in which light quarks are emitted with large energy . bauer et al . @xcite have considered nonleptonic decays using the collinear effective theory , and found that the decay @xmath292 is factorized in the heavy quark limit to all orders in @xmath5 . it will be interesting to look into nonleptonic decays of @xmath4 mesons in the context of the collinear effective theory including higher - order corrections in @xmath3 . what we have not considered here is hard spectator effects , in which spectator quarks interact with the energetic quark through hard gluons . as beneke et @xcite pointed out , this contribution can be as important as the soft contribution to the form factors . if we can analyze the hard spectator contribution also in the scheme of the collinear effective theory , we will have a better understanding of form factors in this kinematic region . this is the next subject to be developed . the authors are supported by the ministry of education grants krf-99 - 042-d00034 , krf-2000 - 015-dp0067 , and hacksim bk21 project . in this appendix , we show explicitly that the effective lagrangian @xmath126 at order @xmath3 in eq . ( [ eftlag ] ) is not renormalized at one loop . the feynman rules for the lagrangian @xmath126 to order @xmath293 is shown in fig . the derivative is of order @xmath78 , and it is replaced by the residual momentum @xmath8 in momentum space . we will concentrate on the first term in @xmath126 , which is of the form @xmath294 which is shown in fig . [ lone ] ( a ) . other terms in @xmath126 contribute to the renormalization of @xmath194 at order @xmath3 along with the radiative corrections of @xmath194 . in order to show that @xmath126 is not renormalized , we have to consider all the radiative corrections for the operators shown in fig . [ lone ] . however , we will concentrate on the renormalization of @xmath194 , since other terms have the same renormalization behavior as @xmath194 at leading logarithmic order . the feynman diagrams to renormalize @xmath194 are shown in fig . [ renl1 ] . and the corresponding diagrams with a soft gluon exchange vanish due to the vertex structure . all the diagrams in fig . [ renl1 ] are zero using dimensional regularization for on - shell external states , and the coefficient of @xmath194 is given by the tree - level value . in order to see the renormalization group behavior , we have to extract the ultraviolet divergent part by putting the external quark off the mass shell by @xmath295 . we will show only the ultraviolet divergent parts here . calculating the feynman diagram in fig . [ renl1 ] ( a ) , ( b ) and ( c ) , we obtain @xmath296 respectively . therefore the sum of all the diagrams is given by @xmath297 when we add the wave function renormalization to this amplitude , the ultraviolet divergences cancels , and the anomalous dimension of @xmath194 is zero . therefore we have shown that the operator @xmath194 is not renormalized at order @xmath5 explicitly . in fact , we have to consider one - loop corrections to the remaining operators in @xmath126 . but no other operators are renormalized though we do not show them here . as a result , the wilson coefficients @xmath298 of the time - ordered products in eq . ( [ ai ] ) come from the wilson coefficients of the operators @xmath215 alone and not from @xmath126 . 99 n. isgur and m. b. wise , phys . b * 232 * , 113 ( 1989 ) ; * 237 * , 527 ( 1990 ) ; h. georgi , phys . b * 240 * , 447 ( 1990 ) ; m. neubert , phys . rep . * 245 * , 259 ( 1994 ) and the references therein . j. chay , h. georgi , and b. grinstein , phys . b * 247 * , 399 ( 1990 ) ; i. i. bigi , m. shifman , n. g. uraltsev , and a. i. vainshtein , phys lett . * 71 * , 496 ( 1993 ) ; b. blok , l. koyrakh , m. shifman , and a. i. vainshtein , phys . rev . d * 49 * , 3356 ( 1994 ) ; * 50 * , 3572(e ) ( 1994 ) ; a. v. manohar , and m. b. wise , _ ibid . _ * 49 * , 1310 ( 1994 ) . a. f. falk , and m. neubert , phys . d * 47 * 2965 ( 1993 ) ; phys . d * 47 * , 2982 ( 1993 ) . m. neubert , phys . d * 49 * , 1542 ( 1994 ) . v. v. sudakov , zh . . fiz . * 30 * , 87 ( 1956 ) [ sov . jetp * 3 * , 65 ( 1956 ) ] ; j. frenkel , and j. c. taylor , nucl . b * 109 * , 439 ( 1976 ) ; b * 155 * , 544(e ) ( 1979 ) ; t. kinoshita , and a. ukawa , phy . d * 13 * , 1573 ( 1976 ) ; * 15 * , 1596 ( 1977 ) ; * 16 * , 332 ( 1977 ) . a. g. grozin , and g. p. korchemsky , phys . d * 53 * , 1378 ( 1996 ) . u. aglietti , g. corbo , and l. trentadue , int . j. mod a * 14 * , 1769 ( 1999 ) . u. aglietti , and g. ricciardi , phys . b * 466 * , 313 ( 1999 ) ; nucl . b * 587 * , 363 ( 2000 ) . c. balzereit , t. mannel , and w. killian , phys . d * 58 * , 114029 ( 1998 ) . v. a. smirnov , phys . b * 404 * , 101 ( 1997 ) ; m. beneke , and v. a. smirnov , nucl . b * 522 * , 321 ( 1998 ) ; v. a. smirnov , and e. r. rakhmetov , theor . phys . * 120 * , 870 ( 1990 ) ; v. a. smirnov , phys . b * 465 * , 226 ( 1999 ) . m. j. dugan and b. grinstein , phys . b * 255 * , 583 ( 1991 ) . c. w. bauer , s. fleming , and m. luke , phys . rev . d * 63 * , 014006 ( 2000 ) . c. w. bauer , s. fleming , d. pirjol , and i. w. stewart , phys . d * 63 * 114020 ( 2001 ) . c. w. bauer , d. pirjol , and i. w. stewart , phys . rev . d * 65 * 054022 ( 2002 ) . c. w. bauer , and i. w. stewart , phys . b * 516 * , 134 ( 2001 ) . g. t. bodwin , e. braaten , and g. p. lepage , phys . d * 51 * 1125 ( 1995 ) . m. luke , and a. v. manohar , phys . b * 286 * , 348 ( 1992 ) . see , for example , a. v. manohar , and m. b. wise , _ heavy quark physics _ , ( cambridge university press , cambridge , 2000 ) . m. beneke and th . feldmann , nucl . phys . b * 592 * , 3 ( 2001 ) . j. charles , a. le yaouanc , l. oliver , o. pene , and j. ch . raynal , phys . rev . d * 60 * , 014001 ( 1999 ) . c. w. bauer , d. pirjol , and i. w. stewart , phys . * 87 * , 201806 ( 2001 ) .

we derive here the friedland - tverberg inequality for positive hyperbolic polynomials . this inequality is applied to give lower bounds for the number of matchings in-regular bipartite graphs . it is shown that some of these bounds are asymptotically sharp . we improve the known lower bound for the three dimensional monomer - dimer entropy . we present ryser - like formulas for computations of matchings in bipartite and general graphs . additional algorithmic applications are given . + 2000 mathematics subject classification : 05a15 , 05a16 , 05c70 , 05c80 , 82b20 keywords and phrases : positive hyperbolic polynomials , friedland - tverberg inequality , lower bounds for sum of all subpermanents of doubly stochastic matrices of fixed order , lower bounds for matchings , asymptotic lower matching conjecture , monomer - dimer partitions and entropies , ryser - like formulas for matchings .

math0603410

the aim of this paper is to explore the connections between the problem of counting the number of partial matchings in graphs and positive hyperbolic polynomials , and ryser - like formulas for partial matchings . given a graph @xmath1 on @xmath2 vertices , i.e. @xmath3 , we want to compute the number of @xmath4-matching , i.e. the number of subsets @xmath5 of edges @xmath6 , where @xmath7 , and no two edges in @xmath5 have a common vertex . our main results are for bipartite graphs @xmath8 , where @xmath9 and @xmath10 . let @xmath11 be the incidence matrix of the bipartite graph @xmath12 . then the number of @xmath4-matchings in @xmath12 is equal to @xmath13 , where @xmath14 is the sum of @xmath15 minors of @xmath16 . for @xmath17 , @xmath18 , the permanent of @xmath19 is the number of perfect matchings in @xmath12 . it is well known that the computation of the number of perfect matching in a general bipartite graph is @xmath20 . see @xcite for the first proof and @xcite for a simplified proof . ryser s algorithm to compute the permanent of any @xmath21 @xcite remains the most efficient exact algorithm , even though it uses around @xmath22 operations . one can speed up significantly the approximate computation of @xmath23 . one knows to compute the permanent of a nonnegative matrix @xmath24\in { \mathbb{r}}_+^{n\times n}$ ] within a simply exponential factor @xcite . in the case that all the entries of the matrix are uniformly bounded below and above by positive constants , one can improve the estimates of the exponential errors @xcite and @xcite . a fully randomized polynomial approximation scheme _ frpas _ for the number of perfect matchings in a bipartite @xmath12 , and more generally for the permanent of a nonnegative matrix is given in @xcite . this result is generalized in @xcite to the number of @xmath4-matchings in bipartite @xmath12 , and more generally to @xmath14 , for any @xmath25 . we now describe our main results for @xmath14 , where @xmath26 is doubly stochastic , and their applications to lower bounds on partial matchings in bipartite graphs . recall that the minimum of the permanent of @xmath27 doubly stochastic matrices , denoted by @xmath28 , is achieved only for the flat matrix @xmath29 , whose all entries equal to @xmath30 . thus @xmath31 for any @xmath32 and this inequality was conjectured by van der waerden @xcite . this conjecture was independently proved by egorichev @xcite and falikman @xcite . we call the above inequality egorichev - falikman - van der waerden ( efw ) inequality . the asymptotic behavior of efw inequaity is captured by the inequality @xmath33 for any @xmath34 . this inequality was shown by the first name author @xcite three years before @xcite . let @xmath35 be the set of all @xmath0-regular bipartite graphs @xmath12 on @xmath36 vertices . for @xmath37 the matrix @xmath38 is doubly stochastic . hence the number of perfect matchings in @xmath12 is at least @xmath39 . thus for @xmath40 , the number of perfect matchings in @xmath0-regular bipartite graphs grows exponentially , which proves a conjecture by erdos - renyi @xcite . schrijver @xcite improved efw inequality to @xmath0-regular bipartite graphs , whose asymptotic growth is best possible . recently , the second name author @xcite improved schrijver s inequality . moreover , the proof in @xcite is significantly simpler and transparent . one of the main tools in the proof in @xcite is the use of the classical theory of hyperbolic polynomials . it was shown by the first name author that @xmath41 for any @xmath42 , and for @xmath43 $ ] equality holds only if and only if @xmath44 @xcite . ( @xmath45 for each @xmath42 . ) this was fact was conjectured by tverberg @xcite , and is called in this paper the friedland - tverberg ( ft ) inequality . ft inequality gives a lower bound on the number of partial matchings in any @xmath46 . we derive here the schrijver type inequalities for @xmath4 matchings in @xmath0-regular bipartite graphs on @xmath36 vertices . this is done using the results and techniques of @xcite . in particular we give a generalized versions of ft inequality to positive homogeneous hyperbolic polynomials , which are of independent interest . the notion of partial matching in @xmath35 can be extended to asymptotic matchings as @xmath47 as follows . given a sequence of @xmath48 we can consider the quantities @xmath49.\nonumber \end{aligned}\ ] ] @xmath50 and @xmath51 can be viewed as the minimal and the maximal exponential growth of matchings of density @xmath52 of the sequence @xmath53 . consider the following special case of the above example . let @xmath54 be a cycle on @xmath4 vertices . note that @xmath54 is bipartite if and only if @xmath4 s even . fix a positive integer @xmath55 and let @xmath56 , be bipartite toroidal grid on @xmath57 vertices . note that @xmath58 . it is shown in @xcite that @xmath59 and this quantity is the exponential growth rate of the number of monomer - dimer tilings of the @xmath55-dimensional cubic grid having sides of length @xmath60 , as @xmath60 tends to infinity and the dimer density , ( fraction of the maximum possible number @xmath61 of dimers ) , in these tilings converges to a fixed number @xmath62 $ ] . ( see also @xcite . ) we denote by this exponential growth by @xmath63 , and call it the @xmath55-dimensional monomer - dimer entropy of dimer density @xmath62 $ ] in the lattice @xmath64 , see @xcite and @xcite . for @xmath65 the rate is known explicitly as a function @xmath52 , whereas for @xmath66 the exact rate is unknown . @xmath63 can be estimated if one can estimate the quantities @xmath67 from below and above for any sequence @xmath68 . lower and upper estimates of @xmath67 are conjectured in @xcite and called the _ asymptotic lower matching conjecture _ and _ asymptotic upper matching conjecture _ , abbreviated here by almc and aumc respectively . for @xmath69 almc and aumc are proved in @xcite . in this paper we apply our lower bounds on the sum of subpermanents of doubly stochastic matrices with @xmath0 nonzero elements in each row to obtain lower bound on @xmath50 . for a fixed integer @xmath40 , we show the validity of almc for the densities @xmath70 where @xmath71 . these inequalities yield new lower bounds for the @xmath55-dimensional monomer - dimer entropy of dimer density @xmath72 $ ] in the lattice @xmath64 . in particular we obtain the best known lower bound for the three dimensional monomer dimer entropy @xmath73 , which combined with the known upper bound in @xcite gives the tight result @xmath74 $ ] . next we discuss briefly the sum of @xmath15 subhafnians of @xmath75 symmetric @xmath76 with nonnegative entries , denoted by @xmath77 . for @xmath78 matrix @xmath76 this is equivalent to the number of @xmath4-matchings in a general graph on @xmath36 vertices . we give ryser - type algorithm for @xmath14 and @xmath77 . unfortunately , the generating function @xmath79 , ( the quadratic for associated with @xmath76 ) , is positive hyperbolic if and only if the second eigenvalue of @xmath76 is nonpositive . we show that for any graph @xmath12 , @xmath80 is positive hyperbolic if and only if @xmath12 is a complete @xmath81-partite graph . the last section is devoted to algorithmic applications related to @xmath14 and @xmath77 . we now list briefly the contents of this paper . in 2 we discuss briefly the notion of positive hyperbolic polynomials and their properties that needed here . in 3 we bring the generalized version of ft inequality for positive hyperbolic polynomials . in 4 we state and discuss the almc and aumc . in 5 we give lower bounds on @xmath50 . we apply these bounds to verify the almc for a countable values of densities for each @xmath40 as explained above . in 6 we discuss the notion of @xmath77 and its connection to the quadratic form @xmath79 . in 7 we discuss the ryser - type formulas for @xmath82 and @xmath77 . we thank uri peled for supplying us with the figures 1 and 2 . * definitions and notations * 1 . a vector @xmath83 is called _ positive _ or _ nonnegative _ , and denoted by @xmath84 or @xmath85 if @xmath86 or @xmath87 for @xmath88 respectively . a nonnegative vector @xmath89 is denoted by @xmath90 . the cone of all nonnegative vectors in @xmath92 is denoted by @xmath93 . a polynomial @xmath94 is called _ positive hyperbolic _ if the following conditions hold : * @xmath52 is a homogeneous polynomial of degree @xmath95 . * @xmath96 for all @xmath97 . * @xmath98 , for @xmath99 , has @xmath4-real @xmath100-roots for each @xmath101 and each @xmath102 . 3 . for any polynomial @xmath103 and any @xmath104 let @xmath105 . 4 . let @xmath106 be the standard basis in @xmath92 . let @xmath107 and denote by @xmath108 the @xmath27 matrix whose all entries are equal to @xmath30 . the following lemma summarizes the basic properties of positive hyperbolic polynomials that needed here . [ propphp ] let @xmath103 be a positive hyperbolic polynomial of degree @xmath109 . then the following properties hold : 1 . let @xmath110 be fixed and denote @xmath111 . assume that @xmath112 . then @xmath113 has @xmath4 real @xmath100 roots . furthermore @xmath114 is a positive hyperbolic polynomial of degree @xmath115 . 2 . let @xmath117 and assume that @xmath118 . then either @xmath119 for all @xmath120 or @xmath121 and @xmath122 . assume that @xmath96 and @xmath113 is not a constant polynomial . then all its roots are real and negative . if @xmath123 is not a zero polynomial then @xmath123 is a positive hyperbolic of degree @xmath115 . if @xmath124 is not identically zero then @xmath125 is a positive hyperbolic of degree @xmath4 in @xmath126 . in particular , @xmath127 + @xmath128 is either a zero polynomial or a positive hyperbolic polynomial in @xmath129 variables of degree @xmath115 . * proof . * we assume that @xmath130 otherwise all the results are trivial 1 . the facts @xmath113 has @xmath4 negative roots and @xmath123 is positive hyperbolic is in @xcite or @xcite . hence @xmath131 on @xmath93 . let @xmath132 and assume that @xmath133 be a sequence of vectors converging to @xmath134 . . therefore @xmath136 on @xmath137 . in particular @xmath138 n @xmath93 for @xmath88 . thus @xmath139 is a nondecreasing function in each variable @xmath140 on @xmath141 . hence @xmath142 . 2 . recall brunn - minkowski theorem that @xmath143 is convex on all positive vectors in @xmath92 ( * ? ? ? * thm 2 , 4 ) ) . since @xmath144 is continuous on @xmath93 , it follows that @xmath144 is nonnegative and convex on @xmath137 . if @xmath121 it follows that @xmath145 on the interval joining @xmath146 . hence @xmath147 assume that @xmath96 . then @xmath148 . if @xmath113 is constant then @xmath149 . assume that @xmath113 is a nonconstant polynomial . let @xmath133 be a sequence of vectors converging to @xmath150 . then @xmath151 . each @xmath152 has @xmath4 real negative zeros . consider the complex projective space @xmath153 with the homogenous coordinates @xmath154 . the @xmath4-roots of @xmath155 correspond to the intersection points of the line @xmath156 with the hypersurface @xmath157 in @xmath153 of degree @xmath4 . all these points are real and are located in the affine part of of the line @xmath158 , i.e. @xmath159 and @xmath160 . as @xmath161 it follows that @xmath162 , counting with the multiplicities if @xmath113 is not constant . thus all roots of @xmath113 are nonnegative . since @xmath148 it follows that all the roots of @xmath113 are negative . hence @xmath163 for @xmath120 . + assume that @xmath164 . recall that @xmath165 . according to part _ 1 _ @xmath166 for each pair @xmath167 . hence @xmath168 . if @xmath169 for some @xmath170 then for any @xmath171 , there exists @xmath172 such that @xmath173 . hence @xmath174 contrary to our assumption . thus @xmath175 . as @xmath176 and @xmath177 is a polynomial of degree @xmath115 , it follows that @xmath177 has @xmath115 real roots for each @xmath178 . thus @xmath123 is positive hyperbolic . 3 . since @xmath179 for @xmath180 on @xmath93 it follows that @xmath181 on @xmath182 , where @xmath183 . since @xmath125 is a homogeneous of degree @xmath4 , from the arguments in part _ 2 _ it follows that either @xmath184 for each @xmath185 or @xmath186 . assume that @xmath187 . then according to part _ 1 _ , @xmath188 has @xmath4 real roots in @xmath100 . assume that @xmath189 . hence @xmath190 . so @xmath191 is positive hyperbolic by _ 2 _ , and by previous arguments @xmath0 is positive hyperbolic on @xmath126 . @xmath192 + the following propositions are well known and we bring their proof for completeness . [ nonnegcphp ] let @xmath103 be a positive hyperbolic polynomial of degree @xmath109 . then the coefficient of each monomial in @xmath52 is nonnegative . * * let @xmath52 be a positive hyperbolic polynomial of degree @xmath109 . let @xmath101 . part _ 1 _ of the above lemma yields that @xmath123 is positive hyperbolic of degree @xmath115 . we prove the proposition by induction on @xmath4 . let @xmath193 . assume that @xmath101 . then @xmath194 . hence @xmath195 and the corollary holds . assume that the result hold for @xmath196 . let @xmath52 be a positive hyperbolic polynomial of degree @xmath197 . let @xmath198 for @xmath88 . let @xmath199 and assume that @xmath200 . hence @xmath201 is positive hyperbolic of degree @xmath202 . by the induction hypothesis the coefficients of all monomials of @xmath201 are nonnegative . let @xmath203 and deduce that the coefficients of all monomials of @xmath204 are nonnegative . hence the coefficients of all monomials of @xmath52 which include the variable @xmath140 are nonnegative . let @xmath88 to deduce that the coefficients of all monomials of degree one at least are nonnegative . as @xmath205 we deduce the proposition . @xmath192 + [ logconsumexp ] for @xmath206 let @xmath207 . let @xmath103 be a nonzero polynomial with such that the coefficient of each monomial is nonnegative . then @xmath208 is a convex function on @xmath92 . let @xmath209 be a line in @xmath92 . then the restriction of @xmath208 to @xmath210 is either an affine function or a strictly convex function . * proof . * the convexity of @xmath208 can be found in @xcite . note that @xmath211 is of the form in @xmath212 , where each @xmath213 and @xmath214 . if @xmath215 then @xmath216 is @xmath217 . otherwise it straightforward to show that @xmath218 is increasing on @xmath219 . @xmath192 + [ convlogpshyp ] let @xmath103 be a positive hyperbolic polynomial of degree @xmath109 . then @xmath208 is a convex function on @xmath92 . let @xmath209 be a line in @xmath92 . then the restriction of @xmath208 to @xmath210 is either an affine function or a strictly convex function . * examples of positive hyperbolic polynomials * 1 . let @xmath220 be a nonnegative matrix , denoted by @xmath221 , where each row of @xmath26 is nonzero . fix a positive integer @xmath222 $ ] . then @xmath223 is positive hyperbolic of degree @xmath81 in @xmath60 variables . 2 . let @xmath224 hermitian , nonnegative definite matrices such that @xmath225 is a positive definite matrix . let @xmath226 . then @xmath139 is positive hyperbolic . * proof . * 1 . first note that @xmath227 for @xmath228 . the hyperbolicity of @xmath229 and @xmath230 is obvious . assume that @xmath231 . let @xmath232 and define @xmath233 . then @xmath234 hence by lemma [ propphp ] @xmath235 positive hyperbolic . this is a standard example and the proof is straightforward . @xmath192 + let @xmath236 be a positive hyperbolic polynomial of degree @xmath109 . as in @xcite define @xmath237 it is possible that @xmath238 . for example let @xmath239 where @xmath240 are nonnegative integer whose sum is @xmath4 and @xmath241 . [ excappos ] let @xmath242 be a doubly stochastic matrix . let @xmath243 $ ] be positive hyperbolic defined as part 1 of the above example . then @xmath244 . let @xmath245 be a matrix with positive entries . then there exists two positive definite diagonal matrices @xmath246 , unique up to @xmath247 , such that @xmath248 is a doubly stochastic matrix @xcite . let @xmath249 be defined as above . then @xmath250 . * * consider first @xmath251 . since @xmath26 is row stochastic @xmath252 . hence @xmath253 . let @xmath254 be a probability vector . then for any @xmath255 the generalized arithmetic - geometric inequality states @xmath256 . use this inequality for each @xmath257 . the assumption that @xmath26 is doubly stochastic yields that @xmath258 . hence @xmath259 . let @xmath260 . then @xmath261 . hence hence @xmath262 . apply the arithmetic - geometric inequality to @xmath263 to deduce that @xmath264 . hence @xmath265 . it is straightforward to show that @xmath266 , where @xmath267 . hence @xmath250 . @xmath192 + the following result is taken from @xcite . [ derest ] let @xmath268 be an integer , @xmath269 and define @xmath270 . let @xmath271 . then @xmath272 for @xmath215 and @xmath273 for @xmath274 . for @xmath274 equality holds if and only if @xmath275 . * proof . * we can assume wlog that @xmath276 ; i.e. that @xmath277 . using the arithmetic - geometric means inequality we get that @xmath278 therefore , by doing basic calculus , @xmath279 which finally gives the desired inequality @xmath280 it follows again from arithmetic - geometric means inequality that the equality holds if and only if @xmath281 . @xmath192 + * definition . * let @xmath282 be a positive hyperbolic polynomial of degree @xmath4 . for each integer @xmath283 $ ] the _ i - th degree _ of @xmath52 is the integer @xmath284 $ ] such that @xmath285 let @xmath286 for @xmath88 . the following proposition follows straightforward from part _ 3 _ of lemma [ propphp ] . [ ranklem ] let @xmath103 be a positive hyperbolic polynomial of degree @xmath4 . let @xmath287 $ ] be an integer . then 1 . 2 . for each integer @xmath289 $ ] @xmath290 is a positive hyperbolic polynomial of degree @xmath291 . 3 . for each integer @xmath292 , j\ne i$ ] , @xmath293 the following result is crucial for the proof of a generalized friedland - tverberg inequality and is due essentially to the second author in @xcite . [ capest ] let @xmath282 be a positive hyperbolic polynomial of degree @xmath109 . assume that @xmath294 . then @xmath295 for @xmath296 . for @xmath297 @xmath298 where @xmath299 . * proof . * it is enough to prove the result for @xmath300 . suppose to the contrary that @xmath52 does not depend on @xmath301 . then let @xmath302 and @xmath303 in ( [ defcap ] ) to deduce that @xmath238 contrary to our assumption . assume that @xmath304 . let @xmath305 . let @xmath306 . proposition [ ranklem ] yields that @xmath307 is a positive hyperbolic function in @xmath129 variables of degree @xmath308 . hence @xmath309 for @xmath310 . thus for @xmath310 @xmath311 the second equality follows from part _ 2 _ of lemma [ propphp ] . assume in addition that @xmath312 . then @xmath313 . apply lemma [ derest ] to the right - hand side of ( [ capest1 ] ) to deduce that @xmath314 . since we assumed that @xmath312 it follows that @xmath315 . @xmath192 + [ concave ] lemma [ derest ] , which is simple but crucial , is a particular case of the following general result : + _ let @xmath316 be a nonnegative function differentiable at zero from the right ; @xmath317 . if @xmath318 and @xmath319 is concave then @xmath320 . on the other hand if @xmath318 and @xmath319 is convex then @xmath321 . _ [ frtver ] let @xmath322 be positive hyperbolic of degree @xmath323 $ ] . assume that @xmath324 $ ] for @xmath296 . rearrange the sequence @xmath325 in an increasing order @xmath326 . let @xmath327 $ ] be the smallest integer such that @xmath328 . then @xmath329 ( here @xmath299 , and the empty product for @xmath215 is assumed to be @xmath330 . ) if @xmath331 and @xmath332 for @xmath333 equality holds if and only if @xmath334 for each @xmath335 . * proof . * suppose that @xmath238 . then part _ 3 _ of lemma [ propphp ] yields that the left - hand side of ( [ frtver1 ] ) is nonnegative and the theorem holds in this case . clearly , it is enough to assume the case @xmath336 . the case @xmath17 is essentially proven in @xcite and we repeat its proof for the convenience of the reader . permute the coordinates of @xmath337 such that @xmath338}\deg_i p\le r_1^*$ ] . assume that @xmath339 . then lemma [ capest ] yields that @xmath340 is positive hyperbolic of degree @xmath129 and @xmath341 . since the sequence @xmath342 is decreasing to have the lowest possible lower bound we have to assume @xmath343 . suppose first that @xmath344 . repeating this process @xmath60 times we get that @xmath345 this inequality to corresponds to the case @xmath346 for @xmath88 . the equality case is discussed in @xcite . let @xmath347 $ ] . put @xmath348 . clearly , @xmath349 is positive hyperbolic of degree @xmath60 . since @xmath350 for each @xmath171 , it follows that @xmath351 . apply ( [ frtver1 ] ) to @xmath349 for @xmath17 to deduce ( [ frtver1 ] ) in the general case . since the equality case for @xmath349 holds if and only @xmath352 it follows that the equality in ( [ frtver1 ] ) holds if and only if @xmath353 . @xmath192 + let @xmath242 be a doubly stochastic matrix . apply this theorem to @xmath229 defined proposition [ excappos ] to deduce the friedland - tverberg inequality for the sum of all @xmath15 permanents of @xmath26 : [ frteverds ] let @xmath354 be a doubly stochastic matrix . then @xmath355 for any @xmath43 $ ] . equality holds if and only @xmath44 . [ gurschr ] ( ) let @xmath354 be a doubly stochastic matrix , such that each column contains at most @xmath356 $ ] nonzero entries . then @xmath357 * proof . * note that for @xmath358 we have that @xmath359 for @xmath88 . apply ( [ frtver1 ] ) to this case , i.e. @xmath17 , @xmath360 and @xmath361 to deduce the theorem . @xmath192 + let @xmath1 be a general graph with the set of vertices @xmath362 and edges @xmath6 . a _ matching _ in @xmath12 is a subset @xmath363 such that no two edges in @xmath5 share a common endpoint . the endpoints of the edges in @xmath5 are said to be _ covered _ by @xmath5 . we can think of each edge @xmath364 as occupied by a _ dimer _ , consisting of two neighboring atoms at @xmath365 and @xmath366 forming a bond , and of each vertex not covered by @xmath5 as a _ monomer _ , which is an atom not forming any bond . for this reason a matching in @xmath12 is also called a _ monomer - dimer cover _ of @xmath12 . if there are no monomers , @xmath5 is said to be a _ perfect matching_. note that if a perfect matching exists then @xmath367 is even . a matching @xmath5 with @xmath368 is called an _ @xmath81-matching_. we denote by @xmath369 be the number of @xmath81-matchings in @xmath12 ( in particular @xmath370 ) , and by @xmath371 the matching generating polynomial of @xmath12 . it is known that all the roots of matching polynomial are real negative numbers @xcite . let @xmath12 be a bipartite graph @xmath1 , where @xmath372 is the set of vertices of @xmath12 and @xmath6 is the set of edges that connect vertices in @xmath373 to vertices in @xmath374 . assume that @xmath375 . we identify @xmath373 and @xmath374 with @xmath376 , where the vertices in @xmath373 and @xmath374 are colored with colors black and white respectively . then @xmath12 is represented by @xmath78 @xmath27 matrix @xmath377 , where @xmath378 if and only if the black vertex @xmath379 is connected to the white vertex @xmath380 . it is convenient to consider multi bipartite graphs . thus , the entries of the representation matrix @xmath381 are nonnegative integers , where @xmath382 is the number of edges from the black vertex @xmath379 to the white vertex @xmath380 . it is straightforward to show that @xmath383 let @xmath35 be the set of bipartite @xmath0-regular multi - graphs , ( each vertex has degree @xmath0 ) , with @xmath384 . let @xmath385 be the set by an @xmath27 nonnegative matrices @xmath26 with integer entries , such that the sum of each row and column is @xmath0 . then each @xmath37 is represented by @xmath386 and vice versa . note for each @xmath387 the matrix @xmath388 is doubly stochastic . corollary [ frteverds ] yields : @xmath389 note that the symmetric group @xmath390 on @xmath60 elements , presented as the group of permutation matrices @xmath391 acts from the left and from the right on @xmath385 , i.e. @xmath392 for each @xmath393 . these actions are equivalent to the action of @xmath390 on @xmath373 and @xmath374 respectively . there is a standard probabilistic model on @xmath35 , which assigns a fairly natural probability measure @xmath394 on @xmath35 @xcite . the measure @xmath394 is invariant under the action of @xmath390 on @xmath373 and @xmath374 as explained above . by abuse of the notation we view @xmath394 also a probability measure on @xmath385 , which is invariant under the left and the right action of @xmath395 . the following result is proven in @xcite : [ mexpcval ] let @xmath394 be the probability measure defined above . then @xmath396 in particular , let @xmath397 $ ] , @xmath398 be a sequence of integers with @xmath399 $ ] . then @xmath400 where @xmath401 the case @xmath17 in ( [ mexpcval1 ] ) is given in @xcite . fix as subset @xmath402 of cardinality @xmath4 : @xmath403 . for @xmath46 let @xmath404 be all @xmath4-matching in @xmath12 that cover the set @xmath405 . for @xmath406 and @xmath407 let @xmath408 $ ] be the submatrix @xmath409 . denote @xmath410=\sum_{i\subset { \lann{\rangle}},\#i = m}{\mathrm{perm\;}}_m a[i|j].\ ] ] then @xmath411 $ ] . use the invariance of @xmath394 under the action of @xmath390 on @xmath374 and the fact that there are @xmath412 distinct subsets @xmath402 of cardinality @xmath4 to obtain : [ mexpcvalcor ] let @xmath394 be the probability measure defined above . then for any @xmath413 @xmath414)=\frac{{n \choose m } r^{2m}m!(rn - m)!}{(rn ) ! } , \quad m=0,\ldots , n.\ ] ] the following conjecture is stated in @xcite . [ lasmc ] for @xmath415 , let @xmath416 , @xmath417 be a sequence of finite @xmath0-regular bipartite graphs with @xmath418 . let @xmath419 $ ] , @xmath398 be a sequence of integers with @xmath420 $ ] . then @xmath421 for @xmath422 this conjecture holds trivially . for @xmath69 this conjecture is proved in @xcite . the inequality ( [ frtverbinmat ] ) implies that under the conditions of conjecture [ lasmc ] the following inequality holds , see @xcite @xmath423 where @xmath424 as usual , we denote by @xmath425 $ ] the algebra of polynomials in @xmath426 with real coefficients , by @xmath427 $ ] the zero polynomial , and by @xmath428 \subset { \mathbb{r}}[x]$ ] the subalgebra of polynomials with non - negative coefficients . we partially order @xmath425 $ ] by writing , for @xmath429 $ ] , @xmath430 when @xmath431 $ ] , and @xmath432 when @xmath431 \setminus { \{0\}}$ ] . clearly , if @xmath433 and @xmath434 , then @xmath435 unless @xmath436 and @xmath437 . let @xmath438 denote the union of @xmath125 complete bipartite graphs @xmath439 having @xmath0 vertices of each color class . it is straightforward to show that any finite graphs @xmath440 satisfy @xmath441 and that @xmath442 the following conjecture is stated in @xcite [ upmatcon ] let @xmath12 be a bipartite @xmath0-regular graph on @xmath443 vertices where @xmath444 . then @xmath445 , equality holding only if @xmath446 . for @xmath447 this conjecture is proved in @xcite . the above conjecture implies the following asymptotic upper matching conjecture @xcite . denote by @xmath448 be the countably infinite union of @xmath439 . let @xmath449 and @xmath450 $ ] be the pressure and the @xmath52-matching entropy associated and the with @xmath448 @xcite : @xmath451 @xmath452 where @xmath453 [ upasmatcon ] + for @xmath415 , let @xmath416 , @xmath417 be a sequence of finite @xmath0-regular bipartite graphs with @xmath418 . let @xmath419 $ ] , @xmath398 be a sequence of integers with @xmath420 $ ] . then @xmath454 equality case holds for the sequence @xmath455 . for @xmath69 the aumc is proven in @xcite . for @xmath456 and any @xmath457 the aumc follows from the proof of minc conjecture by bregman @xcite . some computations performed in @xcite support the almc and aumc . the following plots illustrating the asymptotic matching conjectures for @xmath458 . let @xmath459 a cycle on @xmath60 points , and let @xmath460 be a sequence of @xmath55 dimensional torii . note that each @xmath461 is @xmath462 regular graph . it is a classical result that the following limit exists for any @xmath463 $ ] : @xmath464.\ ] ] @xmath63 is the @xmath55-dimensional monomer - dimer entropy of dimer density @xmath62 $ ] in the lattice @xmath64 @xcite and @xcite . in this case we use the notation @xmath465 } h_d(p)$ ] , ( the quantities @xmath466 and @xmath467 are called the @xmath55-monomer - dimer entropy and the @xmath468-dimer entropy , respectively , in @xcite ) . figure [ fig : h2bounds ] shows various bounds and values for the monomer - dimer entropy @xmath469 of dimer density @xmath62 $ ] in the @xmath470-regular @xmath468-dimensional grid . ft is the friedland - tverberg lower bound @xmath471 of ( [ ftlowerbound ] ) , h2 is the true monomer - dimer entropy equal to @xmath472}h_2(p)$ ] ( it is known to a precision much greater than the picture resolution ) . the crosses marked b are baxter s computed values @xcite . almc is the function @xmath473 of ( [ defghrp ] ) , conjectured to be a lower bound in the asymptotic lower matching conjecture . aumc is the monomer - dimer entropy @xmath474 of dimer density @xmath52 in a countable union of @xmath475 , given by ( [ presk])([pt ] ) and conjectured to be an upper bound by the asymptotic upper matching conjecture . notice that aumc goes a little over h2 : a countable union of @xmath475 has a higher monomer - dimer entropy than an infinite planar grid . -dimensional grid : entropy as a function of dimer density . ft is the friedland - tverberg lower bound , h2 is the true monomer - dimer entropy . b are baxter s computed values . almc is the asymptotic lower matching conjecture . aumc is the entropy of a countable union of @xmath475 , conjectured to be an upper bound by the asymptotic upper matching conjecture.,scaledwidth=75.0% ] figure [ fig : h3bounds ] shows similarly various bounds and values for the monomer - dimer entropy @xmath476 of dimer density @xmath477 $ ] in the @xmath478-regular @xmath479-dimensional grid . ft is the friedland - tverberg lower bound @xmath480 of ( [ ftlowerbound ] ) , h3low and h3high are the best known lower and upper bounds for the true monomer - dimer entropy equal to @xmath472}h_3(p)$ ] . almc is the function @xmath481 of ( [ defghrp ] ) , conjectured to be a lower bound in the asymptotic lower matching conjecture . aumc is the monomer - dimer entropy @xmath474 of dimer density @xmath52 in a countable union of @xmath482 , given by ( [ presk])([pt ] ) and conjectured to be an upper bound by the asymptotic upper matching conjecture . notice that aumc goes a little over h3high : a countable union of @xmath482 has a higher monomer - dimer entropy than an infinite cubic grid . -dimensional grid : entropy as a function of dimer density . ft is the friedland - tverberg lower bound , h3low and h3high are the known bounds for the monomer - dimer entropy . almc is the asymptotic lower matching conjecture . aumc is the entropy of a countable union of @xmath482 , conjectured to be an upper bound by the asymptotic upper matching conjecture.,scaledwidth=75.0% ] in this section we prove the following case of almc : [ palmcc ] let @xmath40 be an integer . then the asymptotic lower matching conjecture ( [ lasmbdfp ] ) holds for @xmath483 . the proof of this theorem follows from the following results . [ frtverv1 ] let @xmath322 be positive hyperbolic of degree @xmath484 . assume that @xmath324 $ ] for @xmath296 . rearrange the sequence @xmath325 in an increasing order @xmath326 . let @xmath485 . let @xmath327 $ ] be the smallest integer such that @xmath486 . then @xmath487 * proof . * let @xmath488 be positive hyperbolic of degree @xmath489 with @xmath490 for @xmath88 and @xmath491 . then @xmath492 is positive hyperbolic of degree @xmath60 with @xmath493 and @xmath494 for @xmath88 . apply theorem [ frtver ] to @xmath144 to deduce @xmath495 where @xmath496 and @xmath497 . let @xmath498 and choose @xmath499 as in ( [ exhpk ] ) . note @xmath500.\ ] ] now take the expected value of the left - hand side of the inequalities ( [ frtver2v1 ] ) corresponding to all @xmath501 . use corollary [ mexpcvalcor ] to deduce that the coefficient of each @xmath502 is @xmath503 . @xmath192 + [ frtverv2 ] let @xmath322 be positive hyperbolic of degree @xmath484 . assume that @xmath504 $ ] for @xmath296 . let @xmath485 and @xmath505 . then @xmath506 [ frtverv3 ] let @xmath507 be a doubly stochastic matrix with at most @xmath0 nonzero entries in each column . let @xmath485 and @xmath505 . then for each @xmath484 @xmath508 * proof . * let @xmath509 as defined by ( [ exhpk ] ) . then ( [ frtver1v3 ] ) follow from corollary [ frtverv2 ] . @xmath192 + let @xmath510 . then @xmath511 is represented by its incidence matrix @xmath512 . let @xmath513 . then @xmath514 is a doubly stochastic matrix where each row and column of @xmath514 has at most @xmath0 positive entries . clearly , the almc conjecture follows from the following stronger conjecture : [ lasrpc ] for @xmath415 , let @xmath515 be a sequence of @xmath27 doubly stochastic matrices , where each column of each @xmath514 has at most @xmath0-nonzero entries . let @xmath397 $ ] , @xmath398 be a sequence of integers with @xmath399 $ ] . then @xmath516 theorem [ palmcc ] follows from the following result : [ palrpc ] let @xmath517 be integers let @xmath515 be a sequence of @xmath27 doubly stochastic matrices , where each column of each @xmath514 has at most @xmath0-nonzero entries . let @xmath397 $ ] , @xmath398 be a sequence of integers with @xmath399 $ ] . then @xmath518 moreover , the asymptotic lower @xmath0-permanent conjecture [ lasrpc ] holds for @xmath483 . * proof of theorem [ palrpc]*. apply the inequality ( [ frtver1v3 ] ) to @xmath514 for @xmath519 . take the logarithm of the both sides of this inequality and let @xmath47 . a straightforward calculation for the right - hand side , using the stirling s formula , yields the inequality ( [ lasrppbp ] ) . assume that @xmath520 . then @xmath521 . then the right - and side of ( [ lasrppbp ] ) is equal to @xmath522 . hence the asymptotic lower @xmath0-permanent conjecture [ lasrpc ] holds for @xmath523 . we now discuss the case @xmath524 , i.e. @xmath525 . let @xmath526 be any @xmath27 nonnegative matrix . denote by @xmath527 the bipartite graph induced by @xmath76 , i.e. the edge @xmath528 is in @xmath6 , if and only if @xmath529 . then @xmath76 induces the weighted graph on @xmath12 , where the weight of the edge @xmath528 is @xmath530 . let @xmath531 . @xmath532 is called the matching polynomial of the weighted graph @xmath12 . heilmann and lieb showed in @xcite that @xmath532 has nonnegative roots . ( see also @xcite . ) hence the arithmetic - geometric inequality for the elementary symmetric polynomials of the nonnegative roots of @xmath532 yields the inequality @xmath533 . ( see @xcite for the case of @xmath4-matchings in bipartite graphs . ) use theorem [ gurschr ] to deduce that @xmath534 . apply the above two inequalities for the sequence @xmath514 and @xmath519 for @xmath417 to deduce the case @xmath456 . @xmath192 + let @xmath535 and @xmath536 . then theorem [ palmcc ] yields that @xmath537 , which implies that @xmath538 } h_3(p)\ge .7845241927 $ ] . this improves the lower bound implied by ( [ lasmbdfp ] ) @xmath539 @xcite . the computations in @xcite yield that @xmath540 . thus @xmath74 $ ] . let @xmath1 be a graph on the set of vertices @xmath362 and the set of edges @xmath6 . assume that @xmath3 . then @xmath12 is represented by a symmetric @xmath78 matrix @xmath541 with @xmath542 diagonal . if @xmath12 has a perfect matching then @xmath543 is even . if @xmath12 is bipartite and @xmath372 , where @xmath544 , we deduce that @xmath545 where @xmath26 is the representation matrix of the bipartite graph @xmath12 . as explained above , the number of @xmath4-matching in the bipartite graph @xmath12 is @xmath82 . in this section we discuss the @xmath4-matching of a general graph @xmath12 , and the related function @xmath77 which counts the number of @xmath4-matching in @xmath12 . let @xmath546 be two nonempty sets of cardinality @xmath547 , @xmath548 , respectively . we then arrange the elements of @xmath549 and @xmath550 in an increasing order : @xmath551 . for @xmath552 we denote by @xmath553\in c^{i\times j}$ ] the submatrix @xmath554 . denote by @xmath555 the space of real valued @xmath556 symmetric matrices , the cone @xmath556 symmetric matrices with nonnegative entries , the subspace of real valued @xmath556 symmetric matrices with zero diagonal , and the subcone of @xmath556 symmetric matrices with zero diagonal and nonnegative entries respectively . let @xmath557 and an integer @xmath558 $ ] . then the @xmath559 _ hafnian _ of @xmath76 is defined as @xmath560.\ ] ] that is if @xmath561 is an @xmath4 matching of a complete graph @xmath562 on @xmath2 vertices , then the product @xmath563 appears exactly once in @xmath77 . since @xmath564 there are @xmath565 choices of @xmath566 and @xmath567 for which this product appear , we need to use the factor @xmath568 in the above definition of @xmath77 . if @xmath541 then @xmath77 gives the number of @xmath4-matching in @xmath12 . note that from the definition of @xmath569 it follows that @xmath570 . equivalently , it is straightforward to show : @xmath571 unfortunately , the quadratic polynomial @xmath79 is not always positive hyperbolic . note that @xmath77 does not depend on the value of the diagonal entries @xmath76 let @xmath572 be the matrix obtained from by replacing the diagonal entries of @xmath76 by zero elements . then @xmath573 . for @xmath574 we denote by @xmath575 the @xmath202 eigenvalues of @xmath76 counted with their multiplicities and arranged in the decreasing order . as usual for @xmath576 we denote @xmath577 if @xmath578 is a nonnegative definite matrix . the maxmin , or minmax charaterization of @xmath579 yields that if @xmath577 then @xmath580 for @xmath581 . in particular , if @xmath582 has nonnegative diagonal entries then @xmath583 and @xmath584 for @xmath581 . the following result is well known and we bring its proof for completeness . [ quadfrmph ] let @xmath585 and @xmath586 . then @xmath79 is positive hyperbolic if and only if @xmath587 and @xmath588 . * proof . * assume that @xmath79 is positive hyperbolic . then proposition [ nonnegcphp ] yields that @xmath589 . since @xmath590 for each @xmath228 @xmath591 . hence @xmath592 . observe next that the positive hyperbolicity of @xmath79 is equivalent to @xmath593 clearly , the above condition holds for any @xmath85 . the perron - frobenius theorem yields that there exists @xmath594 such that @xmath595 . let @xmath596 and assume that @xmath597 . then ( [ quadfrmph1 ] ) yields that @xmath598 . hence @xmath599 . vice versa suppose that @xmath600 and @xmath599 . recall that there exists a permutation matrix @xmath601 such that @xmath602 is a block diagonal matrix @xmath603 , where @xmath604 is irreducible . so @xmath605 unless @xmath606 and @xmath607 . hence our assumptions yield that we may assume that @xmath608 is irreducible and @xmath607 for @xmath609 . thus it is enough to show that @xmath79 is positive hyperbolic for an irreducible @xmath610 , where @xmath586 . clearly @xmath590 for @xmath97 . thus it is left to show that ( [ quadfrmph1 ] ) holds . assume first that @xmath611 . then ( [ quadfrmph1 ] ) follows straightforward . suppose @xmath97 is any vector . then there exists a unique diagonal matrix @xmath612 , with positive diagonal entries such that @xmath613 . that is @xmath614 . replacing @xmath615 by @xmath616 we deduce the inequality second inequality of ( [ quadfrmph1 ] ) . @xmath192 + [ kpartdef ] let @xmath1 be a graph on the set of vertices @xmath362 . then for @xmath274 @xmath12 is called @xmath81-partite , if the following condition holds . there exists a decomposition of @xmath362 to a disjoint union of @xmath81 nonempty sets @xmath617 such that each edge @xmath618 connects @xmath619 to @xmath620 for some @xmath621 . @xmath12 is called a complete @xmath81-bipartite , if there exists a decomposition of @xmath362 to a disjoint union of @xmath81 nonempty sets @xmath617 such that @xmath6 consists of all edges from @xmath619 to @xmath620 for all @xmath622 . note that a complete graph @xmath12 with @xmath586 vertices is complete @xmath60- partite . + we need the following elementary result whose proof is straightforward . [ 3-minor ] let @xmath623 and @xmath624\in { \mathrm{s}}_n({\mathbb{r}})$ ] with @xmath625 on the main diagonal . suppose that and all subsets of cardinality three @xmath626 the principal submatrix @xmath627 $ ] is nonnegative definite . if @xmath628 then @xmath629 . [ 01pathypq ] let @xmath630\in{\mathrm{s}}_{n}({\mathbb{r}}_+)$ ] , i.e. @xmath76 is a real @xmath27 symmetric matrix with nonnegative entries . denote by @xmath631 the graph , ( with no self - loops ) , induced by @xmath76 , i.e. @xmath632 if and only if @xmath633 and @xmath529 . assume that @xmath76 is irreducible and @xmath599 . then @xmath634 is a complete @xmath81-partite graph for some @xmath635 $ ] . * proof . * since @xmath572 is irreducible and @xmath583 , it is enough to prove the lemma in the case where all the diagonal entries of @xmath76 are equal to zero , i.e. @xmath636 . let @xmath637 be the unique positive eigenvector of @xmath76 , corresponding to the maximal eigenvalue @xmath638 , satisfying the condition @xmath639 . then @xmath640 and @xmath641 let @xmath642 . then the zero pattern of the matrix @xmath643 is the same as of the matrix @xmath76 . let @xmath624 : = { \mathbf{1}}{\mathbf{1}}{^\top}- c$ ] , where @xmath644 is the vector of all ones . then @xmath624 \succeq { \mathbf{0}}$ ] . notice that @xmath645 , and for @xmath633 @xmath646 if and only @xmath647 , ( i.e. the vertices @xmath547 are not connected in the graph @xmath634 ) . as any principal submatrix of @xmath648 is nonnegative definite , it follows from proposition [ 3-minor ] that for any triplet @xmath649 such that if @xmath628 then also @xmath629 . in other words , the relation @xmath650 is an equivalence relation . therefore the graph @xmath634 is complete @xmath81-partite , where each equivalence class of vertices corresponds to some class @xmath619 in the @xmath81-partite graph . [ hypkpgrh ] let @xmath12 be a graph on @xmath651 vertices , and denote by @xmath19 the incidence matrix of @xmath12 . then @xmath652 is positive hyperbolic if and only if @xmath12 is a union of a complete @xmath653-partite graph on at least two vertices and of isolated vertices . * proof . * we first show that for a complete @xmath81-partite graph @xmath12 on at least @xmath586 vertices @xmath654 , which is equivalent to the positive hyperbolicity of @xmath652 in view of lemma [ quadfrmph ] . let @xmath29 be a symmetric matrix whose all entries are equal to @xmath330 . then @xmath655 is rank one matrix with @xmath656 and @xmath657 for @xmath658 . let @xmath327 $ ] , @xmath659 and @xmath660 . consider the block diagonal matrix @xmath661 . clearly @xmath662 is a nonnegative definite matrix . it is straightforward to see that renaming the vertices of @xmath12 , we will obtain that @xmath663 for some unique @xmath664 . then minimax characterization of @xmath665 yields that @xmath666 . hence @xmath652 is positive hyperbolic . assume now that @xmath652 is positive hyperbolic . therefore @xmath12 must have at least one edge and @xmath667 . hence @xmath12 has one connected component containing at least two vertices and a union of isolated vertices . without loss of generality we assume that @xmath12 is connected . then @xmath19 satisfies the assumptions of lemma [ 01pathypq ] and @xmath12 is @xmath81-partite . @xmath192 + [ reverse ] let @xmath26 be a symmetric @xmath668 matrix with nonnegative entries . it is straightforward to show that the polynomial @xmath669 is positive hyperbolic if and only if the function @xmath670 is concave on the positive orthant @xmath671 . if @xmath26 is real , symmetric , nonnegative definite then @xmath670 is convex on @xmath672 . in view of remark [ concave ] , it is natural to conjecture that if @xmath26 is @xmath673 real , symmetric , nonnegative definite then the reverse van der waerden bound holds : @xmath674 one of the main purposes of this paper is to construct a generating homogeneous polynomial @xmath139 of degree @xmath60 , for some quantity of interest @xmath675 as hafnian : @xmath676 , sums of subhafnians : @xmath677 , permanent : @xmath23 , sum of subpermanents : @xmath14 , such that @xmath678 . if such polynomial is positive hyperbolic then we can apply the results from @xcite and the results of the previous sections of this paper to get a lower bound on @xmath679 and even to get deterministic polynomial - time algorithms to approximate @xmath675 within simply exponential factor as in @xcite . in the general , ( not positive hyperbolic case ) , we can use this representation to obtain exact algorithm to compute @xmath675 in @xmath680 number of arithmetic operations provided the the generating polynomial @xmath52 can be itself evaluated in @xmath681 number of arithmetic operations . we present below some examples of this approach . [ [ formula - for - sum_1le - i_1-ldotsi_mle - n - fracpartial - m - ppartial - x_i_1ldotspartial - x_i_m - mathbf0 ] ] formula for @xmath682 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ let @xmath686 be the unique solution of the system of linear equations @xmath687 , where the @xmath688 lower triangular matrix @xmath24 $ ] is defined as follows : @xmath689 then the following equality holds : @xmath690 notice that this formula requires @xmath691 evaluations of the polynomial @xmath52 . 1 . let @xmath697 and @xmath698\cap{\mathbb{n}}$ ] . define , as in the proof of theorem 3.1 , the polynomial : @xmath699 + as @xmath700 , the application of [ ryzer ] gives the following ryser - like formula for @xmath77 : @xmath701 where @xmath702 are defined by ( [ defs_i ] ) for @xmath703 . the formula ( [ ryzer - haf ] ) provides @xmath704 , algorithm to compute @xmath705 , where @xmath706 . 2 . for @xmath707 let @xmath708 be the @xmath709 symmetric function of @xmath102 . let @xmath26 be @xmath668 complex matrix and define @xmath710 . then @xmath711 can be evaluated in @xmath712 arithmetic operations and @xmath713 which gives the following algorithm to evaluate @xmath714 @xmath715 where @xmath702 are defined by ( [ defs_i ] ) for @xmath716 . the formula [ ryzer - perm ] provides @xmath717 algorithm to compute @xmath718 . + notice that the naive algorithm , i.e. computing and adding all @xmath719 subpermanents , requires @xmath720 arithmetic operations . in this section we always assume that @xmath721 , @xmath722 . suppose that a positive hyperbolic polynomial @xmath723 has nonnegative integer coefficients and is given as an oracle . i.e. we do nt have a list coefficients , but can evaluate @xmath139 on rational inputs . the number @xmath724 measures the complexity of the polynomial @xmath52 . a deterministic polynomial - time oracle algorithm is any algorithm which evaluates the given polynomial @xmath139 at a number of rational vectors @xmath725 which is polynomial in @xmath60 and @xmath724 ; these rational vectors @xmath726 are required to have bit - wise complexity which is polynomial in @xmath60 and @xmath724 ; and the number of additional auxiliary arithmetic computations is also polynomial in @xmath60 and @xmath724 . + if the number of oracle calls , ( evaluations of the given polynomial @xmath139 ) , the number of additional auxiliary arithmetic computations and bit - wise complexity of the rational input vectors @xmath726 are all polynomial in @xmath60 , ( no dependence on @xmath724 ) then such algorithm is called deterministic strongly polynomial - time oracle algorithm . + the following theorem was proved in @xcite . [ conv ] there exists a deterministic polynomial - time oracle algorithm , which computes for given as an oracle indecomposable positive hyperbolic polynomial @xmath139 a number @xmath727 , satisfying the inequality @xmath728 the algorithm behind theorem 7.1 is based on two observations . first , @xmath731 where we used the notation of proposition [ logconsumexp ] . if the coefficients of the polynomial @xmath52 are nonnegative then the functional @xmath732 is convex , the indecomposabilty of the polynomial @xmath52 is exactly uniqueness and existence of the minimum in [ conv - rel ] . + second point is the inequality : @xmath733 if the positive hyperbolic polynomial @xmath52 is not indecomposable , we need first to check if @xmath734 . if @xmath735 then also @xmath736 . in the case @xmath734 we `` slightly '' perturb the polynomial @xmath52 to get the indecomposability . + theorem 3.1 in this paper provides an analogue of the left inequality in ( [ cap - waer ] ) for positive hyperbolic polynomial @xmath730 of degree @xmath737 . the problem is that in this case we do nt have the right inequality . i.e. it is possible that @xmath738 but @xmath739 . this problem can be easily overcome by the following equivalent reformulation of theorem 3.1 : [ zer check ] let @xmath743 be a positive hyperbolic polynomial of degree @xmath737 given as an oracle . then there exists a strongly polynomial - time , ( in @xmath60 ) , oracle algorithm to check if @xmath744 . * * it follows from theorem [ tver - new ] that @xmath744 if and only if @xmath745 , where @xmath746 is positive hyperbolic polynomial of degree @xmath60 . notice the polynomial @xmath125 is easy to evaluate given an oracle evaluating the polynomial @xmath52 . let @xmath747 , the support @xmath748 , the newton polytope is the convex hull of the support @xmath749 . it was proved in @xcite that an integer vector @xmath750 if and only if @xmath751 . corollary 4.3 in @xcite provides a strongly polynomial ( in @xmath60 ) oracle algorithm for the membership problem @xmath752 for positive hyperbolic polynomial @xmath730 of degree @xmath60 . 1 . there exists a strongly polynomial - time , ( in @xmath60 ) , oracle algorithm to check if + @xmath753 . 2 . there exists a deterministic polynomial - time oracle algorithm which computes a number @xmath754 satisfying the inequality @xmath755 . let the assumptions of corollary [ 2cor ] holds . to approximate @xmath760 we used the identity @xmath761 for @xmath762 . it is natural to ask we can improve our estimates if we choose a different positive hyperbolic @xmath763 such that ( [ identr ] ) holds . denote the set of positive hyperbolic polynomials of degree @xmath4 and in @xmath60 variables as @xmath764 ; and by @xmath765 the subset of polynomials satisfying @xmath766 for all @xmath767 . define @xmath768 assume that @xmath769 . since @xmath0 is positive hyperbolic , all its monomials have nonnegative coefficients . hence @xmath770 for any @xmath171 . in particular @xmath771 . by choosing in @xmath772 definition @xmath773 @xmath774 we deduce straightforward that @xmath775 . we presented in this section one `` natural '' generating polynomial for the hafnian , and described all symmetric boolean matrices such that this polynomial is positive hyperbolic . it is an interesting open problem whether there exists a generating positive hyperbolic polynomial for the hafnians of boolean matrices which can can evaluated in polynomial time . if the answer to this problem is negative it can explain why approximating the hafnian ( number of perfect matchings in general graphs ) is `` harder '' than the same problem for the permanent . it is easy to prove that computing the hafnian of integer symmetric @xmath673 matrices with nonnegative entries and the signature @xmath781 if @xmath782-complete . the results in this paper allow to use sinkhorn s scaling to approximate the hafnian within multiplicative factor @xmath783 in this `` hyperbolic '' case . a. ben - dor and s. halevi , zero - one permanent is @xmath782 complete , a simpler proof , _ theory and computing systems _ 108 - 117 . bregman , some properties of nonnegative matrices and their permanents , _ soviet math . _ 14 ( 1973 ) , 945 - 949 . d.i . falikman , proof of the van der waerden conjecture regarding the permanent of doubly stochastic matrix , _ math . notes acad . 29 ( 1981 ) , 475479 . s. friedland , a lower bound for the permanent of doubly stochastic matrices , _ ann . of math . _ 110 ( 1979 ) , 167 - 176 . hopcroft and r.m . karp , an @xmath784 algorithm for maximum matchings in bipartite graphs , _ siam j. comput . _ 2 ( 1973 ) , 225231 . hammersley , existence theorems and monte carlo methods for the monomer - dimer problem , in _ reseach papers in statistics : festschrift for j. neyman _ , edited by f.n . david , wiley , london , 1966 , 125146 .

the uhf wave function may be written as a spin - contaminated _ pair _ wave function of the apsg form , and the overlap of the alpha and beta corresponding orbitals of the uhf solution can be taken as a proxy for the strength of the correlation captured by breaking symmetry . we demonstrate this with calculations on one- and two - dimensional hydrogen clusters and make contact with the well studied hubbard model . the uhf corresponding orbitals pair in a manner that allows a smooth evolution from doubly occupied orbitals at small distance to one in which wave function breaks symmetry , segregating the and electrons onto distinct sublattices at large distances . by performing spin projection on these uhf solutions , we address strong correlations that are difficult to capture at intermediate distances using a single determinant . approved for public release : la - ur-13 - 22691 .

1304.4457

the proper treatment of what has come to be known as the `` strong correlation '' problem remains a fundamental challenge to our understanding of unconventional superconductivity , frustrated spin lattices , the heavy fermion problem , and the class of materials known as mott insulators . materials are typically considered strongly correlated when conventional density functional approximations fail to qualitatively describe their properties . for example , the local density and generalized gradient approximations ( the lda and gga ) of density functional theory ( dft ) predicted nio to be a ferromagnetic metal . this catastrophically failed to describe the experimental properties : nio behaves as an antiferromagnetic insulator with an optical gap of the order of 4 ev . @xcite this led to statements by some that `` dft can not describe mott insulators . '' recent research has shown that this failure lies in the local or semi - local exchange - correlation approximations typically employed in the field , and not with dft itself . including a fraction of the fully _ non - local _ hf exchange term in the functional remedies many of these problems , and these hybrid functionals yield significantly improved band gaps , lattice constants , and magnetic properties for mott insulators.@xcite the strong correlations responsible for opening a gap in nio do not vanish if it is made metallic by application of pressure , and so conventional local and semi - local approximations to dft have problems with strongly correlated metals as well . mounting evidence suggests that the hybrid functionals do not perform well in this regime either . @xcite theorists have struggled with this problem for decades , and many approaches that build upon traditional dft have attempted to address it . these include the self - interaction - correction ( sic ) to dft,@xcite dft+u approaches,@xcite and many - body approaches such as the gw approximation,@xcite and dynamic mean - field theory ( dmft).@xcite the sic offered a significant improvement over conventional dft in the mott insulator regime but vanishes for a metal , necessitating additional approximations in that realm.@xcite the gw approach , a many - body green s function method based on the random phase approximation , successfully predicted band gaps in many materials and some mott insulators . however , the strength of this method , particularly its focus on the quasiparticle excitation spectrum , also made it difficult to obtain a total energy even though this energy is connected to the rpa ground state correlation.@xcite dmft , while formally a non - empirical embedding method , relies in practice on material dependent parameters for its utilization , as do the dft+u approaches.@xcite a truly predictive , parameter free , _ ab initio _ description of the electronic structure of strongly correlated materials remains a significant challenge for theory . the mott transition bears striking similarities to one of the oldest problems in quantum chemistry : making and breaking chemical bonds . specifically , consider breaking the simplest bond , that of ( see figure [ fig : h2 ] ) . near equilibrium , a symmetry restricted hartree - fock ( rhf ) , doubly occupied @xmath2 molecular orbital ( mo ) , dominates the wave function . this description fails abysmally in the dissociation limit ; the double occupancy implicit in the @xmath3 configuration contributes energetically unfavorable ionic components to the wave function at large bond lengths . in this limit , the bonding and anti - bonding molecular orbitals become nearly degenerate ; when expressed in this symmetry restricted mo basis , a proper wave function requires two configurations , @xmath3 and @xmath4 with approximately equal contributions ( the full configuration interaction , or fci wave function in a minimum basis ) . a superposition of these two configurations with variationally determined weights evolves smoothly from the delocalized mo limit to a localized valence bond ( vb ) description . similarly , the transition from a simple metal to a typical mott insulator requires a wave function which continuously evolves from the independent electron , delocalized bloch orbital description into a strongly correlated , valence bond - like wave function as the localization increases . the success of hybrid functionals in describing mott insulators relies , in part , on relaxing symmetry constraints imposed on the molecular orbitals . in the unrestricted hartree - fock ( uhf ) or kohn - sham ( uks ) approximation each electron occupies its own spatial orbital . @xcite this allows the wave function to localize the electrons when appropriate , one to each atom in at large distance or one per site in an anti - ferromagnet , thereby avoiding the unfavorable ionic component implicit in the restricted approximation . relaxing symmetry constraints sacrifices the quantum numbers associated with both spatial and spin symmetries . nevertheless , the unrestricted ( uhf ) wave function allows a qualitatively ( if not quantitatively ) acceptable description of both the mott insulator and simple metal limits . we should not be surprised that hybrid functionals have difficulty describing the correlated metal regime,@xcite corresponding to the intermediate bond lengths of figure [ fig : h2 ] . in this region , strong correlations expose the problems inherent in single determinant methods , and a successful theory must delicately balance the competition between localization and delocalization . in this work , we focus on understanding this intermediate `` recoupling '' region , in which neither the mo nor vb picture , neither the restricted or unrestricted approximations are appropriate . we begin by reviewing some concepts regarding broken symmetry and pair wave functions , then briefly discuss wave functions which can not be expressed in terms of pair functions . next we describe projection methods which restore the symmetries , illustrated with calculations on finite size hydrogen networks . finally , we conclude by discussing the implications for truly periodic networks . we begin by reviewing the wave function of diatomic hydrogen as a function of bond distance . our discussion utilizes a minimal basis set , the simplest expansion which captures the near degeneracies between the molecular orbitals that occur as the bond is stretched . a minimum basis set also provides a conceptual link with the hilbert space utilized in the widely studied hubbard model,@xcite though we retain the exact hamiltonian in the calculations to follow . these near degeneracies give rise to our primary concern , the so - called left - right , or static correlation . static correlation influences the _ form _ of the wave function , and can generally be accounted for in the hilbert space of the minimum basis . in a condensed matter context this might be referred to as antiferromagnetic correlation . dynamic correlation , or the instantaneous avoidance of one electron by another , constitutes a much weaker effect and , generally speaking , the local and semi - local correlation functionals of dft describe it adequately . that is to say , for a molecule qualitatively well described by a single determinant of doubly occupied orbitals then many - body perturbation theory , coupled - cluster theory , etc . , can describe the dynamic correlation , although it requires an extension of the basis set . to this end , consider a single @xmath5 orbital centered on each of two hydrogen atoms , denoting these as @xmath6 and @xmath7 , assigned to the left and right atomic centers , respectively . for simplicity , we assume orthogonality of these atomic wave functions , though this is not necessary in principle . form molecular orbitals using the standard linear combinations of atomic orbitals , giving gerade ( @xmath2 , bonding ) and ungerade ( @xmath8 , anti - bonding ) orbitals : @xmath9 the rhf determinant with the bonding orbital doubly occupied approximates the exact wave function quite closely at the equilibrium bond length . this wave function can be written as : @xmath10=\mathcal{a}[\sigma_{g}\overline{\sigma}_{g}],\ ] ] where @xmath11 is the anti - symmetrizer , @xmath0 and @xmath1 are the normal spin functions and we have used an overline to indicate beta spin - orbitals . one may then expand this wave function in the original atomic orbital basis : @xmath12 by defining the following spatial functions : @xmath13 these two spatial functions describe the two components of gerade symmetry that contribute to the @xmath14 ground state of . the first term , @xmath15 , describes the covalent portion of the wave function , in the heitler - london ( hl ) sense . the ionic component , @xmath16 , describes the symmetric combination ( @xmath17 ) of charge - transfer terms and . for completeness , define two spatial wave functions of ungerade symmetry that play roles in the excited @xmath18 and @xmath19 states and appear later in the discussion of the uhf wave function : latexmath:[\[\begin{aligned } at long range , the correct wave function for the ground state of contains only the hl term , @xmath15 . the rhf wave function dissociates improperly precisely because the ionic piece imposes an energy cost equal to the ionization potential of one hydrogen atom minus the electron affinity of the other . this quantity is referred to as @xmath21 in the hubbard ubiquitious in the condensed matter literature and @xmath22 in the pariser - parr - pople model familiar to quantum chemists . it is a simple measure of the on - site electron - electron repulsion . qualitatively , what does this mean ? at small distances , delocalizing the electrons over both centers allows the electrons to gain sufficient kinetic energy to overcome the on - site repulsion . at large distances , the kinetic energy obtained from delocalization becomes small , overwhelmed by the associated on - site electron repulsion , and the electrons localize . an acceptable approximation requires a wave function with can evolve smoothly between these two limits . the simplest wave function for that smoothly evolves from a delocalized mo description to a localized vb description is the generalized valence bond ( gvb ) wave function . @xcite for two electrons this may be thought of as a multi - determinant extension of rhf to an open - shell singlet pair ( geminal ) wave function of the form : @xmath23 { |0,0\rangle},\ ] ] where the _ non - orthogonal _ spatial functions @xmath24 must be determined self - consistently at each distance . the non - orthogonality allows each orbital to polarize continuously as a function of bond length , providing enough variational flexibility to successfully break the bond . expanding each of these non - orthogonal orbitals in the original mo basis : @xmath25 introduces a mixing parameter @xmath26 , a real number between -1 and 1 . the spatial overlap of the non - orthogonal orbitals can be calculated : @xmath27 and after some algebra , eq . becomes : @xmath28 this form makes explicit the relationship between the ionic and covalent portions of the wave function , while remaining equivalent to the multi - configuration representation more familiar to quantum chemists : @xmath29{|0,0\rangle}.\ ] ] for a single pair in a minimal basis set , this wave function is exact , with the mixing parameter @xmath26 and the orbitals determined self - consistently . we have belabored this simple two - electron example because we wish to emphasize the usefulness of this construct for strongly correlated materials . pushing the solid - state analogy further , this simple pair function continuously connects an ionic charge - density - wave @xmath30 , a simple metal @xmath31 , and the antiferromagnetic singlet @xmath32 with a single mixing parameter . to extend this pairing concept to systems with more electrons , one writes the wave function for @xmath33 electrons as an anti - symmetrized product of geminals ( apg ) . for an extensive review of geminal wave functions see , for example , the work of @xcite and the references therein . in the apg formalism , electrons not explicitly correlated are grouped in the core and frozen from excitation . each correlated mo @xmath34 pairs with an unoccupied partner @xmath35 in the following manner : @xmath36\prod_{i=1}^{n } \frac{\psi_i\overline\psi_i - \lambda_{i}\psi_{i^{\prime}}\overline\psi_{i^{\prime}}}{\sqrt{1+\lambda_{i}^{2}}}.\ ] ] if one imposes the condition of strong orthogonality between the pairs , the wave function is denoted as an anti - symmetrized product of strongly orthogonal geminals ( apsg ) . in addition to strong orthogonality , enforced by restricting each pair to distinct occupied and virtual orbital spaces , one may further restrict the pairs to pure spin states , a constraint denoted as `` perfect pairing''.@xcite such a wave function explicitly neglects inter - pair correlation effects , as well as _ same - spin _ correlation , that may be important in many cases , particularly for high spin multiplets . the uhf approach eschews the multi - determinant nature of gvb in favor of a single determinant wave function . allowing different spins to occupy different spatial orbitals provides enough flexibility for the electrons to balance the contributions from the covalent and ionic terms . at short distances this approach yields the rhf wave function , but as the bond stretches past some critical distance , known as the coulson - fischer ( cf ) point , the kinetic energy and nuclear attraction terms associated with the ionic portion no longer compensate for the on - site repulsion , and a distinct _ broken - symmetry _ uhf determinant emerges as the ground state . artificially breaking spatial symmetry by placing electrons of different spin in different orbitals finesses the electron correlations that dissociate the bond , projecting away the ionic component of the rhf wave function . breaking symmetry in the presence of orbital near degeneracies should not lower the energy , but the fact that the uhf energy lies lower than the rhf energy lies at the heart of lwdin s symmetry dilemma : one obtains a better approximation to the total energy by artifactually sacrificing good quantum numbers . @xcite returning to the simple two - electron case , the uhf wave function takes the form : @xmath37 where @xmath38 and @xmath39 are the non - orthogonal orbitals defined by the mixing coefficient @xmath26 in equation . in the dissociation limit , these orbitals reduce to the atomic orbitals associated with each atom , and the wave function becomes either : @xmath40 both of which break spatial symmetry . the spin products @xmath41 and @xmath42 are not eigenfunctions of spin , but rather a linear combination of the singlet and triplet functions : @xmath43 where @xmath44 is the @xmath45 component of the triplet : @xmath46 rewriting the full uhf wave function by separating the spatial parts into symmetric and anti - symmetric components , we identify the spatial functions associated with the singlet and triplet spin components : @xmath47 the singlet component ( up to normalization ) is identical to that of the gvb wave function , and the anti - bonding covalent spatial function @xmath48 from equation possesses ungerade symmetry to couple to the @xmath45 triplet spinor . thus , the variational parameter @xmath26 , which becomes non - zero for all bond lengths greater than the cf point , controls the spin symmetry breaking . in the limit where @xmath49 these orbitals simply reduce to @xmath6 and @xmath7 above ; in the limit @xmath50 the wave function becomes identical to rhf ; and in the limit @xmath51 the wave function describes a spin - contaminated charge density wave . while the uhf wave function correctly dissociates the molecule using a single determinant , this comes at a rather high price : a discontinuity in the derivative of the energy with respect to the internuclear separation occurs at the cf point , and at large distances the wave function no longer possesses the correct parity or spin . it is , of course , possible to break further symmetries of a single - determinant wave function . these have been explored in detail by fukutome@xcite as well as stuber and paldus.@xcite in our opinion , this work has not received the attention that it deserves because breaking additional symmetries sacrifices more quantum numbers that , in turn , become quite difficult to recover . an approach of particular note , the generalized hartree - fock approximation ( ghf ) , allows each electron to occupy an orbital without definite @xmath0 or @xmath1 spin . computationally , one typically accomplishes this by allowing the mo coefficients to become complex , i.e. each real atomic orbital gets both a real and imaginary @xmath0 and @xmath1 coefficient . the ghf approach is fraught with other difficulties , as dissociation curves often contain energy crossings and other weird behavior . for these reasons , the community has largely neglected the ghf approach . however , preserving some symmetries of the wave function in a single determinant approach ( e.g. uhf ) comes with an additional price : closed shell molecules ca nt in general be dissociated into the correct open - shell fragments . however , using the ghf approach recovers size consistency in the resulting wave function.@xcite we direct the reader to the rather vast literature on the subject of geminal functions and symmetry for an in depth discussion ( see , for example , ref . ( ) and ( ) and the references therein ) , and have simply presented a few conceptual ideas here which shed light on some of the issues that arise when we encounter spin projection below . the uhf wave function possesses a particularly interesting property , pointed out in different forms by many authors . lwdin proposed the `` pairing '' theorem@xcite which states that the orbitals of a wave function of the uhf type can always be transformed such that the alpha and beta orbitals are mutually orthogonal except for `` corresponding pairs '' of orbitals which have non - zero overlap . this follows from the corresponding orbital transformation of amos and hall@xcite or , alternatively , from the left and right eigenvectors of a singular value decomposition . consider a uhf wave function defined by a set of occupied alpha spin orbitals , @xmath52 , and a set of occupied beta spin orbitals , @xmath53 , orthonormal within themselves , but not mutually so . denoting overlap between the two sets as @xmath54 , they can be brought to bi - orthogonal form by diagonalizing the matrices @xmath55 and @xmath56 withing the subspace of the occupied @xmath0 and @xmath1 electrons , respectively , to generate two new sets of spin - orbitals , @xmath57 and @xmath58 : @xmath59 the overlap matrix becomes diagonal in these subspaces , i.e. : @xmath60 and the eigenvalues @xmath61 are bound between 0 and 1 . thus , each @xmath0 orbital pairs with a single @xmath1 orbital , and the overlap for pair @xmath62 is given by @xmath63 . although a somewhat straightforward result , it does not seem to have been emphasized that because the corresponding orbitals @xmath64 and @xmath65 are obtained from unitary transformations of the occupied @xmath0 and @xmath1 uhf orbitals , the uhf wave function may be written as an apsg : @xmath66\prod_{i=1}^{n } \omega_i(1,2),\ ] ] where the geminals @xmath67 are given by : @xmath68 and the spatial overlap of the non - orthogonal alpha and beta orbitals making up the pair is : @xmath69 this pair function extends the gvb pair idea , retaining the condition of strong orthogonality , but relaxing the perfect pairing constraint and allowing spin contamination from the triplet component of the state @xmath41 associated with each geminal . the non - orthogonal representation of the pair above can of course be transformed back into an orthogonal `` natural orbital '' representation , or expressed in the space of the symmetry restricted molecular orbitals as in equation [ eq : mixorb ] . thus the uhf wave function also contains only intra - pair correlation , though relaxing the perfect pairing constraint allows it the ability to capture more correlation than the gvb approach at the cost of the spin quantum numbers . finally , we wish to emphasize that the ghf wave function is not an apg , as it also captures inter - pair correlation , retaining size consistency . lwdin pointed out the usefulness of the one - particle density matrix ( 1pdm ) for defining the natural spin - orbitals of a system , and noted that the eigenvalues of this matrix were constrained to be between zero and one.@xcite coleman later argued that all two - particle interactions can be described using the one and two - particle density matrices . @xcite we therefore pause for a moment to remark on the structure of the 1pdm of the uhf and ghf wave functions . one can separate the 1pdm into spin blocks and write the 1pdm of a uhf wave function in the following way : @xmath70 the half sum and half difference of the non - zero blocks form the charge ( @xmath71 ) and spin densities ( @xmath72 ) , respectively : @xmath73 the magnetization describes the uhf spin contamination arising from the triplet contribution to the broken symmetry wave function . we call such a wave function _ collinear _ because @xmath74 remains a good quantum number . the 1pdm associated with the uhf wave function must be both hermitian and idempotent since this wave function is a single slater determinant . one can then trivially show the idempotency of each each spin block @xmath75 . thus , the charge density is a linear combination of two idempotent matrices , and consequently has eigenvalues which are either 0 , 1 , 1/2 , or come in `` corresponding pairs '' of @xmath76 and @xmath77.@xcite the natural orbitals of the charge density coupled as `` corresponding pairs '' are simply the natural orbital representation of the non - orthogonal uhf geminals discussed above . in the ghf formalism , even though additional symmetries are broken , the wave function is still a single slater determinant though the 1pdm takes the more general form : @xmath78 where the x and y components of the magnetization are given by : @xmath79 for a ghf wave function , no orbital rotation can bring the 1pdm to the form of equation , and we call such a state _ non - collinear _ , as the state no longer has a good quantum number associated with @xmath74 . the most general ghf wave function can not be written as the sum of two idempotent matrices , and loses the property of corresponding pairs . this results from the fact that the wave function can not be written as an apg . where the uhf formalism can only describe intra - pair correlation , the ghf formalism has additional flexibility to capture inter - pair correlation . since the pairing in the 1pdm is between alpha and beta spins , the uhf approach only describes opposite - spin correlation , whereas through inter - pair coupling the ghf approach possesses the flexibility to describe same - spin correlation as well . the primary drawback of broken symmetry wave functions lies in the difficulty of recapturing quantum numbers associated with the broken symmetries . many attempts have been made to address this problem , @xcite originating with the work of lwdin.@xcite recently a significant breakthrough was made in determining , self - consistently , symmetry adapted wave functions from a single broken - symmetry ( deformed ) determinant . @xcite the resulting wave function possesses multi - reference character calculated with _ mean - field _ computational cost . this projected quasi - particle theory ( pqt ) relies upon the simple idea of deliberately breaking and then self - consistently restoring the symmetries of the hamiltonian with the use of projection operators . a computationally efficient formulation of projected hartree - fock theory based upon pqt recently appeared in the literature . @xcite in that work , breaking and restoring discrete symmetries such as point group and complex - conjugation symmetries were also addressed . in this work , we focus on spin symmetries as they are most familiar to quantum chemists , though the method employs an approach analogous to that of restoration of angular momentum in the context of nuclear physics . @xcite let us begin by taking a moment to review the two types of wave functions we will project . the collinear uhf wave function @xmath80 minimizes the energy functional : @xmath81 sacrifices spatial and spin symmetry , and retains @xmath74 as a good quantum number . the non - collinear ghf wave function minimizes the same functional with the additional sacrifice of the @xmath74 quantum number . we will denote projections of these two wave functions as @xmath82 and @xmath83 , respectively . we can not emphasize enough that these projected wave functions are not single slater determinants . consequently , the density matrices associated with these _ multi - reference _ wave functions are _ not _ idempotent , though they do preserve the symmetries of the hamiltonian . to determine our spin - projected wave functions , we write the energy of a projected , deformed determinant @xmath84 as : @xmath85 where @xmath86 is the hamiltonian , and @xmath87 is the ( hermitian and idempotent ) projection operator that restores the symmetries broken in the deformed determinant @xmath84 . formally , for continuous symmetries such as @xmath88 and @xmath74 , the projection operator takes the form of an integral . in practice , we discretize this integral over a grid of modest size . it can then be shown that the energy expression can be written as a functional of the one - particle density matrix , @xmath89 , associated with the underlying deformed determinant @xmath84 . one then minimizes the resulting functional with respect to the deformed orbitals . the resulting phf equations are qualitatively similar to that of hartree - fock and the method retains _ mean - field _ computation cost . further , as with hartree - fock , any symmetry present in the initial guess for the deformed determinant will be preserved throughout the optimization procedure.@xcite we pause for a moment to remark that the implementation described in ref . ( ) does not strictly restore @xmath88 , as this would require a two - body operator . instead , it restores rotational invariance of the deformed determinant in spin - space . for collinear determinants , we rotate the determinant so that it lies along the z - direction , and then make it rotationally invariant using a projection operator of the form : @xmath90 where @xmath91 . this leaves a particular direction as `` special '' in that the projected wave function ( and its energy ) depends on the choice of the quantum number @xmath92 of the collinear determinant . projecting a non - collinear determinant requires integration over three euler angles and eliminates the dependence on the quantum numbers of the deformed state . interestingly , the projected wave function , though multi - reference in nature , may be characterized by a single , deformed determinant we denote as @xmath84 or , equivalently , the associated one - particle density matrix @xmath89 . we want to stress that @xmath93 or @xmath94 . the determinants @xmath80 or @xmath94 optimize the energy functional of equation ( breaking either @xmath95 , or @xmath95 and @xmath96 , respectively ) , whereas the determinant @xmath84 optimizes the energy functional of equation . the determinant @xmath84 can be of either the uhf or ghf type , depending on what symmetries are broken . finally , we remark that though the suhf wave function and the spin projected extended hartree - fock ( ehf ) method are in fact equivalent , we obtain our results without working in the corresponding orbital basis ( as suggested by mayer and lwdin ) resulting in significantly less computational overhead.@xcite in addition , previous examples of phf in the literature have all been collinear , i.e. restoring only @xmath97 , rather than both @xmath97 and @xmath74 , i.e. previous work did not consider sghf wave functions . as pulay pointed out , @xcite when several orbitals are strongly correlated , the ehf ( equivalently the suhf ) wave function does nt have the flexibility to describe all of the static correlation in the system . the sghf wave function provides a way around this : by breaking and restoring @xmath96 as well as @xmath95 , it captures inter - pair correlation which the suhf wave function can not . we utilize an implementation of the phf equations in the development version of the suite of programs.@xcite a minimal ( sto-3 g ) basis set was deemed sufficient for the purpose of describing the qualitative features of strong correlation associated with near degeneracies in the orbital spectrum . the initial guess for deformed suhf determinants are broken symmetry hartree - fock states . the initial guess for deformed sghf calculations were constructed using an suhf determinant and mixing the alpha and beta orbitals nearest the fermi level with a small mixing angle . we illustrate the evolution of pair wave functions as a function of nearest - neighbor distance ( @xmath100 ) , as well as investigate the errors associated with different types of spin projection ( suhf and sghf ) , by considering the symmetric stretching of rings of four and six hydrogen atoms . there is a topological difference between the orbital spectra of the four and six membered rings , and in general between 4n and 4n+2 membered rings , indicated in figure [ fig : orbital - schematic ] . in general , 4n membered ( anti - resonant ) rings have an effective fermi level which lies within two half - filled degenerate orbitals , whereas the effective fermi level for the 4n+2 membered ( resonant ) rings lies between a pair of fully occupied degenerate orbitals and a pair of unoccupied degenerate orbitals . one expects that the ground state for both systems should be a singlet , though in the case of anti - resonant rings , an open - shell singlet . a schematic depiction of the one - electron molecular orbital energy spectrum for the h@xmath98 ( left ) and h@xmath99 ( right ) rings , with lwdin s pairing scheme indicated by arrows . the effective fermi energy ( e@xmath101 ) lies in the degenerate orbitals for h@xmath98 and in the gap in h@xmath99 . , width=278 ] the ground state of h@xmath98 adopts a @xmath102 configuration , with @xmath103 and @xmath104 defined in the inset of figure [ fig : h4-s - en ] . the open - shell configuration gives rise to three singlet states : @xmath105 , @xmath106 , and @xmath107 , and a triplet state : @xmath108 . placing the electrons in @xmath109 yields the lowest energy . the reason can be understood in the following way : the @xmath110 state takes the form of an open - shell singlet @xmath111 , where one electron resides on each of the atoms lying on one diagonal of the square ( x+y ) , and the other along its complement ( x - y ) . this segregates the alpha and beta spins onto separate sublattices allowing them to avoid double occupancy and the consequent on - site repulsion . the electrons in the lowest ( @xmath112 ) orbital then correlate by mixing with its anti - bonding counterpart ( @xmath113 ) as shown in figure [ fig : orbital - schematic ] . + we present the @xmath114 dissociation curves for several approximations in figure [ fig : h4-s - en ] as a function of nearest - neighbor distance ( @xmath100 ) . while the rhf approximation dissociates improperly , both the uhf and gvb approximations succeed . figure [ fig : h4-s - fci ] shows the energy difference from the exact ( fci ) result and one can see that , in the intermediate coupling regime ( @xmath115 ) , uhf has a significantly smaller error than gvb . this reflects the advantage of allowing orbital spin contamination in the uhf approach . neither of these wave functions are equal to the fci for this system , and the error in the absolute energy is on the order of 1 - 2 ev near equilibrium . in the intermediate recoupling regime where both uhf and gvb still have significant error , the projected wave functions offer significant improvement in the energy , in addition to restoring the correct quantum numbers . this improvement results from the multi - reference nature of the projected wave function . the error in the suhf wave function reduces to the order of mev , and the sghf wave function is identical to the fci for the minimal basis . recall that for both the suhf and the gvb - pp approximations are identical to the exact wave function for the minimal basis set . while they continue to reflect the appropriate intra - pair correlation in the four electron system , they do not treat the inter - pair correlation completely , a problem ultimately associated with the lack of size - consistency in these methods . to understand these results , recall that there exist two distinct singlet spin states for any four - particle fermionic system:@xcite [ eq:4part_singlet ] @xmath116 here , @xmath117 describes the product of two singlet coupled pairs as in the gvb - pp wave function , while @xmath118 describes a linear combination of products of triplet pairs coupling to form an overall singlet state . the dominant contribution to the ground state comes from @xmath117 , though the exact wave function does contain a small contribution from @xmath118 , particularly near the equilibrium bond length . the gvb - pp wave function successfully describes all of the contribution to the total energy associated with the @xmath117 spin state . the uhf singlet obtains a better total energy by breaking spatial symmetry and introducting spin contamination . for the spin projected states , the suhf wave function captures correlation from determinants that couple to the @xmath117 spinor . though both gvb - pp and suhf wave functions are multi - reference and both couple to @xmath117 , suhf captures significantly more correlation because it contains more determinants than the gvb - pp . the first two terms of @xmath118 describe _ same - spin _ correlation , i.e. contributions from the product of @xmath119 two - electron triplets . there are no same - spin correlations present in a deformed @xmath45 singlet determinant ; consequently , the suhf state can not describe correlation from determinants that couple to @xmath118 , resulting in the energy difference from fci seen in figure [ fig : h4-s - fci ] . breaking @xmath74 as well as @xmath95 , allows the sghf wave function to further correlate the two pairs by capturing this same - spin correlation . by deliberately _ breaking and restoring this symmetry , the sghf wave function recovers the full inter - pair correlation energy , i.e. the fci result . + there is another point we wish to make with this simple model exemplified in the triplet potential curves depicted in figure [ fig : h4 t ] . the gvb - pp @xmath120 triplet state arising from the ground state configuration has the wrong symmetry to dissociate properly as evident in figure [ fig : h4-t - en ] . another triplet state , based on the excited electronic configuration @xmath121 ( not shown ) does , but the dissociation curve for this state is purely repulsive and not appropriate at equilibrium . the uhf solution , by breaking spatial symmetry , yields a qualitatively correct curve at all distances , though it predicts a barrier to formation of the bound state . turning now to the projected wave functions , we see that , as in the singlet case , the sghf triplet wave function is equal to fci for this minimal basis . however , we find two distinct suhf triplet wave functions with very different energies ; one arises from projecting a deformed determinant with @xmath45 and a second arises from projecting a deformed determinant with @xmath119 . as we briefly mentioned above , these two states differdue to the fact that we simply assure rotational invariance of the projected state , without respect for a particular axis in spin space . when we restore both rotational invariance as well as @xmath74 ( i.e. sghf ) we do nt have this problem . using this simple four - electron system , we can understand why this occurs . the phf equations preserve the symmetries of the initial guess throughout the scf procedure . the symmetries of the two deformed determinants are different and thus capture different types of correlation . recall that there exist three distinct four - particle triplet spinors:@xcite @xmath122 we obtain the suhf @xmath123 wave function by taking the uhf open - shell triplet state as an initial guess . for distances smaller than the triplet cf point , the deformed determinant has a doubly occupied @xmath124 orbital and the open - shell electrons occupy the @xmath125 and @xmath126 orbitals , delocalizing the un - paired spins across all the atoms . beyond the triplet cf point , the deformed determinant breaks symmetry , mixing the @xmath125 and @xmath126 orbitals as described above to localize the open - shell spins on different atoms . we therefore see that beyond the cf point , the two pairs must form the product spinor @xmath127 , as the open - shell geminal s spinor takes the form of either @xmath128 or @xmath129 , and the remaining geminal couples to the spinor : @xmath130 . this provides only a modest improvement in the energy compared to uhf as symmetry breaking captures most of the correlation . in contrast , we obtain the suhf @xmath45 wave function state by taking the uhf singlet state as an initial guess . this deformed determinant captures contributions from @xmath131 and @xmath132 , and the resulting suhf state captures determinants which couple to product geminals from the open - shell singlet and the paired triplet ( and vice versa ) . as there are significantly more determinants in the fci space which couple to these four - particle spinors , the maximum error drops from almost 2 ev to slightly less than 90 mev ( see figure [ fig : h4-t - fci ] , inset ) . . sharp cusps in the uhf and suhf @xmath133 states are the result of spontaneous symmetry breaking at a cf point.,width=278 ] this has consequences for calculating magnetic properties of materials . we plot the singlet - triplet splitting for the h@xmath134 system in figure [ fig : h4-st - split ] . at large @xmath100 this splitting relates to the magnetic coupling constants of the heisenberg model hamiltonian . the dissociation curves of both the uhf and the suhf @xmath123 state have a cf point , and consequently the singlet - triplet splitting calculated from these states possesses a discontinuity in the derivative at that point . combined with the significant underestimation of the triplet energy , the splitting calculated using the suhf @xmath123 wave function is actually worse than using the uhf single determinant . in contrast , the @xmath45 suhf state gives both a quantitatively as well as qualitatively correct curve . turning now to @xmath135 , two additional electrons fill the degenerate @xmath136 orbitals , leading to a closed - shell singlet ground state . in figure [ fig : h6-pairs ] , we plot the uhf pair functions for the three pairs of @xmath135 ring at three distances : 1.2 , 1.8 , and 5.00 , corresponding to just past the cf point , moderately stretched , and dissociation . the pairs evolve smoothly as the @xmath0 and @xmath1 electrons segregate onto two distinct sublattices . in order to understand the manner in which the pair functions evolve , we project the pair functions back onto the symmetry adapted mos of the system . orbital coefficients for the alpha and beta pairs for this system in the symmetry adapted mo basis at a distance of 1.80 appear in table [ tab : coeff ] , showing how the pairs form by mixing a bonding orbital with its anti - bonding counterpart . in the dissociation limit , the orbital coefficients approach the limiting value of @xmath137 . the pairing ansatz evident in the uhf wavefunction , where each orbital mixes with its corresponding anti - bonding orbital , was suggested as a way to correlate electrons subject to large coulomb interactions by lwdin in 1955 and goes by the name of alternant molecular orbital ( amo ) theory . @xcite for the projected wave functions , the underlying deformed determinant of the suhf state possesses pairing scheme . [ cols="^,^,^,^,^,^,^ " , ] to summarize , we believe these approaches capture the most important static correlation effects . with the suhf wave function , we capture all the intra - pair correlation effects , and the sghf approach captures additional pair - pair correlation . even so , higher order many - body effects are not described properly by these methods , and those terms grow in importance as the system size increases , leading to a lack of size extensivity.@xcite we now consider larger 1d rings of hydrogen as a function of nearest - neighbor bond length . we present results for both 16 and 18 atom rings ; chosen because they are complicated enough that fci calculations become quite cumbersome , yet the system remains small enough that the phf approach can provide a significant improvement over uhf , i.e. we do not approach the thermodynamic ( @xmath138 ) limit . our results simply extend those already seen in the four and six atom rings , and in what follows we focus on the resonant system : . figure [ fig : h18-spectrum ] shows the uhf one - electron molecular orbital energy spectrum for the ring . in the small @xmath100 regime , all the orbitals except for the lowest occupied and highest virtual orbitals remain doubly degenerate . as the bonds stretch , the orbital spectrum compresses as the occupied and virutal mos become energetically degenerate . at dissociation , the orbital eigenvalues collapse into two sets of degenerate levels , an occupied `` lower band '' and the unoccupied `` upper band '' , each containing nine spatial orbitals . the effective fermi energy sits in the gap between these two `` bands . '' + although our calculation uses the exact hamiltonian , we map our results as a function of distance to a hubbard model in an attempt to make contact with the substantial body of work using that approximation . to this end , we approximate the onsite interaction parameter @xmath21 as the uhf `` band gap '' in the limit @xmath139 . in the case of half - filling , this energy difference corresponds to the energy needed to move an electron from one site to another . this `` gap '' has a value of @xmath140 21 ev for both the and rings , not an unreasonably large value for @xmath21 in this minimal basis set . we next extract the effective near - neighbor hopping parameter @xmath141 from the band width of the one - electron spectrum . the effective value we seek should be renormalized by all the terms in the exact hamiltonian except for the on - site @xmath21 term . to this end , at each bond length we compute the uhf eigenvalue spectrum for the high - spin , ferromagnetic state . this singly occupies each orbital , thereby eliminating the onsite hubbard @xmath21 interaction , and the effective hopping parameter @xmath141 is then given by : @xmath142 where @xmath143 - @xmath144 is the splitting between the highest and lowest occupied mos of the ferromagnetic state , and @xmath145 is the number of nearest neighbors . the onset of strong correlations corresponds to the regime where the bandwidth and the on - site repulsion become comparable . for the 1d ring , this occurs at a ratio of @xmath146 , or @xmath147 , shortly past the cf point . as @xmath139 the effective hopping parameter tends toward zero , reflecting an effectively infinite on - site interaction . this transition regime corresponds to distances where the suhf energy differs most greatly from the exact result for h@xmath98 ( the inset of figure [ fig : h4-s - fci ] ) suggesting the importance of inter - pair correlation . in the dissociation curves for the ring shown in figure [ fig : h18-dis ] , the uhf wave function goes to the correct limit by allowing the alpha and beta spins to localize on different sublattices ( see fig [ fig : h18-dis ] , inset ) . notice that the cf point is quite near equilibrium , i.e. the uhf wave function starts to correlate through localization in this region . one might therefore suspect that strong correlations contribute significantly at these distances . in fact , the suhf projected wave function captures @xmath140 3.5 ev of energy from intra - pair correlation , and the sghf wave function captures an additional @xmath140 1.4 ev from inter - pair correlation at the equilibrium near - neighbor distance . overlap of the singlet uhf corresponding orbitals for the ring ( see text ) . all orbitals except for the core are doubly degenerate . the uhf overlap becomes identically 1 for all orbitals at the cf point . ( inset ) overlap of the @xmath0 and @xmath1 homo and core corresponding orbitals of the uhf and deformed suhf determinants of the same system . , width=278 ] the corresponding pairs in and are similar to those discussed for h@xmath134 and h@xmath148 , respectively . at small distances , the orbital overlap of each pair remains identically one ( see figure [ fig : h18-uhf - over ] ) . this follows from the fact that inside the cf point , the uhf wavefunction has not yet broken symmetry and each spin - orbital ( @xmath0 and @xmath1 ) remain spatially identical . the electrons nearest the fermi level localize first as the ring stretches , and at dissociation the overlap of all the orbitals goes to zero , reflecting compete localization of the spins onto separate sublattices . + again , we pause to stress that the multi - reference suhf wave function can be characterized by a single , deformed determinant , distinct from the self - consistent uhf wave function . to illustrate this point , we show the value of the overlap integral ( @xmath149 ) for core and degenerate homo orbitals for both the uhf and broken - symmetry determinant characterizing the suhf wave function in the inset of figure [ fig : h18-uhf - over ] . by _ deliberately _ breaking the symmetry of the underlying determinant in the suhf approach , artificially localizing the electrons at small distances , and projecting the deformed determinant , the suhf approach captures a significant amount of correlation energy near equilibrium . finally returning to the anti - resonant ring , we plot the core and homo pair orbitals in figure [ fig : h16uhforbs ] . these are analogous to the case of h@xmath98 , and the homo orbital localizes one of each spin on separate sublattices at all distances . this creates an open - shell singlet , while the rest of the orbitals retain the freedom to localize or not as correlation effects demand . near the cf point the core orbital remains delocalized , and by 3 it localizes each spin onto different sublattices . the remaining orbitals behave in a similar manner . + + we conclude with a brief discussion of a 2d system : symmetric dissociation of in a square geometry . the square lattice has been previously studied at half - filling by larson and thorson , and calais and coworkers . @xcite though the system studied here does nt have periodic boundary conditions , it still captures the essential hubbard physics . we point out that the uhf single particle energy spectrum differs from that of the ring in that the degeneracy of the homo has lifted and the fermi energy now lies within the gap . this results in a close - shell singlet ground state , which becomes the focus of our discussion . interestingly , we find two distinct stable uhf states , shown in figure [ fig : h16-square - dis ] . the difference between these two states is subtle , amounting to a ninety degree rotation of the localized spins in the center of the plaquette ( see figures [ fig : h16-s1 ] and [ fig : h16-s2 ] ) . we also plot the pair orbitals for the afm state near dissociation , showing the segregation of respective spins onto separate sublattices . the uhf global minimum energy occurs for the afm state , however , the orbital hessian indicates that the sf state exists as a true local minimum , and not a transition state . one might naively hope that projecting these two different uhf singlet states would yield the same multi - reference wave function for the singlet state . unfortunately , this intuition fails for the suhf approach . as mentioned previously , the scf procedure preserves the symmetries of the initial guess for the broken symmetry determinant ; consequently , we find two distinct suhf singlets . using the sghf approach overcomes this problem : breaking @xmath74 allows the dissociation curve to smoothly interpolate between these two different spin densities . the sghf wave function captures the most correlation in the region where these two states become energetically degenerate , as one would expect . note that we have not performed a projection to restore spatial symmetry ; the sghf wave function only restores @xmath88 and @xmath74 . we also find multiple uhf and suhf solutions arising from broken spatial symmetries in larger models such as 8x8 plaquettes . one can in some sense think of these different states as incorporating different types of spin fluctuations . by interpolating between these spin states , the sghf approach captures additional correlation in the region where states of different spatial symmetry become energetically degenerate . ongoing work indicates that breaking and restoring point group symmetry often becomes very important in cases such as these where the sghf wave function does not have enough flexibility to describe all of the static correlation in the system . we conclude this section by pointing out that we may apply the pairing ansatz which segregates spins onto separate sublattices to any alternant lattice : the one - dimensional , square , simple cubic and bcc lattices . for example : in terms of the bloch functions of the original 2d lattice , the pairs mix the configurations @xmath150 with their anti - bonding partner @xmath151 . the wave function may be written as a spin - contaminated apsg if the corresponding pairs at each k - value come from the corresponding orbitals of the spin - polarized solution . the energy of this wave function is identical with that of the nel state . in fact , the energies of all the pure spin states are degenerate , and the nel state a linear combination of them . we have emphasized that the uhf wave function can be written as a spin - contaminated _ pair _ wave function of the apsg form . the overlap of the alpha and beta corresponding orbitals of the uhf solution can be taken as a proxy for the strength of the correlation captured by breaking symmetry . as a function of distance , or the ratio u / t , the uhf corresponding orbitals pair in a manner allowing a smooth evolution from a system with doubly occupied orbitals into one in which the @xmath0 and @xmath1 electrons segregate onto distinct sublattices . in this way the uhf wave function evolves from a regime in which the material behaves as a metal with delocalized spins , to one in which all the spins are localized antiferromagnetically . projecting the spin - contaminated uhf pair wave function recovers additional correlation energy in finite systems . the resulting multi - reference wave function , characterized by a single , deformed determinant , describes a pure spin state . the suhf wave function ( projecting @xmath88 by assuring rotational invariance in spin - space ) captures additional intra - pair correlation beyond uhf . the sghf wave function ( projecting @xmath88 and @xmath74 ) captures additional inter - pair correlation . by _ deliberately _ breaking and then restoring these symmetries , we describe strong correlation even at geometries where the uhf solution reduces to rhf or the ghf solution reduces to uhf . from simple calculations of one and two - dimensional finite lattices of hydrogen atoms , we have shown that the broken symmetry determinant characterizing our suhf solutions mirror the pairing schemes utilized extensively some years ago in amo theory to describe correlation in alternant pi - bonded networks . this early work did not address the questions of what happens as you dope the system away from half - filling , or in what manner the pairs respond . the pairs described here accounting for magnetic correlations are strongly orthogonal to one another , whereas the pairs of bcs theory overlap ; nevertheless , the proximity of superconducting behavior to antiferromagnetic phases suggest this may be an interesting avenue to explore . our work at los alamos national laboratory was supported by the department of energy , office of basic energy sciences , heavy element chemistry program and the ldrd program at lanl . the work at rice university was supported by doe , office of basic energy sciences , heavy element chemistry program under grant de - fg02 - 04er15523 . the los alamos national laboratory is operated by los alamos national security , llc , for the national nuclear security administration of the u.s . department of energy under contract de - ac5206na25396 . de p r moreira , i. ; illas , f. ; martin , r. l. _ physical review b _ * 2002 * , _ 65 _ , 155102 wen , x .- d . ; martin , r. l. ; henderson , t. m. ; scuseria , g. e. _ chemical reviews _ * 2013 * , _ 113 _ , 10631096 henderson , t. m. ; paier , j. ; scuseria , g. e. _ physica status solidi b _ * 2011 * , _ 248 _ , 767774 marsman , m. ; paier , j. ; stroppa , a. ; kresse , g. _ journal of physics : condensed matter _ * 2008 * , _ 20 _ , 064201 paier , j. ; marsman , m. ; kresse , g. _ the journal of chemical physics _ * 2007 * , _ 127 _ , 024103 perdew , j. p. ; zunger , a. _ physical review b _ * 1981 * , _ 23 _ , 50485079 anisimov , v. ; zaanen , j. ; andersen , o. _ physical review b _ * 1991 * , _ 44 _ , 943954 anisimov , v. i. ; lichtenstein , a. i. in _ strong coulomb correlations in electronic structure calculations : beyond the local density approximation _ ; anisimov , v. i. , ed . ; gordon and breach science publishers : amsterdam , 2000 hedin , l. _ physical review _ * 1965 * , _ 139 _ , a796a823 hedin , l. ; lundqvist , s. in _ solid state physics : advances in research and applications _ ; ehrenreich , h. , seitz , f. , turnbull , d. , eds . ; academic press : new york , new york , 1969 ; pp 1181 georges , a. ; krauth , w. ; rozenberg , m. j. _ reviews of modern physics _ * 1996 * , _ 68 _ , 13125 petit , l. ; svane , a. ; szotek , z. ; temmerman , w. m. ; stocks , g. m. _ physical review b _ * 2010 * , _ 81 _ , 045108 snchez - friera , p. ; godby , r. w. _ physical review letters _ * 2000 * , _ 85 _ , 56115614 pople , j. a. ; nesbet , r. k. _ the journal of chemical physics _ * 1954 * , _ 22 _ , 571572 hubbard , j. _ proceedings of the royal society of london series a mathematical and physical sciences _ * 1963 * , _ 276 _ , 238257 hurley , a. c. ; lennard - jones , j. ; pople , j. a. _ proceedings of the royal society of london series a mathematical and physical sciences _ * 1953 * , _ 220 _ , 446455 hunt , w. j. _ the journal of chemical physics _ * 1972 * , _ 57 _ , 738748 moss , b. j. ; bobrowicz , f. w. ; goddard , w. a. _ the journal of chemical physics _ * 1975 * , _ 63 _ , 4632 bobrowicz , f. w. ; goddard , w. a. in _ methods of electronic structure theory _ ; schaefer , h. f. , ed . ; plenum press , 1977 ; pp 79127 surjn , p. in _ correlation and localization _ ; bartlett , r. , bogr , f. , cooper , d. , kirtman , b. , klopper , w. , kutzelnigg , w. , march , n. , mezey , p. , mller , h. , noga , j. , paldus , j. , pipek , j. , raimondi , m. , reggen , i. , sun , j. , surjn , p. , valdemoro , c. , vogtner , s. , eds . ; springer berlin / heidelberg , 1999 ; pp 255272 lykos , p. g. ; pratt , g. _ reviews of modern physics _ * 1963 * , _ 35 _ , 496501 fukutome , h. _ international journal of quantum chemistry _ * 1981 * , _ 20 _ , 9551065 stuber , j. l. ; paldus , j. in _ fundamental world of quantum chemistry volume 1 : a tribute to the memory of per - olov lwdin _ ; brndas , e. j. , kryachko , e. s. , eds . ; springer verlag gmbh : dordrecht , the netherlands , 2003 ; pp 67139 jimenez - hoyos , c. a. ; henderson , t. m. ; scuseria , g. e. _ journal of chemical theory and computation _ * 2011 * , _ 7 _ , 26672674 lwdin , p .- o . _ physical review _ * 1955 * , _ 97 _ , 15091520 amos , a. t. ; hall , g. g. _ proceedings of the royal society of london series a mathematical and physical sciences _ * 1961 * , _ 263 _ , 483493 coleman , a. j. _ reviews of modern physics _ * 1963 * , _ 35 _ , 668686 harriman , j. e. _ the journal of chemical physics _ * 1964 * , _ 40 _ , 2827 rabanovich , v. _ linear algebra and its applications _ * 2004 * , _ 390 _ , 137143 goddard , w. a. _ the journal of chemical physics _ * 1968 * , _ 48 _ , 450 mayer , i. _ chemical physics letters _ * 1971 * , _ 11 _ , 397400 mayer , i. ; ladik , j. ; bicz , g. _ international journal of quantum chemistry _ * 1973 * , _ 7 _ , 583608 mayer , i. _ acta physica academiae scientiarum hungaricae _ * 1973 * , _ 34 _ , 305309 rosenberg , m. ; martino , f. _ the journal of chemical physics _ * 1975 * , _ 63 _ , 53545361 klimo , v. ; tio , j. _ molecular physics _ * 1978 * , _ 35 _ , 17771783 scuseria , g. e. ; jimenez - hoyos , c. a. ; henderson , t. m. ; samanta , k. ; ellis , j. _ the journal of chemical physics _ * 2011 * , _ 135 _ , jimenez - hoyos , c. a. ; henderson , t. m. ; tsuchimochi , t. ; scuseria , g. e. _ the journal of chemical physics _ * 2012 * , _ 136 _ , 164109 sheikh , j. a. ; ring , p. _ nuclear physics a _ * 2000 * , _ 665 _ , 7191 pulay , p. ; hamilton , t. p. _ the journal of chemical physics _ * 1988 * , _ 88 _ , 49264933 frisch , m. j. ; trucks , g. w. ; schlegel , h. b. ; scuseria , g. e. ; robb , m. a. ; cheeseman , j. r. ; scalmani , g. ; barone , v. ; mennucci , b. ; petersson , g. a. ; nakatsuji , h. ; caricato , m. ; li , x. ; hratchian , h. p. ; izmaylov , a. f. ; bloino , j. ; zheng , g. ; sonnenberg , j. l. ; hada , m. ; ehara , m. ; toyota , k. ; fukuda , r. ; hasegawa , j. ; ishida , m. ; nakajima , t. ; honda , y. ; kitao , o. ; nakai , h. ; vreven , t. ; montgomery jr , j. a. ; peralta , j. e. ; ogliaro , f. ; bearpark , m. ; heyd , j. j. ; brothers , e. ; kudin , k. n. ; staroverov , v. n. ; kobayashi , r. ; normand , j. ; raghavachari , k. ; rendell , a. ; burant , j. c. ; iyengar , s. s. ; tomasi , j. ; cossi , m. ; rega , n. ; millam , j. m. ; klene , m. ; knox , j. e. ; cross , j. b. ; bakken , v. ; adamo , c. ; jaramillo , j. ; gomperts , r. ; stratmann , r. e. ; yazyev , o. ; austin , a. j. ; cammi , r. ; pomelli , c. ; ochterski , j. w. ; martin , r. l. ; morokuma , k. ; zakrzewski , v. g. ; voth , g. a. ; salvador , p. ; dannenberg , j. j. ; dapprich , s. ; parandekar , p. v. ; mayhall , n. j. ; daniels , a. d. ; farkas , o. ; foresman , j. b. ; ortiz , j. v. ; cioslowski , j. ; fox , d. j. gaussian development version revision h.01 . http://www.gaussian.com/ , gaussian , inc . , wallingford ct , 2009 amos , r. d. _ molecular physics _ * 1975 * , _ 29 _ , 11251135 the perfect geometric symmetry of the ring was broken slightly by adding a small ( @xmath152 a ) displacement to each succeeding atom along the vertical z - axis . this imposed helical distortion allows us to uniquely define each of the molecular orbitals of the system , as it lifts the perfect degeneracy in the @xmath153 symmetries , and removes the freedom of the diagonalization routine to rotate them in an arbritrary manner . pauncz , r. _ alternant molecular orbital method _ ; studies in physics and chemistry , no . 4 ; saunders , 1967 larson , e. g. ; thorson , w. r. _ the journal of chemical physics _ * 1966 * , _ 45 _ , 15391554 calais , j. l. _ arkiv fr fysik _ * 1965 * , _ 29 _ , 255 calais , j. l. _ arkiv fr fysik _ * 1965 * , _ 28 _ , 539 calais , j. l. _ arkiv fr fysik _ * 1965 * , _ 28 _ , 511 calais , j. l. _ arkiv fr fysik _ * 1965 * , _ 28 _ , 479

we investigate the lhc and higgs factory prospects for measuring the phase in the higgs-- coupling . currently this phase can be anywhere between ( even ) and ( odd ) . a new , ideal observable is identified from an analytic calculation for the channel . it is demonstrated to have promising sensitivity at the lhc and superior sensitivity at the ilc compared to previous proposals . our observable requires the reconstruction of the internal substructure of decaying taus but does not rely on measuring the impact parameter of tau decays . it is the first proposal for such a measurement at the lhc . for the 14 tev lhc , we estimate that about 1 ab data can discriminate-even versus-odd at the level . with 3 ab , the phase should be measurable to an accuracy of . at an higgs factory , we project that a 250 gev run with 1 ab luminosity can measure the phase to accuracy .

1308.1094

the discovery of the higgs boson @xcite has opened a new opportunity in the search for physics beyond the sm . the sm predicts all couplings of the higgs to sm particles completely , and a measured significant deviation of higgs couplings from the sm prediction will be a clear signal of new physics . the most straightforward tests at the moment are comparisons of the higgs production rates times branching ratios to the sm prediction in a variety of final states . thus far , such global fits roughly agree with a sm higgs @xcite . we can go further by testing the @xmath0 properties of higgs couplings . this test has already been done for the coupling of the higgs to electroweak gauge bosons @xcite . in the sm , the higgs couples to the @xmath10 boson as a scalar , @xmath11 . in general , a higgs - like state could couple to @xmath10 bosons as a pseudoscalar , @xmath12 , or with any linear combination of scalar and pseudoscalar couplings , which would imply @xmath0 violation . in the fully leptonic channel for @xmath13 , the azimuthal angle between the decay planes of the two @xmath10 bosons is sensitive to the @xmath10 polarizations , which in turn is sensitive to the @xmath0 structure of the higgs couplings @xcite . current data disfavors a pure pseudoscalar coupling at 99.84% = 3.3 @xmath14 and 99.6% ( 97.8% ) confidence level using cl@xmath15 statistics at cms @xcite and atlas mela ( atlas bdt ) @xcite , respectively . in models where the sm is augmented by heavy new physics , this result is unsurprising . of the two possible interactions mentioned above , the scalar interaction is renormalizable , while the pseudoscalar interaction arises from a dimension six operator . the pseudoscalar coupling is thus expected to be subdominant in higgs decays and corresponding @xmath0 violating effects will be small . while current results favor a pure scalar coupling in the higgs couplings to weak gauge bosons , searches for @xmath0 violation in fermionic decays of the higgs are still highly motivated . such modified couplings can arise from a different source which , in particular , can give a pseudoscalar interaction comparable to a scalar , unlike the higgs-@xmath16/@xmath10 couplings . in this paper we investigate how the @xmath0 structure of the coupling of the higgs to tau leptons can be probed at present and future colliders . the higgs coupling to any fermion generally consists of a @xmath0 even and a @xmath0 odd term , @xmath17 measuring this @xmath0 phase @xmath18 requires knowledge of the spins of the @xmath19 state . tau decays are complex enough to retain non - trivial information about the direction of the tau spin , yet clean enough that the spin information is not washed out by hadronization effects as it is for @xmath20-quark decays @xcite . since the higgs branching fraction to @xmath21 is substantial in the sm ( @xmath22 for @xmath23 gev ) , the @xmath21 decay channel is the best of a limited set of opportunities for @xmath0 violation searches in higgs couplings to fermions . in addition , a pseudoscalar - like coupling of the higgs to taus can conceivably compete with the small tau yukawa coupling , and so @xmath0 violating effects can be sizable . currently , the only direct bound on the higgs - tau coupling is on the net signal strength in @xmath24 channels : @xmath25 @xcite . in this paper , we will maintain @xmath26 and modify only @xmath18 , so this constraint does not apply . we focus on the specific tau decay channel @xmath27 with @xmath28 . this is the most common tau decay sub - channel , with a branching fraction of @xmath29 . moreover , the angular distributions of the tau decay products and subsequent rho decay products are correlated with the original direction of the tau spin , as we will see in section [ sec : heuristic ] . the relative azimuthal orientation of the two hadronic taus , @xmath30 , which we will define precisely in section [ sec : theta ] , contains information about the @xmath0 properties of the higgs coupling to taus . in particular , the @xmath0 phase @xmath31 in the higgs couplings may be read off directly from the @xmath30 distribution . the differential cross section is shown analytically in sections [ sec : eandb ] and c to have the form @xmath32 , and @xmath31 may be measured by finding the minimum of the distribution ( as exemplified in figure [ fig : true_varydelta ] ) . the dominant background for @xmath33 events at the lhc is @xmath34 , which produces a flat @xmath30 distribution . the ability to distinguish scalar versus pseudoscalar higgs couplings in the tau channel has been discussed in @xcite . our work quantitatively improves on these results : our @xmath30 variable is demonstrably more sensitive to the @xmath0 phase of the higgs coupling to taus compared to earlier proposed observables , and our simulation results for the ilc indicate a corresponding increase in sensitivity compared to earlier results . this work is also a qualitative step forward in that we propose a strategy to do this measurement at the lhc . previous studies relied on resolving a displaced vertex in @xmath1 decays which is challenging . we show that our observable retains sensitivity without this . it should be stressed that in order to reconstruct the angle @xmath30 , full knowledge of all four - momenta components in the event is needed , including those of the two neutrinos . we will discuss the challenges that this presents and how they may be addressed . in the context of a higgs factory ( ilc ) , @xmath35 events may be fully reconstructed up to a two - fold ambiguity . furthermore , a favorable signal to background ratio makes our measurement straightforward . at a hadron collider , however , some approximations are needed for the neutrino four - momenta . employing the collinear approximation @xcite , we show that the amplitude of the angular structure in @xmath30 is only reduced by an order one factor for @xmath33 signal events . the challenge for the lhc is thus to increase the signal to background ratio as much as possible in order to produce a statistically significant result . in addition , an improvement over the collinear approximation would make a positive impact on the resulting sensitivity to @xmath31 . our net result is that , using the @xmath30 variable , a measurement of @xmath31 with an accuracy of @xmath36 is possible for a @xmath37 gev @xmath38 collider , assuming 1 ab@xmath5 of luminosity ( without incorporating detector effects , which are expected to be negligibly small ) . this number should be compared with the result of ref . @xcite , which quotes an accuracy of measuring @xmath31 to @xmath39 using the same amount of luminosity but for @xmath40 gev and @xmath41 gev . we also provide the first estimates for sensitivity to @xmath31 at the lhc . without incorporating detector effects or pileup , we find an ideal measurement of @xmath31 to an accuracy of @xmath42 is possible with 3 ab@xmath5 of @xmath43 tev lhc data for a @xmath1-tagging efficiency of 50% . improving the efficiency from 50% to 70% could lead to an accuracy of @xmath44 using the same lhc luminosity . this paper is organized as follows . in section [ sec : cpv_coupling ] we add @xmath0 violation to the higgs coupling to tau leptons . in section [ sec : observable ] we introduce our observable , first in a heuristic analysis that follows every step of the decay , and then rigorously , using the analytic form of the full @xmath45 differential cross section . we present the results of our collider analyses in section [ sec : studies ] . we first present the relevant distributions using monte carlo truth information , then reevaluate in a higgs factory setup , where a twofold ambiguity needs to be considered , and finally consider an lhc setting using the collinear approximation . we conclude in section [ sec : conclusion ] . a weakly - coupled renormalizable model giving rise to @xmath0 violation in the higgs coupling to taus is presented in appendix [ app : uvcompletion ] . [ sec : cpv_coupling ] in our study of the @xmath0 nature of @xmath47 , we use the following phenomenological lagrangian : @xmath48 where @xmath1 and @xmath49 are the physical tau lepton and higgs boson in the mass basis , respectively , @xmath50 is a real parameter parametrizing the magnitude of the @xmath51 coupling , and , most importantly , @xmath52 $ ] is an angle describing the @xmath0 nature of the @xmath51 coupling . can , in fact , take the full range of @xmath53 $ ] . however our technique is not sensitive to a multiplication of the tau yukawa by @xmath54 and so it is sufficient to consider half of this range . resolving this ambiguity would require measuring the interference of higgs with background , which is a tiny effect . ] the @xmath0-even and @xmath0-odd cases correspond to @xmath55 and @xmath56 , respectively , while @xmath57 describe maximally @xmath0-violating cases . the sm corresponds to a special case , @xmath58 with @xmath59 . we will refer to `` @xmath60 '' as a `` @xmath0-violating @xmath46 coupling '' , even though it includes the @xmath0-conserving limits of @xmath55 and @xmath61 . in this work , we focus on the effects of @xmath18 , so we will take @xmath62 while treating @xmath18 as a free parameter . the simplest fully gauge - invariant operator that results in the @xmath0-violating @xmath51 coupling ( [ eq : lpheno ] ) upon electroweak symmetry breaking is given by @xmath63 where @xmath64 and @xmath65 are _ complex _ dimensionless parameters , and @xmath66 is a mass scale taken to be real and positive without loss of generality . to relate the parameters of @xmath67 and @xmath68 , we substitute @xmath69 in ( [ eq : leff ] ) , which yields @xmath70 from which we identify @xmath71 and we have taken @xmath72 to be real and positive ( hence @xmath73 is real and positive ) without loss of generality after suitable redefinition of the phase of @xmath74 . with this phase convention , the @xmath51 coupling in ( [ eq : lpheno ] ) is generally complex : @xmath75 since @xmath76 , new physics at the tev scale ( @xmath77 ) with @xmath78 couplings ( @xmath79 ) can give rise to @xmath18 anywhere in the full range @xmath80 $ ] .. ] this is in stark contrast to the case of a @xmath0-odd / violating higgs coupling to @xmath10 bosons , where tev - scale new physics is expected to give only small corrections to the sm @xmath0-even coupling . [ sec : observable ] to probe the @xmath0-violating @xmath51 coupling in ( [ eq : lpheno ] ) , we will study the following decay process : @xmath81 there are several good reasons to choose this decay chain . first , to minimize the loss of kinematic information due to neutrinos , we want both @xmath82 and @xmath83 to decay hadronically . second , of the hadronic decay modes , we choose @xmath84 , since the subsequent decay , @xmath85 , can be reconstructed at a collider . third , @xmath84 has the largest branching fraction of any individual tau decay mode , @xmath86% , and the following step , @xmath87 , occurs with a nearly 100% probability . finally , the @xmath88 width is sufficiently narrow that it is well justified to consider it on - shell , which makes the process in ( [ eq : theprocess ] ) an analytically tractable sequence of 2-body decays . we begin with a heuristic look at the process in ( [ eq : theprocess ] ) to develop a rough idea of how it can probe the @xmath0-violating @xmath89 coupling ( [ eq : lpheno ] ) . in particular , the highlights of qualitative points to be made in sections [ sec : h_to_tautau ] , a@xmath902 and a@xmath903 are : * measuring @xmath1 helicities can not determine the @xmath0 phase , but the @xmath1 polarizations in directions _ perpendicular _ to the @xmath1 momenta can . * in the tau rest frame the @xmath88 is predominantly longitudinal and is polarized roughly in the direction of the @xmath1 polarization . * the difference between the charged and neutral pion 3-momenta , @xmath91 , is roughly parallel to the respective @xmath92 polarization . therefore , the @xmath0 nature of @xmath93 must be encoded in the orientation of `` @xmath91 '' in the plane perpendicular to the @xmath94 momenta in the higgs rest frame . a precise form of `` @xmath95 '' as well as the best observable to measure the @xmath0 phase @xmath18 will be identified in sections [ sec : eandb ] and c by analytically computing the full matrix element for the sequence of two - body decays in process ( [ eq : theprocess ] ) . [ sec : heuristic ] [ sec : h_to_tautau ] the most general form of the amplitude for the decay @xmath97 is given by @xmath98 where @xmath99 is the probability amplitude of @xmath82 and @xmath83 having helicities @xmath100 and @xmath101 , respectively . lorentz invariance dictates that the proportionality factor omitted in ( [ eq : m : h_to_tautau ] ) has no momentum dependence . in the higgs rest frame , the amplitude ( [ eq : m : h_to_tautau ] ) takes the form @xmath102 where @xmath103 and @xmath104 are the @xmath82 momentum and energy in this frame , while @xmath105 is the linear combination of @xmath99 with angular momentum @xmath106 . in particular , @xmath107 the amplitude in ( [ eq : m : h_to_tautau : higgs_frame ] ) shows that the @xmath0-even contribution ( @xmath108 ) is a spin triplet in a @xmath109-wave , while the @xmath0-odd contribution ( @xmath110 ) is a spin singlet in an @xmath111-wave . this can be understood as a consequence of angular momentum conservation and fermi statistics , with the additional fact that a fermion anti - fermion pair has an odd intrinsic parity . to measure @xmath18 , it is necessary to keep the @xmath112 pair in the above superpositions of @xmath113 and @xmath114 , without projecting the polarizations onto the helicity eigenstates . from ( [ eq : m : h_to_tautau : higgs_frame ] ) and ( [ eq : chi10_chi00 ] ) , we see that the coefficients of @xmath113 and @xmath114 are the complex conjugates of each other , which implies that , regardless of @xmath18 , the probability for both @xmath82 and @xmath83 to be right - handed is always equal to that for both to be left - handed . therefore , to distinguish the two linear combinations in ( [ eq : chi10_chi00 ] ) , we must measure the polarizations in the directions _ perpendicular _ to the momenta , as mentioned in item 1 above . assuming the sm weak interactions for the @xmath1 and @xmath116 , the most general form of the amplitude for @xmath117 is given by @xmath118 with @xmath119 . again , lorentz invariance dictates that the proportionality factor omitted in ( [ eq : m : tau_to_rho - nu ] ) has no momentum dependence . in the @xmath82 rest frame , the amplitude ( [ eq : m : tau_to_rho - nu ] ) has the form @xmath120 where @xmath121 $ ] is the angle between the @xmath122 momentum and the @xmath82 polarization in this frame , and @xmath123 , @xmath124 , and @xmath125 are the polarization vectors for the left - handed , longitudinal , and right - handed polarizations of the @xmath122 , respectively . since @xmath126 , the amplitude ( [ eq : m : tau_to_rho - nu : tau_frame ] ) is dominated by the second term , roughly speaking . thus , we are led to the picture described in the item 2 above , namely , the @xmath122 is predominantly longitudinal ( @xmath127 ) and mostly emitted in the direction of the @xmath82 polarization ( @xmath128 ) . the most general form of the amplitude for @xmath129 is given by @xmath130 the other linear combination , @xmath131 , can not appear here because @xmath132 . again , the proportionality factor omitted in ( [ eq : m : rho_to_pi - pi ] ) can not have any momentum dependence by lorentz invariance . boosting the longitudinal @xmath122 to its rest frame , and neglecting the @xmath133-@xmath134 mass difference , the amplitude ( [ eq : m : rho_to_pi - pi ] ) takes the form @xmath135 where @xmath136 is the angle between the original @xmath122 polarization and the vector @xmath137 in the rest frame . therefore , the momentum difference , @xmath138 , is roughly ( anti-)parallel ( @xmath139 or @xmath140 ) to the original @xmath122 polarization , as we described in the item 3 above . [ sec : eandb ] we now analytically compute the full matrix element for the process ( [ eq : theprocess ] ) to identify the observable that is most sensitive to the @xmath0 phase @xmath18 . combining the amplitudes ( [ eq : m : h_to_tautau ] ) , ( [ eq : m : tau_to_rho - nu ] ) and ( [ eq : m : rho_to_pi - pi ] ) , the full amplitude for the process ( [ eq : theprocess ] ) at tree level is given by @xmath141 where @xmath142 refers to the @xmath134 coming from the @xmath92 decay , respectively , and we have denoted @xmath116 and @xmath143 as @xmath144 and @xmath145 , respectively . the following approximations have been made above : * we neglected the diagram in which the two @xmath134 are exchanged , assuming that we can identify @xmath142 by looking for a @xmath134 flying near @xmath133 , respectively . as the taus from @xmath146 are highly boosted and back - to - back in the higgs rest frame , this should be an excellent approximation . * all intermediate particles are assumed to be on - shell , so the denominators of their propagators have been dropped in ( [ eq : fullm : initial ] ) , as they are just momentum - independent constants @xmath147 . * we neglect @xmath148 throughout the paper . a convenient consequence of this ( very good ) approximation is that the @xmath149 amplitude in ( [ eq : m : rho_to_pi - pi ] ) effectively satisfies a `` ward identity '' , _ i.e. _ , it vanishes upon replacing @xmath150 with @xmath151 : @xmath152 this is why we have dropped the @xmath153 term of the @xmath92 propagators in ( [ eq : fullm : initial ] ) . carefully keeping the combinations @xmath154 intact as suggested by the heuristic analysis of section [ sec : heuristic ] , the amplitude ( [ eq : fullm : initial ] ) can be rewritten as @xmath155 where @xmath156 taking @xmath157 as the set of independent variables ( subject to the constraint @xmath158 ) , let us analyze how the physics depends on these momenta . first , in the square of the amplitude ( [ eq : fullm : final ] ) , the variables @xmath159 and @xmath160 will only enter via the products @xmath161 and @xmath162 . these combinations can be further simplified as @xmath163 where @xmath164 with are respectively equal to @xmath165 used in refs . @xcite . ] @xmath166 in terms of @xmath167 and @xmath168 , the square of the amplitude in ( [ eq : fullm : final ] ) only involves the traces over four @xmath169 matrices , and an elementary computation gives @xmath170 where @xmath171 \,,\\{p_\text{$\mathrm{{\delta } } , { \setbox0=\hbox{$\mathrm{s}$ } \dimen0=\wd0 \setbox1=\hbox{/ } \dimen1=\wd1 \ifdim\dimen0>\dimen1 \rlap{\hbox to \dimen0{\hfil/\hfil } } \mathrm{s } \else \rlap{\hbox to \dimen1{\hfil$\mathrm{s}$\hfil } } / \fi}$}}&\equiv -2 \cos(2{\delta } ) \ , ( k_- { \!\cdot\!}p_{\tau^- } ) ( k_+ { \!\cdot\!}p_{\tau^+ } ) \,,\\{p_\text{${\delta } , \mathrm{s}$}}&\equiv - { \mathrm{e}}^{2{\mathrm{i}}{\delta } } \bigl [ ( k_- { \!\cdot\!}p_{\tau^+ } ) ( k_+ { \!\cdot\!}p_{\tau^- } ) -(p_{\tau^- } { \!\cdot\!}p_{\tau^+ } ) ( k_- { \!\cdot\!}k_+ ) { \nonumber}\\ & \hspace{9.3ex } -{\mathrm{i}}{\epsilon}_{\mu\nu\rho{\sigma } } \ , k_-^\mu \ , p_{\tau^-}^\nu k_+^\rho \ , p_{\tau^+}^{\sigma}\bigr ] . { \label{eq : pgood : original}}\end{aligned}\ ] ] here , @xmath172 is the interesting contribution that depends on both @xmath18 and the @xmath94 spins . on the other hand , @xmath173 is an uninteresting piece since it is independent of @xmath18 . ( it is sensitive to the @xmath94 spins , _ i.e. _ , the relative orientation of the @xmath83 and @xmath82 subsystems , as it involves scalar products like @xmath174 ) . lastly , @xmath175 does depend on @xmath18 but is insensitive to the spins , as it only involves @xmath176 and @xmath177 , which are just scalar quantities of the @xmath83 and @xmath82 subsystems alone . we therefore focus on @xmath172 . to reveal how it depends on the relative orientations of the @xmath94 systems to each other , observe that @xmath172 is antisymmetric under @xmath178 . this suggests that @xmath167 and @xmath168 should be combined into two antisymmetric tensors @xmath179 , one for each @xmath94 system : @xmath180 in terms of these , @xmath172 takes an elegant form : @xmath181 moreover , the fact that @xmath179 are antisymmetric 2nd - rank tensors suggests that the physics is clearest in terms of their `` electric '' and `` magnetic '' components : @xmath182 indeed , @xmath172 then simplifies into just one term : @xmath183 . { \label{eq : pgood}}\end{aligned}\ ] ] we will now develop intuition for @xmath184 and @xmath185 . first , from ( [ eq : e_and_b ] ) , we have @xmath186 where @xmath187 is the 3-velocity of the @xmath94 . thus , @xmath188 in the rest frame of each @xmath94 , respectively , while in all other frames @xmath185 are perpendicular to both @xmath184 and @xmath189 . moreover , in the boosted @xmath94 limit ( @xmath190 ) , we have @xmath191 ) , @xmath184 is given by @xmath192 clearly , @xmath184 takes the simplest form in the @xmath94 rest frame since then @xmath189 in the second term vanishes . let us use @xmath193 to indicate the quantities evaluated in the respective @xmath94 rest frames . then , combining ( [ eq : var : k+- ] ) and ( [ eq : e ] ) in the @xmath94 rest frames , we have @xmath194 \,,{\label{eq : e_rest}}\end{aligned}\ ] ] where we have used @xmath195 . therefore , in an arbitrary frame with a @xmath94 velocity @xmath196 , we have @xmath197^\perp = { \dfrac{e_{\tau^\pm}}{m_\tau } } \ , \vec{e}_{\pm}^\perp{{}^{\!}\bigr|_0}\,,{\label{eq : e_components } } \end{aligned}\ ] ] where @xmath198 and @xmath199 are the components of @xmath184 parallel and perpendicular to @xmath196 , respectively , while @xmath200 ) is that , for a boosted @xmath94 ( @xmath201 ) , we get latexmath:[$|\vec{e}_\pm^\perp| \gg perpendicular to @xmath196 . thus , the relative magnitudes and orientations of @xmath184 , @xmath185 , and @xmath196 in the boosted @xmath94 limit are akin to those of electromagnetic waves . to summarize , we write out @xmath184 and @xmath185 in the higgs rest frame . since the @xmath94 are highly boosted in this frame , we can neglect @xmath203 . then , combining ( [ eq : e_rest ] ) and ( [ eq : e_components ] ) with @xmath204 , we get @xmath205^\perp \,,{\label{eq : e : higgs_frame}}\end{aligned}\ ] ] where @xmath193 on a vector indicates that the vector should be evaluated in the respective @xmath94 rest frame , while @xmath206 denotes the components perpendicular to the respective @xmath94 velocity in the higgs rest frame . recall @xmath185 is given by ( [ eq : b ] ) . the definition of our variable @xmath207 , drawn in the higgs rest frame with the @xmath82 and @xmath83 going in and out of the page respectively . note @xmath207 is taken to be positive if @xmath208 is on the upper - half plane , and negative otherwise , where @xmath209 is fixed to lie along the @xmath210 axis . ] [ sec : theta ] we are ready to evaluate @xmath172 in the higgs rest frame . in this frame , since @xmath211 and @xmath212 are back to back , the @xmath208@xmath213 plane and the @xmath209@xmath214 plane are parallel to each other . thus , we will superimpose them to make a single plane . in this combined plane , let @xmath209 and @xmath214 point to the right and downward , respectively . ( see figure [ fig : theta_definition ] . ) then , we define @xmath207 to be the angle of @xmath208 with respect to @xmath209 , where @xmath215 if @xmath208 is on the upper - half plane , while @xmath216 if on the lower - half plane . is the acoplanarity angle between the @xmath208@xmath211 plane and @xmath209@xmath212 plane , where the orientation of the planes defined by the respective @xmath185 . ] that is , @xmath217 \operatorname{arccos}\biggl [ { \dfrac{\vec{e}_+ { \!\cdot\!}\vec{e}_-}{{\bigl\lvert \vec{e}_{+\ ! } \bigr\rvert } \ , { \bigl\lvert \vec{e}_{-\ ! } \bigr\rvert } } } \biggr],{\label{eq : thetadef}}\end{aligned}\ ] ] where @xmath218 takes values between @xmath219 and @xmath220 . then @xmath213 makes an angle @xmath221 with respect to @xmath209 . the magnitudes of @xmath185 are the same as the respective @xmath184 . putting everything together , the distribution ( [ eq : pgood ] ) becomes @xmath222 where @xmath184 are given by ( [ eq : e : higgs_frame ] ) . the contributions that have been neglected to arrive at ( [ eq : greatformula ] ) from ( [ eq : pgood : original ] ) are only @xmath223% . [ sec : studies ] in this section we develop collider analyses aimed at reconstructing the @xmath30 angle in ( [ eq : thetadef ] ) . from ( [ eq : matrix_element : full ] ) and ( [ eq : greatformula ] ) , the matrix element squared for the @xmath224 decay has a term proportional to @xmath225 : the @xmath30 distribution is thus sensitive to the @xmath0 phase @xmath31 as its minimum is located at @xmath226 . as before , we fix @xmath227 and therefore the only new parameter we introduce is @xmath31 . we implement the @xmath31 phase in ( [ eq : lpheno ] ) and the effective vertices in ( [ eq : m : tau_to_rho - nu ] ) and ( [ eq : m : rho_to_pi - pi ] ) into a feynrules v.1.6.0 @xcite model . we then generate monte carlo events in madgraph 5 @xcite for @xmath228 production at the lhc with @xmath43 tev as well as @xmath229 production at the ilc with @xmath230 gev : in either case , the higgs decays via @xmath231 . in order to retain quantum interference effects , the full @xmath232 body process is simulated . for the lhc study , we also generate a background sample of @xmath233 production with the subsequent decay @xmath234 . we will first study the effectiveness of the @xmath30 distribution at truth level , assuming the neutrino momenta are known : this facilitates a comparison to the @xmath235 variable @xcite , which was previously proposed for studying @xmath0 violation in the higgs coupling to taus . after demonstrating the superior qualities of the @xmath30 variable , we present a sensitivity study for reconstructing @xmath30 at the ilc , where the neutrino four - momentum can be reconstructed up to a two - fold ambiguity . finally , we turn to the lhc , where the neutrinos can not be reconstructed and the irreducible @xmath10 background is significant . in this case , we find that using a collinear approximation @xcite for the neutrino momenta in addition to the standard hard cuts for higgs events still allows the @xmath30 distribution to retain significant discrimination power between different underlying @xmath31 signal models . we do not include pileup or perform any detector simulation in this work , aside from implementing flat efficiencies for @xmath1-tagging for the lhc study . pileup effects are expected to complicate the primary vertex determination necessary for measuring charged pion tracks as well as contribute extra ambient radiation in the electromagnetic calorimeter ( ecal ) , making neutral pion momenta measurements more difficult . furthermore , finite tracking and calorimeter resolutions are expected to smear the @xmath30 distribution . in particular , the ability to distinguish between charged and neutral pion momenta when both pions are overlapping also could affect the @xmath30 measurement . note , however , that because of the magnetic field , the softer @xmath133 and @xmath134 could be separated at the ecal . even if the two pions overlap in the ecal , the @xmath134 momentum can be obtained by subtracting the track momentum from the total momentum measured in ecal , assuming negligible contamination from other sources of energy deposition . we also neglect the neutral pion combinatoric issue , which is justified if the respective parent rho mesons are boosted far apart as a result of the higgs decay . in general , the @xmath133 and @xmath134 coming from the same @xmath92 parent are mostly collinear . this fact has been exploited in the hadronic tau tagging algorithm . for example , the hps algorithm used by cms requires that the charged and neutral hadrons are contained in a cone of the size @xmath236 , where @xmath237 is the transverse momentum of the reconstructed tau @xcite . since the two tau candidates are usually required to be well separated , the combinatorics problem in determining the correct @xmath92 parents can be ignored . recall from ( [ eq : matrix_element : full ] ) and ( [ eq : greatformula ] ) that the minimum of the @xmath30 distribution is located at @xmath238 , and so constructing the @xmath30 distribution allows us to read off the @xmath31 phase of the underlying signal model . in figure [ fig : true_varydelta ] , we show the @xmath30 distribution in @xmath228 events where we have temporarily assumed the neutrinos are fully reconstructed . the various signal models with @xmath239 ( @xmath0-even ) , @xmath240 ( maximal @xmath0 admixture ) , and @xmath241 ( @xmath0-odd ) clearly show the large @xmath242 contribution of the matrix element as seen in ( [ eq : greatformula ] ) . we also superimpose the @xmath30 distribution from @xmath233 event . note that it is flat . clearly , observing the cosine oscillation in experimental data will require both a favorable signal to background ratio as well as a solution for the neutrino momenta that preserves the inherently large amplitude of the @xmath30 oscillation . the @xmath30 distributions ( compare with ( [ eq : greatformula ] ) ) for the higgs with @xmath239 ( @xmath0-even ) , @xmath240 ( maximal @xmath0 admixture ) , and @xmath243 ( @xmath0-odd ) , and the @xmath10 , assuming neutrinos are fully reconstructed . the relative normalization of the @xmath10 line is arbitrary . ] we now compare @xmath30 at truth level with the @xmath235 variable proposed in refs . @xcite : here , @xmath235 is the acoplanarity angle between the decay planes of @xmath244 and @xmath122 in the @xmath245 rest frame . the sign of @xmath235 is defined as the sign of the product of @xmath246 . following @xcite , the events are divided into two classes , @xmath247 and @xmath248 , where the two classes are differ by a @xmath249 phase shift . in order to make a direct comparison with our @xmath30 variable , we combine the @xmath235 distributions of the two classes with a @xmath249 phase shift so the phases of the two classes agree . note that while @xmath235 does not refer to the neutrinos , this classification into the two classes still requires the knowledge of the neutrino momenta ( see ( [ eq : var : y+- ] ) ) . assuming the neutrinos are fully reconstructed , the @xmath30 and @xmath235 distributions for @xmath250 events are shown in figure [ fig : thetavsworek ] with @xmath239 . we readily see that oscillation amplitude of the @xmath30 distribution is larger than that of the acoplanarity angle @xmath235 by about @xmath251 . compared to @xmath235 , the @xmath30 variable thus provides superior sensitivity to the @xmath0 phase @xmath31 . the distributions of our @xmath30 and the @xmath235 variable of ref . @xcite for @xmath239 . the @xmath235 distribution is aggregated from the two @xmath248 and @xmath252 classes as explained in the text to make the direct comparison clearer . ] having considered the case where the neutrinos from the tau decays are fully reconstructed , we next turn to the lepton collider environment , where we will find the neutrinos can be fully reconstructed up to a two - fold ambiguity . at a lepton collider running at @xmath230 gev , such as the ilc , the main production mode for the higgs is via associated production with a @xmath10 boson . our prescribed decay mode for the higgs , @xmath253 , has two neutrinos that escape the detector . we use the known initial four momenta , two tau mass and two neutrino mass constraints to solve for each neutrino momentum component . note we will assume the @xmath10 decays to visible states , which will reduce our event yield by 20% . solving the system of equations for the neutrino momenta gives rise to a two - fold ambiguity , where one solution is equal to the truth input neutrino momenta while the other gives a set of wrong neutrino momenta . note both solutions are consistent with four - momentum conservation and therefore correctly reconstruct the higgs mass . since these solutions are indistinguishable in the analysis , we assign each solution half an event weight . the resulting distribution of @xmath30 for @xmath239 is given in figure [ fig : ilctheta ] , where we superimpose the truth level @xmath30 distribution for @xmath254 events for easy comparison . we can see that the oscillation amplitude at the ilc is degraded from the truth level result by @xmath255 . we also show the reconstructed distribution for @xmath239 , @xmath240 , and @xmath256 in figure [ fig : ilcvarydelta ] . while the two - fold ambiguity for the neutrino momenta solution set does degrade the truth level result , the _ @xmath30 distribution in figure [ fig : ilcvarydelta ] shows significant discrimination power between various @xmath31 signal models . note the amplitude of pseudoscalar distribution ( @xmath256 ) is slightly higher than the scalar amplitude : here , the `` wrong solution '' approximates the correct neutrino momenta on average better than the other @xmath239 or @xmath240 cases . this small effect can be traced back to equation ( [ eq : m : h_to_tautau : higgs_frame ] ) where we derived that a pseudoscalar decays to two taus in the singlet spin state . as a result , in this case the two tau spins point in opposite directions , regardless of the spin quantization axis . in the pseudoscalar case the two tau decays thus tend to occur with opposite orientation and the two neutrinos are slightly more back - to - back and consequently the two solutions for their momenta are closer together . the truth and reconstructed @xmath30 distributions at the ilc for @xmath239 . ] the reconstructed @xmath30 distribution at the ilc for @xmath239 , @xmath257 , and @xmath258 . ] we now discuss the projected ilc sensitivity for measuring @xmath31 . at the ilc , the cross section for @xmath259 production at @xmath230 gev with polarized beams @xmath260 for @xmath261 gev is 0.30 pb @xcite . distribution is insensitive to the polarization of the @xmath262-@xmath263 beams . ] assuming a higgs branching fraction to tau pairs of 6.1% , a @xmath264 branching fraction of 26% , and a @xmath10-to - visible branching fraction of 80% , we calculate the ilc should have 990 events with 1 ab@xmath5 of luminosity . since the solved neutrino momenta correctly reconstruct the higgs mass , the @xmath265 backgrounds are negligible and will be ignored . [ tab : ilc ] to estimate the expected ilc accuracy for measuring @xmath31 , we perform a log likelihood ratio test for the sm hypothesis with @xmath266 against an alternative hypothesis with @xmath267 . in general , the likelihood ratio in @xmath268 bins is given by @xmath269 where @xmath270 , @xmath271 and @xmath272 are the number of background events , signal events assuming @xmath239 , and signal events assuming @xmath267 in bin @xmath273 of the @xmath30 distribution . in our ilc treatment , we neglect @xmath265 and @xmath274 continuum backgrounds and so we set @xmath275 . here , @xmath276 is the usual poisson distribution function , @xmath277 . we parametrize the signal @xmath30 distribution with a @xmath278 fit function , where the offset constant @xmath279 and oscillation amplitude @xmath280 are fixed by the fit of the standard model @xmath30 distribution with @xmath239 , giving @xmath281 and @xmath282 respectively . then , the resulting @xmath283 signal @xmath30 distribution is given by @xmath284 . we construct the binned likelihood bins , though we verified the number of bins is immaterial for our results . ] according to ( [ eq : lr ] ) for various @xmath285 hypotheses to test the discrimination against the sm hypothesis . with 1 ab@xmath5 of ilc luminosity , we find @xmath286 discrimination at @xmath287 , which is a highly promising degree of sensitivity for measuring the @xmath0 phase of the higgs coupling to taus . we summarize our rate estimate and accuracy result in table [ tab : ilc ] . we remark that this sensitivity estimate is only driven by statistical uncertainties , and systematic uncertainties are expected to reduce the efficacy of our result . also , detector resolution effects and sm backgrounds , while expected to be small , will also slightly degrade our projection . based on our results , which surpass earlier accuracy estimates of @xmath39 @xcite , a full experimental sensitivity study incorporating these subleading effects is certainly warranted . we now develop an lhc study for reconstructing the @xmath30 distribution in @xmath250 in the @xmath288 final state . we use the @xmath289 final state for a couple of reasons . first , since hadronic taus can be faked by jets , @xmath290 two hadronic taus faces an immense background from multijet qcd . by requiring another object in the final state , we gain handles to suppress the background . second , the collinear approximation gives ambiguous results if the two taus are back - to - back , so the requirement of an additional object in the event guarantees we are away from this configuration . one option is associated production of a higgs wit a @xmath291 . however this rate is quite small , especially once the branching ratios for @xmath291 into clean final states are taken into account . other possibilities include higgs production via vector boson fusion and in association with a jet . both of these options give promising signal - to - background ratios and both should be considered . for concreteness we will consider @xmath292 here as a demonstration of the feasibility of our technique . as mentioned before , the neutrinos are not reconstructible in the hadron collider environment , and so we will employ the collinear approximation @xcite for the neutrino momenta . in figure [ fig : truthvscollinear ] , we show a comparison between the truth level @xmath30 distribution and the @xmath30 distribution using the collinear approximation for neutrino momenta , for the @xmath239 benchmark . while the collinear approximation reduces the oscillation amplitude of the distribution , the location of the minimum of the distribution does not change . therefore , measuring @xmath31 is a viable possibility at the lhc using the collinear approximation for the neutrino momenta . we remark that in the collinear approximation , @xmath30 is equivalent to the acoplanarity angle @xmath235 @xcite . yet , we are the first feasibility study for measuring @xmath0 violation in the higgs coupling to taus at hadron colliders using prompt tau decays and kinematics . with a more sophisticated scheme than the collinear approximation , the @xmath30 variable will be superior to @xmath235 . the distributions of the truth-@xmath30 and @xmath30 from the collinear approximation for @xmath239 . ] at the lhc , the dominant background for the @xmath293 signal process is the irreducible @xmath294 background , where the @xmath10 decays to the same final state as the higgs . as shown earlier in figure [ fig : true_varydelta ] , the @xmath30 distribution from @xmath10 events is flat : importantly , this is true regardless of possible mass window cuts on the reconstructed @xmath295 resonance . we remark that the @xmath0 phase in the higgs coupling to taus does manifest in the @xmath10@xmath1@xmath1 vertex at one loop . since this effect is suppressed by @xmath296 , whereas the signal to background ratio will be @xmath297 , we can safely ignore the loop induced @xmath0 phase in the @xmath10@xmath1@xmath1 vertex . in addition , we will assume that the qcd background contribution also has a flat @xmath30 distribution , since the qcd contamination in the signal region is not expected to have any particular spin correlations . using our @xmath293 and @xmath294 event samples from madgraph 5 for a 14 tev lhc , we first isolate the signal region with a series of hard cuts . first , we apply a preselection requirement on the leading jet @xmath298 gev with @xmath299 . using mcfm v.6.6 @xcite with these preselection requirements on the leading jet , we obtain a @xmath293 nlo inclusive cross section of 2.0 pb with @xmath23 gev and a @xmath294 nlo inclusive cross section of 420 pb . after applying the appropriate higgs , @xmath10 , and tau branching fractions , we calculate a signal cross section of 8.2 fb and @xmath10 background cross section of 970 fb . , while for the background we use @xmath300 . these scale choices are motivated by agreement with higher order ( nnlo ) calculations ( where they exist ) . ] next , we impose hard kinematic cuts to isolate the signal . motivated by @xcite , we choose the signal region to be : * @xmath301 gev , * @xmath302 gev , * @xmath303 , * @xmath304 gev , where @xmath305 is the reconstructed higgs mass by using the collinear approximation . the hard @xmath305 cut strongly suppresses the @xmath306 background , but is less effective on multijet qcd . to reduce the multijet component and its accompanying uncertainty to less than 10% of the total background we impose a high @xmath307 cut . the net efficiencies for signal and @xmath10 background after these cuts are 18% and 0.24% , respectively . rather than simulate the qcd contribution , we account for qcd contamination in the signal region by increasing the @xmath10 background rate by 10% : a complete treatment of the expected qcd background is beyond the scope of this study . finally , for hadronic @xmath1 tagging efficiency , we consider a standard 50% efficiency and a more optimistic 70% efficiency @xcite . we therefore expect 1100 signal events and 1800 @xmath308 qcd background events with 3 ab@xmath5 of luminosity from the 14 tev lhc , assuming 50% @xmath1 tagging efficiency . these rates are summarized in table [ tab : input ] . [ tab : input ] we note that although we generated signal and background samples independently , there is a small interference between higgs and @xmath10 diagrams in the @xmath309 diagram . our checks of this interference on the @xmath30 distributions for combined signal and background events versus separate signal and background events showed a negligible effect : we thus ignore this interference effect . we now perform a likelihood analysis ( [ eq : lr ] ) to quantify how effectively the @xmath30 distribution distinguishes between signal hypotheses with different @xmath0 phases in the presence of @xmath308 qcd background . first , we test the discrimination between a pure scalar and a pure pseudoscalar @xmath49@xmath1@xmath1 coupling . we find that these two hypotheses can be distinguished at @xmath310 sensitivity with 550 ( 300 ) fb@xmath5 assuming 50% ( 70% ) @xmath1 tagging efficiency . we can attain @xmath311 sensitivity between pure scalar and pseudoscalar couplings with 1500 ( 700 ) fb@xmath5 luminosity assuming 50% ( 70% ) efficiency . we also estimate the possible accuracy for the lhc experiments to measure @xmath31 with an upgraded luminosity of @xmath312 . we adopt the same procedure as with the ilc accuracy estimate described in the previous section , modified to account for the @xmath308 qcd background , which is fixed to be flat in @xmath30 . we find that the accuracy in measuring @xmath31 is @xmath42 ( @xmath44 ) assuming 50% ( 70% ) hadronic @xmath1 tagging efficiency . the scalar versus pseudoscalar discrimination and the accuracy estimates are summarized in table [ tab : lhc ] . [ tab : lhc ] again , these estimates are based only on statistical uncertainties without performing a full detector simulation . the effects from pileup and detector resolution are expected to degrade these projections , but corresponding improvements in the analysis , such as a more precise approximation for the neutrino momenta , improved background understanding ( from other lhc measurements ) or multivariate techniques , could counterbalance the decrease in sensitivity . the promising results of our study strongly motivate a comprehensive analysis by the lhc experiments for the prospect of measuring the @xmath0 phase @xmath31 . [ sec : conclusion ] higgs decays to tau leptons provide a singular opportunity to measure the @xmath0 properties of the higgs - fermion couplings . in this paper , we have studied the decay of @xmath313 followed by @xmath314 . a new observable , @xmath30 , was constructed in ( [ eq : thetadef ] ) using the momenta of the tau decay products . the differential cross section can be written in a form of @xmath315 , hence the @xmath30 distribution can be used to distinguish various @xmath0 mixing as shown in figure [ fig : true_varydelta ] . the @xmath30 variable can be viewed as an acoplanarity angle between the planes spanned by certain linear combinations of the pion and neutrino momenta , and it was demonstrated to be superior to previously proposed acoplanarity angles . at the ilc , where the neutrino momenta can be reconstructed up to a two - fold ambiguity , the advantages of the @xmath30 variable are most evident . we estimate that the @xmath0 phase can be measured to an accuracy of @xmath36 for @xmath230 gev , a substantial improvement over previous results . for the lhc , we have had to rely on the collinear approximation to reconstruct the neutrino momenta and some of the discriminating power of the @xmath30 variable is lost . nevertheless , we find an accuracy of @xmath42 ( @xmath44 ) is possible after 3000 fb@xmath5 of luminosity and assuming a 50%(70% ) tau tagging efficiency . recasting in terms of the parameters introduced in ( [ eq : leff ] ) , a @xmath316@xmath317 deviation from the sm case is equivalent to sensitivity to @xmath318 tev , where @xmath319 is the scale of the dimension six operator in ( [ eq : leff ] ) and @xmath320 is assumed to be @xmath321 . a better approximation scheme for the neutrino momenta will improve these results . in our collider studies we have neglected detector effects and background systematic uncertainties . while adding these effects will worsen our results , this may be offset by better understanding of the backgrounds ( thereby allowing looser cuts ) and with a more sophisticated ( _ e.g. _ mva ) analysis and statistical tools . finally , in this work we picked specific higgs production mechanisms and focused on a single decay channel . to understand the full extent of future colliders sensitivities to the @xmath0 phase of higgs - fermion couplings , additional production channels such as vbf should be explored , both at the lhc and in a higgs factory . in addition , other hadronic decay channels , as well as semi - leptonic channels , of the tau pair might also be sensitive to the @xmath0 properties of the higgs . we would like to thank kaustubh agashe , wolfgang altmannshofer , yuval grossman , uli haisch , josh ruderman , daniel stolarski , raman sundrum , ciaran williams and jure zupan for comments and discussions . we would also like to thank the kavli institute for theoretical physics at ucsb where part of this work was performed . rp is supported by the nsf under grant phy-0910467 . to is supported by the doe under grant de - fg02 - 13er41942 . fermilab is operated by the fermi research alliance , llc under contract no . de - ac02 - 07ch11359 with the united states department of energy . [ app : uvcompletion ] in this appendix we give an example for a uv completion for the dimension-6 operator in ( [ eq : leff ] ) , _ i.e. _ , the term with @xmath65 . our purpose is not to advocate a specific model as particularly well - motivated but to simply provide an existence proof of a weakly - coupled renormalizable theory that can generate the @xmath65 term in ( [ eq : leff ] ) at @xmath322 with an arbitrary @xmath0 phase , without generating other operators that may contradict with experiments . consider an extension of the sm with a second higgs doublet @xmath323 with @xmath324 with the following tree - level lagrangian : @xmath325 where @xmath326 is the sm lagrangian without the tau yukawa coupling . the full quantum lagrangian is @xmath327 , where @xmath328 contains all counterterms _ necessary _ for consistent renormalization at loop level . for simplicity , we neglect neutrino masses and mixings , so @xmath329 possesses an accidental @xmath330 family symmetry , which is then inherited by @xmath328 as well . this immediately implies that there are no lepton flavor changing processes such as @xmath331 . there are no constraints from quark flavor/@xmath0 measurements ; since the couplings of @xmath323 to quarks are absent in @xmath329 and only appear in @xmath328 , they are not only very small ( @xmath332 @xmath333 the corresponding sm yukawa ) but also respect the ckm flavor structure of the sm . similarly , the couplings of @xmath323 to @xmath334 and @xmath335 are inconsequential ; in particular , we have checked that a contribution to the electron elecric dipole moment induced at 2-loop level is negligible . finally , the modification of the coupling of @xmath10 to @xmath1 is also tiny , safely below the lep constraints . in order to see the effects of ( [ eq : uvcomp ] ) on higgs decays let us consider the limit in which @xmath336 and the doublet @xmath323 can be integrated out and we can consider an effective field theory below @xmath337 . at tree level , this generates two dimension-6 interactions : @xmath338 this theory now matches on to the effective theory ( [ eq : leff ] ) with @xmath339 , @xmath340 and @xmath341 . it should be noted that this theory contains in general a @xmath0 violating phase . in particular , the phase of @xmath342 may not be rotated away by field redefinitions . taking @xmath343 and @xmath344 with arbitrary phases in @xmath345 , @xmath346 and @xmath347 can therefore produce an @xmath78 @xmath0-violating phase in higgs decays to tau leptons . other theories , including composite higgs models @xcite and models with vector - like leptons @xcite may also produce the necessary interactions to induce @xmath0 violating higgs decays into taus . in constructing such models one should take care that the coupling of @xmath10 to taus is not modified above the @xmath348 level ( see ref . @xcite or fig . 10.4 of @xcite ) either by construction ( as in the model we just discussed ) or by a cancellation among various contributions . g. aad _ et al . _ [ atlas collaboration ] , phys . b * 716 * , 1 ( 2012 ) [ arxiv:1207.7214 [ hep - ex ] ] . s. chatrchyan _ et al . _ [ cms collaboration ] , phys . b * 716 * , 30 ( 2012 ) [ arxiv:1207.7235 [ hep - ex ] ] . [ atlas collaboration ] , atlas - conf-2013 - 034 . [ cms collaboration ] , cms - pas - hig-13 - 005 . t. aaltonen _ et al . _ [ cdf and d0 collaborations ] , arxiv:1303.6346 [ hep - ex ] . s. chatrchyan _ et al . _ [ cms collaboration ] , phys . lett . * 110 * , 081803 ( 2013 ) [ arxiv:1212.6639 [ hep - ex ] ] . [ atlas collaboration ] , atlas - conf-2013 - 013 . s. bolognesi , y. gao , a. v. gritsan , k. melnikov , m. schulze , n. v. tran and a. whitbeck , phys . d * 86 * , 095031 ( 2012 ) [ arxiv:1208.4018 [ hep - ph ] ] . y. grossman and i. nachshon , jhep * 0807 * , 016 ( 2008 ) [ arxiv:0803.1787 [ hep - ph ] ] . f. bishara , y. grossman , r. harnik , d. robinson , j. shu and j. zupan , in progress . [ cms collaboration ] , cms - pas - hig-13 - 004 . j. r. dellaquila and c. a. nelson , nucl . b * 320 * , 86 ( 1989 ) . j. r. dellaquila and c. a. nelson , nucl . b * 320 * ( 1989 ) 61 . b. grzadkowski and j. f. gunion , phys . b * 350 * , 218 ( 1995 ) [ hep - ph/9501339 ] . g. r. bower , t. pierzchala , z. was and m. worek , phys . b * 543 * , 227 ( 2002 ) [ hep - ph/0204292 ] . m. worek , acta phys . b * 34 * , 4549 ( 2003 ) [ hep - ph/0305082 ] . k. desch , z. was and m. worek , eur . j. c * 29 * , 491 ( 2003 ) [ hep - ph/0302046 ] . k. desch , a. imhof , z. was and m. worek , phys . b * 579 * , 157 ( 2004 ) [ hep - ph/0307331 ] . s. berge and w. bernreuther , phys . b * 671 * , 470 ( 2009 ) [ arxiv:0812.1910 [ hep - ph ] ] . s. berge , w. bernreuther and j. ziethe , phys . lett . * 100 * , 171605 ( 2008 ) [ arxiv:0801.2297 [ hep - ph ] ] . s. berge , w. bernreuther , b. niepelt and h. spiesberger , phys . d * 84 * , 116003 ( 2011 ) [ arxiv:1108.0670 [ hep - ph ] ] . s. berge , w. bernreuther and h. spiesberger , arxiv:1208.1507 [ hep - ph ] . r. k. ellis , i. hinchliffe , m. soldate and j. j. van der bij , nucl . phys . b * 297 * , 221 ( 1988 ) . n. d. christensen and c. duhr , comput . commun . * 180 * , 1614 ( 2009 ) [ arxiv:0806.4194 [ hep - ph ] ] . j. alwall , m. herquet , f. maltoni , o. mattelaer and t. stelzer , jhep * 1106 * ( 2011 ) 128 [ arxiv:1106.0522 [ hep - ph ] ] . c. collaboration _ et al . _ [ cms collaboration ] , jinst * 7 * , p01001 ( 2012 ) [ arxiv:1109.6034 [ physics.ins-det ] ] . ilc technical design report , vol . 2 , [ ilc collaboration ] http://www.linearcollider.org/ilc/tdr j. m. campbell , r. k. ellis and c. williams , jhep * 1107 * , 018 ( 2011 ) [ arxiv:1105.0020 [ hep - ph ] ] . k. agashe , r. contino and a. pomarol , nucl . b * 719 * , 165 ( 2005 ) [ hep - ph/0412089 ] . g. f. giudice , c. grojean , a. pomarol and r. rattazzi , jhep * 0706 * ( 2007 ) 045 [ hep - ph/0703164 ] . j. kearney , a. pierce and n. weiner , phys . d * 86 * , 113005 ( 2012 ) [ arxiv:1207.7062 [ hep - ph ] ] . s. schael _ et al . _ [ aleph and delphi and l3 and opal and sld and lep electroweak working group and sld electroweak group and sld heavy flavour group collaborations ] , phys . rept . * 427 * , 257 ( 2006 ) [ hep - ex/0509008 ] . j. beringer _ [ particle data group collaboration ] , phys . d * 86 * , 010001 ( 2012 ) .

quantum hall bilayer phase diagram with respect to interlayer distance bears a remarkable similarity with phase diagrams of strongly correlated systems as a function of doping , with magnetic ordering on the one end and fermi - liquid - like behavior on the other . moreover , it has been suggested that a bcs correlated state of composite fermions with-wave pairing may exist in the intermediate region . in the same region , an exact diagonalization study in the torus geometry pointed out the existence of state(s ) with pseudospin spiraling order . here we reconcile these two descriptions of the intermediate state by considering the underlying bosonic representation of the composite fermion paired state in the long distance limit , and by performing extensive exact diagonalizations on the torus . we argue that the spiraling states belong to the manifold of degenerate ground state(s ) , and are a consequence of bose condensation of the quasiparticles ( with critical algebraic correlations ) at non - zero momenta in the two pseudospin states . the spiraling states , generated in this way as spin - textures , can be identified with meron - antimeron constructions . thus , merons the fractionally charged vortex excitations of the xy magnetically ordered state constitute some of the topological sectors . it follows that merons are deconfined in the intermediate state , and allow for a smooth transition between the magnetically ordered and fermi - liquid - like phases , in which they are bound in pairs .

1509.01921

multicomponent fractional quantum hall ( fqh ) systems feature rich physics of strongly interacting electrons . in contrast to strong electronic correlations in various ( material dependent ) crystal settings , in fqh systems the kinetic energy is effectively removed from the problem due to the landau quantization in strong magnetic fields . this makes the multicomponent fqh systems a simpler ( better controlled ) setting that is amenable to the investigations of fundamental physics of strongly correlated electrons . one of the most prominent examples of multicomponent fqh systems is the quantum hall bilayer ( qhb ) at total filling factor one . the filling factor @xmath1 is defined as a product of electron density @xmath2 and the magnetic area @xmath3 , where @xmath4 is the emergent length scale in the magnetic field ( magnetic length " ) @xcite . alternatively , the filling factor represents the ratio of the number of electrons @xmath5 to the number of magnetic flux quanta @xmath6 through the system . in the case of the @xmath7 qhb studied here , there are two quantum wells ( layers " ) that we label by @xmath8 , and @xmath9 . note that throughout this work we assume the ordinary electron spin to be fully polarized by the external magnetic field . because of the two - fold layer degree of freedom , in general we need to distinguish the interaction between electrons in the same layer ( intra - layer " ) versus electrons in opposite layers ( inter - layer " ) . the difference between these two interaction strengths can be tuned experimentally via the parameter called the bilayer distance @xmath10 , which makes the physics in this system much richer compared to an ordinary system of electrons with spin . qhb system has indeed attracted much attention since the time of the theoretical prediction @xcite and experimental detection @xcite of the ordered state of excitons ( bosons ) that forms when @xmath11 , despite the fact that the fundamental charge carriers in the system are fermions . perhaps equally remarkable and interesting is the physics at larger ( intermediate ) distances , typically @xmath12 . there , an intermediate phase of paired composite fermions , introduced in ref . [ see eqs . ( [ fpairing1 ] ) and ( [ sdispairing ] ) below ] , is believed to exist . on the other hand , an exact diagonalization study of the system in the torus geometry @xcite also pointed out to the existence of pseudospin state(s ) with spiraling order in the intermediate region . here we reconcile these two descriptions of the intermediate region by considering the underlying bosonic representation of the composite fermion paired state in the long distance limit , and by performing extensive exact diagonalization studies on the torus . this paper is organized as follows . in section [ sec : review ] we start with an overview of the qhb physics from the point of view of model wave functions . we first discuss the limits of small and large bilayer distances which are well - understood and correspond to the exciton superfluid and the product of two decoupled ( composite ) fermi liquid states , respectively . we also review the theoretical evidence for a paired phase in the regime of intermediate bilayer distances , and briefly motivate and summarize our contributions . in section [ sec : picture ] we provide a simple physical picture for the formation of the putative intermediate state . section [ sec : numerics ] contains results of exact diagonalization studies of the qhb on the torus . in section [ sec : boson ] we discuss the torus results from the composite boson viewpoint , and give arguments for meron deconfinement in the intermediate region . we conclude in section [ sec : conc ] with a short summary and discussion . appendices contain further resullts on the intermediate phase . the qhb at the total filling factor one consists of two quantum hall mono - layers separated by distance @xmath10 in the @xmath13-direction ( perpendicular to both layers ) . the tunneling between the layers can be nearly completely suppressed in experiment . therefore , we can assume that @xmath14 and each layer is at filling one half . in the following we first give the basic physical picture of the ground state in the limit of either small ( @xmath15 ) or large distances @xmath16 , followed by the discussion of the intermediate regime @xmath17 . for small distances , the intra- and inter - layer interactions are of similar strength , and favor each electron in one layer to be directly on top of a hole in the other layer , which drives exciton formation . in this simplified picture we can differentiate between up and down excitons , but due to the quantum - mechanical uncertainty in which layer an electron is , we must include the pseudospin ( layer ) degree of freedom . the appropriate model function at total filling factor one then reads @xcite @xmath18 the spatial part of the wave function is written in terms of 2d electron coodinates @xmath19 ( regardless of the layer index ) , and assumes the circular gauge for the magnetic field in the infinite plane . the spatial correlations are encoded via the laughlin - jastrow factor @xmath20 , and the wave function vanishes at large distances because of the gaussian term that originates from the single - electron orbitals @xcite . note that the wave function also contains the pseudospin part @xmath21 . the pseudospin of each electron points to the same direction , due to the combined effect of the pauli principle and the coulomb interaction . the state of each particle @xmath22 is given by the pseudospin @xmath23 i.e. , the pseudospin is polarized in the @xmath24 plane . this happens because of the capacitor effect which enforces the equality of total charges in the up " and down " layer , and breaks the @xmath25 spin symmetry , assumed to exist at @xmath26 , down to @xmath27 . this correlation - condensation was detected in several experiments , e.g. , spielman _ @xcite , although the zero - bias peak " anomaly was never as high and sharp as in the similar experiments on superconductors . more recent experiments using coulomb drag and counterflow measurements are also completely in favor of exciton formation [ for a recent review , see ref . ] . when the layers are far apart , it is well - known that each half - filled layer is in the state of the fermi - liquid - like " state @xcite . for example , the state of the upper layer is given by @xmath28 where we dropped the single - particle gaussian factors . the fermi liquid correlations result from the factor @xmath29 that represents the slater determinant of plane waves . here , @xmath30 is the set of 2d momenta that define the 2d fermi sphere and we have substituted @xmath31 . the total wave function of the system in this case is just a direct product of independent layers , @xmath32 notice that both @xmath33 and @xmath34 contain a part which is the product of the differences of @xmath13 s ( the laughlin - jastrow part ) . this part of the wave function fixes the filling factor of the state , which can be seen as follows . on the one hand , the highest exponent in any of the @xmath13 s is equal to the number of flux quanta through the system , which is the degeneracy of the lowest landau level ( lll ) . on the other hand , this number is equal to the number of electrons in the case of @xmath33 , and twice larger than the number of electrons in the case of @xmath34 , as follows from the filling factors . thus , formally speaking the laughlin - jastrow part is the charge " part of the wave function , while the rest is the neutral " part . in @xmath33 the neutral part is the condensate of pseudospins . in @xmath34 the neutral part contains the slater determinant which introduces the complex conjugate coordinates @xmath35 . thus @xmath34 in general has a non - zero projection on many landau levels and not just the lll . if @xmath34 is to be used a lll trial wave function , in general it must be explicitly projected via a complicated derivative operator @xmath36 to make it a function of @xmath13 s only @xcite . nevertheless , in the long distance regime with the sample length @xmath37 , we may assume that @xmath13 and @xmath35 s commute , and thus in this regime the form of the wave function formally decouples into charge and neutral parts . given that the qhb ground state in the limits of both small and large @xmath10 is in a liquid phase , it seems unlikely that a possible intermediate state is a charge - density wave ( that breaks the translation symmetry ) . a possible way to describe this evolution is to start from the bose condensate which is smoothly modified by two fermi condensates that grow as the distance @xmath10 increases @xcite . in such a study on the sphere @xcite it was found that the two fermi condensates pair in the manner of weakly paired bcs @xmath38 wave cooper pairs . ( we use the phrase weakly paired " to denote cooper pairs that are not exponentially localized . ) thus , at the end of the evolution of the superfluid state one may expect the following wave function , @xcite @xmath39 to describe the system . we use here and below a shortened form of the wave functions with spin , where spatial coordinates carry spin . it is understood that by combining the spatial wave functions with spinor wave functions , and summing with proper sign factors over all ways of choosing which electrons have which spin , we get proper totally antisymmetric wave functions . in eq . ( [ fpairing1 ] ) @xmath40 " denotes the antisymmetrized product of @xmath41 pairs of cooper pair wave functions , @xmath42 , where @xmath43 and @xmath44 . the cooper pair wave function @xmath45 describes @xmath38 pairing , i.e. , at small distances it behaves as @xcite @xmath46 with @xmath47 , and @xmath48 which fixes the relative angular momentum of the cooper pair . based on the study of ref . , other numerical studies ( summarized in sec . [ sec : picture ] ) , and our results ( sec . [ sec : numerics ] ) , we will adopt the view that an intermediate state does exist and can be modeled as a bcs @xmath0-wave paired state of composite fermions . the question of whether in the thermodynamic limit we obtain a critical region ( phase ) or rather a critical point is more subtle because of the limitations of all the existing numerics to small system sizes . therefore , we will use the term intermediate phase " loosely , keeping in mind the possibility of it becoming a critical point in the limit of a very large ( clean ) system . in the remaining part of this section , we will introduce in a formal way a possible alternative description of the intermediate state based on the picture of composite bosons , which we expect to be relevant in the long distance limit . first , we would like to point out that the most natural long - distance weak pairing behavior of the pair function in eq . ( [ fpairing1 ] ) with @xmath48 would be @xmath49 this coincides with the existence of a bcs description with a gap function @xmath50 , i.e. , @xmath51 is an eigenfunction of rotations in 2d with @xmath48 [ see ref . ] . this generic behavior may be used to define a suitable candidate for a projected @xmath0-wave paired composite fermion wave function valid at all distances in the lll , @xmath52 where @xmath53 denotes the necessary projection to the lll ( see appendix [ app : proj ] for the details of the projection ) . here we confine ourselves to the long - distance regime , i.e. , consider eq.([fpairing1 ] ) with ( [ pbehavior ] ) . in this regime , the particles may be considered point - like as the relevant probing wavelengths are much larger than @xmath54 . thus , to obtain the long - distance properties we use eq . ( [ fpairing1 ] ) with ( [ pbehavior ] ) , instead of the proper lll function [ eq . ( [ pfpairing ] ) ] , assuming as usual that the universal ( topological ) properties are insensitive to the microscopic differences between the two wave functions . in the following , we will explain how in this regime of the @xmath0-wave composite fermon pairing we can also consider a composite boson representation , which will enable us to describe the topological features ( degenerate ground states on the torus and quasiparticle content ) of the intermediate state . thus the model wave function for the state into which the superfluid evolves , in the long - distance limit when @xmath13 and @xmath35 commute , is @xmath55 using the cauchy identity , we can rewrite the wave function in the following form @xmath56 again , at long distances the charge and neutral part formally decouple , and the neutral part can be viewed as a wave function of some underlying composite boson quasiparticles . furthermore , the neutral part is the product of two determinants , and a careful conformal field theory ( cft ) analysis @xcite will allow for an existence of a branch of gapless excitations ( see appendix a ) on top of this non - chiral state . these formal arguments also suggest that there is a possibility to interpret the intermediate phase in the qhb in terms of bosonic quasiparticles that correlate in a special way via eq . ( [ twodet ] ) . from eq . ( [ twodet ] ) ( see also appendix a ) we can read off that the one - particle exciton ( or composite boson ) correlation function is @xmath57 therefore , instead of the long - range order that we had in eq . ( [ ferro ] ) , in this ground state we have quasi long - range order . indeed , in the finite - size study performed in ref . , it was noticed that the excitonic superfluid order parameter , i.e. @xmath58 , persists in the state with no correlated bosonic quasiparticles in the manner of eq . ( [ ferro ] ) ( i.e. , the so - called 111 " state @xcite ) . it follows that these critical bosons [ eq . ( [ twodet ] ) ] may exist in a region between a ferromagnetically ordered and a completely disordered phase , and thus we may expect a gapless mode and compressible behavior in the neutral sector . this is supported by exact diagonalization studies on the sphere @xcite , but also by the detection of ordered ( correlated ) behavior in early experiments @xcite , which were performed at distances where we expect the physics of the intermediate state to be relevant . but what makes the intermediate region so attractive , as we will show in greater detail by exact diagonalization studies on torus [ sec . [ sec : numerics ] ] , is the existence of competing ground states at momenta that diverge in the thermodynamic limit , and thus define different sectors of the system . this signals a topological structure that coexists with gapless behavior . a convenient way of describing various sectors , as we will argue , is to take the composite boson view of the ground state pairing correlations . then bose condensates under different boundary conditions @xcite can be identified with the competing ground states from exact diagonalization studies on the torus . as usual in topological systems , the torus ground states can be connected with quasiparticles of the topological phase under consideration . we expect this also to be valid in our gapless system , and by bose condensing in two different pseudospin states , we will be able to connect different ground states with meron spin - textures . this opens a door for a possibility that merons fractionally charged vortices of the @xmath24 ordered state @xcite exist as free ( deconfined or not bound in pairs ) quasiparticles in the intermediate state . the first exact diagonalization study on the torus that pointed out to the existence of the competing ground states with momenta that diverge in the thermodynamic limit was done by park @xcite ( see appendix [ app : park ] for a brief review of the relationship between ref . and our work ) . in ref . a spiraling structure in the competing ground states was identified , and it was argued that the intermediate phase possesses a pseudospin spiral order . therefore , the competing states were interpreted as new developing ground states that replace the @xmath24 ordered state ( valid for smaller distances ) due to a ( classical ) symmetry - breaking ordering . on the other hand , in this work we give numerical and analytical reasons to consider an intermediate state with richer topological content than the @xmath24 ordered state ( which has a simple integer quantum hall effect characterization ) . in the intermediate phase , due to _ topological _ order , new quasiparticles emerge and define additional ground states on the torus which have a spiraling structure . in this way our work suggests a way to reconcile and unify the views of the intermediate phase based on the previous numerical approaches , especially ref . and ref . one may naively estimate that the transition to a disordered phase occurs around @xmath59 , and this estimate is in agrement with several results in the literature that we summarize here . ( i ) the calculated treshold for the disappearance of the long - range correlated composite bosons and the extrapolated estimate for the vanishing of the long - range order parameter in ref . . ( ii ) the change in the dmrg spectrum computed in ref . , i.e. , a change in the nature of excited state " with a sudden decrease in the excitation gap " . ( iii ) an abrupt change in the velocity of the goldstone mode in ref . . ( iv ) exact diagonalization ( ed ) studies ( eds ) presented here as well as in ref . , showing the level crossing for the first excited state around @xmath59 . the physical picture of the evolution of the bilayer is the following : in the ordered @xmath24 phase we have locally equal probability to find an up and down exciton . as we increase the distance between the layers , the intra - layer coulomb interaction becomes dominant and we need to place the up excitons , and separately down excitons , further apart . one natural resolution would be an ising order ( charge density wave ) , but another viable possibility is to slowly disturb the long - range ordering and create spin - textures as ground state(s ) . in the magnetic description of ref . , the tendency to create spin - textures may be recognized by the presence of frustrating next - nearest - neighbor ising couplings . thus the magnetic ( bosonic ) description of the intermediate phase may be useful along with the fermionic description presented in ref . . in this section we present results of exact diagonalization calculations performed on the torus geometry . we model a finite system of @xmath5 electrons moving on a surface spanned by vectors @xmath60 and @xmath61 . we fix the filling fraction to be one , which leads to the condition @xmath62 . the focus of our investigation is on the liquid phases , hence we consider an isotropic torus geometry , i.e. , the length of the two vectors is the same ( aspect ratio equal to one ) . additionally , we can also vary the angle @xmath63 between the two vectors @xmath60 and @xmath61 . unless otherwise specified , we fix @xmath64 , corresponding to centered rectangular symmetry . energy spectrum versus momentum @xmath65 in the odd sector on the torus . panel ( a ) shows the spectrum for @xmath66 electrons , while panel ( b ) is for @xmath67 . in both cases the aspect ratio @xmath68 , @xmath69 and bilayer distance @xmath70 . levels are plotted relative to the corresponding ground state . ] the interaction between electrons consists of an intra - layer coulomb potential , given by the fourier transform @xmath71 and the inter - layer coulomb repulsion which is screened by the bilayer distance @xmath72 the interaction potentials are made appropriately periodic ( i.e. , @xmath73 assumes values @xmath74 times an integer ) to satisfy the periodic boundary conditions . after specifying the geometry and the interaction potential , the basic procedure of exact diagonalization is to represent the hamiltonian of the system as a large finite matrix and diagonalize it on a computer . in order to extract physical information about the energy spectrum and system s eigenstates , it is essential to label them with good quantum numbers . in this case , the complete symmetry classification was first achieved by haldane @xcite ; the primary quantum number that emerges from such an analysis is the two - dimensional momentum that characterizes the invariance under magnetic translations . this momentum is generically conserved for any translationally invariant interaction , and we will use it below to label the eigenstates of the system ( see ref . for a recent overview of the many - body translation symmetry classification in the magnetic field ) . in figs . [ fig : spectrum1315 ] and [ fig : spectrum13dk ] we present ed results on the torus of the bilayer system with odd numbers of particles . in fig . [ fig : spectrum1315 ] we show the energy spectrum versus conserved momentum @xmath75 in the systems with @xmath76 ( a ) and @xmath77 electrons ( b ) . in this figure , the bilayer distance is fixed at @xmath70 , i.e. , in the middle of the transition region . because the total number of particles @xmath5 is odd , we restrict to the the largest sector of the hilbert space with @xmath78 and @xmath79 . more generally , the evolution of the entire spectrum as the bilayer distance is varied in presented in fig . [ fig : spectrum13dk](a ) . in fig . [ fig : spectrum13dk ] ( b)-(d ) we show several snapshots of the energy spectrum in fig . [ fig : spectrum13dk](a ) plotted as a function of momentum . we choose three values of @xmath80 that tentatively correspond to three different regimes : the 111 state , intermediate region , and the onset of decoupled composite fermi liquids . the unit cell in figs . [ fig : spectrum13dk](b)-(d ) is approximately the square , and insets show the same results for the hexagonal geometry . what we can see in both geometries [ figs . [ fig : spectrum13dk](b)-(d ) ] as well as fig . [ fig : spectrum1315 ] , is that at distance @xmath81 two distinct states fall " from the excited ( continuum ) part of the spectrum : for example , the state around @xmath82 and another one at @xmath83 in fig . [ fig : spectrum13dk](c ) . the state with the momentum @xmath82 was first pointed out in the study of park @xcite . our results suggest that this state evolves hand in hand with another one at momentum @xmath83 in fig . [ fig : spectrum13dk](c ) . the presence of these two states is robust and does not depend on system size , for example a larger @xmath67 system shown in fig.[fig : spectrum1315 ] contains the same doublet of states . we notice that these special " states compete with the ground state at @xmath84 as @xmath10 becomes larger . the two special states in fact each comprise a family of 4 states connected by symmetry operations ( simple translations and inversion ) . thus , in fig . [ fig : spectrum1315 ] we find a set of 4 states that correspond to the lower energy excitation ( with @xmath85 ) : @xmath86 in units of @xmath87 where @xmath88 . this set of states is followed by the excitation at @xmath89 with a slightly higher energy ( with momentum @xmath90 ) . we note that we have identified such states in all odd systems we studied , e.g. @xmath91 and @xmath92 . ( color online ) goldstone mode for the bilayer distance @xmath93 . quasicontinuous @xmath94 energy spectrum is obtained by varying the torus angle @xmath63 for a fixed aspect ratio @xmath68 . data is shown for all system sizes between @xmath95 and @xmath96 electrons in total . levels are plotted relative to the corresponding ground state for each @xmath63 and @xmath5 . ] ( color online ) quasicontinuous @xmath94 energy spectrum obtained by varying the torus angle @xmath63 for a fixed aspect ratio @xmath68 . total number of electrons is @xmath66 and bilayer distance is @xmath70 . levels are plotted relative to the corresponding ground state for each @xmath63 . levels @xmath97 and @xmath98 belong to the same excitation branch . ] thus we may conclude that at distances where we may expect the intermediate phase , by examining the torus results we find competing ground states at momenta that diverge in the thermodynamic limit . this may be indicative of the presence of bose condensation and spin textures , but requires additional analysis . studies of the gapless goldstone mode , e.g. the numerical one on the sphere @xcite and the experimental detection in ref . , suggest that even though the numerical work , as summarized in sec . [ sec : numerics ] , points to the formation of the disordered state around @xmath99 , a linear gapless mode persists to exist in this phase . thus although we may notice a small gap structure " in figs . [ fig : spectrum13dk](d ) after @xmath99 , it is likely that it does not represent a paramagnetic phase . because of the finite size effects and geometry , this weak goldstone " mode of , as we will argue , critical bosons , is distorted . despite the distortions , in the hexagonal geometry at @xmath100 , we can see clearly , as we explain in the following section , consequences of the physics of critical bosons in the intermediate state : we find three ( almost ) degenerate ground states ( families with members connected with symmetry operations ) and discernible gapless modes that have beginnings in these states . in order to examine in more detail the question of the existence and description of gapless modes by eds , energy spectrum was obtained by varying the torus angle @xmath63 . in this way , as we sweep @xmath63 continuously between @xmath101 and @xmath102 , we obtain many different resolutions of momenta @xmath103 , which produces a quasicontinuous energy spectrum . in fig . [ fig : goldstone ] we perform such a calculation in the ordered phase ( @xmath93 ) , where we find clear evidence for a linear gapless mode . in fig . [ fig : spectrum13angle ] we perform a similar calculation for @xmath70 , i.e. in the intermediate region . here it is strongly suggestive that the two families of states in eqs . ( [ state1 ] ) and ( [ state2 ] ) indeed evolve together as the system is adiabatically deformed , and likely belong to the same gapless excitation branch . the distortions of a finite system do not give us a rigorous proof of the gapless mode , but we can expect that if such a mode exists it will appear in the manner produced by the probe . thus with the distortion we can just produce a state at the distorted momentum " , or we can access the energy of the state belonging to the excitation branch in the 2d momentum space , at missing ( undistorted ) momentum in the finite size system . in the even sector ( @xmath5 even ) of the eds on torus , fig . [ fig : spectrum12 ] , we can similarly recognize the competing ground states with momenta that diverge in the thermodynamic limit . we will postpone the discussion of these results to sec . [ sec : competing ] and sec . [ sec : topo ] . before that , in the same section , it will be explained how ground states at @xmath104 emerge from the physics of critical bosons . ( color online ) ( a ) energy spectrum versus momentum @xmath65 in the even sector on the torus for @xmath105 electrons in total . aspect ratio @xmath68 , @xmath69 and bilayer distance @xmath70 . levels are plotted relative to the corresponding ground state at @xmath106 . ellipses denote groups of levels in momentum sectors @xmath107 and @xmath108 . total degeneracy inside each ellipse is denoted by @xmath109 . ( b ) the brillouin zone for @xmath105 corresponding to the square unit cell . ] though from the torus results ( fig . [ fig : spectrum1315 ] and fig . [ fig : spectrum12 ] ) we can not claim the ideal degeneracy of the competing states , their configuration is robust and characterizes the intermediate region . ( we have verified this in systems of size up to 15 electrons in the odd sector , and up to 14 electrons in the even sector . ) therefore , the systems analyzed are likely near a critical point , line , or perhaps even a region @xcite in the parameter space where there is ideal degeneracy , implied by the topological structure of the model function [ eq . ( [ fpairing ] ) or eq . ( [ twodet ] ) ] . in the following we explain how the extra " competing ground states at non - zero momenta can be interpreted as pairs of merons ( charged vortices of the superfluid state at small @xmath10 ) , whose spiraling period is commensurate with the length(s ) of the system . these spiraling configurations were considered in the work of park @xcite , but here we would like to point out that on the torus they can be viewed as a consequence of the physics of critical bosons . in this way we will relate the physics of long spiraling on the torus and critical ( composite ) bosons to the existing description in terms of @xmath0-paired states of composite fermions ( see introduction and ref . ) . to capture the critical correlations of bosons , in the following we will consider bosons with the critical bcs pairing function @xmath110 , i.e. , @xmath111 in momentum space . thus we assume that the dominant contribution in eq . ( [ twodet ] ) are pairwise correlations and that the state can be considered as a bcs correlated state of bosons ( where the correlations are expressed in terms of a _ permanent _ and not the more usual determinant ) . the assumption of only pairwise correlations is by no means self - evident , and there are two reasons why it is justified : ( a ) in our system there is no real charge - pseudospin separation or , at least we have to allow for such a possibility , and thus halves " of the ( fused ) quasiparticles [ in appendix [ sec : appa ] , the quasiparticle in eq.([qp2 ] ) is a product of two fused halves " ] in the pseudospin sector are also allowed and should be captured in the critical boson description , and ( b ) the second reason is heuristic : by allowing the reduction to pairwise correlations we can describe the ground states on the torus . the pairing allows the constructions of halves " , e.g. halving " the laughlin quasihole into two non - abelions in the moore - read state @xcite , and by keeping only pairing correlations the degrees of freedom that would be forgotten ( suppressed ) otherwise will be released . thus we assume that the low - energy states of the intermediate phase are in one to one correspondence to the states of @xmath112 bcs paired critical bosons . now we proceed similarly to the case of permanent pairing @xmath113 in ref . . in the effective bcs description the occupation number of bosons at momentum @xmath114 is @xmath115 where we note the minus sign with respect to the fermionic case . to maintain the positivity of the expression with @xmath116 , where @xmath117 is a constant with respect to @xmath114 , we must have @xmath118 . this means we must be in the antiperiodic sector on the torus along at least one direction ( @xmath119 or @xmath120 ) . further , we also must have @xmath121 because @xmath122 . finally , because @xmath123 as @xmath124 , i.e. , no pairing in the thermodynamic limit , we can only have bose condensation ( at @xmath125 ) . let us assume ( as in ref . ) that we are in the @xmath126 sector , i.e. , antiperiodic in @xmath119 direction and periodic in @xmath120 direction . we assume an even number of particles ( i.e. , that we are in the even sector ) , and occupy @xmath127 and @xmath128 with equal probability , i.e. , @xmath129 where @xmath130 , and @xmath131 denote second - quantized bosonic operators . thus in the neutral sector each particle would have a description in the form of a two - component spinor : @xmath132 \label{description}\ ] ] where @xmath133 . on the other hand , in park s states , ( 9 ) in ref . , we have the effective description : @xmath134 , \label{pdescription}\ ] ] that successfully reproduces the momenta that we see in eds on the torus . we may notice that the difference comes from the necessity to multiply the neutral part in eq.([description ] ) by @xmath135 in order to glue " the charge ( periodic ) and neutral ( antiperiodic ) sector , and stay in the periodic sector . gluing " effectively brings non - zero momentum to the state : @xmath136 . in park s picture the momentum count results from excitons turning into dipoles ( not locally neutral particles ) , and each exciton acquiring the momentum @xmath137 . thus , at small length scales , charge and spin are not separated , and this corresponds to the requirement of gluing , i.e. , staying in the charge - periodic sector by an extra connection between charge and spin sectors . at the end , we have a spiral winding with @xmath138 , i.e. , the spins wind exactly once over the length of the system , @xmath139 , in @xmath140 direction . thus , due to the intertwined spin and charge degrees of freedom at small length scales , we obtain ground states at large non - zero momenta . but there is another important relation of the gluing " procedure with what we can infer from eds on torus . the description of the intermediate phase that is based only on the neutral sector is incomplete . as usual in fqh systems , the degeneracy of the ground states on the torus can be connected with large gauge transformations " , which are flux insertions of gauge ( internal ) fields through the two holes of the torus @xcite . this is equivalent to changing the boundary conditions , but also to the creation of particle - antiparticle pairs . these transformations can create large momenta that characterize the degenerate ground states , or alternatively different sectors of the theory [ see , for example , a recent discussion in ref . ] . thus we also have to take into account the charge part of the theory in order to describe the ground state . since the work of moon _ et al . _ @xcite , we know that vortex excitations in the exciton superfluid phase are charged they are merons of unit vorticity @xmath141 and charge @xmath142 . we may then question the cft approach for constructing excitations restricted to the neutral sector [ see appendix [ sec : appa ] ] , or in other words whether the excitations obtained by eq . ( [ qp2 ] ) comprise all possible vortex excitations in the intermediate critical phase . in answering this question we may start from the model function for park s states [ eq . ( 8) in ref . or eq . [ cmeron ] here ] . this construction represents two merons with opposite charge and located at two opposite boundaries [ compare with eq.(135 ) in ref . ] . on the disk , this wave function represents one meron at the center of the disk and the other at the infinity . ( this is somewhat similar to the construction of the non - abelian quasiparticles in the case of moore - read state which also come in pairs @xcite . ) in this way , because the meron boundary constructions define the ground states , we may expect that merons form the quasiparticle content of the critical theory . charge and spin are not separated in these constructions that describe the ground states of the critical phase . thus it was necessary to include the gluing " when discussing the ground state from the point of view that treated the charge and spin sector separately . more simply , on the disk with a fixed number of up and down particles we model a meron as @xmath143 where @xmath144 and @xmath145 denote the ground state and the position of the meron respectively . now we see the need for the square root factors , @xmath146 , in the charge part ( a charge boost " ) to relate to the expression in eq.([qp2 ] ) , which represents a neutral combination of two merons . to understand better the competing ( ground ) states present in eds of the qhb on the torus in the intermediate region , we will consider in more detail the odd and even sectors . as in the previous subsection , we will relate the description of states in terms of bose condensates at non - zero momenta with meron - antimeron constructions . for further discussion concerning the validity of the description by these states , see appendix [ slowlying ] . in the odd sector , for the periodic boundary conditions in both directions , i.e. , in the @xmath147 sector , we have simply bose condensation in @xmath148 momentum . the relevant construction in the odd @xmath126 sector , i.e. antiperiodic in the @xmath119 direction and periodic in the @xmath120 direction , is @xmath149_i \left [ \begin{array}{c } 1 \\ 0 \end{array } \right]_n , \ ] ] where we fixed @xmath150 in the ground state . again , we need the gluing procedure with the boost @xmath151 the translation by @xmath152 in the @xmath119-direction followed by inversion lead to the new relevant state @xmath153 taking also the possibility of the construction along the @xmath154axis , we get a 4-fold degeneracy . in the even sector a similar procedure will lead to only the 2-fold degeneracy for the constructions along any one of two directions . similarly we can consider the following constructions in the odd @xmath155 sector , i.e. antiperiodic in both directions , @xmath156_i \left [ \begin{array}{c } 1 \\ 0 \end{array } \right]_n , \\ \prod_{i=1}^{n-1 } \left [ \begin{array}{c } \mathrm{e}^{\mathrm{i}\frac{\pi}{l}(x_i - y_i ) } \\ \mathrm{e}^{-\mathrm{i}\frac{\pi}{l}(x_i - y_i ) } \end{array } \right]_i \left [ \begin{array}{c } 1 \\ 0 \end{array } \right]_n,\end{aligned}\ ] ] and boosts are @xmath157 and @xmath158 , respectively . by applying the translations by @xmath159 and inversion , we may conclude that in the odd sector we have a 4-fold degeneracy with @xmath160 on the other hand , a similar construction in the even sector leads to a unique state with @xmath161 . thus we can conclude that the prediction for the number of competing ground states in the odd sector , based on the critical boson condensates , is in an agreement with eds in section [ sec : numerics ] , in particular figs . [ fig : spectrum1315 ] and [ fig : spectrum13dk ] and eqs . ( [ state1 ] ) and ( [ state2 ] ) , for the description of the @xmath104 states . on the other hand , fig . [ fig : spectrum12 ] shows the low lying excitation spectrum in the even sector with @xmath162 electrons at distance @xmath163 between the layers . we may notice that with respect to our expectations in the previous considerations ( based on meron - antimeron constructions of ground states ) there are extra low - lying states . this will be discussed in more detail in the next subsection . in the previous subsection we could notice that although the composite boson view is convenient for the description of the low - lying states , it is not complete and sufficient for their description . in connection with this , we may comment that the meron construction in eq . ( [ meron ] ) seems more natural if @xmath144 is given by eq . ( [ fpairing ] ) ( fermionic representation ) than by eq . ( [ twodet ] ) ( bosonic representation ) . in the fermionic case the meron , with the deficit of @xmath164 ( layer ) polarization , may be interpreted as a laughlin quasihole excitation of the condensate of @xmath164 particles . in the absence of pairing , represented by the determinant in the composite fermion wave function eq.([fpairing ] ) , we could consider a bosonic system described by halperin s 220 state @xcite with two components , and merons as half - flux quantum , spinful quasiparticles . for that system we know the number of the degenerate ground states on the torus ; we can consider all possible configurations ( including meron ones ) between the two edges of a cylinder @xcite , and come to the conclusion that there are 4 sectors ( 4 degenerate states on a torus ) . they can be classified as members of the chiral algebra , in which we have the basic sector that we can associate with the identity operation , two sectors due to two kinds of meron - ( anti - meron ) excitations , and a sector that is associated with the one - flux - quantum excitation i.e. laughlin quasihole , which can be viewed as a fusion of the two ( @xmath165 and @xmath166 ) different merons . in this way the four sectors can be viewed as a direct product of two subsystems , each at filling @xmath167 and having two sectors , due to the identity and laughlin excitation in each system . symbolically , if we denote the identity by @xmath168 , and @xmath165 meron by @xmath169 , and @xmath166 meron by @xmath170 , we have a direct product @xmath171 , i.e. , 4 sectors . having the extra determinant as in eq.([fpairing ] ) , beside the ( 220 ) factor , will not change the type of the charged excitations that exist in the intermediate phase , i.e. , we expect merons . thus we obtain a total 4-fold degeneracy of the ground states on the torus . but we have to be aware that our system is far from being a simple bosonic ( 220 ) state . there we could consider excitations of a single boson , @xmath165 or @xmath166 , but this does not create an extra sector in the ( 220 ) system because the two subsystems , @xmath165 and @xmath166 , are uncorrelated , and everything boils down to the direct product of @xmath165 and @xmath166 ( sub)system . in the case of the state in eq.([fpairing ] ) , because of the interlayer correlations , we may assume that a single fermion excitation , which we will denote by @xmath172 for @xmath165 subsystem and @xmath173 for @xmath166 subsystem , is a non - trivial excitation , and may define a sector . thus we may consider @xmath174 as a way to generate all sectors @xcite . but we notice that @xmath175 is equivalent to two @xmath176 excitations , and should not represent an extra sector . thus , in this way we expect an 8-fold degeneracy in the even sector , and this is corroborated by the results in fig . [ fig : spectrum12 ] . this is by no means a usual characterization of a topological phase . the topological characterization is made difficult by the absence of a simple model hamiltonian for the model wave function [ eq . ( [ fpairing ] ) or eq . ( [ twodet ] ) ] . due to ( topologically ) highly intertwined ( pseudo)spin and charge , we might expect a further reduction of the number of relevant states . namely , an operation of equivalence may be introduced for two states that are connected by a gapless branch , like in fig . [ fig : spectrum13angle ] . nevertheless , we find that the low - lying spectrum , in both even and odd sector is robust and independent of system size , up to @xmath177 particles in the even sector , and up to @xmath77 in the odd sector , and it is not in the accordance with the naive expectation of 4 degenerate ground states in the even sector that can be found in the previous literature , or as follows from the charge - pseudospin separated picture of the intermediate state . by comparing the tentative degeneracies on the torus ( 8 states in the even sector and 9 states in the odd sector ) with the predictions in the even and odd sector in section [ sec : competing ] on the basis of @xmath112 bcs paired critical bosons , we can conclude that this view is partially successful . we may expect abelian merons _ and _ composite fermions in the quasiparticle spectrum . in this subsection we summarize the reasons why meron deconfinement is expected in the intermediate state in the qhb . by meron deconfinement " we mean the phenomenon that a meron can exist as a single , gapped excitation of the half - flux quantum and charge @xmath178 in the intermediate phase ( as opposed to being in a bound pair ) . the evolution of the energy spectrum in the odd sector on the sphere as a function of bilayer distance . data is for @xmath179 electrons in total ( left to right ) , for fixed shift @xmath180 corresponding to the 111 state . ] when we compare the evolution of the qhb with respect to distance , on the sphere [ fig . [ sphere ] ] and on torus [ figs . [ fig : spectrum1315 ] , [ fig : spectrum13dk ] ] , we see that on the sphere the lowest goldstone - branch state evolves without level crossing with distance . the calculations on the sphere in fig . [ sphere ] are performed at the fixed shift @xmath180 , which is the same as that of the 111 state . note , however , that the decoupled fermi liquids have a shift of @xmath181 , therefore the sphere geometry can not directly capture the transition without introducing some bias towards one of the phases . in fig . [ sphere ] the ground state smoothly changes around @xmath59 due to the change in the correlations , and also there is a drastic change in the calculated slope ( velocity ) of goldstone mode according to ref . , but the first excited state is smoothly connected to the goldstone branch for any @xmath10 . in contrast , on the torus [ see , e.g. , fig . [ fig : spectrum13dk](a ) ] there are level crossings in the low - lying excited spectrum around @xmath59 . new states appear with lower energies than the first excited goldstone - branch state . these states constitute candidate ground states of the intermediate phase . they are defined by merons , as explained previously in sec . [ sec : competing ] . thus merons belong to the quasiparticle content of the new state . the intermediate state has the interpretation in terms of bcs pairing of composite fermions ( ref . ) , and thus similarly to the moore - read case we expect half - flux quantum ( one flux quantum in the language of superconductivity ) excitations . we presented a description of the intermediate region of the qhb in terms of critical bosons based on detailed exact diagonalizations on the torus . our study complements earlier results on the sphere @xcite and torus @xcite geometry , where an intemediate phase was identified at bilayer distances @xmath182 . these studies have pointed out that the phase involves pairing of composite fermions , and has pseudospin spiral order . our work shows that this state can also be related to critical bosons and is likely gapless . the critical condensates at finite wave vectors of the intermediate phase on the torus can be related to the existence of merons as the lowest lying charged quasiparticles , i.e. , to meron deconfinement . the topological ordering in the intermediate region with meron fractionalization allows a smooth transition ( ordering ) in the adjusted phases in which merons are confined in pairs ; as vortex excitations in the @xmath24 ferromagnet for small d , and , for large d , as bound pairs they make neutral fermions . we caution that the critical boson representation employed here may be less powerful than similar descriptions of the more conventional fqh states where charge and pseudospin ( neutral degrees ) are not so intertwined . in our case , the natural representations in terms of ( composite ) bosons ( exciton superfluid ) and ( composite ) fermions ( composite fermi liquids ) hold in the limiting cases of small and large distances . however , we showed that most of the low - energy features of the intermediate state can be conveniently explained in the picture of critical bosons . assuming the intermediate state develops into a critical point in the thermodynamic limit , the qhb system may offer a realistic application of the ideas of deconfined criticality @xcite . moreover , the presented qhb description may also explain the phenomenon of meron loosening " in experiments @xcite . before invoking any disorder effects , we may assume that merons are liberated ( not bound in pairs with opposite layer indexes ) , and equally contribute to inter and intra dissipative processes that lead to the breakdown of the quantum hall effect . as argued in ref . , the quasiparticles with one - half of the elementary charge are natural ( confined in dipoles ) constituents of the underlying composite fermions in the fermi - liquid - like state ( at filling one - half ) , due to the particle - hole symmetry . in our case these are merons ( of the same layer index ) , which are brought to the existence in the intermediate state , before the two separate fermi - liquid - like states ( each at filling one - half ) set in . we can rewrite the neutral part in eq . ( [ twodet ] ) as @xmath183 thus , the neutral part in eq . ( [ twodet ] ) in the cft language is the correlator of equal numbers of two vertex operators : @xmath184 and @xmath185 where @xmath186 and @xmath187 denote the holomorphic and anti - holomorphic part of the ( non - chiral ) bosonic field , @xmath188 . the basic correlator is : @xmath189 we expect that the excitations are also represented by vertex operators . for their construction we have two possibilities : @xmath190 and @xmath191 and the corresponding conjugates , which in charge - neutral combinations in the correlators with basic particles ( @xmath184 and @xmath185 ) i.e. up and down ( neutral ) excitons , contribute factors , @xmath192 and , with the field , @xmath193 , we have @xmath194 to ensure the single - valuedness of the ensuing electronic wave function we must have @xmath195 . the same argument would allow any rational @xmath196 including @xmath197 . therefore the excitation spectrum contains a branch continuously connected with the ground state and we expect that the state in ( [ twodet ] ) supports a branch of gapless excitations . ( if we allow only @xmath198 we would have the quasiparticle content of the deconfined phase of @xmath199 gauge theory . ) thus the state ( [ twodet ] ) is a critical state with the quasi - long - range order , @xmath200 among up and down excitons . in this appendix we explain the projection to the lll @xcite that defines the lll projected wave function of @xmath0-wave paired composite fermions in eq.([pfpairing ] ) . first we expand and rewrite eq.([pfpairing ] ) as @xmath201 where by @xmath202 we denoted the part with the laughlin - jastrow correlations , @xmath203 to understand the projection , we consider an arbitrary permutation @xmath164 under the sum in eq . ( [ pexpand ] ) . for the fixed permutation , we can express the product @xmath202 as a polynomial in @xmath204 variables . in this way , when we combine @xmath202 with the pairing part with fixed @xmath164 , the projection reduces to the projection of each @xmath205 ( with @xmath206 ) , to the lll . if we choose the symmetric gauge , the basis of the lll is given by @xmath207 , with @xmath208 . here we dropped the gaussian factor , @xmath209 , which is included in the measure defining the scalar product @xcite . thus , @xmath210 it follows that @xmath211 in particular , for @xmath212 , @xmath213 , i.e. , the leading term in the short distance approximation reproduces the behavior of the @xmath0-wave paired composite fermion wave function described by eq.([sdispairing ] ) . the first exact diagonalization study on the torus that pointed out to the existence of the competing ground states with diverging momenta in the thermodynamic limit was done by park in ref . . our numerical results are in agreement with ref . , in particular for @xmath66 electrons and bilayer distance @xmath70 [ fig . [ fig : spectrum1315](a ) ] . ref . pointed out that one energy level ( the lowest excitation at non - zero momentum , which represents four states due to inversion symmetry ) is ripped out " from the quasi - continuum of high momentum states . the momenta that correspond to these states are the ones in eq . ( [ state1 ] ) in the main text , or @xmath214 in units of @xmath87 where @xmath88 . in order to motivate and explain the emergence of the particular momenta above , park discussed the spiraling states " in the landau gauge , @xmath215 here @xmath216 denotes the electron creation operator , with integer @xmath217 labeling the linear momentum along the cylinder , i.e. , @xmath218 . with respect to the ground state of the @xmath24 ferromagnet [ @xmath219 in eq . ( [ pspiral ] ) , or eq . ( [ ferro ] ) if we consider the fixed number of particles representation ] , the acquired momentum is @xmath220 . in the first quantization , in terms of the usual landau gauge states @xmath221 where @xmath222 , @xmath223 , we can express the spiraling state as @xmath224 where we introduced @xmath225 . thus the physics of the neutral sector is essentially given by the factor @xmath226 which describes long spiraling with period @xmath139 in the pseudospin space . park argued that by more elaborate inclusions of the disordering effects in the ground state , similar constructions to the ones in ( [ pspiral ] ) should be relevant for the description of the new ground state(s ) . they will correspond to the new intermediate phase with the pseudospin spiral order " . in this work we point out that there are extra states , as seen in fig . [ fig : spectrum1315 ] falling down , which also belong to the set of candidate ground states of the intermediate phase . given the unusually large momenta ( and long periods ) for spiral order , we argue in sec . [ sec : boson ] that they are very likely an expression of the _ topological _ ordering in the intermediate phase . furthermore , we connect them with merons the spin - texture fractional excitations of the @xmath24 ordered phase which are confined ( bound in pairs ) . the long spiraling in the ground states of the intermediate phase on the torus is thus naturally explained by the spin - texture description of merons . to find and understand the description of the low - lying modes it is convenient to begin with the case @xmath26 ( @xmath25 symmetric case ) in the presence of only @xmath227 pseudo potential for which we can identify the whole family of zero modes ( zero energy states ) . as shown in ref . the basic states ( from which we can get all other zero - energy states ) similarly to edge states in the laughlin case , can be described in the disc geometry as @xmath228 where the sum is over various @xmath229tuples of @xmath165 particles . with the inclusion of higher angular momentum pseudo potentials they will not be degenerate anymore , but will constitute the candidate states for the states of the lowest energy at a given momentum @xmath139 ( with respect to the ground state momentum ) . for example , the state at @xmath230 , @xmath231 , can be associated with the first excited state above ground state in the goldstone branch . this was demonstrated in ref . when a boost of angular momentum @xmath139 acting only on one pseudo - spin projection ( layer index ) in the ground state was identified with the goldstone mode with momentum @xmath232 this ansatz was very satisfactory for the whole range of distances ( from zero to infinity ) in the bilayer for low momenta ( @xmath233 ) . on the other hand the state in eq . ( [ basicstates ] ) when @xmath234 @xmath235 is a meron-(anti - meron at the infinity ) state . this state is a projection of the meron state introduced in ref . @xmath236 where @xmath237 are electron creation operators in the second - quantized notation , to the fixed relative number , @xmath238 of electrons . this state was also used by park @xcite to model the state(s ) that fall(s ) from the higher - momentum region when the distance ( @xmath10 ) increases . the best of overlap in ref . was reached around @xmath15 after that the overlap was rapidly decreasing . as the author already pointed out , the reason for this is that the used construction is based on the 111 state , i.e. the state in eq . [ ferro ] projected to the fixed number of electrons , and we need to include the changes in the ground state that set in around @xmath239 thus we may imitate the approach of ref . and model new state and correlations by @xmath240 where @xmath241 represents the exact ground state at distance @xmath10 that we find in @xmath242 sector . nevertheless , the quantum numbers of relevant states , eq . [ ansatz ] , are the same as from the states they evolve from , eq . [ meronpair ] , and we can use these states i.e. their description given by eq . [ description ] to classify all possible low lying states on the torus . our analysis suggests that the bilayer intermediate state appears as a critical region between a ferromagnetically ordered state and a ( completely ) disordered state . it is tempting to expect that by modifying the parameters in the critical region we can reach the @xmath243 phase ( the deconfined phase of @xmath199 gauge theory , mentioned in sec . [ sec : deconf ] ) . this is one of the simplest topological phases that would be realized by strong pairing of composite bosons . also , it is important to understand better the influence of the various haldane pseudopotentials on the competing ground states in the critical region . in order to explore the phase diagram more broadly , we consider modifications of the bare coulomb potential by the short - range haldane pseudopotential terms @xcite : @xmath244 where @xmath245 . the purpose of these modifications is to model the perturbations that occur in an actual experimental system ( for example , effects due to finite width of the sample in the perpendicular direction , mixing with higher landau levels , etc . ) . note that we are free to separately tune the intra- and inter - layer interactions . adding even @xmath169 pseudopotentials will only have an effect on inter - layer potential because the fermi statistics inside each layer precludes even values of @xmath169 . in fig . [ fig:11pseudo ] we can see the evolution of the spectrum with small departures @xmath246 of the haldane pseudopotentials from their coulomb values . the influence of the even pseudopotentials , @xmath247 and @xmath248 is qualitatively the same ; for stronger positive values the ferromagnetically ordered state is stabilized , and for negative values we expect a compressible behavior . the influence of @xmath249 is according to our expectations ; for stronger positive @xmath249 we see compressible behavior , while for negative @xmath249 the ordered state is stabilized . on the other hand the influence of positive @xmath250 and @xmath251 is interesting as it stabilizes one of the set of states ( competing ground states ) that have characteristic momenta that are compatible with @xmath243 . nevertheless , a closer inspection of the influence of @xmath250 does not lead to a firm ( topological ) characterization of the ground states for positive values of @xmath250 . moreover , the behavior of the gap in this region is strongly dependent on system size , which further suggests the phase is compressible . we acknowledge useful discussions with kwon park . this work made use of the facilities of n8 hpc provided and funded by the n8 consortium and epsrc ( grant no.ep/k000225/1 ) . the work was supported by the ministry of education , science , and technological development of the republic of serbia under projects on171017 and on171027 . g. mller , s. h. simon , and e. h. rezayi , phys . * 101 * , 176803 ( 2008 ) ; g. mller , s. h. simon , and e. h. rezayi , phys . rev . b * 79 * , 125106 ( 2009 ) . k. park , phys . b * 69 * , 045319 ( 2004 ) . r. e. prange and s. m. girvin , _ the quantum hall effect _ , ( springer - verlag , new york , 1990 ) . fertig , phys . b * 40 * , 1087 ( 1989 ) k. moon , h. mori , k. yang , s.m . girvin , a.h . macdonald , l. zheng , d. yoshioka , and s .- c . zhang , phys . b * 51 * , 5138 ( 1995 ) . spielman , j.p . eisenstein , l.n . pfeiffer , and k.w . west , phys . . lett . * 84 * , 5808 ( 2000 ) ; * 87 * , 036803 ( 2001 ) . j. eisenstein , arxiv:1306.0584 . halperin , p.a . lee , and n. read , phys . b * 47 * , 7512 ( 1993 ) ; e. rezayi and n. read , phys . . lett . * 72 * , 900 ( 1994 ) . simon , e.h . rezayi , and m.v . milovanovic , phys . lett . * 91 * , 046803 ( 2003 ) . g. mller and s. h. simon , phys . b. * 77 * , 075319 ( 2008 ) . n. read and d. green , phys . b * 61 * , 10267 ( 2000 ) m. v. milovanovi and z. papi , phys . rev . b * 79 * , 115319 ( 2009 ) . m. freedman , c. nayak , k. shtengel , k. walker , and z. wang , ann . phys . * 310 * , 428 ( 2004 ) . halperin , helv . acta * 56 * , 75 ( 1983 ) . a possible reason for the presence of composite fermions can be seen in a formal way by also noting that , by considering charge and pseudospin sector separately , we have to allow quasiparticle constructions which are not single - valued functions of electron coordinates ( assuming that the gluing of charge and pseudospin will heal this ) . thus in the charge sector we can have a laughlin quasihole , @xmath252 , but also @xmath253 , and in the pseudospin sector , half " , @xmath254 \psi_{gs}$ ] , and thus also @xmath255 \psi_{gs}$ ] ( and other fused quasiparticles , see appendix [ sec : appa ] ) . if we allow for all ( single - valued ) combinations of these basic quasiparticle constructions we may expect @xmath256 ( composite fermion ) , etc . g. mller and s.h . simon , advances in condesed matter physics * 2011 * , 815169 ( 2011 ) . g. moore and n. read , nucl . b * 360 * , 362 ( 1991 ) . d. green , thesis , yale university , arxiv : cond - mat/0202455 . f. d. m. haldane , phys . rev * 55 * , 2095 ( 1985 ) . b. a. bernevig and n. regnault , phys . b * 85 * , 075128 ( 2012 ) . n. shibata and d. yoshioka , j. phys . . jpn . * 75 * , 043712 ( 2006 ) . a. a. burkov and a.h . macdonald , phys . b * 66 * , 115320 ( 2002 ) . c. repellin , t. neupert , b.a . bernevig , and n. regnault , arxiv:1504.00367 m. milovanovi and n. read , phys . b * 53 * , 13559 ( 1996 ) . t. senthil , a. vishwanath , l. balents , s. sachdev , and m.p.a . fisher , science * 303 * , 1490 ( 2004 ) . d. zhang , x. huang , w. dietsche , m. hauser , and k. von klitzing , arxiv:1406.0326 c. wang and t. senthil , arxiv:1507.08290 a.h . macdonald , h.a . fertig , and l. brey , phys . 76 * , 2153 ( 1996 ) . s. m. girvin and terrence jach , phys . rev . b * 29 * , 5617 ( 1984 ) .

let be a kac random polynomial where the coefficients are iid copies of a given random variable . our main result is an optimal quantitative bound concerning real roots repulsion . this leads to an optimal bound on the probability that there is a double root . as an application , we consider the problem of estimating the number of real roots of , which has a long history and in particular was the main subject of a celebrated series of papers by littlewood and offord from the 1940s . we show , for a large and natural family of atom variables , that the expected number of real roots of is exactly , where is an absolute constant depending on the atom variable . prior to this paper , such a result was known only for the case when is gaussian .

1409.4128

let @xmath2 be a real random variable having no atom at 0 , zero mean and unit variance . our object of study is the random polynomial @xmath7 where @xmath1 are iid copies of @xmath2 . this polynomial is often referred to as kac s polynomial , and has been extensively investigated in the literature . the study of real roots of random polynomials has a long history . let @xmath8 be the number of real roots of @xmath4 , sometimes we use the notation @xmath9 to emphasize the dependence of @xmath8 on the distribution of @xmath2 . this is a random variable taking values in @xmath10 . the issue of estimating @xmath8 was already raised by waring as far back as 1782 ( @xcite , @xcite ) , and has generated a large amount of literature , of which we give an ( incomplete and brief ) survey . one of the first estimates for @xmath8 was obtained by bloch and plya @xcite , who studied the case when @xmath2 is uniformly distributed in @xmath11 , and established the upper bound @xmath12 here and later we ll use the usual asymptotic notation @xmath13 or @xmath14 to denote the bound @xmath15 where @xmath6 is independent of @xmath16 . the above bound of bloch and plya was not sharp , as it turned out later that @xmath3 has a remarkably small number of real roots . in a series of breakthrough papers @xcite in the early 1940s , littlewood and offord proved ( for many atom variables @xmath2 such as gaussian , bernoulli or uniform on @xmath17 $ ] ) that @xmath18 with probability @xmath19 , where we use @xmath20 to denote a quantity that goes to @xmath21 as @xmath22 . around the same time , kac @xcite developed a general formula for the expectation of number of real roots @xmath23 where @xmath24 is the probability density for @xmath25 ad @xmath26 . in the gaussian case , one can easily evaluate the rhs and get @xmath27 for non - gaussian distributions , however , kac s formula is often very hard to evaluate and it took a considerable amount of work to extend to other distributions . in a subsequent paper @xcite , kac himself handled the case when @xmath2 is uniformly distributed on the interval @xmath17 $ ] and stevens @xcite extended it further to cover a large class of @xmath2 having continuous and smooth distributions with certain regularity properties ( see @xcite for details ) . these papers relies on and the analytic properties of the distribution of @xmath2 . for discrete distributions , does not appear useful and it took more than 10 years since kac s paper until erds and offord in 1956 @xcite found a completely new approach to handle the bernoulli case . for this case , they proved that with probability @xmath28 @xmath29 in the late 1960s and early 1970s , ibragimov and maslova @xcite successfully refined erds - offord s method to handle any variable @xmath2 with mean 0 . they proved that for any @xmath2 with mean zero which belong to the domain of attraction of the normal law , @xmath30 for related results , see also @xcite . few years later , maslova @xcite showed that if @xmath2 has mean zero and variance one and @xmath31 , then the variance of @xmath32 is @xmath33 . other developments were made in the late 1980s by wilkins @xcite and in the early 1990s by edelman and kostlan @xcite , who evaluated the explicit integral in very carefully and provided a precise estimate for @xmath34 @xmath35 where @xmath36 is an explicit constant ( the value of an explicit , but complicated integral ) . as a matter of fact , one can even write @xmath20 as sum of explicit functions of @xmath37 , which gives a complete taylor expansion . the truly remarkable fact about is that the error term @xmath38 tends to a limit as @xmath37 tends to infinity . the question here is : is this a _ universal phenomenon _ , which holds for general random polynomials , or a special property of the gaussian one ? it is clear that the computation leading to is not applicable for general random polynomials , as the explicit formula in is available only in the gaussian case , thanks to the unitary invariance property of this particular distribution . for many natural variables , such as bernoulli , there is little hope that such an explicit formula actually exists . as a matter of fact , ibragimov - maslova s proof of their asymptotic result for general non - gaussian polynomials ( based on the earlier work of erds - offord ) is a tour - de - force . among others , they followed the erds - offord s idea of using the number of sign changes on a fixed sequence of points to approximate the number of roots . the error term in this approximation , by nature , has a be large ( at least a positive power of @xmath39 ) . on the other hand , numerical evidence tends to support the conjecture that @xmath40 do go to a limit , as @xmath37 to tends to infinity . however , the situation is delicate as this limit seems to depend on the distribution of the atom variable @xmath2 and _ is not _ universal ; see the numerical illustration below . in all three cases . ] from the averages , and the curves seem to converge to different values . ] in a recent work @xcite , the last two authors and oanh nguyen made a first step by showing that the error term in question is bounded . let @xmath2 be a random variable with mean 0 and variance 1 and bounded @xmath41-moment . then @xmath42 . in this paper , we settle this problem in the affirmative for a large and natural class of distributions , as an application of a general theorem concerning the repulsion between real roots of kac s polynomials ( see the next section ) . [ d.uniform-n ] for any positive integer @xmath43 , we say that @xmath2 has uniform distribution with parameter @xmath43 ( or _ type i _ ) if @xmath44 independently , @xmath45 . in particularly , bernoulli random variable has uniform distribution with parameter 1 . [ d.p ] let @xmath46 and @xmath47 . we say that a random variable @xmath2 of mean zero has _ type ii _ distribution with parameter @xmath48 if its has a @xmath49-integrable density function and its @xmath50-moment is bounded . [ theorem : expectedvalue:2 ] let @xmath2 be a random variable with either type i or type ii with fixed parameters . then @xmath51 where @xmath6 is an absolute constant depending on @xmath2 . we would like to mention that due to the abstract nature of our arguments , the exact value of @xmath6 is still unknown , and its determination remains a tantalizing question . in our favorite toy case when @xmath2 is bernoulli , computer simulation suggests that @xmath6 is around @xmath52 , but it looks already difficult to prove that @xmath53 ( say ) . now we are going to present the main technical contribution of our paper , which , together with some other tools , will yield theorem [ theorem : expectedvalue:2 ] as an application . the object of study here is the existence of double roots , or more generally , the repulsion between real roots of kac polynomials . multiple roots is a basic object in theoretical analysis . they also play an important role in practical computational problems . for example , it is a fundamental fact in numerical analysis that the running time of newton s method ( for finding real roots of a function ) decreases exponentially with the presence of a multiple or near multiple root ( i.e. , a place @xmath54 where both @xmath55 and @xmath56 are close to zero ) , see for instance ( * ? ? ? * chapter 8) . intuitively , one expects that a random polynomial does not have multiple or near multiple roots , with high probability . motivated by problems in complexity theory , researchers have confirmed this intuition for the case when @xmath2 is gaussian , in a more general setting ( see ( * ? ? ? * chapter 13 ) and @xcite ) . unfortunately , the methods in these works rely on the invariance property of the gaussian distribution and do not extend to other natural distributions . we are going to introduce a new approach that enables us to fully understand the double root phenomenon , and consequently derive an optimal bound on the probability that a random polynomial has double or near double real roots . for the sake of presentation , let us first consider the toy case when @xmath2 is bernoulli . one expects that the probability of having double roots tends to 0 with @xmath37 ; the question is how fast ? to give a lower bound , let us consider @xmath57 , the probability that the polynomial has a double root at either @xmath58 or @xmath59 ( we stress that @xmath57 denotes one number ) . trivially , the double root probability is at least @xmath57 . a short consideration shows ( see appendix [ appendix:1 ] ) [ fact : arithmetic ] assume that @xmath60 s have the bernoulli distribution . then @xmath61 if @xmath62 and @xmath21 otherwise . we are now ready to state our first theorem , which asserts that in the @xmath62 case , the double root probability is dominating by @xmath57 , while in the other cases , this probability is very small . [ double roots of bernoully polynomials ] [ theorem : dbroot : ber ] let @xmath4 be a bernoulli polynomial . then @xmath63 theorem [ theorem : dbroot : ber ] is a special case of our main result below , which deals with near double roots of general random polynomials . [ main result : roots repulsion ] [ theorem : repulsion : uniform ] assume that @xmath64 are independent and all of them are either type i with the same fixed parameter @xmath43 , or type ii with uniform implicit constants ( which are fixed ) . then for any constant @xmath65 there exists a constant @xmath66 such that for all sufficiently large @xmath37 @xmath67 where @xmath57 denotes the probability that the polynomial has a double root at either @xmath58 or @xmath59 . it is clear that in the type ii setting we have @xmath68 . also , it is not hard to show that @xmath69 for the type @xmath70 setting ( see claim [ claim : anticoncentration ] . ) our proof will provide more detailed information about the location of double and near double roots . for instance , in the discrete case , we will show that with overwhelming probability , any double ( or near double ) root , has to be very close to either 1 or @xmath59 . we present the precise statements in section [ section : prelim ] and with proofs in sections [ section : repulsion : general]-[section : repulsion : edge ] . as an application , we verify theorem [ theorem : expectedvalue:2 ] in section [ section : expectedvalue:2 ] . a few technical statements will be proved in the appendices in order to maintain the flow of the presentation . parallel to this paper , peled et . @xcite proved for random polynomials with the atom variable @xmath2 having support on @xmath71 that the probability of having double roots ( both real and complex ) is dominated by the probability of having double roots at @xmath72 . the method used in @xcite is specialized for the support consisting of @xmath71 and is totally different from the method we used in this paper . the harder case in our study is the discrete one ( @xmath2 is of type i ) . in what follows we first discuss our approach for this case . note that the real roots of @xmath4 have absolute value bounded from above by @xmath73 and below by @xmath74 . it follows that we only need to consider @xmath75 . since @xmath76 and @xmath77 have the same distribution as @xmath4 , it suffices to consider @xmath78 . let @xmath79 be a constant and set @xmath80 , \mbox { and } i_1:= ( 1-n^{-2+{\varepsilon } } , 1].\ ] ] theorem [ theorem : repulsion : uniform ] is a consequence of the following two results , where @xmath2 is assumed to have type @xmath70 . [ theorem : prelim:1 ] let @xmath65 be any constant . then there exists @xmath81 depending on @xmath6 and @xmath82 such that the following holds @xmath83 [ theorem : close : ber ] for sufficiently small @xmath82 , there exists @xmath81 such that @xmath84 in fact , it follows from the proof in section [ section : repulsion : edge ] that one can take @xmath85 and @xmath86 . to prove theorem [ theorem : expectedvalue:2 ] , we will need the following stronger version of theorem [ theorem : prelim:1 ] . [ theorem : repulsion : general ] assume that @xmath2 has uniform discrete distribution with parameter @xmath43 , where @xmath43 is fixed . let @xmath65 be any constant . then there exists @xmath81 depending on @xmath87 and @xmath82 such that the following holds with probability at least @xmath88 . 1 . ( near double roots ) there does not exist @xmath89 such that @xmath90 2 . ( repulsion ) there do not exist @xmath91 with @xmath92 such that @xmath93 3 . ( delocalization ) for any given @xmath94 , there is no @xmath54 with @xmath95 and @xmath96 . we remark that theorem [ theorem : repulsion : general ] might continue to hold for an interval larger than @xmath97 , but we do not try to pursue this matter here . we next turn to the continuous case . for any interval @xmath98 , using hlder s inequality we have @xmath99 since @xmath47 , it follows that @xmath100 . additionally , as @xmath2 has bounded @xmath50-moment , we have @xmath101 . therefore with a loss of at most @xmath102 in probability one can assume that @xmath103 where @xmath104 is a finite constant depending on @xmath49 and @xmath6 . conditioning on this , it can be shown easily that if @xmath105 then @xmath106 and if @xmath107 then @xmath108 . it follows that theorem [ theorem : repulsion : uniform ] follows from the following analogue of theorem [ theorem : repulsion : general ] . we remark that in this theorem , we allow a more general setting where the coefficients @xmath1 are not necessarily iid , which is convenient in the proof . [ theorem : repulsion : general ] assume that @xmath109 have type ii distributions with uniform implicit constants . consider @xmath110 . let @xmath111 be any constant . then there exists @xmath81 depending on @xmath112 and @xmath49 such that ( i),(ii),(iii ) of theorem [ theorem : repulsion : general ] hold with probability at least @xmath88 with @xmath97 replaced by @xmath113 $ ] . in other words , 1 . there does not exist @xmath114 such that @xmath115 2 . there do not exist @xmath116 with @xmath92 such that @xmath117 3 . for any given @xmath118 , there is no @xmath54 with @xmath95 and @xmath96 . in the next section , we discuss the strategy to prove theorems [ theorem : repulsion : general ] and [ theorem : repulsion : general ] . in this section , we first explain our strategy to prove ( the harder ) theorem [ theorem : repulsion : general ] and then deduce theorem [ theorem : repulsion : general ] from this approach . the rest of the proof of theorem [ theorem : repulsion : general ] follows in the next section . our general strategy is to reduce the event of having double ( or near double ) roots to the event that a certain random variable takes value in a small interval . the key step is to bound the probability of the latter , and here our main tool will be a recently developed machinery , the so - called inverse littlewood - offord theory ( see @xcite for an introduction ) . divide the interval @xmath97 into subintervals of length @xmath119 each ( except for possibly the right - most interval which has the same right endpoint as @xmath97 ) , where @xmath119 to be chosen sufficiently small ( polynomially in @xmath37 ) , and @xmath120 is then chosen large enough so that @xmath121 . * near double roots . * assume that there exists a subinterval @xmath70 and an element @xmath122 such that @xmath95 and @xmath123 , then for @xmath124 , the center of @xmath70 , we have @xmath125 for some @xmath126 . in the following , the implicit constants in @xmath127 may depend on @xmath43 unless otherwise specified . on the other hand , as @xmath128 for some @xmath129 . from here , by the trivial upper bound @xmath130 for the second derivative , we have @xmath131 for some @xmath91 with @xmath92 . then there is a point @xmath132 between @xmath54 and @xmath133 such that @xmath134 . thus , for any @xmath135 with@xmath136 , @xmath137 there is a point @xmath138 of some subinterval @xmath70 such that @xmath139 . for this @xmath138 , @xmath140 for some @xmath135 between @xmath54 and @xmath138 . because @xmath54 has distance at most @xmath141 from @xmath133 , @xmath54 also has distance at most @xmath119 from @xmath132 , and so @xmath135 has distance at most @xmath142 from @xmath132 . it follows that @xmath143 [ remark : continuous ] one can also show that the * repulsion * property is a direct consequence of the * near double roots * property by choosing @xmath120 slightly larger if needed . indeed , suppose that @xmath144 for some @xmath91 with @xmath92 , then consider the @xmath132 obtained as above . thus @xmath145 , and by using the trivial bound @xmath146 for the derivative , @xmath147 * delocalization . * assume that @xmath148 and @xmath149 , then @xmath150 for some @xmath132 between @xmath151 and @xmath54 . on the other hand , @xmath152 for any @xmath153 $ ] , it follows that @xmath154 , we will show that the probability that , , hold for any fixed point @xmath54 of @xmath97 is @xmath155 . this definitely takes care of and hence ( iii ) of theorem [ theorem : repulsion : general ] . since there are @xmath156 subintervals @xmath70 , by the union bound we also obtain ( i ) and ( ii ) of theorem [ theorem : repulsion : general ] . in fact we will show the following stronger estimate [ l.almost-zero ] assume that @xmath2 has type i. then there is a constant @xmath157 which depends only on @xmath43 and @xmath82 such that for every @xmath158 sufficiently large the following holds for every @xmath159 and @xmath160 @xmath161 here the implicit constant may depend on @xmath43 and @xmath162 and @xmath104 and @xmath163 . ( the fact that we have an extra factor of @xmath164 or @xmath165 in , , is not an issue here , since these powers could be included as part of the @xmath119 of lemma [ l.almost-zero ] . ) note that by making @xmath162 slightly smaller it suffices to prove the lemma for @xmath166 , i.e. @xmath167 , which we will assume in the following . we will justify this key lemma in the next section . in the rest of this section , we apply our argument to handle distributions of type ii . by following the same argument , and by , it is enough to show the following analog of lemma [ l.almost-zero ] for type ii variables . recall that we are working under the assumption that @xmath64 are uniformly type ii and independent , but they are not required to be identically distributed . [ f.almost-zero ] there is a constant @xmath157 which depends only on @xmath49 such that for every @xmath158 sufficiently large the following holds for @xmath168 @xmath169 thanks to the analytic properties of type ii variables , this statement is much easier to prove than lemma [ l.almost-zero ] ; the details follow . for any @xmath98 , by hlder s inequality @xmath170 thus , by conditioning on @xmath171 , for any @xmath54 we have @xmath172 the desired conclusion follows immediately if @xmath173 . to handle the general case , let @xmath174 denote the density of the distribution of @xmath60 , which is @xmath49-integrable for @xmath47 by the given assumption . since @xmath175 , it follows immediately via convexity that @xmath174 is also @xmath176-integrable for every @xmath177 $ ] and furthermore @xmath178 thanks to the fact that @xmath60 s are uniformly type ii . for convenience , let @xmath179 denote the right hand side in the estimate above . now , let @xmath180 be a large integer that depends only on @xmath49 such that @xmath181 . by young s convolution inequality and an induction over @xmath180 , it is clear that for any family of functions @xmath182 we have @xmath183 consider the random variable @xmath184 . since @xmath60 s are independent , the density of @xmath185 ( which we will denote by @xmath186 ) equals to the convolution of the density of @xmath187 , @xmath188 , , @xmath189 . let @xmath190 denote the density of @xmath191 , clearly @xmath192 , and @xmath193 consequently , @xmath194 therefore for every @xmath195 we have @xmath196 where @xmath163 is a finite constant depending only on @xmath49 and @xmath104 . now , for every @xmath197 ( recall that @xmath180 is a constant ) and @xmath195 we have @xmath198 thus by choosing @xmath199 sufficiently large we obtain the desired estimate ( with @xmath167 and any @xmath200 ) . by making @xmath162 smaller if necessary , it suffices to prove the lemma for @xmath167 where @xmath199 is sufficiently large . also , as indicated before , @xmath120 will be chosen such that @xmath121 ( for instance @xmath201 ) . fix @xmath202 , all the implicit constants below are independent of @xmath54 . we divide @xmath97 into @xmath203 \cup ( 1-\log^2 n / n,1-n^{-2+{\varepsilon}}]$ ] and prove the lemma for @xmath54 inside each interval separately . for the first interval @xmath203 $ ] , we will present a proof for the bernoulli case ( i.e. @xmath204 ) first to demonstrate the main ideas , and then modify the method for uniform distributions later . our treatment for @xmath205 $ ] works for both settings . roughly speaking , the proof exploits the lacunary property of the sequence @xmath207 in this case . let @xmath208 be such that @xmath209 as @xmath210 , we must have @xmath211 note that if @xmath212 then @xmath213 . as the treatment for this case is a bit more complicated , we postpone it for the moment . in the sequel we assume that @xmath214 , and thus @xmath215 . * treatment for @xmath216 . * let @xmath180 be the largest integer such that @xmath217 in other words , @xmath218 where @xmath219 denote the largest integer that does not exceed @xmath54 . as @xmath220 and @xmath37 is sufficiently large , it follows that @xmath221 is strictly less than @xmath37 and @xmath180 is at least @xmath222 , one has the following trivial bound @xmath223 we say that a finite set @xmath224 of real numbers is @xmath225-separated if the distance between any two elements of @xmath224 is at least @xmath225 . [ claim : separation:1 ] the set of all possible values of @xmath226 is @xmath227-separated . ( of claim [ claim : separation:1 ] ) take any two elements of the set . their distance has the form @xmath228 for some @xmath229 . as @xmath230 , this distance is more than @xmath227 . using claim [ claim : separation:1 ] , we have @xmath231 by conditioning on other coefficients @xmath232 s i.e. @xmath233 , it follows that @xmath234 recall that @xmath235 . using the fact that @xmath236 and @xmath237 , we obtain @xmath238 . it follows that @xmath239 therefore @xmath240 as desired . this completes the treatement of the case @xmath216 . * treatment for @xmath241 . * let @xmath242 be a small positive constant . we show that the treatment above also carries over for this range of @xmath54 with a minor modification . as @xmath243 and @xmath235 , we must have @xmath213 for all @xmath54 in this range . recall that the integer @xmath180 was chosen so that @xmath244 by following claim [ claim : separation:1 ] , we again arrive at and . now , as @xmath245 , we have @xmath246 for some small positive constant @xmath247 depending on @xmath242 . as such , using the fact that @xmath180 has order @xmath39 , we have @xmath248 for @xmath37 sufficiently large . it follows that @xmath249 we obtain @xmath250 as desired . * treatment for @xmath251 . * recall that in this case @xmath213 . for this range of @xmath54 we introduce the following improvement of claim [ claim : separation:1 ] . [ claim : separation:2 ] assume that @xmath252 and @xmath242 is sufficiently small . then the set of all possible values of @xmath253 , is @xmath254-separated . ( of claim [ claim : separation:2 ] ) the distance between any two terms is at least @xmath255 > x^{2k}/8,\ ] ] where we used the fact that the factor within the bracket is at least @xmath256 , provided that @xmath242 is chosen sufficiently small . using claim [ claim : separation:2 ] , we obtain the following slight improvement of @xmath257 now , as @xmath180 is order @xmath39 , by taking @xmath37 large we have @xmath258 which implies the desired conclusion . the @xmath204 case was treated before , so we only consider @xmath259 . as before , let @xmath260 be integer such that @xmath261 since @xmath262 , it follows that @xmath263 , and @xmath264 let @xmath180 be the largest integer such that @xmath265 we first show that @xmath266 while @xmath267 . in deed , by definition we have @xmath268 since @xmath269 for every @xmath270 , it follows that @xmath271 therefore @xmath272 since @xmath273 , it follows that @xmath274 for @xmath37 sufficiently large . furthermore , it follows from that @xmath275 therefore @xmath276 hence @xmath277 for some @xmath278 depending only on @xmath43 . consider the sequence @xmath279 where @xmath280 . we ll show the following separation property : [ cl.separation ] the set of all possible values of @xmath279 , where @xmath280 , is @xmath281-separated . take any two terms of the sequence . consider their difference , which has the form @xmath282 for some @xmath229 , and @xmath283 and @xmath284 . as @xmath285 , this difference is more than @xmath281 . it follows that for every @xmath286 @xmath287 now , using the independence of @xmath64 and by conditioning on @xmath60 s with @xmath288 , we obtain @xmath289 thus , by the choice of @xmath180 , we obtain the key bound @xmath290 next , consider two cases : * case 1 : @xmath291 . * since @xmath292 , it follows that @xmath236 . thus , @xmath293 let @xmath294 . as @xmath295 , we obtain @xmath296 since @xmath180 is controlled below by some @xmath297 , we could make @xmath298 arbitrarily close to @xmath58 by taking @xmath37 large ( independent of @xmath299 ) . thus it suffices to show that @xmath300 but it is clear that this holds for every @xmath301 . indeed , consider the function defined on @xmath302 @xmath303 so for @xmath304 we have @xmath305 . * case 2 : @xmath306 . * it follows that @xmath213 . also , @xmath307 let @xmath308 , it follows from that @xmath309 by choice of @xmath180 , @xmath310 , therefore @xmath311 as before , by choosing @xmath37 large ( independent of @xmath54 ) we could ensure that @xmath298 is arbitrarily close to @xmath58 . therefore it suffices to show that @xmath312 which is clear for @xmath259 . our treatment of this interval is more difficult as here the terms @xmath314 are comparable . our main tool is a following theorem , which is an example of the recently developed machinery of inverse littlewood - offord theorems ( see @xcite for an introduction ) . [ theorem : inverse ] fix positive integers @xmath315 and @xmath316 , and assume that @xmath317 for some real numbers @xmath318 , where @xmath1 are iid random variables of uniform distribution with fixed parameter @xmath43 . then for any number @xmath319 between @xmath320 and @xmath37 , with @xmath37 sufficiently large , there exists a proper symmetric generalized arithmetic progression @xmath321 , that is @xmath322 , such that a. @xmath321 is @xmath6-proper , i.e. the elements of the set @xmath323 are all distinct . .1 in b. @xmath321 has small rank , @xmath324 , and small cardinality @xmath325 + where the implied constants here depend on @xmath326 and @xmath327 , and @xmath328 . + .1 in c. for all but at most @xmath319 elements @xmath329 of @xmath330 , there exists @xmath331 such that + @xmath332 + where @xmath333 independent of @xmath6 . + .1 in d. the number @xmath334 , i.e. , there exist @xmath335 , @xmath336 such that + @xmath337 + .1 in e. all steps @xmath338 of @xmath321 are _ integral multiples _ of @xmath339 . we will provide the deduction of theorem [ theorem : inverse ] from ( * ? ? ? * theorem 2.9 ) in appendix [ appendix : inverse ] . it follows from the @xmath6-properness ( i ) of @xmath321 that for any @xmath340 , the equation @xmath341 has a unique solution @xmath342 . now fix @xmath343 . choose the largest @xmath344 so that @xmath345 , thus @xmath346 in the sequel , set @xmath347 and @xmath348 where @xmath299 to be chosen sufficiently small depending on @xmath199 . we will prove the following crucial bound . [ lemma : crucial:1/2 ] we have @xmath349 without loss of generality assume @xmath199 is an integer . assume otherwise that @xmath350 for some sufficiently large constant @xmath104 to be chosen depending on all other parameters . then by theorem [ theorem : inverse ] , all but @xmath320 of the elements are @xmath351-close to a proper gap @xmath321 of rank @xmath352 and size @xmath353 . furthermore , as noticed , the generators of @xmath321 can be chosen to be integral multiples of @xmath351 . let @xmath70 be the collection of indices @xmath354 where @xmath314 can be well - approximated by the elements of @xmath321 as stated in theorem [ theorem : inverse ] . then as @xmath355 , @xmath70 contains a discrete interval of length @xmath356 , which we denote by @xmath357 . ( note that the symbol @xmath97 was used for a different interval previously , which should not be confused with the current setting . ) for any @xmath358 from @xmath97 such that @xmath359 , consider the sequence @xmath360 , together with their approximations @xmath361 from @xmath321 . by the choice of @xmath362 , @xmath363 as such , by ( ii ) @xmath364 with the choice @xmath365 , one guarantees that @xmath366 , and thus the lhs of is strictly positive . after choosing @xmath299 , we choose @xmath367 in theorem [ theorem : inverse ] so that @xmath368 , the constant in the rhs of . next , assume that @xmath369 for @xmath370 . then it follows from the choice of @xmath6 and from that @xmath371 consequently , recalling that all the generators @xmath338 are integral multiple of @xmath351 , there exists @xmath372 such that @xmath373 it thus follows from that @xmath374 in summary , we obtain the following key property for all @xmath375 and @xmath376 , @xmath377 as @xmath378 can not be all zero , without loss of generality , assume that @xmath379 . thus for every @xmath375 such that @xmath380 we have @xmath381 we next require the following observation . [ claim : increasing ] assume that @xmath382 is a sequence of real numbers which satisfy the following inequality for all @xmath383 @xmath384 then there exist @xmath385 such that @xmath386 where @xmath387 depends on @xmath199 . ( proof of claim [ claim : increasing ] ) define @xmath388 by the assumption and pascal s triangle identity @xmath389 it follows that @xmath390 . thus , there are at least @xmath391 consecutive indices @xmath376 such that the corresponding @xmath392 have the same signs and absolute value at least @xmath391 . without loss of generality , we can assume that all of them are at least @xmath391 . repeat the argument with @xmath393 and the above subsequence . after @xmath199 repetitions , we obtain a sub interval @xmath70 of length @xmath394 of @xmath395 $ ] , where all @xmath396 have the same signs and absolute value at least @xmath397 . applying claim [ claim : increasing ] with @xmath398 and @xmath399 , we obtain @xmath400 , and so @xmath401 this contradicts with the bound @xmath402 from theorem [ theorem : inverse ] and with the bound @xmath403 from because @xmath104 is sufficiently large . this completes the proof of lemma [ lemma : crucial:1/2 ] . now we conclude the subsection by proving theorem [ theorem : repulsion : general ] for the interval @xmath404 $ ] . we need to show the existence of @xmath157 ( that depends only on @xmath82 and @xmath43 ) such that for every @xmath158 sufficiently large @xmath405 we will apply lemma [ lemma : crucial:1/2 ] with @xmath406 , and let @xmath299 be the corresponding constant . by making @xmath162 smaller if necessary , it suffices to prove lemma [ l.almost-zero ] for @xmath407 where @xmath199 is sufficiently large ( instead of requiring @xmath167 as before . ) as @xmath408 $ ] , one can verify that @xmath409 thus , by lemma [ lemma : crucial:1/2 ] , @xmath410 for sufficiently small @xmath162 , provided that @xmath199 is sufficiently large . set @xmath411 . then @xmath412 . for every @xmath413 , write @xmath414 where @xmath415 if @xmath416 . consequently @xmath417 @xmath418 notice easily that , with probability @xmath419 , @xmath420 denote this event , on which we will condition for the rest of our argument . thus @xmath421 in particular , with @xmath422 , @xmath423 therefore @xmath424 . but as @xmath1 takes integer values and @xmath425 , therefore by taking @xmath37 large we must have @xmath426 after replacing into @xmath427 , we obtain @xmath428 now we consider the assumption that @xmath429 , from which we infer that @xmath430 @xmath431 we next consider two cases . * case 1 . @xmath432*. it follows easily from and that @xmath433 * case 2 . @xmath434*. as @xmath435 , it follows from that @xmath436 , and also by @xmath437 thus , using the fact that @xmath438 with @xmath85 , it follows that @xmath439 as @xmath1 takes integer values , it follows from both cases that @xmath440 since @xmath2 has no atom at @xmath21 , @xmath441 with probability @xmath58 . first of all , the contribution towards the expectation at the points @xmath442 are negligible owing to the following elementary estimates . [ l.endpoint]we have @xmath443 note that the above claim is trivial in the type ii setting , so it remains to show these estimates for type i. the first estimate clearly follows from the classical erds - littlewood - offord bound . we refer the reader to claim [ claim : anticoncentration ] for a short proof of the remaining two estimates . using claim [ l.endpoint ] , it follows that @xmath444 now , using the fact that the ( real ) zero sets of @xmath4 and @xmath77 have the same distribution , it follows that @xmath445 in the following , we will consider the number of real roots in @xmath446 , and show that @xmath447 for some @xmath448 . to estimate @xmath449 , we consider the random polynomial @xmath450 and let @xmath451 denote its number of real zeros . by definition , we have @xmath452 so we need to estimate the average number of real zeros for @xmath453 in @xmath446 . now , type i distributions are symmetric so in that setting the zero sets of @xmath453 and @xmath3 would have the same distribution , therefore @xmath454 and one obtains the same estimate as above for @xmath449 . for type ii distributions , the argument we use below for estimating @xmath455 could be applied to estimate @xmath456 . most importantly , our result on non - existence of double roots ( theorem [ theorem : repulsion : general ] ) does not require the coefficients @xmath64 to have identical distributions , and could be applied to @xmath453 . also the cited ingredients that we used below ( theorem [ theorem : tv ] and lemma [ lemma : jensen ] ) do not require @xmath64 to have identical distributions . with cosmetic changes , one could use the argument below to @xmath453 and obtains a similar estimate for @xmath456 , which is the same as @xmath449 , and conclude the proof of theorem [ theorem : expectedvalue:2 ] . for the rest of the section we will be focusing on . our first tool is the following result from @xcite . [ theorem : tv ] there is a positive constant @xmath299 such that the following holds . let @xmath457 be an arbitrary small constant and @xmath458 be independent random variables with mean 0 , variance 1 and uniformly bounded @xmath459-moment . there is a constant @xmath460 such that for any @xmath461 and interval @xmath462 $ ] @xmath463 where the implicit constant in @xmath127 depends only on @xmath299 and @xmath82 . next , for convenience , we truncate the random variables @xmath64 . let @xmath464 be a parameter and let @xmath465 be the event @xmath466 . as @xmath2 has mean zero and bounded @xmath50-moment , we have the following elementary bound for @xmath467 @xmath468 in what follows we will condition on @xmath469 . consider @xmath470 and for @xmath471 , we set @xmath472 for any @xmath473 , a generous chernoff s bound yields that for any @xmath474 @xmath475 since @xmath476 it follows that @xmath477 we next compare the roots of @xmath3 and @xmath478 in the interval @xmath479 . [ lemma : comparison ] let @xmath480 and @xmath481 such that @xmath482 , where @xmath483 of theorem [ theorem : repulsion : general ] . then for any subinterval @xmath484 of @xmath485 $ ] one has @xmath486 and the implicit constant depends on @xmath43 and @xmath120 only . to prove lemma [ lemma : comparison ] , we need the following elementary lemma from @xcite . [ approximation ] assume that @xmath487 and @xmath488 are continuous functions satisfying the following properties * @xmath489 and @xmath490 ; .1 in * @xmath491 for all @xmath492 $ ] ; .1 in * @xmath493 . then @xmath494 has a root in @xmath70 . conditioned on @xmath495 , with probability at least @xmath496 the following holds @xmath497 for all @xmath498 and @xmath37 sufficiently large . by theorem [ theorem : repulsion : general ] or theorem [ theorem : repulsion : general ] ( with @xmath499 ) , @xmath500 for all @xmath501 with probability @xmath502 . note that the conditioning on @xmath503 , which holds with probability at least @xmath504 , will not affect the @xmath505 term in this estimate . applying lemma [ approximation ] with @xmath506 , @xmath507 , we conclude that with probability @xmath508 , for any root @xmath509 of @xmath4 in the interval @xmath479 ( which is a subset of @xmath510 ) , there is a root @xmath511 of @xmath512 such that @xmath513 . on the other hand , applying ( 2 ) of theorem [ theorem : repulsion : general ] or theorem [ theorem : repulsion : general ] with @xmath499 , again with probability @xmath514 there is no pair of roots of @xmath3 in @xmath484 with distance less than @xmath515 . it follows that for different roots @xmath509 we can choose different roots @xmath511 . furthermore , by ( 3 ) of theorem [ theorem : repulsion : general ] or theorem [ theorem : repulsion : general ] , with probability @xmath508 , all roots of @xmath4 must be of distance at least @xmath516 from the two ends of the interval . if this holds , then all @xmath511 must also be inside the interval . this implies that with probability at least @xmath514 , the number of roots of @xmath478 in @xmath484 is at least that of @xmath3 . putting together , we obtain @xmath517 where the factor @xmath37 comes from the fact that @xmath3 has at most @xmath37 real roots . switching the roles of @xmath3 and @xmath478 , noting that as @xmath518 , @xmath519 \subset ( 0 , 1 - 1/m).\ ] ] as such , theorem [ theorem : repulsion : general ] and theorem [ theorem : repulsion : general ] are also applicable to @xmath520 . argue similarly as above , we also have @xmath521 it follows that @xmath522 by iterating lemma [ lemma : comparison ] , one can achieve the following . [ cor : comparison : l ] let @xmath523 be sufficiently large , and let @xmath524 . then for any sufficiently large integer @xmath525 ( depending on @xmath483 and @xmath523 ) and @xmath526 @xmath527 where the implicit constant depends only on @xmath528 and the parameter @xmath43 of @xmath2 . let @xmath529 . we assume @xmath530 so that @xmath531 for every @xmath526 . define the sequence @xmath532 with @xmath533 and @xmath534 . by ensuring that @xmath523 is sufficiently large we will have @xmath535 unless @xmath536 . thus , the sequence @xmath537 is decreasing , and let @xmath180 be the first index where @xmath538 . then @xmath539 here the last estimate also holds by applying lemma [ lemma : comparison ] for @xmath540 and @xmath541 , and clearly @xmath542 , and @xmath543 . next , we use ( * ? ? ? * lemma 2 , remark 4)[lemma : jensen ] assume that @xmath458 have mean 0 , variance 1 , and uniformly bounded @xmath50-moments . then there exist constants @xmath544 such that @xmath545 provided that @xmath546 , for each @xmath525 , denote @xmath547 $ ] . by lemma [ lemma : jensen ] , @xmath548 are uniformly bounded provided that @xmath546 , thus there is a subsequence of @xmath549 which tends to a finite limit @xmath550 . by applying corollary [ cor : comparison : l ] , we obtain the following . [ cor : limit ] for any @xmath551 , there exists @xmath552 such that @xmath553 we next turn to the main term by utilizing theorem [ theorem : tv ] and lemma [ lemma : comparison ] . recall the constant @xmath299 from theorem [ theorem : tv ] . let @xmath554 be a small constant ( so that in particular @xmath555 ) , and let @xmath556 to be the constant of theorem [ theorem : tv ] . [ lemma : mainterm ] assume that @xmath557 , then @xmath558 here the implicit constant depends only on @xmath43 , @xmath299 , and @xmath82 . we will show the following equivalent statement : assume that @xmath559 , then @xmath560 and define recursively @xmath561 it is clear that @xmath532 and @xmath562 are , respectively , strictly decreasing and increasing sequences . let @xmath525 be the largest index such that @xmath563 . it follows that @xmath564 , therefore @xmath565 . it also follows that @xmath566 redefine @xmath567 and @xmath568 . it is clear that we still have @xmath569 . for @xmath570 , define @xmath571 $ ] . for every @xmath572 we have @xmath573 $ ] while @xmath574 . thus , by , for any @xmath572 we have @xmath575 by the triangle inequality , @xmath576 for @xmath577 , and @xmath578\subset ( 1-n_i^{-{\varepsilon}},1)$ ] . by theorem [ theorem : tv ] @xmath579 , again by the triangle inequality @xmath580 the right end point of @xmath70 is @xmath581 , and the left end point is @xmath582 . furthermore , by definition of the @xmath583 , it is easy to see that @xmath584 is lacunary , therefore @xmath585 thus , @xmath586 ) - { { \mathbf{e}}}n_{n , n(0,1 ) } ( 1 -c_0^{-{\varepsilon}},1-n^{-{\varepsilon } } ] | = | { { \mathbf{e}}}n_{n,\xi } i - { { \mathbf{e}}}n_{n , n(0,1 ) } i | = o ( { c_0}^{-{\varepsilon}\alpha}).\ ] ] combined with theorem [ theorem : tv ] , @xmath587 proving . it suffices to show that @xmath588 to this end , we first complement the result of lemma [ lemma : mainterm ] by giving an estimate for @xmath589 . [ claim : gaussian ] for every @xmath590 there exists a finite number @xmath591 that depends only on @xmath523 such that @xmath592 here the @xmath593 term is with respect to the limit @xmath594 , and the implied constant depends on @xmath523 . we will use the following formula from @xcite , which asserts that the following equality holds for every interval @xmath595 $ ] : @xmath596 in particular , for every fixed @xmath597 we have @xmath598 thus we can define @xmath599 here recall that @xmath600 is the constant in the asymptotics expansion of the gaussian kac polynomial . it then follows from lemma [ lemma : mainterm ] that @xmath601 and in the @xmath602 term the implicit constant may depend on @xmath603 and @xmath2 . combining with corollary [ cor : limit ] , we obtain @xmath604 where @xmath605 is a number depending on @xmath523 . replacing @xmath523 by @xmath606 and subtract , @xmath607 and sending @xmath594 we obtain @xmath608 which shows that the function @xmath609 tends to a limit as @xmath523 tends to infinity . denote this limit by @xmath610 , it follows that @xmath611 concluding the proof . note that in this application section , we have applied our roots repulsion results ( theorem [ theorem : repulsion : general ] and theorem [ theorem : repulsion : general ] ) only for the interval @xmath612 $ ] . we first address fact [ fact : arithmetic ] . it is clear that when @xmath37 is even then @xmath613 are odd numbers , and hence the polynomial can not have double roots at these points . we next show the same thing for the case @xmath614 . [ fact:4k+1 ] assume that @xmath614 , then @xmath4 can not have double root at @xmath59 or @xmath58 . ( of fact [ fact:4k+1 ] ) assume that @xmath615 , then one has @xmath616 consequently , @xmath617 however , this is impossible as the rhs is an odd number , while the lhs is clearly an even number for any choice of @xmath618 . the non - existence of double root at @xmath59 can be argued similarly ( or just by setting @xmath619 . ) we now give a brief explanation that the probability bound of fact [ fact : arithmetic ] is sharp , up to a multiplicative constant . [ lemma : tight ] let @xmath2 be a bernoulli random variable and @xmath620 be divisible by 4 . then @xmath621 we use the circle method . let @xmath622 , and @xmath623 . let @xmath624 be a sufficiently large prime . we first write @xmath625 observe that @xmath626 , where @xmath627 is the distance to the nearest integer . we are going to analyze the sum @xmath628 basically , if this sum is quite large , then its distribution in is negligible . we state the following elementary claims whose proofs are left to the reader as an exercise . [ claim : integral ] the following holds . * assume that @xmath629 , then @xmath630 .1 in * assume that @xmath631 and @xmath632 , then @xmath630 it follows from claim [ claim : integral ] that the main term of the sum in is governed by @xmath633 and @xmath634 . thus either @xmath635 or @xmath636 and @xmath637 or @xmath638 . as @xmath639 , the interested reader is invited to check that the contribution in of one of these four cases are the same . it is thus enough to work with the case @xmath635 and @xmath637 . we are going to show the following . [ lemma : s1 ] @xmath640 the method below gives an exact estimate on the asymptotic constant , but we will not need this fact . the first equality is trivial , as all cosines are positive in this range of @xmath641 . viewing @xmath642 and @xmath643 as integers with absolute value at most @xmath644 and @xmath645 respectively . we have @xmath646 it follows that as @xmath647 @xmath648 after changing variables @xmath649 and @xmath650 , and using the gaussian identity @xmath651 , we have @xmath652 , completing the proof . to complete the picture , we prove claim [ l.endpoint ] . in fact , more is true : [ claim : anticoncentration ] let @xmath653 and @xmath654 , @xmath655 . assume that @xmath458 are iid copies of a random variable of variance one and bounded @xmath459-moment . then we have * @xmath656 ; .1 in * @xmath657 . as the proof for the second estimate is similar , it suffices the prove the first one . we ll use @xcite . assume otherwise that @xmath658 for some sufficiently large @xmath104 . then by ( * ? ? ? * theorem 2.5 ) most of the @xmath659 belongs to a symmetric gap @xmath321 of rank @xmath660 and size @xmath661 as @xmath662 , @xmath663 , which is smaller than @xmath664 if @xmath104 is sufficiently large , a contradiction to the fact that @xmath321 contains most of the @xmath665 . recall that at the end of the proof of ( * ? ? ? * theorem 2.9 ) ( equation ( 22 ) ) , we obtained the following for the dilated set @xmath666 : for any @xmath667 , there exists a subset @xmath668 of size at least @xmath669 such that @xmath670 now we apply theorem ( * ? ? ? * theorem 1.21 ) , it is implied that the set @xmath687 belongs to a gap @xmath688 of size @xmath689 , where the implied constant depends on @xmath674 and @xmath675 . next , by iterating ( * ? ? ? * theorem 3.40 ) if necessary , one can assume that @xmath688 is @xmath6-proper , while the size @xmath690 remains @xmath689 but the implied constant now also depends on @xmath6 . * @xmath321 has small rank , @xmath696 , and small cardinality @xmath697 ; .1 in * for all but at most @xmath319 elements @xmath151 of @xmath698 , there exists @xmath331 such that + @xmath699 .1 in * @xmath321 is @xmath6-proper ; .1 in * as @xmath700 , there exist @xmath701 such that + @xmath702 .1 in * as @xmath691 , all steps @xmath338 of @xmath321 are _ integral multiples _ of @xmath703 . 99 a. bloch and g. plya , on the roots of certain algebraic equations , _ proc . london math . soc . _ * 33*(1932 ) , 102 - 114 . l. blum , f. cucker , m. shub and s. smale , _ complexity and real computation _ , springer - verlag , new york , 1998 . a. t. bharucha - reid and m. sambandham , _ random polynomials , probability and mathematical statistics_. academic press , inc . , orlando , fla . , 1986 . f. cucker , t. krick , g. malajovich and m. wschebor , a numerical algorithm for zero counting . iii : randomization and condition , _ advances in applied mathematics _ * 48 * , 215 - 248 . a. edelman and e. kostlan , how many zeros of a random polynomial are real ? , _ bull . ( n.s . ) * 32 * ( 1995 ) , 137 . erratum : bull . * 33 * ( 1996 ) , 325 . t. erdlyi , extensions of the bloch - plya theorem on the number of real zeroes of polynomials , _ j. thor . nombres bordeaux _ * 20 * ( 2008 ) , no . 2 , 281-287 . p. erds and a. c. offord , on the number of real roots of a random algebraic equation , _ proc . london math . soc . _ * 6 * ( 1956 ) , 139160 . k. farahmand , topics in random polynomials , _ pitman research notes in mathematics _ , series 393 . longman , harlow , 1998 . i. a. ibragimov and n. b. maslova , the average number of zeros of random polynomials , _ vestnik leningrad . univ . _ * 23 * ( 1968 ) , 171172 . i. a. ibragimov and n. b. maslova , the mean number of real zeros of random polynomials . i. coefficients with zero mean , _ theor . probability appl . _ * 16 * ( 1971 ) , 228248 . i. a. ibragimov and n. b. maslova , the mean number of real zeros of random polynomials . ii . coefficients with a nonzero mean . , _ theor . probability appl . _ * 16 * ( 1971 ) , 485493 . i. a. ibragimov and n. b. maslova , the average number of real roots of random polynomials , _ soviet math . * 12 * ( 1971 ) , 10041008 . m. kac , on the average number of real roots of a random algebraic equation , _ bull . * 49 * ( 1943 ) 314320 . m. kac , on the average number of real roots of a random algebraic equation . london math . soc . _ * 50 * , ( 1949 ) , 390408 . m. kac , probability and related topics in physical sciences , _ lectures in applied mathematics , proceedings of the summer seminar _ , boulder , colo . , 1957 , vol . i interscience publishers , london - new york , 1959 . e. kostlan , on the distribution of roots of random polynomials , chapter 38 , _ from topology to computation : proceeding of the samefest _ , edited by m. w. hirsch , j.e . , marsden and m. shub , springer - verlag , ny 1993 . j. e. littlewood and a. c. offord , on the distribution of the zeros and a - values of a random integral function . i. , _ j. lond _ , * 20 * ( 1945 ) , 120 - 136 . j. e. littlewood and a. c. offord , on the number of real roots of a random algebraic equation . cambridge philos . soc . _ * 35 * , ( 1939 ) , 133 - 148 . j. e. littlewood and a. c. offord , on the number of real roots of a random algebraic equation . sbornik ] n.s . _ * 54 * , ( 1943 ) , 277 - 286 . j. e. littlewood and a. c. offord , on the distribution of the zeros and values of a random integral function . _ ann . math . _ * 49 * ( 1948 ) , 885952 . errata , * 50 * ( 1949 ) , 990 - 991 . 976 ) , 3558 . b. f. logan and l. a. shepp , real zeros of random polynomials . london math . * 18 * ( 1968 ) , 2935 . b. f. logan and l. a. shepp , _ real zeros of random polynomials . london math . * 18 * ( 1968 ) , 308314 . n. b. maslova , the variance of the number of real roots of random polynomials . vero- jatnost . i primenen . _ * 19 * ( 1974 ) , 3651 . n. b. maslova , the distribution of the number of real roots of random polynomials . _ probability appl . _ * 19 * ( 1974 ) , 461473 h. nguyen and v. vu , optimal inverse littlewood - offord theorems , _ adv . _ * 226 * ( 2011 ) , no . 6 , h. nguyen , o. nguyen and v. vu , on the number of real roots of random polynomials , _ submitted_. r. peled , a. sen and o. zeitouni , double roots of random littlewood polynomials , http://arxiv.org/abs/1409.2034 . stevens , the average number of real zeros of a random polynomial . pure appl . math . _ * 22 * ( 1969 ) , 457477 . t. tao and v. vu , local universality of zeros of random polynomials , _ to appear in imrn_. i. todhunter , a history of the mathematical theory of probability , stechert , new york , 1931 . y. wang , bounds on the average number of real roots of a random algebraic equation , _ chinese ann . a 4 ( 1983 ) , 601 - 605 .

we study a bipartite linear chain constituted by spherical metallic nanoparticles , where each nanoparticle supports a localized surface plasmon . the near - field dipolar interaction between the localized surface plasmons gives rise to collective plasmons , which are extended over the whole nanoparticle array . we derive analytically the spectrum and the eigenstates of the collective plasmonic excitations . at the edge of the brillouin zone , the spectrum is of a pseudo - relativistic nature similar to that present in the electronic band structure of polyacetylene . we find the effective dirac hamiltonian for the collective plasmons and show that the corresponding spinor eigenstates represent one - dimensional dirac - like massive bosonic excitations . therefore , the plasmonic lattice exhibits similar effects to those found for electrons in one - dimensional dirac materials , such as the ability for transmission with highly suppressed backscattering due to klein tunnelling . we also show that the system is governed by a nontrivial zak phase , which predicts the manifestation of edge states in the chain . when two dimerized chains with different topological phases are connected , we find the appearance of the bosonic version of a jackiw - rebbi midgap state . we further investigate the radiative and nonradiative lifetimes of the collective plasmonic excitations and comment on the challenges for experimental realization of the topological effects found theoretically .

1611.03349

since veselago @xcite first pondered creating artificial materials to induce unorthodox physical properties in 1968 , metamaterials have been at the forefront of interdisciplinary research to develop novel applications . after the turn of the new millennium , groundbreaking progresses on invisibility and cloaking @xcite , slow - light @xcite and superlensing effects @xcite have been reported . this scientific paradigm has spawned the subfield of nanostructure - based plasmonic metamaterials , which exploit the properties of the surface plasmon , a collective oscillation of valence electronic charges @xcite . the ability for plasmonic materials to focus light at sub - wavelength scales and to generate a large optical density of states are two significant physical properties which may be utilized in future applications @xcite . there have already been numerous proposed implementations of plasmonic metamaterials , including spasers @xcite , optical transmission modulators @xcite , bio - sensers @xcite , switches @xcite and optical storage devices @xcite . there is continued and active interest in schemes enabling long - range propagation in plasmonic waveguides @xcite since they are the building blocks of plasmonic circuitry @xcite , where it is hoped that a similar large bandwith of information as in conventional photonics can be achieved , but without the limitation of diffraction at submicron cross sections . linear , one - dimensional ( 1d ) chains of regularly - spaced metallic nanoparticles , which support localized surface plasmons ( lsps ) confined to each nanoparticle , have been proposed almost two decades ago to act as plasmonic waveguides @xcite . when irradiated with light , the lsp excitations in the nanoparticles forming the array couple to each other through the optical near - field dipole - dipole interaction , which induces collective plasmons @xcite that extend over the whole chain and which guide electromagnetic radiation with strong lateral confinement . since the original proposal by quinten _ et al . _ @xcite , plasmon propagation in nanoparticle chains has been extensively investigated both theoretically @xcite and experimentally @xcite . recently , the potentially nontrivial topological properties of collective excitations in zigzag @xcite and ` diatomic ' @xcite chains of nanoparticles were uncovered . in particular , it has been shown in ref . @xcite by means of a classical quasi - static calculation based on the macroscopic maxwell equations that the collective plasmonic oscillations in the linear array become decidedly topologically nontrivial when the chain becomes dimerized . other 1d or quasi-1d artificial systems such as photonic lattices @xcite , dielectric resonators @xcite , silicon waveguides @xcite , polariton cavities @xcite and metallic nanowires @xcite were also shown to present interesting nontrivial topological features , paving the way to a new exciting field of research coined ` topological photonics ' @xcite . in this work , we present a fully quantum - mechanical treatment of collective plasmons in a linear bipartite chain of metallic nanoparticles , which provides a clear analogy with the nontrivial electronic properties of fermionic topological insulators @xcite . such an analogy remains only elusive within a classical treatment @xcite . moreover , a quantum treatment becomes essential when the size of the nanoparticles composing the chain ( below ca . @xmath0{nm}$ ] in diameter ) is such that quantum - size effects appear in their optical response @xcite . the formed bipartite lattice , given by a regular array with a unit cell consisting of two identical nanoparticles , presents a small gap at the edge of the first brillouin zone in the two - band collective plasmonic bandstructure . we find an effective hamiltonian for the collective plasmons , which at the edge of the brillouin zone corresponds to a 1d dirac equation for massive bosonic quasiparticles . these bogoliubov excitations are accompanied with a nontrivial vacuum state , which is different from the vacuum of the original plasmonic excitations . the dirac - like nature of the quasiparticles associated with our system allows us to propose a setup to observe the plasmonic analogue of klein tunneling @xcite , which is celebrated for enabling transmission with highly suppressed backscattering , as is the case in two - dimensional graphene sheets @xcite . we show that depending on the strength of the dimerization , the system may be in either a topologically trivial or nontrivial phase , which is codified by a topological invariant @xcite . this invariant governs the existence of special , topologically - protected modes at the ends of the chain and identifies the bipartite nanoparticle chain as a ` bosonic topological insulator ' for plasmonic excitations . we unveil several types of edge states , which appear at the midgap frequency of the gapped plasmonic bandstructure . we further investigate the topologically - protected midgap state formed between two dimerized chains , which is reminiscent of the jackiw - rebbi solitonic state @xcite , and which may be utilized to store light at the nanoscale . a major challenge for the effective use of plasmonic circuitry is how to effectively combat losses necessarily present in a system with a high scattering rate of electrons , radiation losses into the far - field and dissipation due to quantum - size effects @xcite . here we estimate , with analytical expressions for both radiation and landau @xcite damping decay rates , the plasmonic losses in the bipartite chain . in particular , we find that the two - band collective plasmonic bandstructure is comprised of one bright and one dark band per polarization , with drastically different responses to radiative losses . this analysis enables us to comment on the experimental challenges to observe topological collective plasmons . the sequel of this paper is organized as follows : section [ sec2 ] presents our model of collective plasmonic excitations in a bipartite chain of metallic nanoparticles coupled to a bath of electron - hole pairs ( responsible for the nonradiative landau damping ) as well as to a photonic bath ( which leads to radiation damping of the collective excitations ) . in sec . [ sec3 ] , we diagonalize the purely plasmonic problem and discuss the properties of the associated plasmonic bandstructure . we detail in sec . [ sec4 ] the mapping of the plasmonic hamiltonian to a 1d dirac equation for massive bosonic quasiparticles , and consider the plasmonic version of klein tunneling . in sec . [ sec5 ] , we study the topological properties of the bipartite chain , codified by a nontrivial zak phase , and illustrate how various types of edge states and topologically - protected midgap states may form . an analysis of plasmonic losses due to both the size - dependent radiation and landau damping is presented in sec . [ sec6 ] , which enables us to comment on the experimental realization of our proposal . finally , we draw conclusions in sec . a few appendices complement the discussion presented in the main text . we consider a linear chain composed of @xmath1 dimers , which comprise the unit cell of two identical spherical metallic nanoparticles . the two corresponding inequivalent sublattices are denoted by @xmath2 and @xmath3 , respectively . this geometry is sketched in fig . [ fig1 ] , showing the nanoparticle radius @xmath4 and the interparticle separations @xmath5 and @xmath6 which build a bipartite chain with a period @xmath7 . along the chain each nanoparticle supports an lsp resonance @xcite , a collective excitation of the electronic center of mass , which couples to the neighboring nanoparticles via the dipole - dipole interaction @xcite . this approximation is valid for interparticle distances @xmath8 @xcite . in what follows , we denote by @xmath9 the plasmonic hamiltonian describing the near - field - coupled lsps along the bipartite chain . each lsp is coupled to vacuum photonic modes described by the hamiltonian @xmath10 through the light - matter interaction @xmath11 ( which includes retardation effects ) , leading to the radiative decay of the collective excitations . moreover , the coupling described by the hamiltonian @xmath12 of the lsps to electron - hole pairs ( with hamiltonian @xmath13 ) in each of the nanoparticles leads to the size - dependent landau damping of the collective modes @xcite . the hamiltonian of our system hence reads @xmath14 dimers , comprised of a pair of metallic nanoparticles denoted @xmath2 and @xmath3 , each supporting localized surface plasmon resonances . ] since the nanoparticles we consider are small enough to be subject to quantum - size effects @xcite , and to elucidate the connection to fermionic quasiparticles in dirac materials , we quantize the classical plasmonic hamiltonian @xmath9 @xcite in the coulomb gauge @xcite , leading to @xmath15 where @xmath16 is an index which denotes the dimer number in the chain ( see fig . [ fig1 ] ) , and the summation over @xmath17 counts the two transverse ( @xmath18 ) and the single longitudinal ( @xmath19 ) polarizations of the collective plasmonic modes . the factor @xmath20 ( @xmath21 ) accounts for the repulsive ( attractive ) interaction between transverse ( longitudinal ) nearest neighbor lsps . we neglect interactions beyond those between nearest neighbors since such interactions only lead to a small quantitative change of the plasmonic dispersion ( see appendix [ appenda ] for details ) . the bosonic operators @xmath22 and @xmath23 ( @xmath24 and @xmath25 ) in eq . annihilate ( create ) an lsp with polarization @xmath26 in the @xmath16th dimer in nanoparticle @xmath2 and @xmath3 , respectively . the hamiltonian displays terms which are reminiscent of a 1d tight - binding model . firstly , the role of the onsite energy is played by the lsp resonance frequency @xmath27 [ here we refer to the first and second terms on the right - hand side of eq . ] . for the case of alkaline nanoparticles in vacuum , this resonance frequency corresponds to the mie frequency @xmath28 , where @xmath29 is the charge on an electron with effective mass @xmath30 and the number of electrons in each nanoparticle is @xmath31 @xcite . the plasma frequency is @xmath32 , where @xmath33 is the electronic density of the considered metal . secondly , the analogue of the hopping terms in eq . , namely those @xmath34 and @xmath35 , are due to the nearest neighbor dipole - dipole interaction . the dipolar interaction gives rise to the coupling constants @xmath36 which alternate along the chain . however , this is where the comparison to a standard tight - binding model ends , due to the appearance of nonresonant terms @xmath37 and @xmath38 which enter the hamiltonian and are also due to the dipolar interaction . such nonrotating - wave terms can not be neglected when calculating plasmon eigenstates and , hence , quantities dependent on the latter , such as damping rates ( see sec . [ sec6 ] ) @xcite . without the nonresonant terms mentioned above , the model encapsulated in the hamiltonian can be seen as the bosonic counterpart to the su - schrieffer - heeger @xcite and rice - mele models @xcite , where electrons hopping in 1d lattices with staggered hopping amplitudes were treated for the spinless and spinful cases , respectively . these simple , though physically rich models were originally developed in studies of conducting polymers such as polyacetylene , and are known to exhibit rather exotic physics including fractionally charged excitations and edge states @xcite . how the presence of the bosonic nonresonant terms in the hamiltonian will affect the formation and properties of edge states is not immediately clear . this issue is addressed in the sequel of our paper . in eq . , the vacuum modes of the electromagnetic field , to which the lsps are coupled to , lead to the radiative damping of the plasmonic collective modes and are described by the photonic hamiltonian @xmath39 where @xmath40 ( @xmath41 ) annihilates ( creates ) a photon with momentum @xmath42 , transverse polarization @xmath43 , and dispersion @xmath44 , with @xmath45 the speed of light in vacuum . in the long - wavelength limit , the plasmon - photon coupling hamiltonian entering eq . has the expression @xmath46 , \end{aligned}\ ] ] where the position of the center of the nanoparticle belonging to sublattice @xmath2 ( @xmath3 ) in the @xmath16th dimer is denoted by @xmath47 ( @xmath48 ) , and where @xmath49 is the quantization volume of the photonic modes . we note that the coupling , together with the static dipole - dipole interaction in eq . , result in a fully retarded description of the plasmon - photon interaction @xcite . the hamiltonian describing the bath of electron - hole pairs in eq . reads as @xmath50 where @xmath51 labels the electron and hole states with energy @xmath52 in nanoparticle @xmath53 , and where @xmath54 annihilates ( creates ) the one - body state @xmath55 . in the hard - wall approximation for the mean - field potential felt by the electrons @xcite , the collective plasmon - electron - hole coupling leading to the landau damping of the collective plasmons reads @xcite @xmath56,\end{aligned}\ ] ] where @xmath57 . we start by fully characterizing the fundamental properties of the plasmonic bipartite chain in absence of the damping mechanisms . to this end , we diagonalize exactly in the sequel the hamiltonian by means of bosonic bogoliubov transformations . we provide analytic expressions for the plasmonic dispersion relation and its associated eigenstates . using periodic born - von karman boundary conditions and introducing the bosonic operators in momentum space @xmath58 and @xmath59 through the fourier transforms @xcite [ eq02 ] @xmath60 the real space hamiltonian becomes @xmath61,\end{aligned}\ ] ] where we have introduced the quantity @xmath62 the quadratic hamiltonian is diagonalized with the bosonic bogoliubov transformation @xcite @xmath63 where the index @xmath64 ( @xmath65 ) corresponds to the upper ( lower ) collective plasmon band . in eq . , the bogoliubov coefficients read [ eq06 ] @xmath66 and @xmath67 and the auxiliary bosonic operators are @xmath68 thus we arrive at the diagonalized form of eq . @xcite , @xmath69 where the eigenfrequencies of the collective plasmonic modes are @xmath70 ) and longitudinal ( blue and cyan solid lines , @xmath71 ) modes for both the upper ( @xmath72 ) and the lower ( @xmath73 ) bands . the undimerized case ( thinner lines ) with @xmath74 and a typical dimerized case ( thicker lines ) with @xmath75 and @xmath76 are both displayed . ] the dispersion relation is plotted in fig . [ fig2 ] for the regular chain limit where @xmath77 ( thinner lines ) and a typical dimerized chain with @xmath78 ( thicker lines ) for both the transverse ( dashed red and orange lines ) and longitudinal ( solid blue and cyan lines ) collective plasmons . one notices an immediate analogy to the energy bands in a peierls chain @xcite , as the dimerization has the effect of opening up a small gap of the order of @xmath79 at the edges of the first brillouin zone ( @xmath80 ) . the longitudinal collective plasmon mode has a gap which is about twice as large as compared to the transverse modes because the attractive interaction between pairs of dipoles parallel to the chain is twice as much as between dipoles perpendicular to the chain which experience a repulsive interaction [ cf . the factor @xmath81 appearing in eq . ] . in appendix [ appenda ] we calculate the collective plasmon dispersion where the dipole - dipole interaction is considered beyond nearest neighbors . such additional terms in the near - field interaction only lead to a minor correction to the dispersion relation shown in fig . [ fig2 ] , justifying the nearest - neighbor approximation to the near - field interaction which we use in the remainder of this work . now that the plasmonic hamiltonian has been diagonalized in sec . [ sec3 ] , we are in a position to study in more details the properties of its associated eigenstates . in this section we show that at the edge of the brillouin zone , the system can effectively be described by a 1d dirac - like hamiltonian , which gives rise to a klein tunnelling phenomenon for the collective plasmons . we further comment on the experimental setups required to observe the aforementioned tunneling effect . to analyze the physics occurring at the edge of the brillouin zone , we taylor - expand the plasmon dispersion in the vicinity of @xmath82 , with @xmath83 . in the regime @xmath84 , we find that the collective excitations are governed by a pseudo - relativistic spectrum @xmath85 which is reminiscent of the spectrum of massive dirac particles , @xmath86 , where @xmath87 distinguishes the electron and positron particle species , and @xmath88 and @xmath89 are the momentum and mass of the particle , respectively . the pseudo - relativistic spectrum indicates a connection to a dirac - like hamiltonian , as was first noticed in the continuum quantum field theory study of the ( fermionic case ) of diatomic polymers @xcite . working in the regime of first order in @xmath84 , one finds from eq . that @xmath90 and @xmath91 , where @xmath92 . it follows that for small dimerizations @xmath93 , @xmath94 , such that the plasmonic hamiltonian can be expressed in terms of the auxiliary operators of eq . , @xmath95 . performing the summation over @xmath96 , we obtain a @xmath97 matrix hamiltonian @xmath98 , with @xmath99 where @xmath100 and where @xmath101 . the associated eigenvectors are in the form of a dirac spinor @xmath102 which is reminiscent of eigenvectors found in 1d dirac materials such as single - wall carbon nanotubes @xcite . the index @xmath96 denotes whether the excitation can be thought of ( in the language of semiconductor physics ) as hole - like ( @xmath103 ) or electron - like ( @xmath104 ) , as is the case with the spinor of graphene @xcite . for the dirac - like nature of the hamiltonian to become apparent , we utilize the unitary transform @xmath105 , to reach @xmath106 , with @xmath107.\ ] ] here , @xmath108 is the @xmath97 identity matrix and @xmath109 , @xmath110 and @xmath111 are pauli s spin matrices . the parallels with the 1d form of the dirac hamiltonian @xmath112 are evident , up to the unimportant constant energy shift @xmath113 . a striking feature in the effective hamiltonian is that the role of the mass term in a typical dirac hamiltonian is played by a quantity @xmath114 which can be modified in both magnitude and sign through the interparticle distances @xmath115 and @xmath116 [ see eq . ] . this tunability has significant implications for both pseudo - relativistic klein tunneling ( see appendix [ appendb ] ) and the topological properties of bipartite metallic nanoparticle chains ( cf . sec . [ sec5 ] ) . the present realization of 1d dirac - like bosons encapsulated in the hamiltonian complements the recent discovery of their counterparts in 2d honeycomb metasurfaces @xcite . a prominent feature of excitations described by a dirac equation is their ability to be transmitted through a potential barrier with highly suppressed backscattering @xcite , the so - called klein paradox . for the system of nanoparticles we investigate here , a plasmonic version of klein tunneling can be realized by arranging a dimerized chain into two distinct sections , which we distinguish with the notation @xmath117 ( for left ) and @xmath118 ( for right ) . these two sections have different resonance frequencies @xmath119 to mimic the application of an electrostatic potential step , as can be inferred from the first term in the right - hand side of the dirac - like hamiltonian ( see appendix [ appendb ] for a detailed calculation of the corresponding transmission probability ) . such a scenario may be experimentally achieved by the three following different means : ( i ) using two different metals ( for example silver and gold ) ; ( ii ) modifying the electrostatic environment of one chain with respect to the other , using an embedding dielectric medium ; or ( iii ) using nanoparticles of different sizes , for which quantum - size effects such as the spill - out of the electronic density outside the nanoparticle @xcite as well as coupling to the electronic environment @xcite influence the position of the surface plasmon resonance . the appearance of a dirac equation with its associated spinor wavefunctions in the description of the collective plasmons in the dimerized chain ( see sec . [ sec4 ] ) indicates that a nontrivial berry phase may arise in this system . in this section we study the implications of such physics on bosonic quasiparticles , classify the topological properties which are found to arise and investigate the existence of edge and soliton - like states . as a result , we uncover our system as a ` plasmonic topological insulator ' , which has the ability to both harbor and exploit topologically - protected midgap states . the berry phase @xcite when applied to the dynamics of electrons in periodic solids in 1d is known as the zak phase @xcite . the latter is connected to a @xmath120 topological invariant , which characterizes exotic properties such as fermionic charge fractionalization @xcite . for our purposes here , its utility arises because it is related to the existence of edge states , in a manifestation of a bulk - edge correspondence principle @xcite . the zak phase is defined as an integration over the first brillouin zone of the berry connection @xmath121 , explicitly @xmath122 which requires knowledge of the plasmonic eigenstates @xmath123 . we may write the plasmonic eigenstates via identification with the bogoliubov coefficients of eq . , namely @xmath124 in the basis @xmath125 @xcite . then , proceeding to calculate the zak phase from eq . with eq . , we find @xmath126 defined modulo @xmath127 ( since the berry connection @xmath128 is not gauge invariant ) . thus the dimerized chain has a trivial zak phase @xmath129 when @xmath130 [ @xmath131 , cf . eq . ] and a nontrivial one @xmath132 when @xmath133 ( @xmath134 ) , with a topological phase transition at the crossover point of the regular chain @xmath77 , where the bandgap closes ( see fig . [ fig2 ] ) . equation alludes to the presence of highly confined edge states in the nontrivial phase , corresponding to topologically - protected localized collective plasmons , which we investigate in detail in the remainder of this section . ( red dots , thin gray line : guide for the eye ) at each site in a chain of @xmath135 nanoparticles , where @xmath136 . inset : sketch of a typical chain comprised of an even number of nanoparticles , where the loops depict the favored dimerization . ] we now search for a midgap collective plasmon state at the frequency @xmath27 supported by the finite bipartite chain described by the hamiltonian and composed of an even number of nanoparticles ( for completeness , we consider in appendix [ appendd ] the case where the chain is composed of an odd number of constituants ) . to this end , we work in real space with open boundary conditions . due to the requirement of preserving bosonic statistics , diagonalization of eq . requires the framework of paraunitary transforms @xcite . the result of this numerical procedure is shown in fig . [ fig3 ] , where we plot the probability density associated with the midgap state at each nanoparticle site in the regime @xmath137 ( @xmath133 ) . as foreshadowed by the nontrivial zak phase @xmath138 found in sec . [ sec : zak ] in that case , both edges are highly plasmonically active . we have also checked numerically that , as expected , there is no midgap state when @xmath139 ( @xmath130 ) for which one finds a vanishing zak phase [ cf . eq . ] . to better understand the critical factor determining the formation of the midgap edge states , we neglect the nonresonant terms in eq . and maintain only the regular ` hopping ' terms in the following analytic analysis . this rotating wave approximation ( rwa ) is justified post hoc due to the excellent comparison to the results found via full numerical diagonalization of the hamiltonian in the limit of a large number of nanoparticles . we thus start from the real space hamiltonian in the form @xmath140 , with @xmath141 writing the above hamiltonian as @xmath142 where @xmath143 , @xmath144 is an even , symmetric , tridiagonal matrix . the diagonal elements of @xmath144 are all equal to @xmath27 , while the off - diagonal elements alternate between @xmath145 and @xmath146 . diagonalizing the matrix @xmath144 amounts to the following system of equations for the coefficients @xmath147 associated with the operator @xmath148 : [ eqbig1 ] @xmath149\ ] ] @xmath150\ ] ] @xmath151 @xmath152 searching for midgap states at @xmath27 produces from eqs . and the boundary conditions @xmath153 at the chain edges . in the bulk , the recurrence relations formed from eqs . and asymptotically satisfy these boundary conditions as long as @xmath137 ( @xmath133 ) , with the probability densities on each site given by [ eq88 ] @xmath154}\left ( \frac{\omega_1}{\omega_2 } \right)^{2 ( n-1 ) } , \\ \left|b^\sigma_{n}\right|^2 & = \frac{(\omega_1/\omega_2)^2 - 1}{2\left[(\omega_1/\omega_2)^{2\mathcal{n}}-1\right]}\left ( \frac{\omega_1}{\omega_2 } \right)^{2 ( \mathcal{n}-n)},\end{aligned}\ ] ] with @xmath155 $ ] . a plot of the expressions above is indistinguishable from the one of the full numerical solution in fig . [ fig3 ] for long enough chains and so is not displayed in the figure @xcite . the exponential behavior of the plasmonic probability density associated with the midgap state at frequency @xmath27 can be understood with the help of the simple picture in the inset of fig . [ fig3 ] , where we sketch a representative chain consisting of an even number of nanoparticles . the @xmath137 dimerization favors a pairing - up of nanoparticles as depicted with the drawn loops , leaving seemingly isolated nanoparticles at both ends of the chain which are both free to harbor the majority of the plasmonic activity . a completely different edge state as the one previously discussed may form at the interface between two topologically distinct bipartite chains ( both considered for simplicity to have the same resonance frequency @xmath27 ) . at the formed ` domain wall ' , a topologically protected kink - like bound state arises at the midgap frequency @xmath27 . in this way , the state is reminiscent of a jackiw - rebbi solitonic mode @xcite , an exponentially pinned eigenstate which arises at the boundary where the mass term changes sign in a dirac equation with position dependent mass . let us consider bringing together two bipartite chains , identified by their different zak phases @xmath156 and @xmath157 , and characterized by their coupling constants @xmath158 we choose the left - hand side chain to be topologically trivial @xmath159 and the right - hand side chain to be topologically nontrivial @xmath160 . furthermore , the connection of these two chains is described with the interface coupling constant @xmath161 . for simplicity , we investigate the case of an odd total number of nanoparticles , comprised of @xmath162 nanoparticles in the left hand side chain and @xmath163 nanoparticles in the right hand side chain . the resulting real space hamiltonian can be written down in analogy to the hamiltonian . at each site in a chain of @xmath164 nanoparticles , for three different choices of the coupling constants . in the two symmetric setups @xmath165 , with the interface coupling constant being either @xmath166 ( red dots ) or @xmath167 ( blue squares ) . in the asymmetric setup @xmath168 , @xmath169 , and @xmath166 ( green triangles ) . the thin gray lines are guides for the eye . ] as alluded to in sec . [ sec : edge_states ] , one may diagonalize quadratic , bosonic hamiltonians by employing paraunitary transforms @xcite . following this approach , we obtain the probability density of the midgap state at frequency @xmath27 as a function of the nanoparticle site , which is plotted in fig . we consider three fundamental varieties of solitonic state which can be obtained by expedient choices of inter - nanoparticle separations . firstly , we consider symmetric states with coupling constants @xmath165 . with the interface coupling constant @xmath166 ( red dots ) the soliton is single - peaked and centered on the nanoparticle corresponding to the @xmath2 sublattice in the @xmath170st dimer ; whilst for @xmath167 ( blue squares ) the state is double - peaked , with equal maxima on the nanoparticles corresponding to the @xmath2 sublattice in the @xmath171th and @xmath172st dimer . finally , a typical asymmetric midgap state is shown in fig . [ fig4 ] with the parameters @xmath168 , @xmath169 , and @xmath166 ( green triangles ) . in this case , the significant asymmetry causes the midgap state to be highly confined to the nanoparticle in sublattice @xmath2 in the @xmath172st dimer . notably , due to the experimental possibility to position metallic nanoparticles in any desired location , the system provides some tunability over the size of the localization length of the soliton - like state . the topologically - protected midgap states above can be well described using the rwa , which disregard the nonresonant terms which appear in the governing hamiltonian . the validity of this conjecture is , as in sec . [ sec : edge_states ] , proved after the fact by comparing the analytic expressions obtained in the rwa with the outcome of the numerical diagonalization of the full hamiltonian . such an rwa analysis proceeds in a similar manner to that carried out in sec . [ sec : edge_states ] . one finds the complete extinction of plasmonic activity on the @xmath3 sublattice , @xmath173 , with @xmath174 $ ] . meanwhile on the @xmath2 sublattice , we find an exponentially pinned probability density profile [ eq : probsoliton ] @xmath175 for @xmath176 $ ] and @xmath177 for @xmath178 $ ] in terms of the probability density on the final @xmath2 site of the left hand side chain , @xmath179^{-1}. \end{aligned}\ ] ] since these analytic expressions are imperceptible from the numerically exact solution displayed in fig . [ fig4 ] , they are not plotted in the figure . equation defines a bosonic version of a solitonic midgap state . its existence at the domain wall boundary only requires the connection of two chains of different zak phases . the precise details of the chain are unimportant , so it corresponds to a topologically - protected plasmonic bound state . the analysis above provides a physical realization of various bosonic jackiw - rebbi - like states , and gives rise to the chance of witnessing such exotic states experimentally , especially when compared to fermionic systems such as conducting polymers where such an observation is a challenging task . now that the topological properties of the purely plasmonic problem encapsulated in the hamiltonian have been discussed , here we analyze in detail the influence of both the photonic and electronic environments [ eqs . and , respectively ] onto plasmon lifetimes . in particular , we calculate the radiative and size - dependent nonradiative ( landau damping ) losses encountered by the collective plasmons in bipartite chains of metallic nanoparticles , and comment on the various challenges to experimentally observe their topological properties . the plasmonic system sketched in fig . [ fig1 ] is subject to radiative losses , which arise due to the coupling of the collective plasmons along the nanoparticle chain to vacuum electromagnetic field modes described by eq . and entering the system hamiltonian . treating the plasmon - photon coupling as a perturbation , the fermi golden rule result for the radiative decay rate of the collective plasmon with momentum @xmath180 in the branch @xmath96 and polarization @xmath26 reads @xmath181 the above expression describes the transitions between the plasmonic excited state @xmath182 and the ground state @xmath183 by spontaneous emissions of photons ( @xmath184 ) . using eq . together with the bogoliubov transformation , we arrive at @xmath185 where the array factor is defined as @xmath186d},\ ] ] with @xmath187 being the @xmath19 component of the photon wavenumber @xmath42 . equation can easily be evaluated , yielding @xmath188d/2)}}{\sin^2{([q - k_z]d/2)}},\ ] ] which , in the large chain limit ( @xmath189 ) , simplifies to @xmath190 d \right ) $ ] . using that @xmath191 , working in the continuum limit ( @xmath192 ) , and replacing the sum over @xmath42 by a three - dimensional integral in eq . , we finally arrive at the result @xmath193 \nonumber\\ & \times \left [ 1 + \tau \operatorname{sgn}{\{\eta_{\sigma}\ } } \frac{\omega_1 \cos ( q d_1 ) + \omega_2 \cos ( q d_2)}{\sqrt{\omega_1 ^ 2 + \omega_2 ^ 2 + 2 \omega_1 \omega_2 \cos ( q d ) } } \right ] \nonumber\\ & \times \theta \left ( \omega_q^{\sigma \tau } - c |q| \right)\end{aligned}\ ] ] for the radiation damping of the collective plasmons in terms of the single particle result @xmath194 , where @xmath195 . ) and longitudinal ( blue and cyan solid lines , @xmath71 ) collective plasmonic modes for @xmath196 . the undimerized case ( thinner lines ) with @xmath74 and a typical dimerized case ( thicker lines ) with @xmath197 and @xmath198 are both displayed . only the results for the bright bands , i.e. , the lower @xmath73 band for the longitudinal polarization and the upper @xmath72 band for the transverse modes , are displayed in the figure as the dark bands provide vanishingly small contributions . ] we plot in fig . [ fig5 ] the radiation damping decay rate . the results for the regular chain limit with @xmath77 ( thinner lines ) and a typical dimerized chain with @xmath199 ( thicker lines ) are both plotted , each for the transverse ( dashed red / orange lines ) and longitudinal ( solid blue / cyan lines ) polarizations . notably , even when the chain is in the dimerized regime , there remains only one band per polarization that provides a non - negligible contribution to the decay rate . namely , the longitudinal modes in the upper @xmath72 band and the transverse modes in the lower @xmath73 band are weakly coupled to light and are thus referred to as ` dark bands ' . these polarization - dependent low - loss bands will inevitably have an increased plasmonic lifetime due to being protected against radiation damping . the dark modes arise due to destructive interferences in the far field , with an almost vanishing net dipole moment . in contrast the bright modes ( the lower @xmath73 band for the longitudinal polarization and the upper @xmath72 band for the transverse modes ) contribute a non - vanishing dipole moment and are thus subject to radiation damping . as can be seen in fig . [ fig5 ] , dimerization has a weak effect on radiation damping , as the latter follows qualitatively the @xmath180-dependence of the simple chain result @xcite . outside of the light cone @xmath200 the decay rate is vanishing , whilst inside the light cone @xmath201 the modes are superradiant , with a magnitude significantly larger than that of a single nanoparticle . the coupling of the collective plasmons to a bath of electron - hole pairs , described by the hamiltonian and appearing in the system hamiltonian , gives rise to a size - dependent nonradiative decay of a purely quantum mechanical origin : the landau damping @xcite . assuming that the plasmon - electron - hole hamiltonian acts perturbatively , one can write down the following fermi golden rule for the nonradiative decay rate of the collective plasmon with momentum @xmath180 in the branch @xmath96 and with the polarization @xmath26 , @xmath202 with @xmath203 denoting electron - hole states in the nanoparticle belonging to the dimer @xmath16 and the sublattice @xmath204 . the above expression describes the transitions between the plasmonic excited and ground states by nonradiative emission of electron - hole pairs ( @xmath205 ) . evaluating eq . with the help of the bogoliubov transformation yields @xmath206 in terms of the summation over electron and hole states ( assumed to be the same in each nanoparticle ) @xmath207 the quantity @xmath208 can be analytically calculated using a semiclassical expansion of the angular - momentum - restricted density of states @xcite , leading to @xmath209 where @xmath210 and @xmath211 are the fermi velocity and energy of the metal , respectively . an explicit expression of the monotonically decaying function @xmath2120,1]$ ] can be found in eqs . ( 107 ) and ( 108 ) of ref . @xcite or in eq . ( b2 ) of ref . @xcite . ) and longitudinal ( blue and cyan solid lines , @xmath71 ) collective plasmonic modes for @xmath213 . the thinner ( thicker ) lines correspond to @xmath74 ( @xmath197 and @xmath198 ) . ] the landau damping decay rate is plotted in fig . [ fig6 ] , scaled with the single nanoparticle result @xmath214 @xcite . the results for the regular chain with @xmath77 ( thinner lines ) and a typical dimerized chain with @xmath199 ( thicker lines ) for both the transverse ( dashed red / orange lines ) and longitudinal ( solid blue / cyan lines ) polarizations are shown . the landau damping linewidths of the coupled plasmons are significantly different in magnitude depending on the band index @xmath96 due to the quartic dependence on the spectrum . whilst the @xmath103 bands show decreasing landau damping with increasing magnitude of the momentum , the opposite is true for the @xmath104 bands . therefore the center of the brillouin zone displays the greatest disparity in decay rate between each of the two bands per polarization , being either appreciably higher or lower than the result of a single nanoparticle . the preceding quantitative studies of both the radiative and landau decay rates in secs . [ sec : radiative ] and sec . [ sec : landau ] , respectively , allow us to investigate the optimal parameters of the bipartite chain of nanoparticles in order to maximize the lifetime of the collective plasmonic excitations . in fig . [ fig6b ] , we plot these aforementioned decay rates , scaled by the mie frequency @xmath27 , as a function of both @xmath215 ( @xmath216 denotes the fermi wavevector ) and the scaled momentum @xmath217 . the parameters used in the figure correspond to a chain of silver nanoparticles with @xmath218 , @xmath219{\aa}$ ] , @xmath220{nm}$ ] , and with interparticle separations @xmath221 and @xmath222 . the panels ( a)-(d ) in fig . [ fig6b ] display the competition between the radiative and landau losses for each of the four plasmonic bands : ( a ) the upper transverse band ( @xmath223 , @xmath104 ) , ( b ) the lower transverse band ( @xmath223 , @xmath103 ) , ( c ) the upper longitudinal band ( @xmath224 , @xmath104 ) and ( d ) the lower longitudinal band ( @xmath224 , @xmath104 ) . the drastic effects of radiative losses scaling with the nanoparticle size as @xmath225 [ see eq . ] can be seen in the bright plasmonic bands , shown in panels ( a ) and ( d ) , in stark contrast to the dark plasmonic bands , shown in panels ( b ) and ( c ) . the two bright bands also exhibit the striking feature of a complete supression of radiative losses outside the light cone @xmath226 , as was explicitly derived in eq . . the common feature across all four types of plasmonic band is the quantum size effect of strong landau damping for small nanoparticles , which graphically illustrates the result of eq . and its distinctive @xmath227 scaling . it may be seen from figs . [ fig6b](a),(d ) that the optimal size of nanoparticle to probe bright modes of collective plasmonic excitations ( of either polarization ) is of the order of @xmath228 , corresponding to @xmath229{nm}$ ] . an estimate of the total plasmonic lifetime can be made by considering @xmath230 , where the total damping rate @xmath231 includes the momentum - independent bulk ohmic decay rate @xmath232 . with a typical ohmic damping rate @xmath233 for silver nanoparticles @xcite , the bright band plasmonic lifetime is approximately @xmath234{fs}$ ] for @xmath235{nm}$ ] , such that experimental detection is a somewhat challenging task . however , an enhancement of the lifetime by a factor of two can be achieved by employing metallic nanoparticles with gain in order to compensate for ohmic losses @xcite . various methods of introducing gain into plasmonic systems exist , including embedding the nanoparticles in a dielectric environment with gain @xcite or by cladding the nanoparticle core with a gain material shell @xcite . therefore , we conclude that the phenomena discussed in this work are within experimental accessibility , particularly for setups with optimally sized metallic nanoparticles and appropriate gain media . in this theoretical work , we have analyzed in detail the fundamental properties of collective plasmonic excitations in bipartite chains of spherical metallic nanoparticles . we have revealed that this system harbours an unusual form of collective excitations : topological bosonic bogoliubov quasiparticles . near the edge of the brillouin zone , we have shown that these quasiparticles are described by a one - dimensional massive dirac hamiltonian . therefore , we suggest bipartite chains of plasmonic nanoparticles as an ideal testbed to experimentally probe 1d dirac - like bosons , which undergo exotic phenomena such as klein tunneling . whilst our proposal complements the hosting of 2d dirac - like bosons in honeycomb lattices @xcite , it also has the notable advantage that the bipartite linear lattice may be easier to realize experimentally . having derived the full quantum eigenstates describing the bosonic bogoliubov excitations , we were able to calculate the associated zak phase of the system . this has allowed us to characterize two topologically distinct regimes , depending on the dimerization of the chain , and foreshadowed the presence of plasmonic edge states . we have studied the various types of topologically - protected edge states that may be formed and derived transparent formulas describing their behavior , which opens up a new avenue to manipulate light at the nanoscale . since combating radiative and nonradiative losses is a considerable challenge in plasmonics , we have undertaken a detailed study of the decay rates of plasmonic excitations in bipartite nanoparticle chains . we have provided analytic formulas describing losses in the system due to both radiation damping , which has uncovered the existence of both bright and dark plasmonic bands , and landau damping , a quantum size effect fundamental to the study of quantum metamaterials . we have therefore been able to estimate the lifetime of the collective plasmonic excitations in the system and hence the optimal parameters required for experimental detection . with the current trend of technological advances in the fabrication of increasing small metallic nanoparticle arrays with gain media , the topological and dirac - like phenomena we have discussed should be within experimental reach . we are grateful to d. weinmann for illuminating discussions and for his careful reading of the manuscript . we thank t. j. sturges for technical support . we acknowledge financial support from the cnrs through the pics program ( contract no . 6384 apag ) and from the anr ( grant no . anr-14-ce26 - 0005 q - metamat ) . in the plasmonic hamiltonian , we consider the interaction between nearest neighbors on the 1d bipartite lattice . however , as the near - field dipolar interaction decays as one over the cube of the interparticle distance , it is necessary to check the robustness of our results against the inclusion of interactions beyond nearest neighbors only . in this appendix , we show that the interaction between the nearest neighbors alone captures the relevant physics of the problem and that the plasmon dispersion is only slightly quantitatively modified by interactions beyond nearest neighbors . in the following , we evaluate the collective plasmon dispersion classically by taking into account the interaction between an arbitrary number of neighbors on the bipartite lattice . specifically , we consider the dipole - dipole interaction between a nanoparticle on site @xmath236 belonging to sublattice @xmath204 and dimer @xmath16 up to sites @xmath237 and @xmath238)$ ] with @xmath239 and where @xmath240 @xmath241 for @xmath242 @xmath243 . the plasmonic dispersion can thus be straightforwardly evaluated to yield the general result @xmath244 with @xmath245 and @xmath246 ^ 3 \mathrm{e}^ { \mathrm{i } ( m-1 ) q d } \right . \nonumber\\ & + \left . \left[\frac{a}{(m-1)d+d_2 } \right]^3 \mathrm{e}^ { - \mathrm{i } m q d}\right\}.\end{aligned}\ ] ] , @xmath247 and @xmath248 , corresponding to the longitudinal collective plasmon polarization . ] we plot the bandstructure in fig . [ fig7 ] for different values of the number of neighbors taken into account ( for clarity , we show only the longitudinal mode @xmath224 , since similar conclusions can be drawn from the two transverse modes @xmath223 ) . the nearest neighbor case @xmath249 [ corresponding to eq . ] is shown by solid blue lines , whilst the cases with successively more nearest neighbors included , i.e. , @xmath250 , @xmath251 , and @xmath252 are represented by dashed green , dotted orange and dash - dotted magenta lines , respectively . as can be seen from the figure , dipole - dipole interactions beyond the nearest neighbor contribution do not lead to significant qualitative changes in the collective plasmon dispersion relation . quantitatively , when we discuss physics occurring close to the edge of the brillouin zone we should renormalize the resonance frequency by a small amount @xmath253 . for simplicity , we thus limit ourselves to the discussion of dipole - dipole interaction effects between nearest neighbors in the main text . in his appendix we consider the transmission problem of massive dirac - like collective plasmons at the edge of the first brillouin zone between two semi - infinite bipartite chains of metallic nanoparticles , denoted by l ( for left ) and r ( for right ) , and described by two different resonance frequencies @xmath119 , respectively . therefore , the four associated coupling constants arising from the dipole - dipole interaction @xmath254 are also ( in general ) different . for simplicity we here consider the case where the interparticle spacings @xmath115 and @xmath116 are the same in the left and right chains . the dirac - like collective plasmons at the edge of the first brillouin zone @xmath255 are described by the spinor , which pick up a label @xmath256 in each section of the bipartite chain , such that the quantity entering eq . becomes @xmath257 in our proposed setup , a notable contrast to the traditional klein potential step scenario is that the change in on - site energy between the left and right parts of the chain @xmath119 is intrinsically linked to a change in the mass term , as the coupling constants are directly proportional to @xmath119 . therefore , we effectively deal here with a combined ` electrostatic - mass step ' , meaning in the hamiltonian both the customary diagonal terms @xmath258 and the mass term @xmath259 change as the plasmonic excitation propagates from the left to the right section of the chain . the transmission probability @xmath260 can be calculated using the spinor wavefunctions describing incident , reflected and transmitted plasmonic excitations , respectively . from conservation of energy , the planewaves wavevectors are @xmath261 here @xmath262 is the frequency of the excitation propagating from left to right in a chain with @xmath263 , and @xmath264 arises to ensure conservation of the sign of the group velocity , which is not necessarily aligned with the wavevector . applying the boundary conditions of continuity of both wavefunction components of the eigenspinors at the domain interface ( @xmath265 ) and solving for the transmission coefficient , we arrive at the transmission probability @xmath266 , @xmath75 , @xmath267 , and @xmath248 . inset : sketch of the spectrum of the quasiparticles . the excitations may be transmitted in three regimes ( dotted lines ) corresponding to : ( i ) electron - like to electron - like , ( ii ) electron - like to hole - like and ( iii ) hole - like to hole - like states . ] we plot in fig . [ fig8 ] the transmission probability of a longitudinal excitation ( @xmath224 ) as a function of excitation frequency in a chain with @xmath263 . the problem can be partitioned into three different regimes of interest , as sketched in the inset of fig . [ fig8 ] . in region i ( iii ) , above @xmath268 ( below @xmath269 ) the tunneling is from electron ( hole)-like to electron ( hole)-like states . however , in the interval ii , where @xmath270 , the effective quasiparticle species changes from electron - like to hole - like , thus relying on the klein physics to transmit . most notably , due to the small effective masses the transmission probabilities are nearly always approaching unity , whilst no transmission takes place in the gapped regions . in the main text , we consider bipartite chains consisting of an even number of nanoparticles and described by the hamiltonian . while bulk properties do not depend on this restriction ( see secs . [ sec3 ] and [ sec4 ] ) , the formation of edge states is highly influenced by the parity of the chain . for the sake of completeness , we derive in this appendix expressions for the probability densities at each nanoparticle site corresponding to the midgap state at frequency @xmath27 for the case of a chain comprised of an odd number of nanoparticles . proceeding as in the case of an even number of nanoparticles ( cf . [ sec : edge_states ] ) but adding an @xmath2 site on the right side of the chain , we now deal with a symmetric tridiagonal matrix @xmath271 of odd order , which is always guaranteed to support one eigenvalue at the midgap frequency @xmath27 @xcite . here the boundary conditions arising at the edges of the chain are @xmath272 , which together with the recurrence relation for @xmath3 sites imply the complete absence of a midgap plasmonic mode on one half of the nanoparticle sites . on the @xmath2 sites , nonvanishing probability densities arise as follows : @xmath273 @xmath274 with @xmath275 $ ] . thus , highly localized collective plasmons occur at one edge only : either the left [ eq . ] or right [ eq . ] end of the chain , determined by the critical ratio @xmath276 , as exemplified in fig . such an outcome can be naturally explained with the aid of the sketches in the inset of of the figure . as one nanoparticle is always left over when the chain coalesce into dimers , depending on the coupling ratio @xmath276 one can choose the remaining unpaired nanoparticle , which primarily supports the edge state , to be at either end of the chain . at each site in a chain of 101 nanoparticles , for @xmath136 ( red dots ) and @xmath277 ( blue crosses ) . the thin gray lines are guides for the eye . inset : sketches of a typical chain comprised of an odd number of nanoparticles , where the loops depict the favored dimerization for @xmath278 ( top ) and @xmath139 ( bottom ) . ] d. schurig , j. j. mock , b. j. justice , s. a. cummer , j. b. pendry , a. f. starr , and d. r. smith , metamaterial electromagnetic cloak at microwave frequencies , http://dx.doi.org/10.1126/science.1133628[science * 314 * , 977 ( 2006 ) ] . d. j. bergman and m. i. stockman , surface plasmon amplification by stimulated emission of radiation : quantum generation of coherent surface plasmons in nanosystems , http://dx.doi.org/10.1103/physrevlett.90.027402[phys . lett . * 90 * , 027402 ( 2003 ) ] . r. f. oulton , v. j. sorger , d. a. genov , d. f. p. pile , and x. zhang , a hybrid plasmonic waveguide for subwavelength confinement and long - range propagation , http://dx.doi.org/10.1038/nphoton.2008.131[nat . photon . * 2 * , 496 ( 2008 ) ] . m. quinten , a. leitner , j. r. krenn , and f. r. aussenegg , electromagnetic energy transport via linear chains of silver nanoparticles , http://dx.doi.org/10.1364/ol.23.001331[opt . lett . * 23 * , 1331 ( 1998 ) ] . m. l. brongersma , j. w. hartman , and h. a. atwater , electromagnetic energy transfer and switching in nanoparticle chain arrays below the diffraction limit , http://dx.doi.org/10.1103/physrevb.62.r16356[phys . b * 62 * , r16356 ( 2000 ) . ] s. y. park and d. stroud , surface - plasmon dispersion relations in chains of metallic nanoparticles : an exact quasistatic calculation , http://dx.doi.org/10.1103/physrevb.69.125418[phys . b * 69 * , 125418 ( 2004 ) ] . j. del pino , j. feist , f. j. garcia - vidal , and j. j. garcia - ripoll , entanglement detection in coupled particle plasmons , http://dx.doi.org/10.1103/physrevlett.112.216805[phys . lett . * 112 * , 216805 ( 2014 ) ] . a. brandstetter - kunc , g. weick , c. a. downing , d. weinmann , and r. a. jalabert , nonradiative limitations to plasmon propagation in chains of metallic nanoparticles , phys . b ( to appear ) , https://arxiv.org/abs/1607.02999[arxiv:1607.02999 ] . j. r. krenn , a. dereux , j. c. weeber , e. bourillot , y. lacroute , j. p. goudonnet , g. schider , w. gotschy , a. leitner , f. r. aussenegg , and c. girard , squeezing the optical near - field zone by plasmon coupling of metallic nanoparticles , http://dx.doi.org/10.1103/physrevlett.82.2590[phys . 82 * , 2590 ( 1999 ) ] . s. a. maier , m. l. brongersma , p. g. kik , and h. a. atwater , observation of near - field coupling in metal nanoparticle chains using far - field polarization spectroscopy , http://dx.doi.org/10.1103/physrevb.65.193408[phys . b * 65 * , 193408 ( 2002 ) ] . s. a. maier , p. g. kik , h. a. atwater , s. meltzer , e. harel , b. e. koel , and a. a. g. requicha , local detection of electromagnetic energy transport below the diffraction limit in metal nanoparticle plasmon waveguides , http://dx.doi.org/10.1038/nmat852[nat . * 2 * , 229 ( 2003 ) ] . d. solis jr . , b. willingham , s. l. nauert , l. s. slaughter , j. olson , p. swanglap , a. paul , w - s . chang , and s. link , electromagnetic energy transport in nanoparticle chains via dark plasmon modes , http://dx.doi.org/10.1021/nl2039327[nano lett . * 12 * , 1349 ( 2012 ) ] . a. apuzzo , m. fevrier , r. salas - montiel , a. bruyant , a. chelnokov , g. lerondel , b. dagens , and s. blaize , observation of near - field dipolar interactions involved in a metal nanoparticle chain waveguide , http://dx.doi.org/10.1021/nl304164y[nano lett . * 13 * , 1000 ( 2013 ) ] . s. j. barrow , d. rossouw , a. m. funston , g. a. botton , and p. mulvaney , mapping bright and dark modes in gold nanoparticle chains using electron energy loss spectroscopy , http://dx.doi.org/10.1021/nl5009053[nano lett . * 14 * , 3799 ( 2014 ) ] . a. poddubny , a. miroshnichenko , a. slobozhanyuk , and y. kivshar , topological majorana states in zigzag chains of plasmonic nanoparticles , http://dx.doi.org/0.1021/ph4000949[acs photonics * 1 * , 101 ( 2014 ) ] . c. w. ling , m. xiao , c. t. chan , s. f. yu , and k. h. fung , topological edge plasmon modes between diatomic chains of plasmonic nanoparticles , http://dx.doi.org/10.1364/oe.23.002021[opt . express * 23 * , 2021 ( 2015 ) ] . a. kanshu , c. e. rter , d. kip , v. shandarov , p. p. believ , i. ili , and m. stepi , observation of discrete gap solitons in one - dimensional waveguide arrays with alternating spacings and saturable defocusing nonlinearity , http://dx.doi.org/10.1364/ol.37.001253[opt . lett . * 37 * , 1253 ( 2012 ) ] . c. poli , m. bellec , u. kuhl , f. mortessagne , and h. schomerus , selective enhancement of topologically induced interface states in a dielectric resonator chain , http://dx.doi.org/10.1038/ncomms7710[nat . * 6 * , 6710 ( 2015 ) ] . a. p. slobozhanyuk , a. n. poddubny , a. e. miroshnichenko , p. a. belov , and y. s. kivshar , subwavelength topological edge states in optically resonant dielectric structures , http://dx.doi.org/10.1103/physrevlett.114.123901[phys . 114 * , 123901 ( 2015 ) ] . a. blanco - redondo , i. andonegui , m. j. collins , g. harari , y. lumer , m. c. rechtsman , b. i. eggleton , and m. segev , topological optical waveguiding in silicon and the transition between topological and trivial defect states , http://dx.doi.org/10.1103/physrevlett.116.163901[phys . lett . * 116 * , 163901 ( 2016 ) ] . d. d. solnyshkov , a. v. nalitov , and g. malpuech , kibble - zurek mechanism in topologically nontrivial zigzag chains of polariton micropillars , http://dx.doi.org/10.1103/physrevlett.116.046402[phys . lett . * 116 * , 046402 ( 2016 ) ] . g. weick , c. woollacott , w. l. barnes , o. hess , and e. mariani , dirac - like plasmons in honeycomb lattices of metallic nanoparticles , http://dx.doi.org/10.1103/physrevlett.110.106801[phys . rev 110 * , 106801 ( 2013 ) ] . g. weick , r. a. molina , d. weinmann , and r. a. jalabert , lifetime of the first and second collective excitations in metallic nanoparticles , http://dx.doi.org/10.1103/physrevb.72.115410[phys . b * 72 * , 115410 ( 2005 ) ] . a. brandstetter - kunc , g. weick , d. weinmann , and r. a. jalabert , decay of dark and bright plasmonic modes in a metallic nanoparticle dimer , http://dx.doi.org/10.1103/physrevb.91.035431[phys . b * 91 * , 035431 ( 2015 ) ] ; http://dx.doi.org/10.1103/physrevb.92.199906[*92 * , 199906(e ) ( 2015 ) ] . it is known , e.g. , from studies of crystalline structure with more than one atom per unit cell [ see c. bena and g. montambaux , remarks on the tight - binding model of graphene , http://dx.doi.org/10.1088/1367-2630/11/9/095003[new j. phys . * 11 * , 095003 ( 2009 ) ] ] , that whilst the expectation values of operators describing physical quantities is independent of the basis chosen , the form of the operators may depend on the basis . here we made this choice of transformation to ensure that we arrive at the most transparent form of both the zak phase , which is dependent on the choice of the unit cell , and also the effective dirac hamiltonian which both arise later in this work . the very form of the bogoliubov transformation appearing in the hamiltonian exposes the unusual nature of the excitations as a nontrivial mixture of both creation and annihilation operators of the original plasmonic particles . indeed , the vacuum of the bogoliubov excitations @xmath281 is not the vacuum of the original @xmath282 plasmonic excitations . instead , @xmath283 involves pairs of @xmath282 excitations which are virtually excited by the dipolar interaction . r. jackiw and g. semenoff , continuum quantum field theory for a linearly conjugated diatomic polymer with fermion fractionization , http://dx.doi.org/10.1103/physrevlett.50.439[phys . lett . * 50 * , 439 ( 1983 ) ] . s. banerjee , j. fransson , a. m. black - schaffer , h. agren , and a. v. balatsky , granular superconductor in a honeycomb lattice as a realization of bosonic dirac material , http://dx.doi.org/10.1103/physrevb.93.134502[phys . b * 93 * , 134502 ( 2016 ) ] . g. weick , g .- ingold , r. a. jalabert , and d. weinmann , surface plasmon in metallic nanoparticles : renormalization effects due to electron - hole excitations , http://dx.doi.org/10.1103/physrevb.74.165421[phys . b * 74 * , 165421 ( 2006 ) ] . we note that these bosonic eigenstates are normalized according to an inner product with a metric , @xmath284 , where @xmath285 . this prescription arises due to the necessity of observing bosonic statistics , see refs . @xcite . a method to diagonalize a general bosonic quadratic hamiltonian , which utilizes paraunitary matrices in order to conform to the bosonic commutation relations , is described in refs . the required paraunitary matrices are obtained by a cholesky decomposition of the bosonic hamiltonian written in bogoliubov-de gennes form . charl , w. schulze , and b. winter , the size dependent shift of the surface plasmon absorption band of small spherical metal particles , http://dx.doi.org/10.1007/bf01427000[z.phys . d * 12 * , 471 ( 1989 ) ] . m. a. noginov , g. zhu , m. bahoura , j. adegoke , c. small , b. a. ritzo , v. p. drachev , and v. m. shalaev , the effect of gain and absorption on surface plasmons in metal nanoparticles , http://dx.doi.org/10.1007/s00340-006-2401-0[appl . b * 86 * , 455 ( 2007 ) ] . m. a. noginov , g. zhu , a. m. belgrave , r. bakker , v. m. shalaev , e. e. narimanov , s. stout , e. herz , t. suteewong , and u. wiesner , demonstration of a spaser - based nanolaser , http://dx.doi.org/10.1038/nature08318[nature * 460 * , 1110 ( 2009 ) ] .

the local histogram transform of an image is a data cube that consists of the histograms of the pixel values that lie within a fixed neighborhood of any given pixel location . such transforms are useful in image processing applications such as classification and segmentation , especially when dealing with textures that can be distinguished by the distributions of their pixel intensities and colors . we , in particular , use them to identify and delineate biological tissues found in histology images obtained via digital microscopy . in this paper , we introduce a mathematical formalism that rigorously justifies the use of local histograms for such purposes . we begin by discussing how local histograms can be computed as systems of convolutions . we then introduce probabilistic image models that can emulate textures one routinely encounters in histology images . these models are rooted in the concept of image occlusion . a simple model may , for example , generate textures by randomly speckling opaque blobs of one color on top of blobs of another . under certain conditions , we show that , on average , the local histograms of such model - generated - textures are convex combinations of more basic distributions . we further provide several methods for creating models that meet these conditions ; the textures generated by some of these models resemble those found in histology images . taken together , these results suggest that histology textures can be analyzed by decomposing their local histograms into more basic components . we conclude with a proof - of - concept segmentation - and - classification algorithm based on these ideas , supported by numerical experimentation . local histogram , occlusion , texture , classification , segmentation

1105.4069

a _ local histogram _ of an image is a histogram of the values of the pixels that lie in a neighborhood of a given pixel s location . it indicates the particular combination of pixel intensities or colors that appear in that neighborhood . when used as features in an image classification scheme , such histograms can help distinguish one texture from another . we , in particular , use them in automated segmentation - and - classification algorithms for digital microscope images of biological tissues . to be precise , the work presented here was motivated by the need to identify and delineate the various tissues exhibited in images of histological sections of teratoma tumors derived from embryonic stem cells , such as the one given in figure [ figure.histology](a ) . this image was provided by dr . carlos castro of the university of pittsburgh and dr . john a. ozolek of the children s hospital of pittsburgh , who grow and image such teratomas to gain greater insight into tissue development . in this image , which is purple - pink from hematoxylin and eosin ( h&e ) staining , even a layman can discern several distinct textures , each corresponding to a distinct tissue type . for each image under study , drs . castro and ozolek make use of their years of medical training and experience to identify what tissues are present , and to what degree . moreover , when provided with a point - and - click interface , they can manually segment the image according to tissue type , resulting in per - pixel labels such as those given in figure [ figure.histology](b ) . though straightforward for medical experts , such tasks are nevertheless tedious and time - consuming , leading to inconsistencies when working with large data sets . it is therefore our goal to automate as much of this process as is possible . our current algorithm is given in @xcite and builds upon previous work given in @xcite . our use of local histograms was motivated by the unique image features found in histology images of teratomas derived from primate embryonic stem cells . in layman s terms , these tumors begin as masses of undifferentiated cells that are implanted in laboratory animals . over time , these tumors grow and their cells differentiate into many various types bone , cartilage , skin , etc.until a point at which they are excised , sectioned , stained and viewed under a microscope , resulting in images such as the one in figure [ figure.histology](a ) . as such , these images exhibit a wide variety of tissue types , arranged in a seemingly random fashion . indeed , to a casual observer such images can appear as a jumbled mess . in truth however , the arrangement of these tissues is not completely random , and is rather the result of not yet well - understood biological mechanisms . castro and ozolek believe that by looking at many such images many sections of many teratomas they can gain greater insight into these mechanisms . here , spatial context is crucial : one must identify which particular tissue is present at any given point in order to estimate the total amount of each type , as well as the degree to which any given type is adjacent to other types . in light of these facts , we seek an algorithm which assigns a tissue label to each pixel , thereby segmenting ( delineating ) and classifying ( identifying ) the image at the same time . indeed , such an algorithm would be useful in a broad class of digital pathology applications beyond the teratoma problem @xcite . while designing such an algorithm , we must keep in mind that often no single pixel contains enough information to uniquely determine a label . rather , the decisions will be made based on features computed over some fixed neighborhood of every given pixel location . to determine which specific features to use , it helps to have a closer look at each individual tissue . for example , for the @xmath0 image given in figure [ figure.histology](a ) and thumbnailed in figure [ figure.histograms](a ) , we zoom in on three tissue types cartilage , connective tissue and pseudovascular tissue resulting in the @xmath1 subimages given in figure [ figure.histograms](b ) , ( c ) and ( d ) , respectively . each of these three tissue types exhibits a unique aperiodic texture . for instance , the cartilage texture can be regarded as a light purple field speckled with darker reddish - purple blobs ; each blob represents an individual cell s nucleus . meanwhile , connective tissue appears as dark purple blobs over a light pink field ; pseudovascular tissue is similar to connective tissue , but contains additional reddish - pink structures . in particular , these three textures exhibit distinct distributions of color , a fact which can quantitatively be confirmed by computing the two - dimensional histograms of their red - blue ( rb ) pixel value pairs , as depicted in figure [ figure.histograms](f ) , ( g ) and ( h ) . + as certain tissues can be distinguished from others based solely on the distributions of their pixel values , we propose to use histograms as image features in a segmentation - and - classification scheme . these histograms must be computed locally over a fixed neighborhood of every pixel location since global histograms , such as the one depicted in figure [ figure.histograms](e ) derived from figure [ figure.histograms](a ) , destroy spatial context by mixing all of the individual distributions together . a similar issue arises in time - frequency analysis : spectrograms preserve spatial context while fourier transforms do not . indeed , local histograms are philosophically similar to spectrograms : in a neighborhood of a given point , the local histogram transform estimates the _ frequency of occurrence _ of a given value while the spectrogram estimates frequency in the traditional sense . the purpose of this paper is to provide a mathematically rigorous justification for the use of local histograms in this fashion . to be precise , we regard our images as functions from a finite abelian group @xmath2 of pixel locations into a second finite abelian group @xmath3 of pixel values . that is , our images @xmath4 are members of the set @xmath5 . for example , the @xmath0 , @xmath6-bit red - green - blue ( rgb ) image given in figure [ figure.histology](a ) has @xmath7 and @xmath8 , where @xmath9 denotes the cyclic group of integers modulo @xmath10 . for purple - pink h&e - stained images , we often omit the green channel for the sake of computational efficiency , at which point @xmath3 becomes @xmath11 . the local histograms of an image @xmath4 are defined in terms of a _ weighting function _ , that is , a nonnegatively - valued @xmath12 whose values sum to one . specifically , the _ local histogram transform _ of @xmath4 with respect to @xmath13 is the function @xmath14 , @xmath15 where @xmath16 if @xmath17 and is otherwise zero . for any fixed @xmath18 , the corresponding cross - section of this function , namely @xmath19 , counts the number of instances at which @xmath4 obtains a given value @xmath20 in a @xmath13-neighborhood of @xmath21 . in this paper , we show that local histogram transforms are well - suited to the analysis of a particular class of textures . in short , we want a rigorous explanation of the following hypothesis : say for the sake of argument that 80% of the cartilage texture in figure [ figure.histograms](b ) consists of background " light purple pixels while the remaining 20% of pixels lie in a foreground " of darker - reddish purple blobs ; we then expect a local histogram computed over a portion of cartilage to be a mixture convex combination of @xmath22 of the background pixels distribution with @xmath23 of the foreground pixels distribution . other tissues arise from other distinct decompositions . for example , looking at the pseudovascular tissue of figure [ figure.histograms](d ) , we might guess it to be @xmath24 light pink , @xmath25 dark purple and @xmath25 reddish - pink . we rigorously show that such decompositions of local histograms indeed exist for textures arising from a certain class of probabilistic image models ; our long - term goal is to exploit this fact in a segmentation - and - classification algorithm . to see how to formalize these ideas , it helps to consider a toy example : imagine that at any given pixel location , a coin is flipped , with heads " resulting in a pink pixel value , and tails " resulting in a purple one . one expects that , on average , the local histogram at any point will consist of two peaks : one in the pink portion of @xmath3 , and one in the purple . such an image can be regarded as the result of _ occluding _ a solid purple image @xmath26 with a solid pink one @xmath27 : at each pixel , the flip of a coin determines whether @xmath27 lies on top of @xmath26 at that point , or vice versa . more generally , the _ occlusion _ of a set of @xmath10 images @xmath28 in @xmath29 with respect to a given _ label function _ @xmath30 is : @xmath31 that is , at any pixel location @xmath21 , the label @xmath32 determines which of the potential pixel values @xmath33 actually appears in the composite image @xmath34 at that point . the main results of this paper are concerned with when the local histograms of a composite image are related to the local histograms of the individual @xmath35 s . though it is unrealistic to expect a clean relation for any fixed @xmath36 , we can show that these quantities are indeed closely related , provided one averages over all possible label functions @xmath36 . indeed , denoting the probability of getting heads " in the above toy example as @xmath37 $ ] , we would expect the volumes of the pink and purple peaks of the composite image s local histograms to be @xmath38 and @xmath39 , respectively . that is , @xmath40 should be @xmath41 , on average . we generalize this idea so as to permit more realistic textures with more colors and with spatially - correlated pixels . to be precise , fix a set of source images @xmath28 and consider the set @xmath42 of all possible composite images obtained by letting @xmath36 be any one of the @xmath43 elements of @xmath44 , where @xmath45 denotes the cardinality of @xmath2 . we refer to a random method for choosing one of these composites as an _ occlusion model _ @xmath46 . formally speaking , @xmath46 is a random variable version of @xmath36 , meaning there exists a probability density function @xmath47 $ ] such that @xmath48 . for example , imagine three @xmath1 images @xmath26 , @xmath27 and @xmath49 which exhibit a nearly constant shade of pink , purple and red , respectively . given any label function @xmath50 we can produce a corresponding @xmath51 composite image @xmath52 whose pixels are some mixture of pink , purple and red . for some choices of @xmath36 the resulting composites will look like the pseudovascular tissue texture given in figure [ figure.histograms](d ) . however , even in this small example , there are an enormous number of such possible composites one for each of the possibilities for @xmath36and only a few of these will look like pseudovascular tissue ; most will appear as pink - purple - red static . the role of the occlusion model @xmath46 is to assign a probability to each of these possible @xmath36 s in a manner that emphasizes those textures one expects to appear in a given tissue while de - emphasizing the rest . in this paper , we provide a sufficient hypothesis on the occlusion model @xmath46 so as to ensure that the local histograms of a composite image can , on average with respect to @xmath53 , be decomposed in terms of the local histograms of the individual images . in particular , we focus on the special case where the occlusion model @xmath46 is _ flat _ , meaning that on average , the probability that @xmath46 chooses label @xmath54 at a given pixel location @xmath21 is equal to the probability of choosing @xmath54 at any other @xmath55 ; formally , @xmath46 is _ flat _ if there exists scalars @xmath56 such that : @xmath57 that is , @xmath46 is flat if the marginal distributions obtained by fixing any given @xmath18 are identical . note that for any fixed @xmath18 , summing over all @xmath54 yields that @xmath58 . indeed , at any given pixel location @xmath21 , the value @xmath59 represents the probability that the random label function @xmath46 will have label @xmath54 at that @xmath21 . in our toy example where the values of @xmath36 are determined by @xmath45 spatially - independent coin flips , the probability of getting any particular @xmath60 is @xmath61 ; substituting this expression into , the binomial theorem implies that this model is indeed flat with @xmath62 and @xmath63 . note that , if @xmath64 , the resulting random image @xmath65 will be more pink than purple ; flatness does not mean that each label is equally likely , but rather that the chance of being pink at any given pixel location is the same as at any other location . these concepts in hand , we present one of our main results , which formally claims that , on average , the local histograms of composite images produced from flat occlusion models are but mixtures of the local histograms of the source images : [ theorem.flatness ] if @xmath46 is flat as in , then the expected value of the local histogram transform of a composite image is a convex combination of the local histograms of each individual ] from the point of view of our motivating application , the significance of theorem [ theorem.flatness ] is that it gives credence to a certain type of segmentation - and - classification algorithm . to be precise , given a set of training images which are manually segmented and labeled by medical experts , we , for any given tissue type , can compute local histograms over regions which are labeled as that type . in light of theorem [ theorem.flatness ] , it is reasonable to _ demix_decompose into convex combinations the local histograms of that type into a set of more basic distributions . for example , we expect that the local histograms of pseudovascular tissue ( figure [ figure.histograms](d ) ) can be demixed into three simpler distributions one pink , another purple and a third reddish - pink while those of connective tissue ( figure [ figure.histograms](c ) ) are mixtures of only the first two . once sparse demixings of each tissue type have been found , we then use them to segment and classify : given a new image , we assign a label at any given point by determining which particular set of learned distributions its local histogram is most consistent with . the remainder of our main results are in support of this interpretation of theorem [ theorem.flatness ] . specifically , the next section contains several basic results on local histograms . in section [ section.occlusion ] , we prove theorem [ theorem.flatness ] and also a generalization of it theorem [ theorem.expected value of lh of occ]to the non - flat case . in section [ section.flatness ] , we provide various methods theorems [ theorem.pj ] , [ theorem.star product ] and [ theorem.overlay]for constructing flat @xmath46 s , and some of these produce textures that resemble those found in digital microscope images of histological tissues . the final section discusses a preliminary segmentation - and - classification algorithm inspired by theorem [ theorem.flatness ] in which local histograms are decomposed using principal component analysis ( pca ) . both local histograms and probabilistic image occlusion models have long been subjects of interest . theorem [ theorem.propertiesoflhs ] below details how local histograms can be computed as systems of convolutions ; a similar result is given in @xcite , and both @xcite and @xcite discuss how such a computation can be implemented in optical hardware . recently , local histograms have been used in an active contour - based segmentation scheme @xcite ; this algorithm partitions an image into two smoothly bounded regions whose pixel values are maximally separated with respect to the wasserstein ( earth mover s ) distance . local histograms have also recently been used as smoothing filters @xcite . though the work we present here focuses exclusively on local histograms of the pixel values themselves , an alternative approach is to first pass the image through a filter bank and then compute histograms of the resulting values @xcite . local histograms , like time - frequency transforms , preserve global spatial context while obscuring all local spatial context , and as such they are well - suited to the processing of _ locally orderless _ images @xcite . we use local histograms to analyze a class of textures generated by a certain probabilistic occlusion model ; this model , like the _ dead leaves model _ @xcite , generates these textures via a sequential superposition of random sets . our contribution to this body of literature is a formalism that unifies the theory of local histograms with that of occlusion models and permits us to rigorously prove that local histograms are indeed a useful transform for the analysis of a particular class of textures . in this section , we discuss an efficient means of computing local histograms and discuss several of their basic properties . computing local histograms can be time consuming , especially as @xmath2 and @xmath3 become large . in particular , for a general window @xmath13 , a direct computation of requires @xmath66 operations : @xmath67 operations for each @xmath18 and @xmath68 . a more efficient method is given in theorem [ theorem.propertiesoflhs ] below : can be computed as a system of @xmath69 convolutions over @xmath2 , which only requires @xmath70 operations if discrete fourier transforms are used . in particular , we filter the _ characteristic function _ of the graph of @xmath4 , namely @xmath71 , @xmath72 with the _ reversal _ of @xmath12 , namely @xmath73 . this method for computing local histograms is illustrated in figure [ figure.graph ] . an example of how to compute local histograms using theorem [ theorem.propertiesoflhs](a ) . for the sake of readability , larger numerical values are represented by darker shades throughout . the source image ( far left ) @xmath4 is @xmath74 and has grayscale values ranging from @xmath75 to @xmath76 . that , is @xmath77 where @xmath78 and @xmath79 . its characteristic function is a @xmath80-valued @xmath81 data cube whose cross - sections ( left column ) indicate those locations at which @xmath4 attains any given value . by theorem [ theorem.propertiesoflhs](a ) , the @xmath82 data cube that contains the local histograms of @xmath4 ( far right ) may be computed one level at a time ( right column ) by filtering these binary - valued cross - sections with a real - scalar - valued weighting function ( middle column ) . in this simple example , the weighting function is @xmath83 , where the origin lies in the upper left - hand corner of the grid . ] alternatively , can be computed as a single convolution over @xmath84 ; here , the tensor product of @xmath12 with @xmath85 is defined as @xmath86 , @xmath87 . [ theorem.propertiesoflhs ] for any @xmath12 , @xmath85 , @xmath77 , @xmath18 , and @xmath68 : 1 . local histograms can be evaluated as a system of @xmath69 convolutions over @xmath2 : @xmath88 . 2 . alternatively , may be computed as a single convolution over @xmath84 : @xmath89 . + in particular , taking @xmath90 gives @xmath91 . for ( a ) , replacing @xmath55 with @xmath92 , and substituting the relation @xmath93 into yields : @xmath94 for ( b ) , the definition of @xmath95 gives : @xmath96(x , y ) = \sum_{(x',y')\in{\mathcal{x}}\times{\mathcal{y}}}(\delta_0\otimes\omega)(x',y')({\mathrm{lh } } _ w f)(x - x',y - y ' ) = \sum_{y'\in{\mathcal{y}}}\omega(y')({\mathrm{lh } } _ w f)(x , y - y ' ) . \label{equation.3dconvresult}\ ] ] substituting into and using , gives our result : @xmath97(x , y ) & = \sum_{y'\in{\mathcal{y}}}\omega(y')({{\tilde { w}}*{1 _ { f^{-1}\{y - y'\}}}})(x)\\ \nonumber & = \sum_{y'\in{\mathcal{y}}}\omega(y')\sum_{x'\in{\mathcal{x}}}\tilde { w}(x')1 _ { f^{-1}\{y - y'\}}(x - x')\\ \nonumber & = \sum_{(x',y')\in{\mathcal{x}}\times{\mathcal{y}}}(\tilde { w}\otimes\omega)(x',y')1 _ f(x - x',y - y')\\ & = [ { { ( \tilde { w}\otimes\omega)}*{1 _ f}}](x , y).\qedhere\end{aligned}\ ] ] the next result summarizes several other basic properties of local histograms , the proofs of which are given in @xcite . [ proposition.lhresults ] for any @xmath12 and @xmath77 : 1 . the levels of a local histogram transform sum to @xmath98 : for any @xmath18 , @xmath99 . local histograms commute with spatial translation @xmath100 : for any @xmath18 , @xmath101 . 3 . adding constants to images shifts their local histograms along @xmath3 : for any @xmath68 , @xmath102 . 4 . quantizing an image will bin its local histograms : for any @xmath103 , @xmath104(x , y')=\sum_{\substack{y\in{\mathcal{y}}\\q(y)=y'}}({\mathrm{lh } } _ w f)(x , y).\ ] ] this basic understanding of local histograms in hand , we turn to the theory of applying them to textures generated by the probabilistic image occlusion models discussed in the introduction . in this section , we rigorously confirm our intuition regarding local histograms of textures generated via random occlusions : if a texture , such as that found in the pseudovascular tissue of figure [ figure.histograms](d ) , is some sufficiently - spatially - random combination of @xmath105 pink pixels , @xmath106 purple pixels and @xmath106 red pixels , then its local histograms should , on average , be a mixture of three simpler distributions , namely a convex combination of @xmath24 of a purely pink distribution with @xmath25 purely purple and red ones . to do this , fix any set of @xmath10 source images @xmath28 and let @xmath46 be any occlusion model as defined in the introduction . that is , let @xmath46 be a random variable version of a label function @xmath107 , as defined by a probability density function @xmath47 $ ] where @xmath48 . in the results that follow , a useful quantity to consider is the expected value with respect to @xmath53of the characteristic function @xmath108 obtained by letting @xmath109 in : @xmath110 essentially , @xmath111 is the probability that a random label function @xmath36 generated by the occlusion model @xmath46 will assign label @xmath54 to pixel location @xmath21 . when compared with the definition of flatness , we see that @xmath46 is flat if and only if there exist scalars @xmath56 such that @xmath112 for all @xmath18 and @xmath113 . that is , @xmath46 is flat if and only if @xmath111 is constant with respect to pixel location @xmath21 . having this concept , we present one of our main results : [ theorem.expected value of lh of occ ] for any sequence of images @xmath114 , weighting function @xmath13 and any @xmath10-image occlusion model @xmath46 , the expected value of the local histogram of the composite image with respect to @xmath13 is : @xmath115 where the error term @xmath116 is bounded by @xmath117 . moreover , @xmath118 and so states that , on average , the local histograms of the composite image @xmath119 can be approximated by convex combinations of local histograms of each individual image @xmath35 . the expected value of the local histogram of a composite image is : @xmath120 for any fixed @xmath36 , @xmath21 , and @xmath55 , we have @xmath121 for exactly one @xmath54 . for any fixed @xmath21 , @xmath55 and @xmath20 , we can therefore split a sum of @xmath122 over all @xmath54 into one summand where @xmath123 and the remaining @xmath124 summands for which @xmath125 : @xmath126 where the final equality follows immediately from . substituting into and using yields : @xmath127 rewriting in terms of @xmath128 $ ] gives our first claim : @xmath129 for the second claim , we bound @xmath116 using the triangle inequality and the fact that @xmath130 : @xmath131}\biggr| } \leq\sum_{n=0}^{n-1}\sum_{x'\in{\mathcal{x } } } w(x'){|{\overline1_\phi(x+x',n)-\overline1_\phi(x , n)}|}.\ ] ] finally , to prove our third claim , note that for any fixed @xmath18 , gives : @xmath132 since as previously noted we have @xmath133 for exactly one @xmath54 , becomes : @xmath134 . an example illustrating the direct computation of the left - hand side of is given in figure [ figure.elh ] . an example of how to compute the left - hand side of explicitly as a probability - weighted sum . for the sake of readability , larger numerical values are represented by darker shades throughout . we consider two @xmath135 , @xmath136-bit images , namely @xmath28 in @xmath29 where @xmath137 , @xmath138 and @xmath139 . in this particular example , the values of the @xmath35 s are all distinct , with @xmath26 assuming values @xmath140 and @xmath27 assuming values @xmath141 ( far left ) . there are distinct label functions @xmath142 ( left column ) each yielding a composite image @xmath143 ( center column ) ; in accordance with , we take values from @xmath26 in places where @xmath36 is white and values from @xmath27 where @xmath36 is black . each of these composites has a @xmath144 local histogram transform ( right column ) . since occlusion is nonlinear , there is no clean relationship between the local histograms of any single composite and the local histograms of the source images @xmath26 and @xmath27 . nevertheless , under certain hypotheses , we can say something about the average of these local histograms ( far right ) with respect to some probability density function @xmath53 on the set @xmath145 of all possible @xmath36 s . in particular , if the occlusion model @xmath46 is flat , meaning in this case that the probability - weighted - sum of all @xmath36 s is a constant function of @xmath21 , then theorem [ theorem.flatness ] states that this average is a convex combination of the local histograms of @xmath26 and @xmath27 as depicted in figure [ figure.elh2 ] . ] note that theorem [ theorem.expected value of lh of occ ] implies that the error term @xmath116 in will be small provided the probability @xmath111 of assigning label @xmath54 to @xmath21 changes little as @xmath21 varies over regions smaller than the the support of @xmath13 . the extreme case of this is when the occlusion model @xmath46 is flat , meaning @xmath111 is constant with respect to @xmath21 . in this case , @xmath116 vanishes entirely , leading to theorem [ theorem.flatness ] as given in the introduction : if @xmath46 is flat , @xmath146 for all @xmath147 . the error bound in theorem [ theorem.expected value of lh of occ ] then gives @xmath148 . denoting @xmath111 as @xmath59 in thus yields our claim . that is , when @xmath46 is flat , simplifies to , and so the in - depth computation of figure [ figure.elh ] can be replaced by the much simpler one depicted in figure [ figure.elh2 ] . a continuation of the example of figure [ figure.elh ] . when the occlusion model @xmath46 is flat , theorem [ theorem.expected value of lh of occ ] becomes theorem [ theorem.flatness ] , with simplifying to . though each of the @xmath149 distinct composite images shown in figure [ figure.elh ] has a distinct local histogram transform , the average of these @xmath149 local histogram transforms with respect to @xmath53 is but a convex combination ( right ) of the local histograms ( center ) of the two source images ( left ) . that is , when @xmath46 is flat , the average - over - all - composites local histogram computed in figure [ figure.elh ] is equal to the average - over - all - sources local histogram computed above.,title="fig : " ] thus , flatness is indeed an important theoretical assumption for the analysis of local histograms of textures generated via random occlusions . it nevertheless remains to be shown that flatness is also a realistic assumption from the point of view of our motivating application ; this is the topic of the next section . theorem [ theorem.flatness ] gives some insight into the behavior of the local histograms of images generated via random occlusions . however , this result only holds when @xmath46 is flat , namely when its average characteristic function @xmath111 , as defined in , is constant with respect to pixel location @xmath21 , but is still permitted to vary with label value @xmath54 . in this section , we demonstrate that flatness is a reasonable assumption . in particular , we provide a variety of methods for constructing flat occlusion models . some of these models produce textures similar to those encountered in digital microscope images of histological tissues . our first method involves the translation operator @xmath150 , @xmath151 . to be precise , we show that an occlusion model @xmath46 is flat if it is _ translation - invariant _ , meaning that its probability density function @xmath53 satisfies : @xmath152 [ theorem.pj ] if @xmath46 is translation - invariant , then @xmath46 is flat . we begin by placing an equivalence relation @xmath153 on @xmath44 , letting @xmath154 when there exists some @xmath18 such that @xmath155 . letting @xmath156 denote a set of representatives from the corresponding equivalence classes , we have that for all @xmath157 , there exists a unique @xmath158 such that @xmath155 . as such , @xmath159 now , fix any @xmath158 , and consider the subgroup @xmath160 of the finite abelian group @xmath2 . letting @xmath161 denote a fixed set of coset representatives of @xmath2 with respect to @xmath162 , we claim that @xmath163 , @xmath164 is a bijection . indeed , to show @xmath165 is one - to - one , note that if @xmath166 , then @xmath167 , implying @xmath168 ; since @xmath21 and @xmath55 are both coset representatives of @xmath161 , this is a contradiction unless @xmath169 . meanwhile , to show @xmath165 is onto , take any @xmath154 , and consider a corresponding @xmath55 such that @xmath170 . taking the unique @xmath171 and @xmath172 such that @xmath173 , we have : @xmath174 . invoking this claim , along with the assumed translation - invariance of @xmath46 , yields : @xmath175 again , writing any @xmath176 as @xmath173 , where @xmath171 and @xmath172 , gives : @xmath177 substituting into gives : @xmath178 since @xmath179 substituting into gives : @xmath180 implying @xmath46 is flat , since the value of @xmath111 depends only on @xmath54 and is independent of @xmath21 . theorem [ theorem.pj ] indicates that flatness is not too strong of an assumption . indeed , one method for producing a flat model @xmath46 is to generalize the coin - flipping example given in the introduction : given any random method for picking a number from @xmath9a probability spinner produce @xmath36 by conducting @xmath45 independent spins . the resulting model @xmath46 is translation - invariant , and therefore flat , since @xmath181 is solely determined by the number of times that @xmath36 achieves each given value @xmath54 . other translation - invariant examples abound . for instance , for any fixed @xmath182 , we can assign equal probability to @xmath182 and each of its translates , and assign probability @xmath75 to all others ; if the source images @xmath183 are constant , the composite images produced by such a model are all translates of a single image . more generally , we can always partition the @xmath43 elements of @xmath44 into translation - invariant equivalence classes and assign any fixed probability to the members of each class , provided we ensure that in the end they all sum to one . for example , for the case @xmath137 and @xmath138 depicted in figure [ figure.elh ] , we may partition the @xmath149 possible @xmath36 s into @xmath184 such classes , and pick any probabilities @xmath185 such that @xmath186 , @xmath187 , @xmath188 , @xmath189 , @xmath190 . armed with one method translation - invariance for producing flat models @xmath46 , we now turn to ways of combining known models to produce more complicated and realistic ones . digital microscope images of histological tissues often contain randomly distributed blobs . these blobs correspond to biological structures : cells , nuclei , etc . the nature of these processes guarantees that the distribution of such structures is roughly uniform , both spatially and in terms of color : two cells can not occupy the same space ; cells will usually grow and reproduce so as to occupy any empty space ; cells in a given tissue all have approximately the same size and color patterns . we want to construct flat occlusion models that emulate such textures , since in light of theorem [ theorem.flatness ] , doing so would formally justify the demixing of local histograms as part of a segmentation - and - classification algorithm . note that there is a natural method for randomly generating a set of roughly uniformly - distributed points : flip a coin at each point @xmath21 . here , we explore the idea of _ expanding _ each of these randomly generated points into a given blob . to be precise , let @xmath60 indicate a set of randomly generated points . for each of the points @xmath18 for which @xmath191 , we will replace it with a blob whose shape is indicated by some @xmath192 . the new texture will be the union of all such blobs . formally , given any @xmath60 and @xmath193^{\mathcal{x}}$ ] , we define the _ expansion _ of @xmath36 by @xmath194 to be @xmath195 , @xmath196 two examples of this expansion operation are given in figure [ figure.expansion ] . + note that expansion itself is not an occlusion model . indeed , is but a way of combining functions in @xmath197 to produce other ones , whereas an occlusion model is a random variable @xmath46 defined by a probability density function @xmath53 over @xmath197 . this fact notwithstanding , the expansion operation on label functions @xmath36 and @xmath194 does in fact induce a parallel operation on their random variable cousins @xmath46 and @xmath198 . to be precise , given two occlusion models @xmath46 and @xmath198 from @xmath2 into @xmath199 , we define the expansion of @xmath46 by @xmath198 to be the occlusion model @xmath200 whose probability density function is @xmath201 $ ] , @xmath202^{\mathcal{x}}\\{{\varphi}\star{\{\psi_x\}_{x\in{\mathcal{x}}}}}=\sigma}}{\mathrm{p}}_\phi(\varphi)\prod_{x\in{\mathcal{x}}}{\mathrm{p}}_\psi(\psi_x).\ ] ] note that the probability that @xmath200 will produce a given label function @xmath203 depends on the ways in which @xmath203 can be written as @xmath204 and , moreover , the probability that @xmath46 and @xmath198 will produce those particular @xmath36 s and @xmath205 s , respectively . in the next result , we verify that indeed defines a probability density function on @xmath197 . we further show that if @xmath46 is translation - invariant , then @xmath200 is translation - invariant which implies that @xmath200 is flat by theorem [ theorem.pj ] . in particular , image models which produce collections of blobs similar to those found in biological tissues will indeed be flat provided the distribution that produces the centers " of these blobs is translation - invariant . moreover , if the flatness of @xmath200 is all that is desired , we can weaken the requirement that @xmath46 be translation - invariant so as to only require that @xmath46 is itself flat , provided @xmath46 and @xmath198 are _ effectively disjoint _ : @xmath206 put another way , means that there is only at most one way , with nontrivial probability , in which the @xmath21 in can be written as @xmath207 where both @xmath208 and @xmath209 . [ theorem.star product ] if @xmath46 and @xmath198 are occlusion models from @xmath2 into @xmath199 , then their expansion @xmath200 , with probability density function , is as well . moreover , if @xmath46 is translation - invariant , then @xmath200 is translation - invariant . furthermore , if @xmath46 and @xmath198 are effectively disjoint and either @xmath46 or @xmath198 is flat , then @xmath200 is flat . we first show that defines a probability density function , namely that values of @xmath210 over all @xmath203 in @xmath197 sum to @xmath98 . since @xmath53 is a probability density function by assumption , we have : @xmath211 similarly , for any fixed @xmath18 , we have : @xmath212 where the subscript @xmath21 " on @xmath213 indicates that this particular @xmath213 is intended to expand @xmath36 at the particular point @xmath21 as opposed to at some other point . taking the product of with the product of over all @xmath21 yields : @xmath214^{\mathcal{x}}}}{\mathrm{p}}_\phi(\varphi)\prod_{x\in{\mathcal{x}}}{\mathrm{p}}_\psi(\psi_x),\ ] ] where the final quantity in contains all of the cross terms resulting from distributing the product over all sums of the form . now , since for each choice of @xmath36 and @xmath194 there is exactly one resulting @xmath215 , we can rewrite in terms of the definition of @xmath216 , obtaining our claim : @xmath217^{\mathcal{x}}\\{{\varphi}\star{\{\psi_x\}_{x\in{\mathcal{x}}}}}=\sigma}}{\mathrm{p}}_\phi(\varphi)\prod_{x\in{\mathcal{x}}}{\mathrm{p}}_\psi(\psi_x ) = \sum_{\sigma\in\ell({\mathcal{x}},{\mathbb{z}}_2)}{\mathrm{p}}_{{{\phi}\star{\psi}}}(\sigma).\ ] ] thus , indeed defines a probability density function , as claimed . we next show that the occlusion model @xmath200 is translation - invariant , if @xmath46 is translation - invariant . to do this , we claim that if @xmath218 then @xmath219 . to see this claim , note that @xmath220 if and only if there exists some @xmath55 , @xmath221 in @xmath2 such that @xmath207 , @xmath208 , and @xmath209 . letting @xmath222 , we thus have that @xmath223 if and only if @xmath224 , where @xmath225 and @xmath226 , implying @xmath219 , as claimed . having the claim , implies : @xmath227^{\mathcal{x}}\\{{\varphi}\star{\{\psi_x\}_{x\in{\mathcal{x}}}}}={\mathrm{t}}^{\tilde{x}}\sigma}}{\mathrm{p}}_\phi(\varphi)\prod_{x\in{\mathcal{x}}}{\mathrm{p}}_\psi(\psi_x ) = \sum_{\substack{\varphi\in\ell({\mathcal{x}},{\mathbb{z}}_2)\\\{\psi_x\}_{x\in{\mathcal{x}}}\in[\ell({\mathcal{x}},{\mathbb{z}}_2)]^{\mathcal{x}}\\{{({\mathrm{t}}^{-\tilde{x}}\varphi)}\star{\{\psi_{x+\tilde{x}}\}_{x\in{\mathcal{x}}}}}=\sigma}}{\mathrm{p}}_\phi(\varphi)\prod_{x\in{\mathcal{x}}}{\mathrm{p}}_\psi(\psi_x).\ ] ] to continue , we make the change of variables @xmath228 and @xmath229 : @xmath230^{\mathcal{x}}\\{{(\varphi')}\star{\{\psi'_x\}_{x\in{\mathcal{x}}}}}=\sigma}}{\mathrm{p}}_\phi({\mathrm{t}}^{\tilde{x}}\varphi')\prod_{x\in{\mathcal{x}}}{\mathrm{p}}_\psi(\psi'_{x-\tilde{x}}).\ ] ] since @xmath46 is translation - invariant and @xmath231 , we have : @xmath230^{\mathcal{x}}\\{{(\varphi')}\star{\{\psi'_x\}_{x\in{\mathcal{x}}}}}=\sigma}}{\mathrm{p}}_\phi(\varphi')\prod_{x\in{\mathcal{x}}}{\mathrm{p}}_\psi(\psi'_{x } ) = { \mathrm{p}}_{{{\phi}\star{\psi}}}(\sigma),\ ] ] and so @xmath200 is indeed translation - invariant , as claimed . for our final claim , we assume that @xmath46 and @xmath198 are effectively disjoint and that either @xmath46 or @xmath198 is flat . to do so , it is helpful to characterize the flatness of an arbitrary occlusion model @xmath46 from @xmath2 to @xmath199 in terms of the corresponding function @xmath232 . indeed , for any @xmath233 , may be rewritten as @xmath234 and so : @xmath235 in light of , we claim that @xmath46 is flat if and only if @xmath236 is constant . indeed , if @xmath46 is flat , then there exists @xmath237 such that @xmath238 for all @xmath18 . conversely , if @xmath239 is constant , then there exists @xmath237 such that @xmath240 for all @xmath18 ; by , this further implies that @xmath241 for all @xmath18 and so @xmath46 is flat . having this claim , we show that @xmath200 is flat by showing that @xmath242 is constant . to do this , we show that if @xmath46 and @xmath198 are effectively disjoint then @xmath243 where @xmath244 " denotes standard convolution over @xmath2 . according to the definition of @xmath200 we have : @xmath245^{\mathcal{x}}\\{{\varphi}\star{\{\psi_x\}_{x\in{\mathcal{x}}}}}=\sigma}}{\mathrm{p}}_\phi(\varphi){\biggl({\prod_{x\in{\mathcal{x}}}{\mathrm{p}}_\psi(\psi_x)}\biggr)}({{\varphi}\star{\{\psi_x\}_{x\in{\mathcal{x}}}}}).\ ] ] since any particular choice of @xmath36 and @xmath246 produces a unique @xmath203 via @xmath247 we can simplify to @xmath248^{\mathcal{x}}}}{\mathrm{p}}_\phi(\varphi){\biggl({\prod_{x\in{\mathcal{x}}}{\mathrm{p}}_\psi(\psi_x)}\biggr)}({{\varphi}\star{\{\psi_x\}_{x\in{\mathcal{x}}}}}).\ ] ] moreover , since @xmath46 and @xmath198 are effectively disjoint we have @xmath249 meaning becomes : @xmath250^{\mathcal{x}}}}{\mathrm{p}}_\phi(\varphi){\biggl({\prod_{x\in{\mathcal{x}}}{\mathrm{p}}_\psi(\psi_x)}\biggr)}{\biggl({\sum_{\substack{x'\in{\mathcal{x}}\\\varphi(x')=1}}{\mathrm{t}}^{x'}\psi_{x'}}\biggr)}\\ \label{equation.proof of star product 7 } & = \sum_{\varphi\in\ell({\mathcal{x}},{\mathbb{z}}_2)}{\mathrm{p}}_\phi(\varphi)\sum_{\substack{x'\in{\mathcal{x}}\\\varphi(x')=1}}{\mathrm{t}}^{x'}{\biggl[{\sum_{\{\psi_x\}_{x\in{\mathcal{x}}}\in[\ell({\mathcal{x}},{\mathbb{z}}_2)]^{\mathcal{x}}}{\biggl({\prod_{x\in{\mathcal{x}}}{\mathrm{p}}_\psi(\psi_x)}\biggr)}\psi_{x'}}\biggr]}.\end{aligned}\ ] ] now , for any fixed @xmath176 such that @xmath208 , we factor the corresponding innermost sum in into a product of @xmath45 distinct sums one for each @xmath18to obtain : @xmath251^{\mathcal{x}}}{\biggl({\prod_{x\in{\mathcal{x}}}{\mathrm{p}}_\psi(\psi_x)}\biggr)}\psi_{x ' } = { \biggl[{\prod_{x\neq x'}{\biggl({\sum_{\psi_x\in\ell({\mathcal{x}},{\mathbb{z}}_2)}{\mathrm{p}}_\psi(\psi_x)}\biggr)}}\biggr]}\sum_{\psi_{x'}\in\ell({\mathcal{x}},{\mathbb{z}}_2)}{\mathrm{p}}_\psi(\psi_{x'})\psi_{x ' } = { \biggl({\prod_{x\neq x'}1}\biggr)}\overline\psi = \overline\psi.\ ] ] substituting into then gives : @xmath252 thus , the effective disjointness of @xmath46 and @xmath198 indeed implies @xmath243 . as such , if we further assume that either @xmath46 or @xmath198 is flat , then either @xmath236 or @xmath253 is constant , implying in either case that @xmath242 is constant and so @xmath200 is flat . above , we discussed how the expansion of a binary - valued occlusion model @xmath46 with another such model @xmath198 is a new model @xmath200 that randomly generates label functions of the form @xmath215 as defined in . under certain hypotheses , theorem [ theorem.star product ] gives that such models @xmath200 are flat , meaning their local histograms can be understood in terms of theorem [ theorem.flatness ] . moreover , some examples of these models produce textures that resemble those encountered in histological tissues : if @xmath26 and @xmath27 are roughly constant light purple and dark purple fields , respectively , then the composite image @xmath143 obtained by picking @xmath36 as in figure [ figure.expansion](f ) bears some similarity to an actual image of cartilage , such as the one given in figure [ figure.histograms](b ) . taken together , these facts provide some theoretical justification for the use of local histograms for the analysis of such tissues . there is however a deficit with this theory : due to the nature of the construction , models produced by expansion can only be binary - valued , and as such are insufficient to emulate textures that exhibit three or more distinct color modes , such as the pseudovascular tissue depicted in figure [ figure.histograms](d ) . in this subsection , we discuss a method for _ laying _ one occlusion model over another which , amongst other things , permits us to build multivalued models out of binary - valued ones . to be precise , for any @xmath254 , @xmath255 and @xmath256 , we define the _ overlay _ of @xmath36 over @xmath213 with respect to @xmath203 to be @xmath257 , @xmath258 essentially , an overlay is the result of cutting holes out of an image of @xmath36 and laying it on top of an image of @xmath213 ; the location of these holes is indicated by @xmath203 and the values of @xmath213 are increased by a factor of @xmath259 so that they can not be confused with those of @xmath36 . examples of this overlay operation are given in figure [ figure.overlay ] . + in a manner similar to the relationship between and , we have that naturally induces a parallel operation on occlusion models : given probability density functions @xmath53 , @xmath260 and @xmath261 on @xmath262 , @xmath263 and @xmath264 , respectively , we define the overlay of the occlusion model @xmath46 over @xmath198 with respect to @xmath265 to be the new occlusion model @xmath266 whose probability density function is @xmath267 $ ] , @xmath268 in the next result , we verify that indeed defines a probability density function , and moreover that the corresponding model @xmath266 is flat provided @xmath46 , @xmath198 and @xmath265 are flat , meaning that the local histograms of composite images produced by such a model will behave according to theorem [ theorem.flatness ] . [ theorem.overlay ] if @xmath46 , @xmath198 and @xmath265 are occlusion models on @xmath262 , @xmath263 and @xmath264 , respectively , then defines a probability density function on @xmath269 . moreover , if @xmath46 , @xmath198 , and @xmath265 are flat , then @xmath270 is flat . to show that defines a probability density function on @xmath269 , note that : @xmath271 noting that for each fixed @xmath36 , @xmath213 , and @xmath203 , there exists exactly one @xmath272 such that @xmath273 , becomes : @xmath274 as claimed . for the second conclusion , assume that @xmath46 , @xmath198 , and @xmath265 are flat . our goal is to show that @xmath270 is flat , meaning that for any @xmath275 , we want to show that there exists a scalar @xmath59 such that : @xmath276 for all @xmath18 . to see this , note that for any such @xmath21 and @xmath54 , we have : @xmath277 now , in the special case where @xmath278 , gives that @xmath279 if and only if @xmath133 and @xmath280 . as such , in this case becomes : @xmath281 if , on the other hand @xmath282 then gives that @xmath279 if and only if @xmath283 and @xmath284 . in this case , becomes : @xmath285 thus , for any @xmath18 we either have or meaning @xmath270 is flat , as claimed . in this section , we present a proof - of - concept segmentation - and - classification scheme that is inspired by theorem [ theorem.flatness ] . we emphasize that for the algorithm presented here , local histograms are the only image features that are computed . that is , the decision of which label to assign to a given pixel is based purely on the distribution of color in its surrounding neighborhood . we do this to demonstrate the validity of the concept embodied by theorem [ theorem.flatness ] as an image processing tool . for algorithms intended for real - world use , such color information should be combined with morphological data size , local and global shape , orientation and organization in order to obtain better classification accuracies . an example of such an algorithm , accompanied by thorough testing and comparisons against other state - of - the - art methods , is given in the sister article @xcite to this one ; these facts are not reprinted here . the concept of theorem [ theorem.flatness ] is that the local histograms of certain textures should , on the whole , be able to be decomposed in terms of more basic distributions . indeed , it is reasonable to expect a local histogram computed over a region of cartilage ( figure [ figure.histograms](b ) ) to be a mixture of @xmath22 of a light purple " distribution a distribution mostly supported in portions of @xmath3 that correspond to light purple with @xmath23 of a darker reddish - purple one . meanwhile , local histograms of other tissues will correspond to distinct mixtures of other distributions . for example , local histograms computed over a region of pseudovascular tissue ( figure [ figure.histograms](d ) ) might be a mixture of @xmath24 of a light pink distribution , with @xmath25 of a dark purple one and @xmath25 of a reddish - pink one . the algorithm we present here exploits this concept . the first step is to train our classifier . to do so , let @xmath286 be the number of distinct tissue types found in a training image such as figure [ figure.classification](a ) or ( d ) . for any tissue type @xmath287 , we compute local histograms @xmath288 about pixel locations @xmath289 that have been labeled as being of that type by medical experts . each @xmath290 is a nonnegatively - valued function over @xmath3 that sums to one . there are several ways to pick the @xmath291 s . one approach is to have the expert choose each point individually . alternatively , if the expert has manually segmented and labeled the entire image ( figure [ figure.classification](b ) ) , then the @xmath291 s can be chosen at random from regions of type @xmath292 . the number @xmath293 of local histograms that we compute for type @xmath292 is somewhat arbitrary ; we used repeated experimentation to find a sample size large enough to guarantee reliably - decent performance . + in light of theorem [ theorem.flatness ] , it would be nice to demix the training local histograms @xmath288 in terms of a type - dependent class of more basic distributions . that is , we would like to find nonnegatively - valued functions @xmath294 over @xmath3 that sum to one and have the property that for each training local histogram @xmath290 there exists nonnegative scalars @xmath295 that themselves sum to one and such that : @xmath296 unfortunately , computing the @xmath294 s that minimize the approximation error in is a nontrivial optimization problem . as such , we leave this approach for future work , and instead consider a mathematically - simpler problem in which the @xmath297 s and @xmath294 s are permitted to be arbitrary real scalars and vectors , respectively . that is , we perform pca for each tissue type @xmath292 . to be precise , for each type , we form a @xmath298 matrix @xmath299 whose columns are the ( vectorized ) local histograms @xmath290 less their average @xmath300 : @xmath301 we then compute the singular value decompositions @xmath302 and identify those left - singular vectors that correspond to some experimentally - determined number @xmath303 of dominant singular values @xmath304 . in this setting , the approximation is replaced by : @xmath305 the classical theory of pca states that the approximation error in is optimally small in the sense that these specific @xmath306 s span the particular @xmath303-dimensional subspace of @xmath307 whose orthogonal projection operator @xmath308 minimizes the total squared - error @xmath309 . the vectors @xmath300 and in hand , we store them in memory , completing the training phase of our classification algorithm . to segment and label a given image @xmath310 , we compute its local histograms , obtaining local distributions of color @xmath311 , @xmath312 about every pixel location @xmath18 . at any given @xmath21 , we then assign a tissue label @xmath313 by finding the tissue type @xmath292 whose shifted subspace @xmath314 is nearest to @xmath315 . specifically , we let : @xmath316 in implementation , we compute the summations over @xmath3 in as running sums , looping over all @xmath68 . this computational trick greatly reduces our memory requirements : at any given time , we only store a single level of @xmath317 . by theorem [ theorem.propertiesoflhs ] , such a level can be obtained by filtering an indicator function ; in the following experimental results , we avoided edge artifacts by using a weighted noncyclic method of filtering , namely the @xmath247-convolution of @xcite . without such a trick , one must store the entire local histogram transform in memory , a daunting task for even modestly - sized images : the full local histogram transform of the @xmath0 , 8-bit rgb image given in figure [ figure.histology](a ) is a @xmath318 array . further computational advantages may be gained by quantizing the image and reducing the dimension of the color space . for our particular set of histology images , we experimentally found that we could still obtain good accuracies even if we discard the green channel of our purple - pink images , and moreover quantize the @xmath6-bit red and blue channels down to @xmath136-bits apiece . that is , we quantize @xmath3 from @xmath319 to @xmath320 . by proposition [ proposition.lhresults ] , this is equivalent to binning the original @xmath318 local histogram array down to a new one of size @xmath321 . the quantized version of figure [ figure.histology](a ) is given in figure [ figure.classification](a ) ; for the sake of readability , a @xmath136-bit quanitized version of the unused green channel was included in this rendering . as a result of this quantization , it only takes a few seconds to assign per - pixel labels to a @xmath0 histology image using a matlab - based implementation of , running on standard desktop hardware . for this particular set of images , further color quantization , such as using @xmath322-bit colors ( @xmath323 ) or converting the original image to grayscale ( @xmath324 ) , results in an unacceptable loss in classification accuracy , as do attempts at spatial quantization ( @xmath325 ) . two runs of this classification algorithm are depicted in figure [ figure.classification ] . in the first run , we train the classifier on the @xmath136-bit @xmath0 red - blue image given in figure [ figure.classification](a ) . for the sake of simplicity , we restrict ourselves to @xmath326 tissue types : cartilage , connective tissue and pseudovascular tissue ; all other tissue types are ignored in the confusion matrices given below . for each type @xmath327 , we randomly choose @xmath328 points of that type , making use of a small number of the @xmath329 ground truth labels given in figure [ figure.classification](b ) ; edge artifacts are avoided by not picking points near the border . for each type , we then perform pca on the @xmath330 local histograms @xmath290 of that type , computing an average local histogram @xmath300 as well as the dominant left - singular vectors of @xmath299 . for the sake of simplicity , in a given experiment we will use the same number of principal components for each of the three types , that is , @xmath331 for @xmath327 . at the same time , we experiment with this number itself , letting @xmath10 be either @xmath98 , @xmath322 , @xmath136 or @xmath76 . with the training complete , we then segment and classify figure [ figure.classification](a ) using the decision rule , resulting in per - pixel labels such as the ones given in figure [ figure.classification](c ) for @xmath332 . comparing figure [ figure.classification](c ) and the ground truth of figure [ figure.classification](b ) , we see both the power and limitations of local histograms : color is a big factor in determining tissue type , but by ignoring shape , we suffer from oversmoothing . the accuracy percentages for various choices of @xmath10 are given by a confusion matrix : @xmath333 here each row of the matrix tells us the percentage a certain tissue was labeled as cartilage ( ca ) , connective tissue ( co ) , and pseudovascular tissue ( ps ) . in particular , the first three entries of the first row of this table tell us that when using a single principal component , those points labeled as cartilage by a medical expert in figure [ figure.classification](b ) are correctly labeled as such by our algorithm @xmath334 of the time , while @xmath335 of it is mislabeled as connective tissue and @xmath336 of it is mislabeled as pseudovascular tissue . note here that we have trained and tested on the same image ; such experiments indicate the feasibility of our approach in a semi - automated classification scheme in which a medical expert handpicks @xmath330 points of each given type and lets the algorithm automatically assign labels to the rest . the second run of this algorithm is almost identical to the first , with the exception that we use a distinct image in the training phase . to be precise , for each of the three tissue types , we perform pca on the local histograms of @xmath330 randomly - chosen points of that type in figure [ figure.classification](d ) , making use of its ground truth labels ( not pictured ) . we then apply the principal components obtained from figure [ figure.classification](d ) to generate labels ( figure [ figure.classification](e ) ) for figure [ figure.classification](a ) using the decision rule . compared to the first run , the algorithm s performance here is a better indication of its feasibility as a fully automated classification scheme , and is summarized by the following confusion matrix : @xmath337 though the performance in the second run is understandably worse than that of the first , it nevertheless demonstrates the real - world potential of the idea exemplified by theorem [ theorem.flatness ] : the local histograms of certain types of textures can be decomposed into more basic distributions , and this decomposition can serve as an image processing tool . the authors are extremely grateful to dr . carlos castro and dr . john a. ozolek for introducing us to the motivating application and for providing us with raw image data and manually segmented ground truth labels used throughout this article . this work is supported by nsf dms 1042701 and ccf 1017278 , afosr f1ata01103j001 and f1ata00183g003nih , nih - r03-eb009875 and 5p01hd047675 - 02 , the pa state tobacco settlement and the kamlet - smith bioinformatics grant . parts of the work were presented at isbi 2010 @xcite and sbec 2010 @xcite . the views expressed in this article are those of the authors and do not reflect the official policy or position of the united states air force , department of defense , or the u.s . government . r. bhagavatula , m. fickus , j. a. ozolek , c. a. castro , j. kovaevi , automatic identification and delineation of germ layer components in h&e stained images of teratomas derived from human and nonhuman primate embryonic stem cells , proc . ieee int . ( 2010 ) 10411044 . a. chebira _ et al . _ , multiresolution identification of germ layer components in teratomas derived from human and nonhuman primate embryonic stem cells , proc . ieee int . biomed . ( 2008 ) 979982 .

squeezed states , a special kind of entangled states , are known as a useful resource for quantum metrology . in interferometric sensors they allow to overcome the `` classical '' projection noise limit stemming from the independent nature of the individual photons or atoms within the interferometer . motivated by the potential impact on metrology as wells as by fundamental questions in the context of entanglement , a lot of theoretical and experimental effort has been made to study squeezed states . the first squeezed states useful for quantum enhanced metrology have been proposed and generated in quantum optics , where the squeezed variables are the coherences of the light field . in this tutorial we focus on spin squeezing in atomic systems . we give an introduction to its concepts and discuss its generation in bose - einstein condensates . we discuss in detail the experimental requirements necessary for the generation and direct detection of coherent spin squeezing . two exemplary experiments demonstrating adiabatically prepared spin squeezing based on motional degrees of freedom and diabatically realized spin squeezing based on internal hyperfine degrees of freedom are discussed .

1203.5359

interferometers , which are among the most precise measurement devices known today , are based on the interference of two coherent atomic or photonic modes . the quantity to be measured is mapped to the relative phase between the two modes , explicitly using the wave - like nature of the resources ( atoms or photons ) . however , in the actual detection process their particle - like character becomes important ; the population of the two modes can be at best be counted one by one . traditional interferometers use uncorrelated resources in two coherent modes which results in a poissonian distribution of the population difference at the interferometer output . the standard deviation of this distribution @xmath0 is the so called projection noise or shot noise limit . the best theoretically possible phase precision for uncorrelated resources and linear interferometers is given by the classical cramer - rao bound of parameter estimation @xcite . in the context of ( classical ) interferometric phase estimation this is known as the standard quantum limit . it can be reached with coherent states , the quantum states that are naturally generated by coherent splitting of a quantum mode , as it is routinely done by beam splitters in optics ( with one input port unused ) . indeed , modern interferometers measuring for example time @xcite , position @xcite or magnetic fields @xcite are limited by this quantum noise leading to a maximal interferometric phase precision of @xmath1 . several strategies have been proposed to overcome this standard quantum limit for metrology by using quantum correlated , that is entangled , resources within the interferometer @xcite . one of these strategies is to use squeezed quantum states @xcite which have been successfully generated in quantum optics @xcite . the analogue for atom interferometry are spin squeezed states @xcite . spin squeezed states can be obtained from coherent spin states by a continuous transformation which gradually redistributes the quantum noise among two orthogonal spin directions . this property makes them promising candidates for their experimental realization since the degree of squeezing can be balanced with the increased sensitivity to decoherence . interferometric precision is increased to @xmath2 for spin squeezed states with a squeezing parameter @xmath3 . many - body quantum mechanics sets an ultimate limit to the phase precision for general interferometric measurements . this is the so - called heisenberg limit , for which an @xmath4 scaling of the interferometric precision can be achieved @xcite . the heisenberg limit can be reached with spin squeezed states in the limit of infinite squeezing where spin squeezed states become identical to twin fock states . the potential gain in phase precision is enormous : in an atomic clock , for example , one measures an energy difference @xmath5 for a certain time @xmath6 ( @xmath7 is planck s constant ) . the `` classical '' uncertainty scales as @xmath8 , while heisenberg limited metrology would allow for an error of @xmath9 . assuming a fictitious heisenberg limited measurement with @xmath10 atoms lasting one second , a `` classical '' projection noise limited apparatus would need @xmath11 days to obtain the same level of precision . + in this tutorial we attempt to provide a basic introduction into the concept of spin squeezing . we focus on the theoretical concept of spin squeezing , its connection to entanglement and applications in quantum metrology . experimental techniques to generate and detect spin squeezing in bose - einstein condensates ( bec ) are explained in detail . finally we shortly discuss two exemplary experiments in which spin squeezing has been observed in becs . spin squeezing is a quantum strategy introduced in 1993 by kitagawa and ueda @xcite which aims to redistribute the fluctuations of two orthogonal spin directions among each other . in 1994 it was theoretically shown that spin squeezed states are useful quantum resources to enhance the precision of atom interferometers @xcite and in 2001 the connection between spin squeezing and entanglement was pointed out @xcite . in the field of quantum optics squeezed states have been proposed to overcome the standard quantum limit in optical interferometry already in 1981 @xcite . these states are similar to spin squeezed states however differences arise from the different variables , quadratures versus collective spin directions @xcite . the connection of spins to interferometry becomes obvious when realizing that an interferometer is essentially a two mode system with modes @xmath12 and @xmath13 . each particle in the interferometer can be regarded as an elementary spin with spin @xmath14 , with two possible longitudinal eigenstates @xmath15 corresponding to the two interferometer modes . the relative phase measured by the interferometer corresponds to the spin component in the plane spanned by the transversal spin directions @xmath16 and @xmath17 . the generalization of this picture to @xmath18 particles , which results in large collective spin vectors @xmath19 , is outlined below . an extensive theoretical review of two - mode systems also detailing the pseudo spin description can be found in @xcite . the mathematical concept of a spin algebra with total spin @xmath20 is a powerful tool to describe very different physical systems . any observable within such a spin @xmath20 system can be expressed by the three spin operators @xmath21 , @xmath22 , @xmath23 and the identity operator . the collective spin operators @xmath24 can be defined as the sum over all elementary spin operators ( pauli matrices ) @xmath25 , where @xmath26 : @xmath27 the three orthogonal spin components are non - commuting operators . their commutation relation is @xmath28=i \,\epsilon_{ijk } { \ensuremath{\hat{\textsf{\ensuremath{j}}}}}_{k}$ ] , where @xmath29 is the levi - civita symbol . therefore , any pair of spin operators obeys a heisenberg uncertainty relation which , for @xmath30 and @xmath31 , is given by @xmath32 and @xmath33 is the variance in @xmath34 direction . a basis of the general problem can be obtained as the tensor product of all @xmath18 bases of the individual components , each of dimension @xmath35 the dimension of the hilbert space is huge @xmath36 and grows exponentially with the number of elementary spins . the length of the collective spin @xmath20 is smaller or equal than half the number of elementary spins . the spin length in interferometry is usually large such that @xmath37 is a good approximation . in the symmetric case , that is , when all operations done on the spin ensemble affect each spin in the same way ( such that there is exchange symmetry ) the problem simplifies considerably . for example , each elementary spin can be prepared in the @xmath38 state and maximum collective polarization @xmath39 can be reached . hence , the total spin length is given by @xmath40 and the dimension of the hilbert space dramatically reduces to @xmath41 , linearly dependent on the number of elementary spins . one possible choice of a basis are the symmetric dicke states @xcite @xmath42 with @xmath43 . each of the dicke states corresponds to a perfectly defined particle number difference between the two modes @xmath12 and @xmath13 and since the total number of particles @xmath18 is fixed the dicke states correspond to twin fock states in the modes @xmath12 and @xmath13 . due to their exchange symmetry the elementary spins can be effectively described as bosons , the schwinger bosons @xcite . employing the second quantization formalism the creation ( @xmath44 , @xmath45 ) and annihilation operators ( @xmath46 , @xmath47 ) of the two modes can be related to the different spin components @xcite @xmath48 with @xmath49 and @xmath50 . identical bosonic particles in two modes can be described by the symmetric spin model and the schwinger representation given above can be used . when it comes to experiments , however , an a priori symmetry assumption is a strong constraint . its validity has to be shown experimentally . coherent spin states are the most `` classical - like '' pure quantum states of a symmetric ensemble of @xmath18 elementary @xmath51 spins or of @xmath18 two - mode bosons @xcite . they are constructed by placing all @xmath18 particles in the same single particle state , in any superposition of the two modes @xmath52^{n } { \ensuremath{\left| \mathrm{vac } \right\rangle}}\ ] ] where @xmath53 is the vacuum state . any coherent spin state @xmath54 can be constructed from the state @xmath55 ( all n bosons in mode @xmath13 ) by a suitable unitary rotation with the angles @xmath56 @xcite . one important aspect of the coherent spin states is that no quantum correlations between the particles are present . therefore , a coherent spin state features equal variance in any direction @xmath57 orthogonal to the mean spin direction @xmath56 which , according to the gaussian addition of variances , is given by the sum of the variances of the @xmath58 elementary spin @xmath59 particles . the perpendicular variances @xmath60 of individual elementary spins are by definition isotropic around @xmath56 since there are no subsystems that could cause any correlations ( figure [ fig : css]a ) @xcite . the heisenberg uncertainty relation ( [ eq.spinheisenberg ] ) for a single elementary spin @xmath61 pointing in @xmath62 direction is @xmath63 leading to an isotropic variance of @xmath64 for the collective coherent spin state , which identifies this quantum state as a minimal uncertainty state since @xmath65 ( see equation ( [ eq.spinheisenberg ] ) ) . we refer to the perpendicular spin fluctuations of a coherent spin state @xmath66 as the shot noise limit . + we go back to the first quantization formalism in order to obtain the probability distribution over different sets of basis states especially the two dicke state bases in the directions orthogonal to the mean spin direction . in order to develop a more detailed understanding of the coherent spin state and its fluctuations we start with the discussion of a special case where each particle is in a @xmath67 superposition of the two modes with @xmath68 relative phase . each elementary spin points in @xmath62 direction and its quantum state is @xmath69 , where we have chosen the dicke states in @xmath70 direction as the basis states . the probability to observe the elementary spin in state up or down is equal @xmath71 . + the @xmath18 atom coherent spin state is a collection of these independent elementary spins @xmath72^{\otimes n}$ ] and hence the measurement of the @xmath73 spin component is equivalent to @xmath18 measurements on a single spin . the probability distribution over the dicke states is binomial . we could have chosen equally the dicke states in @xmath74 direction to describe the spin state which shows again that the spin fluctuations in the directions perpendicular to @xmath75 , the mean spin direction , are isotropic . + a general coherent spin state @xmath76 in the dicke basis @xmath42 is described by @xcite : @xmath77 the coefficients @xmath78 follow a binomial distribution peaked around @xmath79 : . + [ t ] due to the fundamental uncertainty in the orthogonal spin directions the mean direction ( @xmath80 ) of any spin state can not be defined with arbitrary precision . for a coherent spin state the isotropic angular uncertainty @xmath81 , defined by the ratio of the uncertainty of the perpendicular spin directions @xmath82 to the mean spin length @xmath20 , is given by : @xmath83 as outlined above this limit arises as the classical statistical limit in a system consisting of @xmath18 independent constituents @xcite . it is this angular uncertainty which limits the interferometric precision in a ramsey interferometer . quantum correlations between the elementary spin @xmath59 particles of a collective spin @xmath20 can cause anisotropic fluctuations of the collective spin in the directions perpendicular to the mean spin ( figure [ fig : css]b ) . in reference @xcite quantum states are considered spin squeezed if the variance of one spin component is smaller than the shot noise limit @xmath84 for a coherent spin state : @xmath85 definition ( [ eq.numbersqueezing ] ) does not take the second , perpendicular spin direction into account . due to the heisenberg uncertainty relation ( [ eq.spinheisenberg ] ) reduction of the variance in one direction causes an increase of fluctuations in the other . real life strategies to obtain spin squeezing might also involve states that are not minimal uncertainty states . for these states , especially for experimentally very important non - pure quantum states , the variance in some other direction than the squeezed direction can be much larger than given by the heisenberg uncertainty relation . this leads to a reduction of the effective mean spin length @xmath86 . + standard metrologic applications , for example ramsey interferometry , require a large mean spin length . in order to measure the usefulness of spin squeezed states for these applications another definition of the squeezing parameter was introduced in reference @xcite @xmath87 whose inverse @xmath88 measures the precision gain in a ramsey interferometric sequence relative to the standard quantum limit ( [ eq.shotnoise ] ) . spin squeezing among @xmath18 constituents is related to many - body entanglement . in this context a third spin squeezing criterion was defined @xcite : @xmath89 entanglement is detected by the inequality @xmath90 as detailed below . here we explicitly use the standard assumption throughout this tutorial that the mean spin points in @xmath75 direction and the direction of minimal variance if not explicitly mentioned is the @xmath73 direction . + @xmath91 can be used equivalently to @xmath92 to quantify spin squeezing and precision gain in interferometry and we refer to it as coherent number squeezing or coherent spin squeezing . both quantities remain a useful parameter to detect entanglement or to measure metrologic precision gain even in a non - symmetric situation . to demonstrate entanglement in a collective spin system without presumptions not only number squeezing is required but also the coherence of the quantum state has to be shown . employing the q - representation @xcite , collective spin states can be conveniently visualized on a generalized bloch sphere with radius @xmath20 . in order to describe the most general spin state , i.e. pure states and statistical mixtures , the density matrix formalism is used @xcite . the density operator @xmath93 in coherent spin state basis is given by @xmath94 , where the integral covers the full solid angle and @xmath95 . the quasi - probability distribution @xmath96 is normalized to one . the q - representation uses the diagonal elements of the density operator to represent the quantum state @xmath97 . it gives the probability of finding the system specified by @xmath93 in the coherent spin state characterized by @xmath98 . the interpretation of this representation on generalized bloch spheres differs from the usual single spin @xmath51 bloch sphere . in that case the dimension of the hilbert space is two - dimensional and the quantum state representation on the surface of a sphere is exact . however for symmetric collective spin systems the dimension of the hilbert space is @xmath99 such that an exact mapping to the surface of a sphere is not possible . the position @xmath100 on the spin @xmath59 bloch sphere describes the full quantum state , while the position on the generalized bloch sphere gives only the mean spin direction and , within the following constraints , its fluctuations : the q - representation projects the density matrix on minimal uncertainty states , in particular coherent spin states . the most obvious consequence is that the minimal extension of a quantum state in @xmath101 on the bloch sphere is given by the uncertainties of the basis states . + the q - representation for visualizing collective spins is rather easily implemented with modern plotting tools . however , there exist more elaborate ways to visualize the quantum state of the two - mode system as described in @xcite . the bloch vector visualization described therein has the advantage that the spin noise in all three components of the bloch vector is directly visible . also the fragmentation of the quantum state , that is , if the @xmath18 bosons do not occupy the same single particle state @xcite can be seen directly by a bloch vector that lies inside the bloch sphere . entanglement is only a meaningful concept if the subsystems relative to which entanglement is defined are clearly specified @xcite . coherent spin squeezing detects entanglement among the elementary spins the subsystems are the individual spin @xmath59 particles @xcite . for @xmath18 distinguishable particles the definition of a separable state , i.e. non - entangled state , is that its @xmath18-body density matrix @xmath102 can be written as a direct product of single particle density matrices @xmath103 : @xmath104 @xmath105 is a probability distribution to account for incoherent mixtures . entanglement in many - body systems ( for a general review see @xcite ) is defined as the non - separability of the density matrix @xmath102 meaning the equality in equation ( [ eq.separability ] ) does not hold . + due to technical limitations the individual elementary spins can not be addressed before the measurement in many experiments . however , it is important to note that the elementary spins ( the subsystems ) have to be in principle distinguishable in order to define entanglement among them in a meaningful way @xcite . for example , for @xmath18 particles in a bec the distinguishability is a priory not given . however , by a gedanken local operation one can pinpoint each particle in space at the time of measurement without affecting the spin properties of the system . distinguishability is now given by the position of each particle . if entanglement is detected in the system , it must have been present in the system before the localization , since local measurements can not generate entanglement @xcite . given that the atoms in the bec are spaced by more than one wavelength of the detection light ( which is usually fulfilled ) , this gedanken local operation means to overcome the technical limitations for addressability and detection of the individual elementary spins . the fictitious pinning of the particles at the time of measurement ( the photon scattering destroys the bec ) effectively assigns a spatial mode to each particle by which they become distinguishable @xcite . the authors of reference @xcite show that the notion of entanglement in such a collective spin system can be generalized even without such a fictitious pinning of the individual particles . + not surprisingly different entanglement criteria can be found when regarding the two modes that define the spin @xmath59 , as the subsystems @xcite . it has been argued that this kind of mode entanglement can not be extracted in experiments without a suitable reference frame @xcite . recently , this problem of a missing reference frame for atomic fields has been addressed experimentally and mode entanglement has been shown in a system of ultracold bosons @xcite . without the possibility to address the individual elementary spins before the measurement tests for entanglement , that is , experimentally accessible criteria , based on the collective spin variables are necessary . the observables in most experiments so far are limited to first and second order moments of the distributions functions in different spin directions due to rather small counting statistics and technical noise . based on these , a complete set of inequalities , that is fulfilled for any separable quantum state , has been found @xcite . complete in this sense means that assuming the only information available are first @xmath106 and second moments @xmath107 of the distribution functions . note that these inequalities allow for experimental tests of entanglement , but they are no quantitative measures for entanglement . such entanglement criteria only allow for the statement : `` yes , that state is entangled with respect to these subsystems . '' or `` no , that state is not entangled with respect to these subsystems . '' the coherent spin squeezing criterion given in equation ( [ eq.coherentspinsqueezing ] ) corresponds to a less general form of such an inequality : all separable states fulfill the inequality @xmath108 , but a subgroup of entangled states violate it @xcite . coherent spin squeezing as an entanglement criterion does not require any symmetry assumption . however , in the fully symmetric subspace many entanglement criteria simplify . for example , the detection of spin fluctuations in one direction below the shot noise limit implies entanglement @xcite , that is @xmath109 holds for any separable symmetric state . equation ( [ eq.sqsymmetric ] ) is identical to the spin squeezing parameter of kitagawa and ueda ( [ eq.numbersqueezing ] ) showing that in the symmetric two - mode case entanglement is necessary to redistribute the fluctuations of orthogonal spin components . to avoid misunderstandings about the different definitions of spin squeezing we refer to @xmath110 as number squeezing . entanglement in collective spin systems is not only interesting from a conceptual perspective but it has also been shown to provide a useful quantum resource . in 1994 wineland et al . @xcite pointed out , that in particular spin squeezed states can be used to overcome the standard quantum limit in metrology . they can be directly used to boost the precision of conventional ramsey interferometers . prominent applications of ramsey interferometry @xcite are the definition of the time standard @xcite , high precision magnetometry @xcite or inertia measurements @xcite . reviews covering atom interferometry both experimentally and theoretically can be found in references @xcite . + figure [ fig : ramsey_bs ] shows the ramsey interferometric sequence illustrated on generalized bloch spheres . the parameter @xmath111 to be measured couples via a @xmath23 term in the hamiltonian resulting in a rotation of the spin around the @xmath34 axis during the evolution with rotation frequency @xmath111 . the accumulated relative phase @xmath112 by which the initial spin state rotated around the @xmath34 direction of the bloch sphere is finally converted into an observable population difference . the phase resolution of the interferometer @xmath113 is determined by the projection noise @xmath114 at the time of readout and the local slope of the ramsey fringe @xmath115 ( see figure [ fig : ramsey_prec ] ) . the point of maximum sensitivity ( taking finite environmental noise into account ) is reached where the mean population difference is zero and the slope is maximal @xmath116 . the visibility @xmath117 measures the mean spin length @xmath118 . + the amount of precision gain ( or loss ) relative to standard quantum limit is given by @xmath119 , or equivalently by @xmath120 . the measure can be expressed in visibility @xmath117 and spin noise in @xmath73 direction at readout @xmath30 @xmath121 the phase uncertainty measured as the root mean square deviation is @xmath122 . + the phase @xmath123 appears in two connected contexts in this tutorial . on the one hand , for coherent spin states , it is a variable characterizing the full quantum state of the system ( together with @xmath79 ) . on the other hand , it has the meaning of a relative phase between the two modes of the interferometer ( or more general between the modes @xmath12 and @xmath13 of the two - mode system ) . in both cases the phase @xmath123 can be interpreted as the mean angle of the spin state in longitudinal direction ( that is , in the @xmath124 , @xmath125 plane ) on the bloch sphere as schematically shown in figure [ fig : ramsey_bs ] . the angular uncertainty in this direction finally limiting the precision of a ramsey interferometers is @xmath126 ( cf . section [ sec.css ] ) . spin squeezed states feature reduced noise in one of the spin directions but excess noise in another direction can be present either due to a non - heisenberg limited quantum state or due to an incoherent mixture of several quantum states . the former might limit precision in standard ramsey interferometry , but specific correlated quantum states enable even enhanced interferometric precision in a generalized interferometer @xcite . the latter is easily limiting interferometric precision at a level above the standard quantum limit and experimentally it requires a large effort to prevent decoherence due to technical noise from the environment or due to finite temperature in the system . large noise quantum or classical even in a spin direction that is not directly measured has a degrading effect on interferometric precision in a standard interferometer which arises due to the curved surface of the bloch sphere . as soon as the noise amplitude is large enough such that the area of uncertainty can no longer be approximated by a plane , the mean spin is effectively shortened and the visibility decreases @xmath127 . + entanglement can be used as a quantum resource in a ramsey interferometric sequence in different ways . in order to increase the phase sensitivity either the slope of the signal @xmath128 has to be increased or the projection noise @xmath30 has to be decreased . slope increase can be reached by schrdinger cat type entanglement which involves maximally entangled states that are very fragile to decoherence . therefore they have been realized so far with very few particles only @xcite . + spin squeezing aims to decrease the projection noise . this is possible in gradual steps meaning that depending on the amount of spin squeezing the precision is gradually increased . therefore at least for moderate levels of spin squeezing these states are less fragile and they have been realized with a large number of particles but only with a relatively small squeezing factor @xcite . interferometric sensitivity for a coherent spin state and a coherent spin squeezed state is illustrated in figure [ fig : ramsey_prec ] . for the coherent spin squeezed state the decreased quantum fluctuations @xmath114 reduce the projection noise while the increased fluctuations @xmath129 cause a slight decrease of the mean spin length and therefore of the visibility of the ramsey fringe . nevertheless , the ratio of projection noise and slope of the ramsey fringe and therefore the phase sensitivity is increased . the ultimate limit for metrologic precision is the heisenberg limit @xcite , where the phase estimation error @xmath130 is given by @xmath131 for @xmath18 resources used in a single measurement . this fundamental limit can in principle be reached with both approaches mentioned above , schrdinger cat type entanglement or spin squeezing . in the context of quantum metrology the schrdinger cat state is sometimes called a noon state @xcite . its name originates from its form in fock states basis @xmath132 . it is a coherent superposition of all atoms in state @xmath12 and zero atoms in state @xmath13 and vice versa . in spin representation the noon state is the superposition of the two maximal dicke states @xmath133 . the increase of the signal slope for a noon state is obvious since the phase acquired between the two components @xmath134 is @xmath18 times larger than for a coherent spin state @xcite . experimentally it is important to note that the readout of the interferometer can not be realized by measuring @xmath135 . the reason is the vanishing mean spin length @xmath136 of this state . it has been shown that the parity of the state is a useful experimental observable to make use of noon states in interferometry and to reach the heisenberg limit @xcite . further proposals to achieve heisenberg limited interferometry involve twin - fock states @xmath137 @xcite . the mean spin length for these states is also vanishing such that special readout schemes are required @xcite . recently such states have been produced experimentally @xcite and their usefulness for quantum enhanced metrology has been shown @xcite . spin squeezed states allow for the optimization of interferometry gain demanding a finite mean spin length such that standard interferometric readout can be used . the optimum @xmath92 for a given mean spin length was found numerically in reference @xcite . an experimental protocol to generate spin squeezed states close to the heisenberg limit was proposed in reference @xcite . the authors show that the heisenberg limit can be reached asymptotically in the number of atoms . the precision of ramsey type interferometric measurements increases with evolution time within the interferometer ( cf . section [ sec.intro ] ) . the same inter atomic interactions , that can be used to generate entangled states initially and to overcome the standard quantum limit ( as shown in this tutorial ) , however , might spoil the precision of the interferometer if present during this evolution time @xcite . hence , interaction control ( for example using feshbach resonances ) is crucial for future applications that use standard interferometric readout . recently it has been emphasized that the fisher information is the most general criterion to measure phase sensitivity @xcite since it saturates the quantum cramer - rao bound @xcite . calculating the fisher information for a coherent spin state state evolving under the non - linear hamiltonian @xmath138 , where @xmath139 parametrizes the nonlinearity , heisenberg like scaling for the phase precision was found @xcite after long evolutions times where the potential precision gain in a standard interferometer was already decreased . the quantum state here is neither necessarily a noon state nor a coherently spin squeezed state . however standard interferometric readout can not be used to extract the phase information and a new type of bayesian readout might be employed which was experimentally demonstrated in @xcite . strong effort has been put into the experimental generation and detection of spin squeezed states in large collective spin systems . early experiments detected spin noise beyond the shot noise limit in a cold ensemble of cesium atoms prepared by quantum state transfer from non - classical light to the atoms @xcite , or in a cesium vapor cell at room temperature by quantum non - demolition measurement @xcite . indirect measurements of number squeezing of ultracold atoms in optical lattices and double well systems followed , however , the direct detection of the fluctuations was not possible @xcite . several experiments generating and detecting coherent spin squeezing of large collective spins in atomic gases were done in last four years @xcite . full ramsey interferometers working beyond the standard quantum limit have been realized using coherent spin squeezed states within the interferometer @xcite . different methods have proven useful to generate spin squeezed states in atomic samples . many of them involve light , for example by quantum state transfer , quantum non - demolition measurements or cavity induced interactions @xcite . in contrast , ultracold quantum gases offer the possibility to directly use the intrinsic interaction between the atoms . these interactions lead to nonlinear terms in the hamiltonian necessary to generate spin squeezed states @xcite . we will take intrinsic s - wave interactions among atoms in a two - mode bec as an example here and discuss how spin squeezing can be generated . examples for such two - mode systems based on internal or external degrees of freedom are shown in figure [ fig : example_twomode ] . in such two - mode becs the inter atomic interactions lead to a nonlinear term in the hamiltonian proportional to @xmath140 @xcite . a review on the experimental realization of the external two mode system can be found in @xcite and for the internal system details are given in @xcite . the theoretical description for both cases , including the possibility for fragmentation of the quantum state @xcite , has been reviewed in detail recently @xcite . thus , we only sketch the connection between the general hamiltonian and the pseudo spin description for the external double well case in the following . the general hamiltonian is @xmath141 from which , by making the two - mode ansatz @xmath142 , the two - mode bose - hubbard hamiltonian is obtained ( for improved readability we did not write the spatial dependence of the operators in equation ( [ eq.genham ] ) explicitly ) . @xmath143 here we defined three parameters in the hamiltonian , the nonlinearity @xmath139 characterizing the inter atomic interactions , the rabi type coupling between the two modes @xmath144 and a differential energy shift @xmath111 . the remaining step is to write the hamiltonian ( [ eq.bosehubbard ] ) in spin language by using the schwinger representation introduced in equation ( [ eq.schwinger ] ) : @xmath145 [ t ] and @xmath146 define the modes . * b * another example for a two - mode bec is a spinor bec with two populated hyperfine spin states @xmath12 and @xmath13 and identical spatial wave function @xmath147 . ] note that the hamiltonian in both cases , external and internal degrees of freedom , can be written in the form of equation ( [ eq.spin ] ) . the actual calculation of the parameters involves spatial integrals over the wave functions and differs in both cases . in this context it is important to note that the spatial wave functions @xmath147 are not the single particle ground state wave functions in the confining potential , but the mean field wave functions obtained by coupled gross - pitaevskii equations for the two modes ( for several different approaches see @xcite ) . details on the calculation of the parameters in the hamiltonian can be found for example in @xcite . + three regimes can be distinguished by looking at the quantum fluctuations in the orthogonal spin directions for the ground state of the hamiltonian in equation ( [ eq.spin ] ) ( see figure [ fig : adiabatic_scheme ] ) . for the following discussion we assume an unbiased situation , that is @xmath148 . the important parameter measuring the relative importance of tunnel coupling and interaction is @xmath149 . this can be seen directly by normalizing the spin operators to spin @xmath59 such that the total number of atoms appears explicitly . then @xmath150 is the ratio of the nonlinear to the linear coefficient . the rabi regime is the noninteracting limit where @xmath151 . here the ground state is a coherent spin state with maximal mean spin length @xmath152 and equal spin fluctuations in the orthogonal directions @xmath153 . for increasing interactions fluctuations in the relative atom number ( equivalent to fluctuations in @xmath23 ) become energetically unfavorable and the system enters the josephson regime ( @xmath154 ) . its name originates from the analogy of the hamiltonian ( [ eq.spin ] ) to the josephson hamiltonian in solid state physics @xcite . the ground state is a coherent spin squeezed state with reduced spin fluctuations in @xmath34 direction at the cost of increased fluctuations in @xmath125 and hence a reduced mean spin length . the tunnel coupling is negligible in the fock regime ( @xmath155 ) . the ground state is an extremely spin squeezed state with vanishing mean spin length . it finally approaches the twin fock state , an eigenstate of @xmath23 with eigenvalue zero . there are two possibilities to control the regime parameter @xmath150 . for a given total atom number @xmath18 either the tunnel coupling @xmath144 or the nonlinearity @xmath139 has to be changed . while the former is straight forward , there are again several possibilities to change the nonlinearity @xmath139 . the rather direct approach is to control the scattering properties @xcite , which is directly possible when using quantum states with accessible feshbach resonances @xcite . the second method , which has also been realized experimentally , is to control the wave function overlap between the two states @xcite . assuming the weakly interacting regime can be realized experimentally there are two strategies to generate coherent spin squeezed states . the first is an adiabatic approach where the regime parameter is changed slowly driving the system deeper into the josephson regime @xcite . for some experimental settings a fast non - adiabatic approach might be preferable @xcite where spin squeezing develops during the evolution after a quench in @xmath150 . in the following we discuss the advantages and disadvantages of the two strategies . [ t ] one of the advantages of ultracold atom systems is the ability to control the parameters of the hamiltonian dynamically . starting from a strong tunnel coupling the regime parameter @xmath150 can be adiabatically changed . such an adiabatic transformation of the quantum state is a robust way to generate coherent spin squeezed states . the tunnel coupling is present throughout the transformation , making the system less prone to phase noise ( fluctuations in @xmath111 ) . this can be intuitively understood by the concept of the spin echo technique @xcite : for small differential energy shifts @xmath156 the continuous tunnel coupling causes a continuous sequence of @xmath157 echo pulses counteracting the dephasing . another advantage of the adiabatic technique is its usefulness in a finite temperature environment , where the initial state is a thermal mixture of the lowest many - body eigenstates of the hamiltonian ( [ eq.spin ] ) . in the josephson regime the adiabatic increase of @xmath150 results in an effective `` cooling '' of the system @xcite . the term cooling should be used with care here since no entropy is removed from the system . with increasing parameter @xmath150 each of the lower eigenstates of the system features reduced fluctuations in @xmath23 , however at the cost of increased fluctuations in @xmath22 and a loss in coherence . nevertheless , the gain in @xmath34 direction outweighs the loss of coherence for small enough temperatures , such that the coherent spin squeezed regime can be reached by this method . it has been theoretically shown that the heisenberg limit can be reached asymptotically in the number of particles by the adiabatic technique @xcite . experimentally , however , it is a major problem to assure adiabaticity during the full range of the parameter change . at least when changing the regime parameter by tuning the tunneling coupling ( for fixed and typically small nonlinearity @xmath139 ) the required timescale for adiabaticity eventually diverges setting a limit to the attainable coherent spin squeezing factor @xmath3 . the long times required to generate coherent spin squeezing adiabatically are the main disadvantages of this approach . for some experimental systems the robust but slow adiabatic approach to generate spin squeezing might not be optimal . a diabatic technique , in which the hamiltonian parameters are quenched ( changed non - adiabatically ) , can produce spin squeezing in a much shorter time . the idea of the diabatic one - axis twisting was introduced in 1993 by kitagawa and ueda . starting with a coherent spin state with mean spin in @xmath124 direction the tunnel coupling is suddenly switched off such that the nonlinear term in the hamiltonian governs the evolution @xmath158 the nonlinearity causes a shearing of the uncertainty region on the bloch sphere since the rotation speed around @xmath34 is itself @xmath34 dependent . the scheme is similar to kerr effect based squeezing protocols in quantum optics where a material with a finite kerr nonlinearity is used such that the refractive index @xmath159 is proportional to the light intensity @xmath160 . the light experiences intensity dependent phase modulation within this medium resulting in quadrature squeezing @xcite . [ t ] the one axis twisting evolution of an initial coherent spin state can be nicely visualized . earlier in this manuscript it has been discussed how a coherent spin state on the equator of the bloch sphere @xmath161 can be described as a coherent superposition of several dicke states where the probability distribution over these basis states is binomial . each dicke state @xmath42 composing the coherent spin state rotates with a different frequency around the @xmath73 axis where the difference in rotation frequency between next neighboring dicke states is @xmath139 ( figure [ fig : diabatic_scheme]a ) . for short evolution times @xmath162 this shearing effect results in coherent spin squeezing under an axis rotated by the angle @xmath163 with respect to the equator of the bloch sphere ( figure [ fig : diabatic_scheme]b ) . @xmath164 is the dephasing time , after which the mean spin length has dropped from a value close to unity to @xmath165 due to the interaction induced spread of the state around the bloch sphere @xcite . in general @xmath166 is inversely proportional to the extension of the quantum state over the dicke basis @xmath167 which , for a coherent spin state on the equator of the bloch sphere , is @xmath168 . + at evolution times @xmath169 reduced quantum fluctuations under a certain axis are still present but the mean spin length is strongly reduced such that the quantum state is no longer optimally coherently spin squeezed . after even longer times the dynamics show highly non - classical interference effects @xcite finally resulting in a revival of the mean spin length at @xmath170 when each neighboring pair of dicke states is in phase again @xcite . the best achievable noise suppression in general interferometry still increases with evolution time even if coherent spin squeezing degrades . it has been recently shown @xcite that a new type of bayesian interferometer readout can be employed to make use of these quantum states . as typical for macroscopic superposition states the highly entangled states occurring for long evolution times are extremely sensitive to decoherence . for example , particle loss leads to an exponential decay of the revival contrast with a @xmath18 times enhanced decay constant @xcite . the maximum squeezing achievable with the diabatic one axis twisting technique is @xmath171 @xcite . this has to be contrasted to the maximum squeezing that can be generated using the adiabatic technique which approaches the heisenberg limit @xmath172 at the boundary to the fock regime where the mean spin length is still reasonable high @xcite . the maximum squeezing is better in the adiabatic case however the adiabaticity criterion requires long evolution times @xmath173 . the best squeezing in the one axis twisting protocol is achieved after @xmath174 , a factor of @xmath175 faster favoring the diabatic protocol for certain applications . the generation of spin squeezing might be optimized using different hamiltonians . stronger maximal squeezing , for example , can be reached by a two axis twisting scheme @xcite . another example , optimizing the required time to achieve squeezing , is to use evolution under the hamiltonian ( [ eq.spin ] ) after a quench to a small but finite tunnel coupling ( into the josephson regime ) @xcite . for initial states with relative phase @xmath176 the evolution leads first to coherent spin squeezing and for longer times to schrdinger cat like states @xcite . when discussing the maximum achievable coherent spin squeezing one has to keep in mind that with decreasing @xmath177 the generated states become more and more sensitive to decoherence . for the exemplary case of the one axis twisting evolution this problem has been studied in detail taking into account particle loss , random dephasing ( for example due to noise in @xmath111 , see equation ( [ eq.spin ] ) ) and multi mode effects due to finite temperature which are always present in real condensates @xcite . including these effects the achievable coherent spin squeezing is no longer proportional to @xmath178 but tends to a finite but small constant value in the thermodynamic limit @xmath179 , showing that a large metrology gain can still be reached @xcite . reference @xcite gives a review on decoherence effects in dynamically generated spin squeezing with becs and points out the effective dephasing effect of the different sources of decoherence . both approaches the adiabatic and diabatic one outlined in the last section require precise control over the experimental parameters . especially so far neglected uncontrolled differential energy shifts @xmath111 are the main experimental challenge to observe spin squeezing . another important issue , mainly in experiments with becs in double well traps , is finite temperature which easily limits the observable spin squeezing . the detection of atomic noise at the shot noise level in a small optically thick cloud is challenging and requires careful calibration of the imaging system . furthermore , a method to detect the mean spin length has to be available experimentally in order to detect coherent spin squeezing . in this section we will discuss these challenges in detail , starting with the detection issues . when working with internal degrees of freedom the measurement of the mean spin length can be done by a ramsey experiment @xcite . a ramsey scheme is usually relatively easy to implement given the availability of precise microwave and radio frequency sources ( or raman lasers ) to control the tunnel coupling @xcite . this coupling @xmath144 usually can be made much stronger than the nonlinearity @xmath139 such that the rabi regime is reached . when working with two photon transitions , which is often required to use states less sensitive to magnetic field fluctuations , care has to be taken to accurately determine the resonance . the resonance is shifted during the coupling with respect to the free evolution times due to the ac - stark shift induced by the coupling fields @xcite . the readout of the ramsey interferometer requires precise detection of the atom number difference in the two internal states over the full dynamic range . the mean spin length @xmath180 is then inferred from the visibility @xmath117 of the ramsey signal . an example for such a ramsey fringe revealing a visibility of @xmath181 is shown in figure [ fig_ramseyfringe ] @xcite . [ t ] for external degrees of freedom , for example in a double well trap for becs the mean spin length can be revealed from the interference pattern after time of flight expansion @xcite . additional complications might arise from a finite imaging resolution or due to excitations of additional spatial modes when expanding the condensate , both of which limit the visibility of the interference pattern in a single shot . care has to be taken when extracting the average visibility @xmath117 . in general the analysis method chosen has to include the effects of the possibly reduced single shot visibility , which might also be caused by multi mode effects in the initial double well @xcite . exemplary interference measurements for a bec released from a double well trap are shown in figure [ fig_numberphase]b later in this manuscript . the detection of number squeezing requires an accurate calibration of the deduced total atom number and the linearity of the imaging system . simultaneously a high spatial resolution is often required @xcite . to realize sufficiently large nonlinearities @xmath139 in the mesoscopic atom number regime a high atomic density and therefore tight confinement of the atoms is necessary , leading to optically think samples . this complicates the use of in principle single atom sensitive fluorescence detection methods @xcite which suffer from light induced collision losses if the individual atoms are not spatially separated during the detection . an alternative is to use absorption imaging @xcite in the high saturation intensity regime @xcite which provides good signal to noise ratio and high spatial resolution , however , its calibration is non trivial as outlined below . in high intensity absorption imaging @xcite the full beer - lambert absorption formula is necessary to calculate the atomic column density @xmath182 from the counts per pixel @xmath183 on the picture containing the absorption information and on the reference picture @xmath184 ( the ccd camera pixels but it is mostly omitted below to ensure better readability ) . for the image analysis the mean light intensity on the reference picture has to be normalized to the mean light intensity on the picture with the atomic signal . for its determination the area containing the atomic signal is not taken into account . not only the optical density @xmath185 but also the difference in the counts @xmath186 contribute to the signal : @xmath187 here @xmath188 is the linear extension of the ccd pixel taking the magnification into account , the resonant cross section , @xmath189 the clebsch - gordan coefficient , @xmath190 a correction factor obtained from a comparison to simulations as explained below and @xmath191 is the imaging pulse duration . the factor @xmath192 contributes to the linear part of the formula where @xmath193 is the total quantum efficiency and @xmath194 the gain factor of the camera . @xmath195 is the light angular frequency and @xmath196 the saturation energy of the transition . the gain factor @xmath194 is determined by measuring the noise curve @xmath197 of the ccd camera . it can be measured over the full dynamic range using an incoherent light source ( e.g. a led ) such that the light hitting the active region of the ccd is uncorrelated featuring shot noise limited noise characteristics . the factor @xmath194 follows from the slope of the ccd s noise curve via a linear fit in the working region where the camera noise is not read out noise dominated . the calibration of the @xmath198 factor is based on the quantum efficiency measurement which itself relies on the power meter calibration . furthermore , for a typical numerical aperture around @xmath199 approximately @xmath200 of the solid angle is covered by the imaging lens such that a non negligible fraction of the scattered photons reaches the camera . this effect is not taken into account in equation ( [ eq.imaging ] ) . a possible method to cross check the inferred number of atoms is to prepare the bec in a very well known trap configuration and image it with very high imaging intensity ( typically @xmath201 ) . in this regime only the linear part of the beer - lambert formula contributes , but the signal to noise ratio is poor . imaging is done with a very short pulse ( @xmath202s ) to minimize diffusive broadening of the profile due to photon scattering . the extracted average profile along an axis much bigger than the optical resolution is calculated for different correction factors @xmath190 and compared to theoretical profiles obtained from three dimensional gross - pitaevskii equation simulations varying the total numbers of atoms ( figure [ fig : calibration]a ) . the correction factor is finally found by minimizing the quadratic deviation between the simulated and measured profile . the expected corrections are small such that @xmath190 is should be close to unity . [ t ] atoms yields a factor of two smaller slope while the absolute number is underestimated by only @xmath203 . this results in a strong bias of the measured atom number fluctuations in this regime . the reference atom number in this figure was measured for @xmath204s expansion time and with high imaging intensity @xmath205 ( black circles ) @xcite . this figure was originally published in the supplementary online material to @xcite . ] calibration of the imaging is completed by the measurement of the clebsch - gordan coefficient @xmath189 . this is done by imaging the atoms with varying imaging intensities @xmath206 . a correct clebsch - gordan factor means intensity independent atom number measurements ( see figure [ fig : calibration]b ) . + the total atom number follows from direct summation over the region where the atoms are detected ( not by a fit which might bias fluctuation measurements ) . the size of this region is typically chosen to three standard deviations as obtained from a gaussian fit to the atomic cloud . [ t ] [ t ] quantitative measurements of the atom number using absorption imaging is limited to atomic clouds with a size larger than the optical resolution of the imaging system . for to small clouds the non - linear part of equation ( [ eq.imaging ] ) causes a systematic error in the detected atom number . this can be easily understood in the limit where the optical resolution is much larger than the pixel size . the detected column density @xmath207 per pixel is calculated from photon counts integrated over the size of the pixel @xmath208 where we omitted the linear term in equation ( [ eq.imaging ] ) . correct atom number calculation @xmath209 however requires to calculate @xmath210 in general @xmath211 holds such that the calculated atom number is wrong . when the pixel size is smaller than the point spread function @xmath212 of the imaging system the same argument holds , but the main averaging effect is due to the convolution with the point spread function , e.g. the replacement @xmath213 has to be made , where @xmath214 means the convolution of the functions @xmath212 and @xmath215 . figure [ fig : nonlinear_effects ] shows an example of this effect . the nonlinearity in the atom number calculation for high optical densities and small atomic clouds is clearly visible . it leads not only to an underestimation of the absolute atom number but also to an underestimation of the fluctuations in the high density region due to the smaller local slope . by slightly expanding the atomic clouds before imaging this effect can be minimized . the atoms are detected via their resonant interaction with the probe light whose noise characteristics add to the atomic noise of interest . at least the shot noise of the detection light adds to the detected noise , but especially in absorption imaging interference fringes easily increase this minimal background noise level . these fringes originate from motion of the imaging systems optics or air motion in the light path . special care has to be taken to eliminate these effects . the detection light shot noise contribution can be inferred pixel by pixel on both , the reference and absorption picture , using the camera noise calibration curve . standard error propagation of equation ( [ eq.imaging ] ) allows for a conversion of the photonic variance @xmath216 into atomic variance : @xmath217 here the photon shot noise contribution @xmath218 is calculated by a expansion up to second order , assuming a gaussian distribution for the photon statistics @xmath219 . photon shot noise on different pixels ( i , j ) is uncorrelated such that the total contribution in a given area of the ccd chip @xmath220 can be obtained by summation over the variance per pixel . the photon shot noise is independent of the atomic noise @xmath221 . thus , the best estimate of the true atomic fluctuations can be obtained by subtracting it from the detected fluctuations @xmath222 . the uncertainty in the estimated amount of photon shot noise is typically in the order of a few percent . + in the strong saturation regime the imaging light intensity @xmath223 controls the transparency of the atomic cloud @xcite . the imaging signal to noise ratio can be optimized by choosing the proper light intensity depending on the atomic density and the camera parameters . from the discussion above it is clear that correct calibration of the imaging system is non - trivial . therefore it is important to have an independent test of the deduced atomic variances . one possibility is to hold an atomic sample in the trap until a large fraction of the atoms is lost by background gas collisions . this uncorrelated one - body loss tends to restore a poissonian atom number distribution . however , the poissonian limit is only reached asymptotically and the fluctuations might be super - poissonian for an initial distribution governed by experimental instabilities or slightly sub - poissonian if the loss is dominated by two or three body collisions @xcite . + when working with internal degrees of freedom coherent spin states can be prepared by precise electromagnetic coupling pulses . their well known noise characteristics , that is , relative atom number fluctuations at the shot noise limit , can be directly used for the imaging calibration . figure [ fig_imaginglinearity ] shows such a calibration check which reveals a correctly calibrated imaging over a large range of total atom numbers @xcite . in the discussion of the squeezing generation we assumed vanishing external noise during the evolution . in real experiments , however , fluctuations of the differential energy shift @xmath111 coupling via the @xmath23 term in the hamiltonian ( [ eq.spin ] ) cause phase noise @xcite . both , the adiabatic and non - adiabatic approach to generate coherent spin squeezed states suffer from this noise problem . especially the non - adiabatic method is very sensitive to dc phase noise ( fluctuations from shot to shot ) . a spin echo technique @xcite can be used to effectively remove the sensitivity to these shot to shot fluctuations , however , the sensitivity to low but finite frequency fluctuations remains problematic . in general the spectral sensitivity is dependent on the duration of the free evolution and knowing the noise spectrum a specially designed spin echo technique can be used to optimize the spectral sensitivity . figure [ fig_phasenoise ] shows an exemplary sensitivity curve versus frequency of the environmental noise with and without spin echo . [ t ] between optical dipole trap and optical lattice . in equilibrium the chemical potential @xmath224 is the same in both wells which results in an occupation number difference between the left and right mode dependent on the differential energy shift @xmath225 between them . ] in the adiabatic case the final relative atom number fluctuations are also influenced by the noise in the differential energy shift , however , the main contribution is due to shot to shot fluctuations . the continuous coupling of the two states results effectively in continuous spin echo pulses counteracting dephasing . the remaining dc sensitivity can be understood intuitively for the example of the external double well system , but the argumentation also holds for an analogous internal system . the system response to an imbalance in the offset energy for the two wells ( a dc differential shift ) is to adapt the relative atomic population such that the overall chemical potential is constant . this means an increased population of the lower state such that the extra interaction energy balances the external energy offset . an estimation for the noise sensitivity can be obtained by the following consideration where the main experimental challenge , the relative movement of the dipole traps that generate the trapping potential @xcite becomes obvious . figure [ fig_trapshift ] illustrates this situation . the optical dipole trap generates harmonic confinement in axial direction @xmath226 , where @xmath227 is the atomic mass . fluctuations of the energy difference @xmath225 between the two potential minima separated by @xmath188 due to fluctuations of the barrier position @xmath228 is given by @xmath229 . the overall chemical potential in local density approximation @xcite is @xmath230 and in equilibrium it is the same for the two wells . the contribution due to inter atomic interactions @xmath231 balances the change in @xmath232 : @xmath233 , where we used @xmath234 @xcite and @xmath235 . to give an estimation of the required position stability of the trapping potentials we calculate the actual numbers for the trapping parameters of the double well trap in reference @xcite . the nonlinearity is @xmath236hz and the longitudinal trapping frequency @xmath237hz . we obtain @xmath238 for the position fluctuations leading to an extra noise of the same order as the shot noise limit @xmath239db for @xmath240 atoms . since these technical induced fluctuations add to the variance caused by the atomic quantum state , their magnitude has to be much smaller than the shot noise level in order to measure a reasonable amount of number squeezing . as a figure of merit , position fluctuations of @xmath241 nm between different experimental realizations limit the best observable number squeezing to @xmath242db . this corresponds to roughly one percent of the double well spacing in this experiment @xcite . [ t ] the influence of finite temperature is very different in the two settings discussed here . while the effective temperature in the spin degree of freedom is nearly zero in the internal case , finite temperature effects easily dominate the fluctuations when dealing with a bec in a double well trap @xcite . the difference emerges from the very different initial energy gap between the two modes comprising the pseudo spin . when using two hyperfine states their energy separation is typically in the order of several gigahertz corresponding to temperatures of a few hundred millikelvin , much larger than the temperature in the bec of a few tens of nanokelvin . thus , using almost noise free coupling fields and short coupling pulses much faster than any thermalization timescale , spin states with almost zero temperature can be realized ( see references @xcite for the effect of finite temperature in this system . ) . [ t ] the situation is very different in the external double well potential , where the pseudo spin in comprised of the two center of mass modes in the two potential wells . in a unbiased ( @xmath148 ) situation , there is no or a negligibly small energy gap separating the two modes such that several collective modes are excited even at nanokelvin temperatures . this leads to non pure initial spin states and to increased spin fluctuations as compared to a coherent spin state . nevertheless the adiabatic spin squeezing technique results in an effective cooling of the spin fluctuations such that coherent spin squeezed states can be realized @xcite . figure [ fig_adiabaticisothermal ] shows the number fluctuations and mean spin lengths for different initial temperatures comparing the adiabatic method to the isothermal case . parameters are the same as for the experiments reported in @xcite . note that the fluctuations for the ground state would be orders of magnitude better for the same hamiltonian parameters , showing the dominating effects of finite temperature . in the previous sections we took two settings as examples for the generation of adiabatic and non - adiabatic spin squeezing in a bec . we will briefly discuss some results of these experiments @xcite in the following . a more detailed description of these experiments can be found in @xcite . the experiments discussed in this section have been originally published in @xcite and in @xcite . approximately @xmath240 condensed rubidium atoms were loaded into a symmetric double trap where initially the potential barrier separating the two modes was small . for the initial parameters the system was already in the josephson regime due to the rather strong nonlinearity . next the potential barrier was increased adiabatically , driving the system deeper into the josephson regime . the decrease of the tunnel coupling resulted in a decrease of the relative number fluctuations directly corresponding to the spin fluctuations in @xmath34 direction . the fluctuations were redistributed to the orthogonal spin direction such that the mean spin length decreased . the simultaneous change in the number fluctuations and mean spin length can be seen directly on the raw data ( figure [ fig_numberphase ] ) . predicting the optimal time required for an adiabatic change of the parameters is not trivial in such an experiment , especially because additional aspects such as particle losses or heating limit the duration of the experiments . therefore , the number squeezing was measured versus ramp up time of the double well potential while the final barrier height was kept constant . figure [ fig_adiabaticity ] shows the results of this experiment . in order to explain the dependence of number squeezing versus ramp up time both finite temperature and finite fluctuations of the trap position had to be taken into account . they dominate the observed fluctuations ( details of this experiment can be found in @xcite ) . around the feshbach resonance @xcite . the black solid line is a fit to the real part of @xmath245 @xcite revealing an inelastic width of @xmath246 mg in accordance with the theoretical prediction @xcite . the elastic width extracted from the fit is @xmath247 mg . the center of the resonance is at @xmath248 g , but the absolute magnetic field has an uncertainty of approximately @xmath249 mg in this measurement @xcite . the background scattering length @xmath250 was chosen as a free parameter and the fit reveals @xmath251 . ] despite of these limitations coherent spin squeezed states in the double well trap were realized with a maximal squeezing factor of @xmath252db @xcite . however , the implementation of a full atom interferometer beyond the standard quantum limit remains an experimental challenge with such a setup even though the feasibility of the production of coherent spin squeezed states was demonstrated . the analog of electromagnetic coupling pulses rotating the spin state non - adiabatically needs to be implemented in order to realize a ramsey sequence with the spin squeezed state as input state . the required non - adiabatic changes are difficult to realize without the possibility to change the nonlinearity directly ( which could be done by using a feshbach resonance @xcite ) since fast changes in the external trapping potential lead to unwanted excitation of the condensate . + using internal hyperfine states to define the two modes instead of the center of mass modes these limitations can be overcome as presented in the following section . when working with internal hyperfine states of the atoms fast and precise spin rotations are readily available using electromagnetic coupling . the experimental challenge is to obtain a sufficiently high nonlinearity such that spin squeezed states can be realized within the experimental coherence time . several ideas to increase the nonlinear interaction have been experimentally realized reaching from cavity mediated interactions in thermal gases @xcite to state dependent microwave traps on atom chips which allow for the control of the spatial overlap of the two spin modes @xcite . we focus here on another method exploiting an interstate feshbach resonance between two hyperfine states of rubidium around @xmath253 g @xcite and @xmath254 hyperfine states . ] . such a feshbach resonance may be used to increase the nonlinearity but at the cost of increased particle losses @xcite . figure [ fig_feshbach ] shows the measured resonantly enhanced effective nonlinearity @xcite . the simultaneously enhanced losses limit the available time to generate coherently spin squeezed states to a few tens of milliseconds favoring the fast non - adiabatic `` one axis twisting '' method proposed in @xcite for the generation of spin squeezing @xcite . coherent spin squeezed states with inferred squeezing factors up to @xmath255db could be realized within an evolution time of @xmath256ms @xcite . the use of a spin echo technique and an actively stabilized magnetic field @xcite was crucial here in order to minimize the dephasing effects . given the possibility of precise spin rotations a full interferometric sequence can be implemented . in a prototypical measurement enhanced interferometric sensitivity of @xmath203 beyond the standard quantum limit has been demonstrated @xcite using coherent spin squeezed states within the ramsey interferometer . figure [ fig_interferometer ] shows a direct comparison of the ramsey interferometer performance for coherent spin states as input state versus coherent spin squeezed states . in this direct comparison the performance increase due to the spin squeezing is even larger ( @xmath257 ) . however , it should be made clear that experimental noise is included here , that is , the comparison is not relative to the standard quantum limit . measured is the increase in performance due to spin squeezing in this particular interferometric setup . several other experiments recently succeeded in the realization of quantum enhanced interferometers . an interferometer with @xmath258db precision gain has been realized using a cold sample of cesium atoms where coherent spin squeezing has been generated by quantum non - demolition measurements @xcite . using cavity mediated interactions in a cold rubidium gas to generate coherent spin squeezing the time dependent performance of a similar interferometer has been characterized in terms of the allan variance @xcite . these different experiments show that atomic coherent spin squeezing can nowadays be realized with a variety of techniques and in quite different systems . it remains an open challenge to realize a large amount of coherent spin squeezing with atom numbers comparable to those used in modern linear interferometers . however , the basic concept of spin squeezing and its experimental feasibility has been demonstrated such spin squeezing might help to push the limit of precision measurements in the future . i thank the bec team at the kirchhoff institute for physics in heidelberg for their support and practical help . especially i thank markus oberthaler for his great support and for his supervision of my phd thesis . special thanks goes to andreas weller , eike nicklas , tilman zibold and jrme estve for the joint work on our squeezing experiments .

we predict and theoretically investigate the new coherent effect of nonlinear quantum optics spatial propagation of rabi oscillations ( rabi waves ) in one - dimensional quantum dot ( qd ) chain . qd - chain is modeled by the set of two - level quantum systems with tunnel coupling between neighboring qds . the space propagation of rabi waves in the form of traveling waves and wave packets is considered . it is shown , that traveling rabi waves are quantum states of qd - chain dressed by radiation . the dispersion characteristics of traveling rabi waves are investigated and their dependence on average number of photons in wave is demonstrated . the propagation of rabi wave packets is accompanied by the transfer of the inversion and quantum correlations along the qd - chain and by the transformation of quantum light statistics . the conditions of experimental observability are analyzed . the effect can find practical use in quantum computing and quantum informatics .

0907.3191

rabi oscillations are periodical transitions of a two - level quantum system between its stationary states under the action of an oscillatory driving field , see e.g. @xcite . the phenomenon was theoretically predicted by rabi on nuclear spins in radio - frequency magnetic field @xcite and was firstly observed by torrey @xcite . afterwards , rabi oscillations were discovered in various physical systems , such as electromagnetically driven atoms@xcite ( including the case of rydberg atomic states @xcite ) , semiconductor quantum dots ( qds ) @xcite , and different types of solid - state qubits ( superconducting charge qubits based on josephson junctions @xcite , spin - qubits @xcite , semiconductor charge - qubits @xcite ) . in real physical systems the ideal picture @xcite of rabi effect can be essentially modified by additional features , such as the time - domain modulation of the field - matter coupling constant @xcite , the phonon - induced dephasing @xcite and the local - field effects @xcite . new phenomena appear in rabi oscillators with broken inversion symmetry @xcite and in systems of two coupled rabi oscillators @xcite . in spatially extensive samples with a large number of oscillators the propagation effects come into play . as a result , the mechanism responsible for rabi oscillations causes also a number of nonstationary coherent optical phenomena , such as optical nutation , photon echo , self - induced transparency , etc @xcite . in low - dimensional systems the propagation effects also take place . for example , the numerical modeling of the coherent intersubband rabi oscillations in a sample comprising 80 algaas / gaas quantum wells @xcite shows that the population dynamics depends on the quantum well position in the series . this result demonstrates strong radiative coupling between wells and , more generally , significant difference in the picture of rabi effect for single and multiple oscillators . another aspects of such a difference , namely , the effects of quantum interference and correlations between photons in multiatom fluorescence , are demonstrated in @xcite ( on the example of atomic chains in nanofibers ) . from practical point of view , the effect of rabi oscillations is a key ingredient for realization of binary logic and optical control in quantum informatics and quantum computing . the theoretical analysis of rabi oscillations is highly diversified both in form and content . the common feature is the impossibility of consideration of electromagnetic field influence as a small perturbation . different ways of description are used in the analysis of the model problem of rabi oscillations in a single two - level atom @xcite . one such a way , the probability amplitude method , consists in solving of the schrdinger equation for wavefunction @xmath0 , which is the superposition of various atom - photon states . the second one is the heisenberg operator method which is based on the analysis of the photonic and atomic operators time evolution . and finally , the third way is the unitary time - evolution operator technique . as is demonstrated in @xcite , all these methods lead to identical solutions ; the choice of the concrete one is determined by the convenience considerations . taking into account the quantum nature of the electromagnetic field has the special significance in the analysis of rabi oscillations in complex systems . two different cases can be marked out . the first could be called quasiclassical one @xcite . in this case external field has classical nature . however , the true field is supposed to be concordant with the quantum motions of the particles . so , the field should contain the contribution of the induced polarization , which has a quantum nature . another case is really quantum and takes into consideration the photon structure of the electromagnetic field @xcite . during the process of energy - level transitions in the atom the photon structure of the field is also transformed , therefore the atom - photon dynamics should be considered self - consistently . in this paper we build for the first time a theoretical model of a distributed system of coupled rabi oscillators and predict a new physical effect : propagation of rabi oscillations in space in the form of traveling waves and wave packets . quantum oscillators in the system that exhibit rabi - wave propagation interact strongly enough to stare and exchange the e - h pair excitation between light emission and light absorption acts . often in quantum optics an `` all - matter '' picture is employed , where the dynamics of electromagnetic field is integrated out , for example in the optical bloch equations @xcite . in @xcite the light - matter interaction is treated in an `` all - light '' picture ( lippmann - schwinger equation approach ) . in our model the quantum nature of light is also fully accounted . as the theoretical approach the _ probability amplitude method _ , generalized for the case of 1d - chain , is convenient . the paper is organized as follows . in sec.[sec : theory ] the model is formulated and equations of electron - photon dynamics both for discrete chain and continuous limit are obtained . in sec.[sec : trav_w ] the plane rabi waves are considered ( their dispersion equations , eigenmode structure , dispersion curves properties ) . sec.[sec : wave_pack ] is devoted to the rabi wavepackets investigation , their spatial - temporal dynamics is analyzed , experimental observability of rabi waves is discussed . in sec.[sec : class_lim ] the rabi waves in the case of classic light are examined and distinctive features of semiclassical consideration are treated of . in sec.[sec : local_field ] the role of the local - field effects and conditions of this effects negligibility are discussed . in sec.[sec : correlators ] the space - time structure of electron - electron and electron - photon correlators is considered . the main results of the work are formulated in sec.[sec : conclusion ] . schematic picture of the qd - chain interacting with single - mode electromagnetic field ( a ) and the energy levels diagram for @xmath1-th qd ( b).,scaledwidth=45.0% ] consider an infinite periodical one - dimensional chain of identical qds , containing an electron in the size - quantized conduction band . it is assumed that in each @xmath1-th qd there are at least two one - electron orbital states , ground @xmath2 and excited @xmath3 , with transition frequency @xmath4 . neighboring qds are coupled via electron tunneling @xcite , i.e. the electron can go from state @xmath3 to state @xmath5 and from @xmath2 to state @xmath6 . transitions between ground state and excited state belonging to different qds are neglected : @xmath7 . let the qd chain be exposed to a plane wave of quantum light , which is incident on the chain at an angle @xmath8 , @xmath9 , where @xmath10 , @xmath11 are photon creation and annihilation operators , @xmath12 , @xmath13 is the normalizing volume , @xmath14 is the unit polarization vector , wavenumber @xmath15 . the dependence of the field on transverse coordinates is negligibly small because of electrical smallness of qds . the case of oblique propagation is of interest due to the possibility of wavenumber variation with variation of incidence angle @xmath8 at constant frequency @xmath16 . then the hamiltonian of the system in the rotating wave approximation @xcite reads @xmath17 where @xmath18 is hamiltonian of qd - chain in the absence of electron tunneling and qd - field interaction , @xmath19 , @xmath20 is the hamiltonian of the free electromagnetic field@xcite . the component @xmath21 describes qd - field interaction , where @xmath22 is interaction constant , @xmath23 is the qd dipole moment . note that all transition dipole moments in the chain are assumed to be real values and to have fixed orientation . the operators @xmath24 and @xmath25 are the transition operators from ground to excited states and vice versa for @xmath1-th qd . the term @xmath26 takes into account interdot electron tunneling ; @xmath27 are the electron tunneling frequencies for the excited ( @xmath28 ) and ground ( @xmath29 ) states of the qds . the term @xmath30 corresponds to the local - field effects originated from the dipole - dipole electron - hole intradot interaction @xcite ; in the mean - field approximation this term is given by @xmath31 here @xmath32 is the depolarization tensor ( see eq . ( 18 ) in ref . @xcite ) , @xmath33 is the volume of qd . the state vector of the `` qd - chain+light '' system may be represented in terms of the eigenstates of isolated qds as @xmath34 here @xmath35 , @xmath36 , where @xmath37 is the light fock state with @xmath38 photons , @xmath39 , @xmath40 are the probability amplitudes . in the interaction picture the evolution of the system is described by the nonstationary schrdinger equation @xmath41 , where the interaction hamiltonian has the form @xmath42 . this schrdinger equation leads to the following equations for the probability amplitudes @xmath43 where @xmath44 is the local - field induced depolarization shift @xcite . obtaining this equations we have taken into account that the interaction can cause the transitions between the states @xmath45 , @xmath46 only . as it is seen from eqs. , two competitive mechanisms manifest themselves in the light - qd - chain coupling : the local - field induced nonlinearity and the dispersion spreading due to the tunneling . to this point we have taken no account of the decay processes inside the qds . to take this processes into account one should allow for processes of interaction of e - h pair in qd with phonon bath . such interaction can be described by conception of quantum trajectories @xcite of e - h pair , which are the superpositions of deterministic evolutions and random jumps under the action of lindblad operators . in this case the hamiltonian should be replaced by the efficient non - hermitian hamiltonian @xmath47 , where @xmath48 is lindblad operator for @xmath1-th qd . suppose that the density of phonon states of the bath has the form of lorentz line with frequency @xmath4 and width @xmath49 . if the bath is weakly coupled with qd then @xmath50 ( see refs @xcite ) . it means that the decay can be taken into account by substitution @xmath51 into eq.[system_diskr_a ] and @xmath52 into eq.[system_diskr_b ] . let us restrict here consideration to linear regime of the carrier motion and omit the terms @xmath53 in . in the sec.[sec : local_field ] we shall analyze the limitations imposed by omission of nonlinear terms . by this means , omitting the terms @xmath53 , we arrive at the set of difference - differential equations for vectors @xmath54 $ ] as follows : @xmath55 \bm{\psi}_{p , n}\\ + i\hat{\xi}\left(\bm{\psi}_{p-1,n}+\bm{\psi}_{p+1,n}\right ) , \end{split}\ ] ] where @xmath56 denotes 2d unit operator , @xmath57 $ ] , @xmath58/2 $ ] . coefficients in are expressed in terms of pauli matrices @xmath59 acting on vectors @xmath60 . equations are the recurrent ordinary differential equations . in some cases it is more convenient to transform them into the system of partial differential equations . to do this we should turn to the continuous limit , making the standard substitutions ( see , for example , @xcite ) @xmath61 , @xmath62 , @xmath63 , @xmath64 . then the system leads to @xmath65\bm{\psi}_n \\ + ia^2\hat{\xi}\partial^2_x\bm{\psi}_n . \end{split}\ ] ] formula represents the system of partial differential equations with the variable coefficient @xmath66 . this is the basic system for most of our further calculations . consider elementary solution of system in the form of damping traveling waves : @xmath67 , @xmath68 , where @xmath69 is a given wave number and @xmath70 is the eigenfrequency to be found , @xmath71 , @xmath72 are the unknown constant coefficients . substituting them into equations and omitting common factors , we obtain @xmath73 where @xmath74 , @xmath75 , % \\ % \label{theta_2 } % \vartheta_2(h)=\xi_2[2-a^2(h - k/2)^2].\end{aligned}\ ] ] system has a nontrivial solution if its determinant is equal to zero , i.e. we obtain quadratic equation which connects @xmath70 and @xmath69 . solving the equation with respect to @xmath70 , we determine the eigenfrequencies of system as @xmath76.\ ] ] here @xmath77 it follows from that there are two eigenmodes for each value of photon number @xmath38 , namely @xmath78 where @xmath79 is the number of the eigenmode , and coefficients @xmath80 , @xmath81 are defined by the expressions @xmath82 and @xmath83 where @xmath84 are normalizing constants , @xmath85 , the values @xmath86 are expressed by the eqs . , , respectively . each mode corresponds to the electron - photon entangled state , which partial amplitudes oscillate both in time and space . the expression with or describes the _ state of radiation - dressed qd - chain _ since the entangled state includes field states with different photon numbers ( @xmath38 and @xmath87 ) . this states are the generalization of the single atom dressed states @xcite for the case of distributed system . the qualitative distinction of this case is the space - time modulation of dressing parameter : it propagates along the qd - chain according to the traveling wave law @xmath88 $ ] . entanglement and dressing are caused by the interaction of light with qd chain and vanish in the limit of @xmath89 . in that case , describes qd - chain electron - hole pairs in equilibrium and inverse states , respectively . spatial oscillations of the partial amplitudes are due to qd - coupling . they vanish in the limit of @xmath90 . generally , both the modes are excited simultaneously . however , any of them can be excited separately by a proper choice of initial conditions . similarly to the other coherent excitations in condensed matter , rabi waves and introduce a new family of quasi - particles ( we name them _ rabitons _ ) . one can apply to them the standard secondary quantization technique . similarly to the effect of the self - induced transparency , the rabi wave propagation can be interpreted as the motion of a precessing pseudo dipole @xcite . however , the coherence mechanisms in these two cases are principally different : in the rabi wave the coherence is settled by the dispersion relation while in the case of the self - induced transparency it has solitonic character . note that the eigenmodes ( [ mode1 ] ) and ( [ mode2 ] ) each comprise traveling waves with _ different _ wave numbers @xmath91 . physically , it means that the rabi wave propagates in an effective periodically inhomogeneous medium formed by spatially oscillating ( with period @xmath92 ) electric field . therefore , the diffraction appears in the system . in the limit @xmath93 , the medium turns homogeneous and the diffraction effect vanishes . reflections of rabi waves and their mutual transformations at the field inhomogeneities become possible . thus one obtain a unique possibility to control the processes of the reflection and transmission of rabi waves by varying the spatial structure of the light . note , that for the waves under consideration _ the reciprocity conditions break down _ : @xmath94 . this is due to the presence of preferential direction , which is determined by the direction of photon mode propagation along the qd - chain ( sign of the value @xmath95 ) . the form of the dispersion characteristics for typical values of parameters is presented on fig.[fig_disp_char ] . dispersion curves for rabi waves for different parameters values . a ) @xmath96 , @xmath97 , @xmath98 , @xmath99 ; b ) @xmath100 , @xmath101 , @xmath102 , @xmath103 , @xmath104 ; c ) @xmath100 , @xmath105 , @xmath106 , @xmath98 , @xmath99 . , title="fig:",scaledwidth=48.0% ] dispersion curves for rabi waves for different parameters values . a ) @xmath96 , @xmath97 , @xmath98 , @xmath99 ; b ) @xmath100 , @xmath101 , @xmath102 , @xmath103 , @xmath104 ; c ) @xmath100 , @xmath105 , @xmath106 , @xmath98 , @xmath99 . , title="fig:",scaledwidth=48.0% ] dispersion curves for rabi waves for different parameters values . a ) @xmath96 , @xmath97 , @xmath98 , @xmath99 ; b ) @xmath100 , @xmath101 , @xmath102 , @xmath103 , @xmath104 ; c ) @xmath100 , @xmath105 , @xmath106 , @xmath98 , @xmath99 . , title="fig:",scaledwidth=48.0% ] to interpret them one should note that in the limit @xmath107 both of branches have cross points , two in general case and one at @xmath108 . modes under examination are characterized by continuous spectrum ( the value @xmath69 varies continuously ) . the values @xmath109 are real for all real @xmath69 . therefore , the system being investigated is stable @xcite . dispersion equation for - can be solved with respect to wave number @xmath110 for given real - valued frequency @xmath70 . in this case for every mode there are such critical values @xmath111 that for @xmath112 the values @xmath113 become complex . physically , it means opacity of the qd - chain for rabi waves with such frequencies@xcite . for @xmath114 the qd - chain is opaque only for one of the modes and for @xmath112 it is non - transparent for both of them . one can choose the system parameters in such a way that in some range of @xmath70 the all real solutions @xmath115 are either negative or positive ( as an example see fig.[fig_disp_char]a ) . this implies one - way opacity of the system ( rabi waves could propagate in one direction only ) , which is caused by nonreciprocity of the qd - chain . the critical frequencies are determined from the condition @xmath116 , where @xmath117 it is significant , that critical values @xmath111 depend on photon number @xmath38 ( see fig.[critical_fr ] ) . this gives an important result , namely : _ the conditions of transparency of the qd - chain are different for rabi waves with different photon numbers_. note , that @xmath118 increases and @xmath119 decreases with increasing of @xmath38 . dependence of critical frequencies of rabi waves on the photon number , for the input parameters as follows : @xmath100 , @xmath120 , @xmath121 , @xmath98 . , scaledwidth=48.0% ] usually both rabi waves demonstrate the normal dispersion and propagate in the _ same direction_. but for a special choice of the parameters the _ abnormal dispersion region _ for one of the modes appears . for this mode phase and group velocities are oppositely directed ( see fig.[fig_disp_char]a ) . the dispersion characteristics have special behavior for absolutely nontransparent ground - state barrier : @xmath122 ( fig.[fig_disp_char]c ) . in this case the group velocity @xmath123 of one of the waves is very small for all @xmath69 , except for the narrow vicinities of the cross - points . due to this fact the modes exchange their places when passing the cross points ( e.g. on fig.[fig_disp_char]c @xmath124 for @xmath125-mode between cross points @xmath126 and for @xmath127-mode at all other values of @xmath69 ) . thus , in current case for a wide range of @xmath69 one of the modes practically does not transfer the energy along the qd - chain . to find the general solution of the system let us introduce the new variable @xmath128 . it leads to the system of partial differential equations with coefficients independent of qd - position @xmath129 : @xmath130&\bm{\phi}_n\\ + a^2k\hat{\sigma}_z\hat{\xi}\partial_x\bm{\phi}_n - ia^2&\hat{\xi}\partial_x^2\bm{\phi}_n=0 , \end{split}\ ] ] where @xmath131 . this system can be solved by using the fourier transform with respect to @xmath129 : @xmath132 substituting eq . into eq . , we obtain the system of ordinary differential equations @xmath133 where @xmath134/2 $ ] . the system can be easily integrated , the solution is given by @xmath135 where @xmath136t}e^{i\hat{\sigma}_z \delta t } e^{i[\bm{m}\hat{\bm{\sigma}}\omega_n(h)]t},\ ] ] @xmath137 , @xmath138 is the unit vector of the form @xmath139 . the expression can be written in matrix form as follows : @xmath140,\ ] ] @xmath141 , \end{aligned}\ ] ] @xmath142 and @xmath143 are determined by and , correspondingly , and @xmath144 can be obtained from initial conditions by the inverse fourier transform : @xmath145 thus , for the vector of probability amplitudes @xmath146 we have the following expression : @xmath147t/2\ } } } dh . \end{split}\ ] ] expression is the central result of this section . it allows one to characterize such observables as inversion , as well as the exciton - exciton and exciton - photon correlators for different initial states of light and qds . in this section we will analyze the propagation of rabi waves with spatially localized initial state . the qd - chain issupposed to be initially either in the ground state ( @xmath148 for all @xmath38 ) , or in the excited state ( @xmath149 for all @xmath38 ) , or in the mixed state ( arbitrary superposition of ground and excited states ) . we describe the spatial - temporal dynamics of rabi oscillations by the spatial density of the inversion ( the inversion per single qd ) @xmath150.\ ] ] space - averaged temporal evolution of rabi oscillations is characterized by the integral inversion @xmath151 let us approximate the initial spatial distributions by the gaussian beams @xmath152 $ ] , where @xmath153 are normalization constants , @xmath154 are the positions of the beams and their widths , respectively ( the indexes @xmath155 refer to excited and ground state of the electron in the qd - chain , respectively ) . the rabi wavepacket evolution can be qualitatively illustrated with the help of the dispersionless approximation for eqs .. according to we obtain for @xmath156 $ ] : @xmath157 where @xmath158 is an arbitrary photonic distribution . one can see from that the spatial spectrum of @xmath159 is given by two gaussian peaks with peak widths @xmath160 and positions of the peak centers @xmath161 , respectively . if @xmath162 are sufficiently large , the largest contribution to the integration in comes from the narrow vicinities of points @xmath163 . assuming @xmath164 and @xmath165 ( the dispersion effects neglecting ) and doing standard calculations , we obtain from eq . : @xmath166e^{-ikx/2}\right.\\ & + \left.\left[b_{n+1}(x+v_1^+t,0)\eta_1 ^ -e^{i\theta_1^+t}+b_{n+1}(x+v_1 ^ -t,0)\eta_1^+e^{i\theta_1 ^ -t}\right]e^{ikx/2}\right\},\\ \label{appr_sol_b } \nonumber b_{n+1}(x , t)&=e^{-i(kx-\omega t)/2}e^{-\lambda t } \left\{\left[a_{n}(x+v_2^+t,0)\eta_2 ^ -e^{i\theta_2^+t}+a_{n}(x+v_2 ^ -t,0)\eta_2^+e^{i\theta_2 ^ -t}\right]e^{-ikx/2}\right.\\ & + \left.\left[b_{n+1}(x+v_1^+t,0)\zeta_1 ^ -e^{i\theta_1^+t}+b_{n+1}(x+v_1 ^ -t,0)\zeta_1^+e^{i\theta_1 ^ -t}\right]e^{ikx/2}\right\},\end{aligned}\ ] ] where the velocities @xmath167 are defined as @xmath168 @xmath169 and @xmath170 and @xmath171 , @xmath172 are introduced to denote the frequency shifts and amplitude factors , respectively : @xmath173-v^{\pm}(h_{1,2}^{(0 ) } ) , \ \\ \label{zeta } \zeta_{1,2}^{\pm}&=\frac{\omega_n(h_{1,2}^{(0 ) } ) \pm \delta_{eff}(h_{1,2}^{(0)})}{2\omega_n(h_{1,2}^{(0 ) } ) } , \ \\ \label{eta } \eta_{1,2}^{\pm}&=\pm \frac{g\sqrt{n+1}}{\omega_n(h_{1,2}^{(0)})}. \ \\end{aligned}\ ] ] as is apparent from , generally , any probability amplitude in the rabi - wave packet is the _ superposition of four subpackets_. two of them ( first and second terms in expressions ) correspond to the excited initial state and two another ( third and fourth terms in ) correspond to the ground initial state . the partial subpackets in eqs . are characterized by the different frequency shifts @xmath170 and different velocities of motion @xmath167 , given by , . this velocities coincide with the group velocities of traveling rabi waves when @xmath174 : one can elementary verify , that @xmath175 , @xmath176 . it is essential that velocities @xmath167 as well as the frequency shifts @xmath170 depend on photon number @xmath38 . it means that the spatial propagation of wavepacket is accompanied by the change of quantum light statistics ( for example , in initially coherent light incoherent component appears ) . under the some specific conditions the number of subpackets could decrease . two mechanisms of such decrease are possible : the first one is the tending to zero the subpacket amplitude and the second one is the confluence of the subpackets velocities . as is seen from , if @xmath177 for one pair of the subpackets the velocity synchronism condition @xmath178 is fulfilled . notice that eq . is the analogue of the ordinary synchronism condition in single two - level system @xmath179 . the velocity synchronism condition can be fulfilled by fitting the value of the detuning @xmath180 . if this is the case , there are three subpackets instead of four them due to the confluence mechanism acting . note , that @xmath181 and @xmath182 can not be equal to zero together at @xmath183 . if the system is initially prepared in the stationary state , and the synchronism condition is fulfilled in the corresponding point ( for example @xmath184 for any @xmath38 and @xmath185 ) , only one subpacket is preserved . in this case the expression for the inversion density becomes rather simple : @xmath186.\ ] ] expression sums up contributions of different photonic states ( terms with different @xmath38 ) . each contribution is the product of two multipliers describing different optical processes . the second multiplier describes rabi oscillations with the frequencies @xmath187 , occurring in the qd - chain , while the first one shows that the region of rabi oscillations moves therewith in space with velocity @xmath188 . one should note , that in the case of the exact synchronism @xmath189 does not depend on @xmath38 . that is why in this partial case the quantum statistics of light is not distorted with propagation of the wavepacket . in particular , the initial coherency of quantum light persists in time . it is immediately follows from eq . and normalization condition that the integral inversion @xmath190 corresponding to the inversion density oscillates in time similarly to the rabi - oscillations in the single two - level system ( see eq.(6.2.21 ) in @xcite at @xmath102 ) . generally , each pair of the subpackets propagates due to the tunneling through its own potential barrier the velocities of the subpackets motion are proportional to the barrier transparency ( @xmath191 ) . if one of the barriers becomes absolutely opaque , the corresponding pair of subpackets does not move : @xmath192 at @xmath193 . the light propagation _ along the qd - chain _ is another necessary condition of rabi subpackets motion : @xmath192 at @xmath194 . for more detail analysis of rabi wavepackets dynamics ( in particular for taking into account the diffraction spreading ) the integrals need to be calculated numerically . the results of calculations is analysed in the following section . consider the case of an excited initial state with space distribution in form of single gaussian beam : @xmath195{\pi\sigma^2}$ ] , @xmath184 . assume that initially light is prepared in coherent state , so photon distribution is given by the poisson law : @xmath196 , where @xmath197 defines the average photon number . the spatial - temporal dynamics of inversion density for this case is depicted on fig.[fig1]a . as is seen from it , original gaussian packet oscillates in time and moves along the chain . oscillations collapse to zero quickly , but after a while they revive _ in another area of space_. the phenomenon of rabi oscillations collapses and revivals , caused by discreteness of photon distribution , is well studied @xcite , but the spacing the collapse and subsequent revival is qualitatively new effect . space - time distribution of the inversion density ( a ) and temporal dependence of the integral inversion ( b ) in the qd chain for a coherent initial state of light ( @xmath198 ) . the initial state of qd - chain is a single gaussian wavepacket . @xmath195{\pi\sigma^2}$ ] , @xmath184 , @xmath199 , @xmath200 , @xmath121 , @xmath201 , @xmath202 , @xmath203.,title="fig:",scaledwidth=48.0% ] space - time distribution of the inversion density ( a ) and temporal dependence of the integral inversion ( b ) in the qd chain for a coherent initial state of light ( @xmath198 ) . the initial state of qd - chain is a single gaussian wavepacket . @xmath195{\pi\sigma^2}$ ] , @xmath184 , @xmath199 , @xmath200 , @xmath121 , @xmath201 , @xmath202 , @xmath203.,title="fig:",scaledwidth=48.0% ] space - time distribution of the inversion in the qd chain for the case of two counterpropagating wavepackets . @xmath204/\sqrt[4]{4\pi\sigma^2}$ ] , @xmath205/\sqrt[4]{4\pi\sigma^2}$ ] . the light is initially in coherent state ( @xmath206 ) , @xmath207 , @xmath208 , @xmath209 . a ) @xmath210 , @xmath211 ; b ) @xmath202 , @xmath203.,title="fig:",scaledwidth=48.0% ] space - time distribution of the inversion in the qd chain for the case of two counterpropagating wavepackets . @xmath204/\sqrt[4]{4\pi\sigma^2}$ ] , @xmath205/\sqrt[4]{4\pi\sigma^2}$ ] . the light is initially in coherent state ( @xmath206 ) , @xmath207 , @xmath208 , @xmath209 . a ) @xmath210 , @xmath211 ; b ) @xmath202 , @xmath203.,title="fig:",scaledwidth=48.0% ] although a variation of the inversion density , depicted in fig.[fig1]a , in arbitrary point of the space occupied by the rabi wavepacket is not too large , the integral inversion , presented in fig.[fig1]b oscillates between -1 and 1 , thus indicating presence of strong light - qd coupling . temporal dependence of the integral inversion for a coherent initial state of light at the input parameters as follows : @xmath206 , @xmath202 , @xmath203 ( solid line ) ; @xmath206 , @xmath210 , @xmath211(dotted line ) @xmath212 , @xmath202 , @xmath211(dashed line ) . in all cases , @xmath208 , @xmath207 , @xmath209 . , scaledwidth=48.0% ] the asymmetry of the inversion space - time distribution for two counterpropagating wavepackets in case of the coherent state of light with small average number of the photons @xmath212 . @xmath204/\sqrt[4]{4\pi\sigma^2}$ ] , @xmath205/\sqrt[4]{4\pi\sigma^2}$ ] , @xmath208 , @xmath207 , @xmath209 , @xmath202 , @xmath211.,scaledwidth=48.0% ] let us consider now the case of a mixed initial state ( both @xmath213 and @xmath214 are nonzero ) . as it has been noted above , the synchronism condition can not be fulfilled for both pair of subpackets ( except the case when one of the pairs is motionless , see below ) . as a consequence , we have four subpackets ( or three ones if @xmath215 for one of the pairs ) . if @xmath216 ( potential barriers with equal transparency ) and @xmath179 , then both packet pairs move with pairwise equal velocities in opposite directions ( @xmath217 ) . if the initial spatial distributions of probability amplitudes for ground and excited states are spaced , the subpackets will collide ( see fig.[double ] ) . initial gaussian profiles of packets deform with time due to the difference in subpacket velocities . one should note that for the case of quantum external field in coherent state qd - chain can not be saturated , i.e. the integral inversion could not be a constant ( see fig . [ int_inv2 ] ) . this is due to the fact that @xmath218 and @xmath219 are asymmetrical with respect to @xmath38 and as a result @xmath220 is not a constant for any form of space dependence of @xmath218 , @xmath219 . this asymmetry exists due to the presence of vacuum term in the sum and increases with decreasing of the average photon number @xmath197 ( compare fig.[double ] and fig.[double_3 ] ) . in other limiting case , for @xmath122 ( fully opaque ground - state barrier ) , the velocities @xmath221 and one pair of subpackets do not moves along the chain . if , moreover , @xmath222 , then @xmath223 , and the synchronism condition is fulfilled for both packet pairs ( see fig.[double2 ] ) . space - time distribution of the inversion in the qd chain for the case of the fully opaque ground - state barrier . @xmath204/\sqrt[4]{4\pi\sigma^2}$ ] , @xmath205/\sqrt[4]{4\pi\sigma^2}$ ] , @xmath199 , @xmath122 , @xmath224 , @xmath201 , @xmath202 , @xmath203 . the light is initially in the coherent state with @xmath206,scaledwidth=48.0% ] let us now consider the interaction of the qd - chain with another important state of initial electromagnetic field , namely , the fock qubit state , i.e. a superposition of two fock states : @xmath225 , where @xmath226 is an arbitrary fixed number , @xmath227 satisfy the normalization conditions . as in the previous case , for initial spatial distributions the gaussian beams will be used . the spatial - temporal dynamics of the inversion density for the initial state of field @xmath228 is shown in fig . [ fig_fock_qubit]a . it can be seen from the figure that rabi wavepacket does not collapse and revive , but oscillates in a complicated manner ( see also integral inversion dynamics , fig.[fig_fock_qubit ] ) . space - time distribution of the inversion ( a ) and temporal dependence of the integral inversion ( b ) in the qd chain for the field in the fock qubit initial state . @xmath228 , @xmath229{\pi\sigma^2})$ ] @xmath230 , @xmath148 for @xmath231 , @xmath149 for all @xmath38 , @xmath199 , @xmath200 , @xmath121 , @xmath202 , @xmath209 , @xmath203.,title="fig:",scaledwidth=48.0% ] space - time distribution of the inversion ( a ) and temporal dependence of the integral inversion ( b ) in the qd chain for the field in the fock qubit initial state . @xmath228 , @xmath229{\pi\sigma^2})$ ] @xmath230 , @xmath148 for @xmath231 , @xmath149 for all @xmath38 , @xmath199 , @xmath200 , @xmath121 , @xmath202 , @xmath209 , @xmath203.,title="fig:",scaledwidth=48.0% ] in the end of this section we consider the vacuum initial state of external electromagnetic field . it is well known that for this case the vacuum harmonic rabi oscillations occuring due to spontaneous emission can take place @xcite . the corresponding rabi - wave packet in the initially excited qd - chain is illustrated at fig.[fig_fock_qubit ] . this excitation couples single mode electromagnetic vacuum state with one - photon state only and can be imagined as a wave beam which is described by the monochromatic rabi - frequency spectrum and continuous spatial spectrum at the same time . however , it should be noted that one need to consider an interaction of the qd - chain with a multi - mode field for full treatment of such effects @xcite . this problem is a subject for future considerations . space - time distribution of the inversion in the qd chain for the vacuum initial state of the field . @xmath232{\pi\sigma^2})$ ] @xmath230 , @xmath148 for @xmath233 , @xmath149 for all @xmath38 , @xmath234 , @xmath235 , @xmath121 , @xmath210 , @xmath201 , @xmath203.,scaledwidth=48.0% ] the case of a coherent state of electromagnetic field with the large photonic number @xmath38 is especially interesting for us . in this case one can neglect the quantum nature of the light and consider electromagnetic field as a classical one replacing the electric field operator with its expectation value @xmath236\}$ ] . in doing this we neglect any changes of the field state . omitting the local - field effects as before , we can write the hamiltonian of the system `` qd - chain - electromagnetic field '' in the form @xmath237 , where the term @xmath238\ ] ] describes rabi oscillations in non - interacting qds and @xmath239 is the rabi frequency @xcite , the interdot interaction mechanism is independent on light properties , so the term @xmath240 is defined as before by equation . the state vector of the system has the form : @xmath241 where @xmath242 , @xmath243 are unknown functions . now the equations of motion for them can now be obtained using approach of sec.[sec : mot_eq ] . introducing the decay factor @xmath49 into the nonstationary schrdinger equation and making continuous limit transition in the qd - chain similar to the sec.[sec : cont_lim ] , we obtain the system of equations as follows : @xmath244\bm{\psi}\\+ia^2\hat{\xi}\partial^2_x\bm{\psi } , \end{split}\ ] ] where @xmath245 $ ] . note , that spontaneous emission can be taken into account in the same way as in [ sec : mot_eq ] , i.e. by describing it as interaction of qds with photon bath . in this case the damping constant @xmath49 in eq . will be the sum of two partial constants . the system can be considered as a particular case of the general system and can be solved in the same way . the solution reads @xmath246t/2\ } } } dh . \end{split}\ ] ] where the values @xmath247 , @xmath248 , @xmath249 , @xmath250 , @xmath251 , @xmath143 are determined , respectively , by equalities , , , , , if one makes the substitution @xmath252 ( in doing that the photonic number dependence in @xmath247 , @xmath253 , @xmath254 , @xmath255 disappears ) . the vector function @xmath144 is determined by the initial conditions similar to eq .. space - time distribution of the inversion in the qd in the classical light chain for a single gaussian wavepacket @xmath256{\pi\sigma^2})$ ] @xmath230 , @xmath257 , @xmath258 , @xmath259 , @xmath121 , @xmath209 , @xmath260 . a ) @xmath261 ; b ) @xmath262,title="fig:",scaledwidth=48.0% ] space - time distribution of the inversion in the qd in the classical light chain for a single gaussian wavepacket @xmath256{\pi\sigma^2})$ ] @xmath230 , @xmath257 , @xmath258 , @xmath259 , @xmath121 , @xmath209 , @xmath260 . a ) @xmath261 ; b ) @xmath262,title="fig:",scaledwidth=48.0% ] a typical space - time distribution of the inversion density @xmath263 $ ] is shown in fig . [ fig2 ] . as is seen from the figure , the space - time dynamics of rabi wavepacket is similar to the one for the case of vacuum oscillations ( but it is of substantially different physical nature : electromagnetic field is in coherent state with photon number @xmath264 and this state does not change when the system oscillates between ground and excited states in contradiction to the case of vacuum rabi oscillation ) . since the photon distribution does not change , collapses and revivals are absent . however there is another phenomenon of similar nature . one can see in fig . [ fig3 ] , where the plots of the integral inversion are presented , that the oscillations of this quantity at @xmath265 , @xmath266 decrease with time even at @xmath267 , whereas such a damping is absent at @xmath268 , @xmath108 and integral inversion oscillates harmonically in the range from -1 to 1 ( dotted curve in fig . [ fig3 ] ) . temporal dependence of the integral inversion in the classical light for the input parameters as follows : @xmath102 , @xmath268 , @xmath269 ( dotted line ) ; @xmath270 , @xmath271 , @xmath269 ( solid line ) ; @xmath121 , @xmath271 , @xmath272 , @xmath273 ( dashed line ) . in all cases @xmath274 , @xmath211 . , scaledwidth=48.0% ] such a behaviour is the consequence of dependence of the effective detuning @xmath143 ( and therefore the rabi frequency @xmath251 ) on @xmath69 . but if @xmath268 and @xmath108 , then @xmath275 and @xmath255 does not depend on @xmath69 . the damping rate is controlled by values of wavenumber @xmath95 and coupling constants @xmath27 . if the initial state of the system is the excited state ( @xmath276 ) , the damping rate is predominantly determined by product @xmath277 ( and damping is practically absent if @xmath278 ) , for the ground initial state ( @xmath279 ) it is defined by @xmath280 . in the case of a mixed initial state all three quantity are important . note , that more narrow wavepacket spreads faster , than the wider one ( compare fig . [ fig2]a and fig . [ fig2]b ) . in the weak coupling limit the indicated dephasing mechanism is analogous to the landau damping in plasma space - time distribution of the inversion density in the qd chain in the classical light for two counterpropagating identical gaussian wavepackets : @xmath281/\sqrt[4]{4\pi\sigma^2}$ ] , @xmath282/\sqrt[4]{4\pi\sigma^2}$ ] , @xmath102 , @xmath209 , @xmath261 , @xmath283 , @xmath260.,scaledwidth=48.0% ] interaction of two counterpropagating identical gaussian rabi wavepackets colliding at @xmath284 is shown in fig . [ fig4 ] . unlike the case of quantum field , the space - time distribution of the inversion density is fully symmetrical with respect to the line @xmath284 and integral inversion does not oscillate : this quantity equals zero for all @xmath285 and arbitrary values of @xmath286 . local fields are depolarization fields arising inside qds due to the dipole - dipole electron - hole interaction . as a result , the field acting on the e - h pair differs from mean field @xcite . the local - field action is described by eq . and leads to nonlinear terms in the equations of motion ( last terms in eqs . the presence of nonlinearity results in the qualitatively new features of rabi oscillations in single qds @xcite . the influence of the local - fields on rabi waves is the subject for future investigations . in this section we consider some aspects of this influence only and discuss some sufficient conditions of the local - field negligibility . let us consider finite ( of the length @xmath287 ) qd - chain interacting with classical light . the relaxation processes are neglected ( i.e. @xmath211 ) . the equations of motion follow from eqs.- in the same way as eq . reduce to eq . and have the form @xmath288 we use the periodic boundary conditions ( born - von karman conditions ) : @xmath289 suppose that electromagnetic field are also satisfy periodic conditions , i.e. @xmath290 , where @xmath291 is integer number . then the problem under consideration has a clear physical interpretation : we are dealing with ringed qd - chain of the radius @xmath292 . qd - chain interacts with electromagnetic mode of whispering gallery type , which is excited in microcavity with rotation axis ( @xmath293 , @xmath294 , @xmath295 is the azimuthal coordinate ) . let us seek partial solution of the system - in the form of traveling wave : @xmath296 , @xmath297 , where @xmath298 , @xmath299 are constant amplitudes satisfying the normalization condition for wavefunction , @xmath69 and @xmath70 are real values . ssubstituting @xmath300 and @xmath301 into eqs.- we obtain @xmath302u_0 + \frac{\omega_r}{2}v_0=0 , \\ \rule{0in}{4ex } \label{char_eq_2b } & \left[\nu+\phi_2-\delta\omega|u_0|^2\right]v_0 + \frac{\omega_r}{2}u_0=0,\end{aligned}\ ] ] where @xmath303 . from boundary conditions one can obtain the quantization condition for @xmath69 : @xmath304 , @xmath38 is integer . expressing @xmath299 from eq . and substituting into eq . , we obtain @xmath305 the wavefunction normalization condition results in @xmath306 eqs.- form the closed system of equations with respect to @xmath70 and @xmath307 , which defines the dispersion law of rabi waves @xmath308 with regard to the local - field effects . it should be noted that only solutions with real - valued @xmath70 have physical meaning ( otherwise the system - contradicts to the initial equations - ) . for @xmath309 , @xmath310 the system - reduces to the dispersion equation @xcite for rabi waves in the infinite qd - chain interacting with classical light . dispersion curves for rabi waves subject to local - field effects . a ) @xmath311 , b ) , c ) @xmath312 . in all cases @xmath269 , @xmath313 , @xmath102 , @xmath314 . , title="fig:",scaledwidth=48.0% ] dispersion curves for rabi waves subject to local - field effects . a ) @xmath311 , b ) , c ) @xmath312 . in all cases @xmath269 , @xmath313 , @xmath102 , @xmath314 . , title="fig:",scaledwidth=48.0% ] dispersion curves for rabi waves subject to local - field effects . a ) @xmath311 , b ) , c ) @xmath312 . in all cases @xmath269 , @xmath313 , @xmath102 , @xmath314 . , title="fig:",scaledwidth=48.0% ] the dispersion characteristics of rabi waves for two essentially different values of @xmath287 but identical @xmath315 and other parameters are shown in fig.[nonlin_disp ] . the curves depicted in fig.[nonlin_disp]a corresponds to the qd - chain of large radius . the nonlinear effects in this case are negligibly small . dispersion curves do not feel nonlinear effects and are practically identical to obtained in @xcite . for qd - chain of small radius the situation becomes principally different due to nonlinear effects . the _ closed loop _ appears in the vicinity of any cross point and the number of the modes increases to four ( fig.[nonlin_disp]b ) . the size of this loop decreases with decreasing the ratio @xmath316 . the detailed structure of closed loop is shown in fig.[nonlin_disp]c . when @xmath69 ( and , correspondingly , @xmath38 ) changes adiabatically slow the motion along the loop takes place ( it is shown by arrows in fig.[nonlin_disp]c ) . the transitions from one branch to another are followed by simultaneous changes of rabi frequency and rabi wave group velocity . the trajectories of motion in different directions proves to be nonidentical . thus , the loop has an hysteresis character . interaction of the local - fields and rabi waves leads to effect of bistability of quantum states in the qd - chain . the detailed analysis of different dispersion branches stability is the subject for future investigations . let us now discuss the conditions of the local field effects neglecting in equations . note , that for @xmath317 the system transforms to equations ( 24 ) in @xcite , where rabi oscillations in single qd are described with account of local field effects . an analysis of equations ( 24 ) in @xcite has shown that local field effects are negligibly small for the case of sufficiently large rabi frequencies , namely , @xmath318 , where @xmath319 , @xmath197 is average photon number . if this condition is fulfilled for coherent state , the conventional picture of rabi oscillations ( the alternation of collapses and revivals ) takes place . in the opposite case qualitatively new regime of oscillations arises . this regime is due to the local field effects and is described in detail in@xcite . in this work we are interested only in first regime . in our case the role of @xmath320 is played by frequencies @xmath321 , which are determined by eq .. therefore it is reasonable to suggest that for traveling rabi waves the local fields effects are negligible under the condition @xmath322 where @xmath315 is determined by eq .. in the case of rabi - wavepackets general line of reasoning is the same as for eq . and leads to the following condition of the local - field negligibility : @xmath323 . the space - time dependence of normalized correlation function @xmath324 for light in fock state with @xmath325 and the input parameters as follows : @xmath326 , @xmath327 , @xmath274 , @xmath328{\pi\sigma^2}$ ] , @xmath329.,scaledwidth=48.0% ] in this section we will calculate exciton - exciton and exciton - photon multi - time correlation functions . for this purpose it is convenient to pass to the heisenberg picture and to use the discrete model of qd - chain ( transition to the continuous limit will be made on the final stage ) . thus , we should look for a time dependence of operators @xmath330 , i.e. we shall seen to find the evolution operator . it can be obtained from ( for details see appendix [ sec : evol_operator ] ) and has the form @xmath331\\ & \left.+{\varphi}_n^+ ( h , t ) e^{i\delta_h^-t } \left|b_p , n+1\right\rangle\left\langle b_q , n+1\right| \right\}e^{ih(p - q)a}dh . \end{split}\ ] ] here @xmath332 , @xmath333 , @xmath334 are given by formulas , , . quantities @xmath335 , @xmath143 are defined by expressions , if one substitutes @xmath336 $ ] . note , that if one makes approximation @xmath337 \cong1-(h\pm k/2)^2a^2/2 $ ] ( this is the usability condition for continuous limit ) , the quantities @xmath335 , @xmath143 coincide with expressions , . from expression we obtain by using the formula @xmath338 the exciton - exciton correlation functions @xmath339 as follows : @xmath340\\ [ \varphi_{n+1}^+(h , t)\varphi_{n+1}^-(g , t')u_{l , n+1}^*(0)u_{r , n+1}(0)+ \psi_{n+1}(h , t)\psi_{n+1}(g , t')v_{l , n+2}^*(0)v_{r , n+2}(0 ) \\ \rule{0in}{3ex } \nonumber+\varphi_{n+1}^+(h , t)\psi_{n+1}(g , t')u_{l , n+1}^*(0)v_{r , n+2}(0)- \psi_{n+1}(h , t)\varphi_{n+1}^-(g , t')v_{l , n+2}^*(0)u_{r , n+1}(0)]dh'dhdg,\end{gathered}\ ] ] where @xmath341 , @xmath342 . the continuous limit transition can be made by means of the following substitutions : @xmath343 , @xmath344 , @xmath345 , @xmath346 . one can see , that in general case @xmath347 can not be represented as @xmath348 or @xmath349 . it indicates the absence both spatial and time homogeneity . let us consider case of @xmath350 and @xmath351 . then @xmath352 , @xmath353 and taking into account that @xmath354 , we obtain rather simple expression for correlation function : @xmath355 or , in continuous limit , @xmath356 plot of normalized correlation function @xmath357 for @xmath358 , @xmath351 is presented on fig.[correl ] . the external field is supposed to be in a single - photon initial state and the electron in qd - chain is in the excited initial state ( @xmath184 ) . the correlation function of the polarization operator @xmath359 with the operator of the external field @xmath360 ( exciton - photon correlators ) appears as @xmath361 \\ \nonumber -\psi_{n+1}(h',t)\sqrt{n+2}\left[\psi_n(h , t)v^*_{r , n+1}(0)u_{l , n}(0)+\varphi_n^+(h , t)v^*_{r , n+1}(0)v_{l , n+1}(0)\right ] \ } e^{ipa(h - h')}e^{i(h'r - hl)a}e^{i(kpa-\omega_0 t)}e^{i\omega(t - t')}.\end{gathered}\ ] ] the main results , obtained in this work , can be summarized as follows : \i ) in qd - chain with tunneling coupling in strong electron - photon coupling regime the space propagation of rabi oscillations ( rabi waves ) takes place . for propagation of rabi waves the wave vector of the photon mode must have a nonzero component along the chain . characteristics of the rabi waves depend strongly on relations between parameter of electron - photon coupling , frequency deviation and transparency factors of tunneling barriers for both of levels . \ii ) traveling rabi waves are quantum states of the qd - chain dressed by radiation . the qualitative distinction of this states from states of single dressed atom @xcite is the space - time modulation of dressing parameter by traveling wave law . traveling rabi waves can be interpreted as entangled states of e - h pair and photons . these states are characterized by the dependence of energy on quasi - momentum and can be treated as new type of quasiparticles ( in this paper they are called rabitons ) . \iii ) there are two traveling rabi modes with different frequencies of rabi oscillations at a given value of wavenumber . the range of rabi oscillations frequencies is limited by the critical value , which is different for both modes . the qd - chain is opaque for rabi modes with frequencies lesser than critical one . the critical frequencies as well as dispersion characteristics depend on number of photons in the mode . \iv ) in general case the propagation of rabi wavepackets occurs in the form of four partial packets , which are characterized by different amplitudes and velocities of motion . rabi wavepackets transfer energy , inversion , quasimomentum , electron - electron and electron - photon quantum correlations along the chain . particularly , in the case of qd - chain interacting with quantum light in the coherent state the known collapses - revivals picture @xcite is drastically modified due to propagation effect : collapses and revivals take place in different points of space . \v ) one pair of rabi wavepackets exists due to the ground state contribution in initial state of the system , another one appears due to the contribution of excited state ( generally , initial state is arbitrary superposition of ground and excited states ) . each pair of packets moves through its own potential barrier . if one of the barriers is completely opaque , then corresponding pair of wavepackets does not move and does not spread . \vi ) rabi wavepackets propagation along the qd - chain is followed by the transformation of quantum light statistics ( for example , in initially coherent light incoherent component appears ) . if specific conditions are fulfilled then incoherent part can be decreased to zero . \vii ) in particular cases a number of rabi wavepackets can be diminished . there are two mechanisms of such diminution : tending to zero of partial wavepacket amplitude and merging of the packets due to their velocities equality . rabi waves can take place in a number of other distributed systems strongly coupled with electromagnetic field . the example are superconducting circuits based on josephson junctions , which are currently the most experimentally advanced solid - state qubits @xcite . it is evident , for example , that the chain of qubits placed inside a high - q transmission - line resonator due to the qubit - qubit capacitance coupling will support the rabi waves propagation similar to described in this article . rabi waves effect can be practically used in nanoelectronics , quantum computing , quantum informatics . the work of g. ya . slepyan was partially carried out during the stay at the institut fr festkrperphysik , tu berlin , and supported by the deutsche forschungsgemeinschaft ( dfg ) . authors are grateful to dr . s. a. maksimenko , dr . d. s. mogilevtsev , dr . j. haverkort and dr . a. m. nemilentsau for stimulative discussions . in this appendix , the evolution operator for a discrete qd - chain ( eq . ) is obtained from eq . let us introduce new variables @xmath362 , @xmath363 . it leads to the system coefficients in this system do not depend from qd - number @xmath1 . it allows us to seek the solution in the form @xmath365 , @xmath366 , @xmath367 $ ] . then the equations for different qds become independent and may be written as @xmath368 u_{p , n}-i g\sqrt{n+1}v_{p , n+1 } e^{i\delta t},\\ \label{system_diskr4_b } & \partial_tv_{p , n+1}=2i\xi_2 \cos [ ( h - k/2)a]v_{p , n+1}-i g\sqrt{n+1}u_{p , n } e^{-i\delta t}.\end{aligned}\ ] ] expressing @xmath369 from and substituting into , one obtains the set of independent second - order ordinary differential equations for the variables @xmath370 @xmath371 where @xmath372 + 2\xi_2 \cos [ ( h - k/2)a]$ ] . the solution of reads @xmath373t/2}+c_2^n(0)e^{i\left[\delta_h-\omega_n(h)\right]t/2}\ ] ] with the constants @xmath374 to be determined from initial conditions . writing the latter in the form @xmath375 , @xmath376 , we rewrite as @xmath377t}+\beta_ne^{i\left[\delta_h^++\omega_n(h)/2\right]t}\right ] e^{ihpa}\ ] ] and @xmath378t}+\delta_ne^{-i\left[\delta_h^--\omega_n(h)/2\right]t}\right ] e^{ihpa},\ ] ] where @xmath379 we now can represent the required wavefunction as fourier integral with basis functions @xmath381 : @xmath382 where @xmath383 is unknown weighting function . to determine it let us take into account , that substituting into , we obtain @xmath389 , with @xmath390 given by the eq .. these calculations allow one to pass from schrdinger picture to the heisenberg one . they are used in the calculations of different types of correlators . in hamiltonian @xmath391 the constant term describing the photon vacuum contribution is omitted . in singlemode assumption its influence identically equal to zero . in so doing we are not interested in effects of spontaneous decay and lamb shift . this effects are caused by interaction of the system with photon vacuum and can be taken into account in multimode assumption only .

we determine the prospects for finding dark matter at the tevatron and lhc through the production of exotic 4th generation quarks that decay through , where is dark matter . the resulting signal of has not previously been considered in searches for 4th generation quarks , but there are both general and specific dark matter motivations for this signal , and with slight modifications , this analysis applies to any scenario where invisible particles are produced in association with top quarks . current direct and indirect bounds on such exotic quarks restrict their masses to be between 300 and 600 gev , and the dark matter s mass may be anywhere below . we simulate the signal and main backgrounds with madgraph / madevent - pythia - pgs4 . for the tevatron , we find that an integrated luminosity of will allow 3 discovery up to and 95% exclusion up to . for the 10 tev lhc with , the discovery and exclusion sensitivities rise to 490 gev and 600 gev . these scenarios are therefore among the most promising for dark matter at colliders . perhaps most interestingly , we find that dark matter models that can explain results from the dama , cdms and cogent collaborations can be tested with high statistical significance using data already collected at the tevatron and have extraordinarily promising implications for early runs of the lhc .

1002.3366

one of the great hopes for current and future particle colliders is that they will be able to produce dark matter . in this study , we determine the prospects for finding dark matter through the production of exotic 4th generation quarks @xmath0 that decay through @xmath1 , where @xmath2 is dark matter . current direct and indirect bounds on such exotic quarks restrict their mass range to @xmath10 . our analysis is valid for all dark matter masses up to @xmath11 , although there are special reasons to be interested in very light @xmath2 particles , with @xmath12 . there are both general and specific dark matter motivations for this signal . starting with the general motivation , one of the few things that is absolutely certain about dark matter is that it must be long - lived on cosmological time scales . this is typically achieved by giving dark matter a charge under an unbroken discrete or continuous symmetry , which makes it absolutely stable . none of the unbroken symmetries of the standard model ( sm ) will do for this purpose , so the dark matter particle must be charged under a new unbroken symmetry . there are then two options . the dark matter , with its stabilizing `` dark charge , '' may have only gravitational interactions with the sm . in this case , dark matter may have interesting astrophysical signals @xcite , but it can not be discovered at colliders . alternatively , the dark matter may be coupled to sm particles @xmath13 through connector particles @xmath14 that have both dark and sm charges to make @xmath15 couplings possible . @xmath2 may or may not have sm weak interactions . however , @xmath14 necessarily has sm charge . it can therefore be produced at colliders , and so dark matter can be discovered through @xmath14 production followed by @xmath16 . since the energy frontier is dominated by hadron colliders for the foreseeable future , the most promising case is where @xmath14 is strongly interacting . supersymmetric ( universal extra dimension ) models provide a concrete example of this , where the dark matter is neutralinos @xcite ( kaluza - klein ( kk ) gauge bosons @xcite ) , the connector particles are squarks ( kk quarks ) , and the stabilizing symmetry is @xmath17-parity ( kk - parity ) . here we consider the case where the dark matter has no sm gauge interactions , the connector particles are exotic quarks , and the stabilizing symmetry may be either discrete or continuous @xcite . note , however , that with minor modifications , our analysis applies much more generally , both to the supersymmetric and extra dimensional scenarios just mentioned , as well as to many other dark matter scenarios motivated by the general chain of reasoning given above . these scenarios also have a more speculative , but at the same time more specific and tantalizing , dark matter motivation . the dama experiment sees an 8.9@xmath6 signal in the annual modulation of scattering rates that can potentially be explained by dark matter @xcite . uncertainties from both astrophysics @xcite and detector response @xcite open the possibility that dama can be explained without conflicting with other experiments by a light dark matter particle with mass @xmath18 elastically scattering off nucleons with spin - independent cross section @xmath19 @xcite . this explanation is supported by unexplained events recently reported by the cogent collaboration @xcite , which , if interpreted as a dark matter signal , are best fit by dark matter with @xmath20 and @xmath21 . the low mass and high cross sections preferred by dama and cogent are consistent with the recent bounds and results from cdms @xcite . such large cross sections are several orders of magnitude larger than those of typical weakly - interacting massive particles ( wimps ) . as we review below , however , they are easily obtained if the dark matter particle scatters through @xmath22 , where @xmath23 is an exotic 4th generation quark . furthermore , this scenario naturally emerges in wimpless dark matter models , where dark matter not only can have the correct @xmath24 and @xmath25 for dama and cogent , but also naturally has the correct thermal relic density @xcite . this scenario is a special case of the general framework described above , and , as discussed in sec . [ sec : model ] , particularly motivates the case where dark matter couples to third generation quarks and the @xmath2 particles are light . given all of these motivations , we explore here the detection prospects for exotic 4th generation quarks decaying directly to dark matter . this signal differs from most supersymmetry searches , which typically assume that decays to dark matter are dominated by cascade decays . the 4th generation quarks examined here also differ from the 4th generation quarks that are typically studied , because they are charged under a new symmetry under which sm particles are neutral . this forbids decays to sm quarks , such as @xmath26 and @xmath27 , which are the basis for most standard 4th generation quark searches . instead , all @xmath23 decays must necessarily produce hidden sector particles charged under the new symmetry , with the lightest such particle being @xmath2 , the dark matter . this leads to the signal of @xmath28 , which has previously not been considered in searches for 4th generation quarks . in addition , as emphasized earlier , this type of signals appear in a general set of dark matter motivated models , as well as other new physics scenarios , such as little higgs models with @xmath29-parity conservation @xcite and models in which baryon and lepton number are gauge symmetries @xcite . the results of our analyses can be easily applied to these other models . we will find that @xmath0 pair production followed by @xmath30 leads to @xmath31 signals that may be discovered at the tevatron with @xmath32 of integrated luminosity , or at the lhc with integrated luminosities of as little as @xmath33 . in sec . [ sec : model ] we detail the dark matter motivations and define the model we explore . in sec . [ sec : constraints ] we summarize the current theoretical and experimental constraints on exotic 4th generation quarks . we then describe the details of our signal and background simulations and cut analysis in sec . [ sec : simulation ] . in sec . [ sec : discovery ] we present the prospects for discovering or excluding these exotic 4th generation quark scenarios with tevatron and early lhc data . we conclude with a discussion of future prospects in sec . [ sec : conclusions ] . as discussed above , in this work we consider the case where dark matter has a charge under some new symmetry , but no sm gauge interactions . this dark matter is therefore not a typical wimp , but it may nevertheless naturally appear in theories motivated by the gauge hierarchy problem and have the correct thermal relic density . this is the case for wimpless dark matter models @xcite , supersymmetric models where the effects of a supersymmetry - breaking sector are transmitted to both the minimal supersymmetric sm ( mssm ) sector and a hidden sector through gauge - mediation . the hidden sector superpartner mass scale is @xmath34 where @xmath35 is the hidden sector gauge coupling , @xmath36 is the supersymmetry - breaking scale squared , and @xmath37 is the mass scale of the messenger particles . because the mssm superpartner masses are also generated by gauge - mediation from the same susy - breaking sector , we find @xmath38 the ratio @xmath39 sets the annihilation cross section of a particle through gauge interactions , which in turn determines the thermal relic density of a stable particle @xcite . the `` wimp miracle '' is the remarkable coincidence that , for a stable wimp with mass @xmath40 and coupling @xmath41 , this thermal relic density is roughly that required by astrophysical observations . equation ( [ thermalsigmascaling ] ) shows that our hidden sector candidate automatically has approximately the same annihilation cross section , and thus the same relic density . if there are connectors @xmath14 with both dark and sm charge , wimpless dark matter may have observable interactions through couplings @xmath15 , where @xmath13 are sm particles . in this case , wimpless dark matter has many of the virtues and implications commonly associated with wimps . in contrast to wimps , however , the wimpless dark matter s mass need not be at the electroweak symmetry - breaking scale . it may be treated as a free parameter , which in turn determines the gauge coupling strength of the hidden sector . this freedom opens new possibilities for dark matter model parameters and new experimental search windows . the dama dark matter signal of annual modulation in direct detection has motivated a variety of explanations @xcite . the canonical possibility , where wimp dark matter with mass @xmath42 elastically scatters , is excluded , as the required scattering cross section is in conflict with other experiments . inelastic scattering @xcite , in which dark matter is assumed to scatter to another state that is @xmath43 heavier , has also been explored . such scattering alleviates the conflict between different direct detection experiments , but is tightly constrained by neutrino bounds on dark matter annihilation in the sun @xcite . an alternative explanation is elastic scattering of a light dark matter particle with @xmath44 and large spin - independent nucleon scattering cross section @xmath19 @xcite , a region also supported by recent results from cogent @xcite . such explanations are possible if the dama signal is enhanced by populations of dark matter in tidal streams @xcite or detection thresholds are lowered by channeling @xcite , and may also require an unusual background spectrum for consistency @xcite . in typical wimp models , the required small masses and large cross sections are possible @xcite , but not at all generic . in contrast , wimpless dark matter provides a natural setting for the low mass explanation . in the example presented in refs . @xcite , the dark matter particle in the hidden sector couples to the sm through yukawa couplings @xmath45 \ . \label{couplings}\ ] ] each term may have a different coupling , but we assume equal couplings for simplicity . in eq . ( [ couplings ] ) , @xmath2 is the dark matter , a complex scalar could also be a real scalar . ] charged under a discrete symmetry ( hidden parity ) ; @xmath46 , @xmath47 , and @xmath48 are the third generation quarks of the sm ; and @xmath49 , @xmath50 , and @xmath51 are the connectors , exotic 4th generation quarks . the @xmath23 have hidden parity and are in the sm su(3)@xmath52su(2)@xmath52u(1)@xmath53 representations @xmath54 the subscripts @xmath55 and @xmath17 refer to su(2 ) doublets and singlets , respectively , not chirality ; the chirality of the @xmath56 , @xmath50 , and @xmath51 fields is opposite to their sm counterparts , and they are therefore mirror quarks . finally , the @xmath23 receive mass through electroweak symmetry breaking . the couplings of eq . ( [ couplings ] ) imply scattering through @xmath57 , where @xmath58 . this induces a coupling to the gluons of the nucleon at one - loop @xcite . as shown in ref . @xcite , for @xmath12 , @xmath59 , and @xmath60 , the coupling to @xmath61 quarks produces a cross section @xmath25 in the right range to explain dama and cogent . for example , the best fit point for the cogent data can be obtained with @xmath20 , @xmath62 and @xmath63 . the large cross section can be understood as follows : spin - independent scattering requires a chirality flip on the fermion line , which is typically suppressed by a small yukawa coupling . but if the dark matter is a scalar , than a mass insertion on the @xmath23 propagator provides the necessary chirality flip without suppressing @xmath25 , since the @xmath23 are heavy and their yukawa couplings are large . the explanation therefore requires that @xmath2 is a scalar and the connectors are chiral fermions . in fact , this mechanism is overly efficient . if the 3rd generation quarks are replaced by 1st generation quarks in eq . ( [ couplings ] ) , the dark matter couples to nucleons at tree - level , and the desired @xmath25 is achieved for couplings @xmath64 @xcite . this is also perfectly acceptable and worth studying @xcite , but the required coupling for @xmath61 quarks appears to be somewhat more natural and , in the general case where there are couplings to more than one generation , less constrained by flavor - changing neutral currents . in this study , we assume negligible couplings to 1st and 2nd generation quarks and focus on the collider phenomenology of the case where the @xmath23 decay directly to 3rd generation quarks . if the dark matter is stabilized not by a discrete symmetry , but by a continuous symmetry , @xmath56 and @xmath65 must have opposite dark charges to allow them to get a mass . the yukawa couplings of eq . ( [ couplings ] ) must therefore be generalized to @xmath66 \ , \ ] ] where @xmath67 and @xmath68 are two complex scalars with opposite dark charges . in general , @xmath67 and @xmath69 will mix to form mass eigenstates @xmath70 and @xmath71 . the lighter state is the dark matter particle and can couple to both @xmath56 and @xmath65 . despite slight additional complications , we therefore recover the discrete symmetry case , although there may now be additional decays @xmath72 . for simplicity , we focus in the rest of this study on the discrete symmetry case with couplings given in eq . ( [ couplings ] ) . as with sm quarks , the 4th generation quarks receive their mass through electroweak symmetry breaking , and so @xmath73 , where @xmath74 . perturbativity places an upper bound on @xmath75 ; requiring @xmath76 implies @xmath77 . 4th generation quark masses are also constrained by precision electroweak data . these constraints are not modified by the exotic and mirror features of the quarks we consider , and they imply @xmath78 , where some non - degeneracy is required @xcite . 4th generation quarks may also have many beneficial effects , for example , raising the higgs boson mass in supersymmetric theories through their loop corrections , and enhancing higgs boson production rates @xcite . direct searches place lower bounds on @xmath75 . these searches are rather independent of the details of the couplings to the hidden sector . @xmath0 and @xmath79 production is dominated by qcd processes . in addition , the coupling @xmath80 only affects the @xmath0 and @xmath79 lifetimes . for all but extremely small @xmath80 , the @xmath0 and @xmath79 decay promptly . with the model framework and assumptions given in sec . [ sec : model ] , the possible decays of the @xmath23 are @xmath81 , @xmath82 , @xmath83 , and @xmath84 . if @xmath85 , the decays @xmath86 and @xmath87 are strongly suppressed by kinematics . in the type of scenarios we are interested in here , the coupling between the @xmath23 and @xmath2 states is furthermore rather strong , which means that the decays @xmath81 and @xmath82 completely dominate . in our analysis below , we assume that @xmath88 . @xmath79 pair production followed by @xmath82 leads to a signature of @xmath89 , which is identical to the final state of bottom squark pair production followed by @xmath90 . searches for this supersymmetric signal have been carried at both cdf and d at the tevatron . the d analysis , based on an integrated luminosity of @xmath91 from run ii , implies @xmath92 ( 95% cl ) for @xmath93 @xcite ; the corresponding cdf result using @xmath94 is @xmath95 ( 95% cl ) @xcite . taking into account the difference in @xmath96 and @xmath97 cross sections , the dresults imply @xmath98 . a later search for gluino pair production with @xmath99 and @xmath100 has been carried out by the cdf collaboration using @xmath101 collected luminosity . candidate events were selected requiring two or more jets , large @xmath31 , and at least two @xmath61-tags @xcite . using neural net analyses , cdf found @xmath102 ( 95% cl ) for large mass splitting @xmath103 , and about 340 gev for small @xmath104 . their result in the case of small mass splitting @xmath105 can be applied to the @xmath106 search , implying roughly @xmath107 . finally , there are also projections for squark searches at the lhc based on the @xmath108 signal @xcite . it is hard , however , to apply their results to the @xmath109 signal , since no @xmath61-tagging is used in that analysis . in our analysis , we will focus on @xmath0 pair production , @xmath110 , with the distinctive , but more complicated , final state of a top quark pair plus missing energy from the @xmath2 particles . the cdf collaboration has reported a search for the analogous supersymmetric process of top squark pair production based on an integrated luminosity of @xmath111 , using the purely leptonic final states from @xmath112 , followed by @xmath113 @xcite . the data are consistent with the sm , leading to the constraint @xmath114 , where the exact limit depends on @xmath115 , @xmath116 and @xmath117 . similar signals also appear in other new physics scenarios such as little higgs models with @xmath29-parity @xcite . the @xmath118 signature at the lhc in the semi - leptonic channel has also been studied in ref . @xcite ; however , that study focused on higher masses with larger integrated luminosity . the hadronic mode has been analyzed at the parton level in ref . @xcite , focusing on the prospects for spin determination and mass measurements . in contrast , our study is performed at the detector level and is focused on the exclusion and discovery potential of both the tevatron and early lhc data . to investigate the discovery and exclusion prospects , we have simulated production and decay of the new particles at the tevatron and at the lhc with @xmath119 , as well as the main backgrounds . all simulations have been done using madgraph / madevent - pythia 6.4.20 - pgs4 @xcite with the @xmath120-ordered pythia showers and the cdf or atlas detector cards for pgs4 . matrix element / parton shower matching has been applied both for signal and backgrounds , and its validity has been double - checked by comparing different maximum multiplicity samples . the parton distribution function set used is cteq6l1 , and factorization and renormalization scales are set to @xmath121 for the centrally - produced particle pair . we do not apply @xmath122-factors for higher order qcd effects to either signal or background . these @xmath122-factors are expected to enhance these cross sections and be similar for the signal and top pair production ( the dominant background after cuts ) , and hence the effect of including them would only be to increase the signal significance . in addition , there are uncertainties in the cross sections of both signal and backgrounds due to parton distribution functions @xcite . for @xmath123 , the @xmath0 can not decay to an on - shell @xmath124 . these parameter points have therefore been simulated in madgraph / madevent using off - shell top decay , @xmath125 . this procedure guarantees that finite width effects are correctly accounted for . note that qcd multi - jet backgrounds have not been simulated . instead , we refer to the studies of refs . @xcite and apply similar cuts , in particular @xmath126 cuts , @xmath31 cuts , and cuts on number of jets , which should be enough to suppress the qcd multi - jet backgrounds to negligible levels . the signatures of @xmath127 are determined by the decays of the top quark pair and can therefore be divided into hadronic , semi - leptonic and purely leptonic channels . since the @xmath0 decays will always generate missing transverse momentum from the invisible @xmath2 particles in the final state , the most relevant backgrounds are those with significant missing energy from @xmath128 or @xmath129 boson decays into neutrinos . we will focus on the semi - leptonic and hadronic channels . the dilepton channel has suppressed cross section because of the small leptonic decay branching ratios . the semi - leptonic decay has the advantage that the presence of an isolated lepton ( electron or muon ) makes it easier to suppress qcd backgrounds , while for the fully hadronic channel this requires additional @xmath126 cuts . this is particularly important for early running of the lhc , where the missing energy resolution might not yet be fully under control . on the other hand , the hadronic decay mode has larger branching ratio . moreover , the main backgrounds , @xmath130 and @xmath131 production , only have substantial @xmath31 in association with leptons ; in fact , the only truly irreducible background to the fully hadronic decay mode is @xmath132 + jets ( and the negligible @xmath133 background ) . to distinguish signal from background in the semi - leptonic case , we look for large @xmath31 and large transverse mass of the leptonic @xmath128 candidate , defined to be @xmath134 for the background is mainly from @xmath128 decay , we expect @xmath135 for most of the background , while much of the signal extends beyond this limit . for the background surviving the @xmath31 and transverse mass cuts , we expect a significant fraction to be due to decays of the second top into hadronic @xmath136 leptons and missing energy . we therefore expect fewer jets for the background than for the signal . to further suppress the background , we also require the presence of a second , hadronically - decaying @xmath128 . to implement this strategy , we employ the following precuts ( differences between the tevatron and the lhc are noted where they apply ) : * one isolated electron or muon with @xmath137 . * no additional isolated leptons with @xmath138 . * minimum missing transverse energy : @xmath139 . * minimum transverse mass : @xmath140 . * at least 4 jets with @xmath141 ( tevatron ) or @xmath142 gev ( lhc ) . * at least one jet pair with invariant mass within the @xmath128 mass window @xmath143 . we also use additional cuts to achieve the best signal significance : * additional @xmath144 cut : @xmath145 ( tevatron ) or @xmath146 ( lhc ) . * additional @xmath31 cuts : @xmath147 ( tevatron ) ; @xmath148 ( lhc ) . * @xmath149 cuts : @xmath150 ( tevatron ) ; @xmath151 ( lhc ) . * combinations of the cuts above . the relevant backgrounds for the semi - leptonic channel are @xmath130 ( semi - leptonic and purely leptonic decays ) and leptonically - decaying @xmath131 + jets production . top pairs are the main background . because this has the same number of @xmath61-quarks as the signal , no @xmath61-tagging information is employed , since this would suppress signal and background by the same amount , leading to a reduced signal significance . for completeness , we also simulated @xmath129 + jets ( where the largest contribution comes from @xmath152 with one of the taus decaying leptonically ) and @xmath153 , but these processes both turned out to be negligible after precuts . we show the distributions for missing transverse energy @xmath31 , transverse mass @xmath144 , number of jets @xmath154 , and jet pair invariant mass @xmath155 for example signal parameters and backgrounds for the 10 tev lhc in fig . [ fig:1lep - distributions ] . each observable is plotted after the cuts coming before it in the list , and the position of the precut is marked with a vertical dashed line . for clarity , we have split the @xmath130 background into components : semi - leptonic decays ( to electron or muon ) , decays with at least one tau lepton , purely leptonic decays ( where both @xmath128 s decay to electron or muon ) and fully hadronic decay ( which is negligible with these cuts ) . the @xmath155 plot shows the invariant mass for the jet pair closest to the @xmath128 mass . the corresponding distributions for the tevatron are qualitatively similar . distributions of missing transverse energy @xmath156 , transverse mass @xmath144 , number of jets @xmath154 , and jet pair invariant mass @xmath155 for signal and backgrounds for the 10 tev lhc in the semi - leptonic channel . each of the observables has been plotted after the precuts coming before it in the list , and the chosen precut has been marked by a vertical line . for signal , the masses @xmath157 , @xmath158 , and @xmath159 have been chosen for illustration . the @xmath128 and @xmath129 samples were simulated with a cut on @xmath160 and at least 3 jets in the parton - level generation . see text for details . , title="fig : " ] distributions of missing transverse energy @xmath156 , transverse mass @xmath144 , number of jets @xmath154 , and jet pair invariant mass @xmath155 for signal and backgrounds for the 10 tev lhc in the semi - leptonic channel . each of the observables has been plotted after the precuts coming before it in the list , and the chosen precut has been marked by a vertical line . for signal , the masses @xmath157 , @xmath158 , and @xmath159 have been chosen for illustration . the @xmath128 and @xmath129 samples were simulated with a cut on @xmath160 and at least 3 jets in the parton - level generation . see text for details . , title="fig : " ] + distributions of missing transverse energy @xmath156 , transverse mass @xmath144 , number of jets @xmath154 , and jet pair invariant mass @xmath155 for signal and backgrounds for the 10 tev lhc in the semi - leptonic channel . each of the observables has been plotted after the precuts coming before it in the list , and the chosen precut has been marked by a vertical line . for signal , the masses @xmath157 , @xmath158 , and @xmath159 have been chosen for illustration . the @xmath128 and @xmath129 samples were simulated with a cut on @xmath160 and at least 3 jets in the parton - level generation . see text for details . , title="fig : " ] distributions of missing transverse energy @xmath156 , transverse mass @xmath144 , number of jets @xmath154 , and jet pair invariant mass @xmath155 for signal and backgrounds for the 10 tev lhc in the semi - leptonic channel . each of the observables has been plotted after the precuts coming before it in the list , and the chosen precut has been marked by a vertical line . for signal , the masses @xmath157 , @xmath158 , and @xmath159 have been chosen for illustration . the @xmath128 and @xmath129 samples were simulated with a cut on @xmath160 and at least 3 jets in the parton - level generation . see text for details . , title="fig : " ] the combined background cross section after precuts is 2.4 fb for the tevatron , and 82 fb for the 10 tev lhc . typical signal efficiency for the precuts is 2 - 4% at the tevatron and 1 - 2% at the lhc . the cross sections after cuts , for the main backgrounds and some example signal parameters , are found in the appendix in tables [ table : tevsemi ] and [ table : lhcsemi ] . for the fully hadronic case , the background is mainly leptonic @xmath128 decays ( from @xmath128+jets and @xmath161 ) , where the lepton is either missed ( or non - isolated ) or a @xmath136 lepton has been mistagged as a jet , and @xmath129 + jets , where the @xmath129 decays to neutrinos . we therefore expect the background to have fewer jets than the signal , which mainly consists of fully hadronic top decays , with missing energy from the invisible @xmath2 particles . to be sure to avoid qcd multi - jet background , we also need to apply @xmath126 cuts between the hardest jets and the missing energy . in fact , this also helps to reduce the @xmath162 background , since with large @xmath31 cuts , the @xmath128 tends to be boosted , while the tau jet tends to be in the direction of the missing energy . the signal is furthermore expected to have larger @xmath163 than the background . for the fully hadronic channel , we use the following precuts : * no isolated electrons , muons or tau - tagged jets with @xmath164 . * minimum missing transverse energy : @xmath139 . * at least 5 jets with @xmath141 ( tevatron ) or @xmath165 ( lhc ) . * minimum @xmath126 for the leading jets : @xmath166 and @xmath167 ( tevatron ) ; @xmath168 for the first , second and third leading jets ( lhc ) . we also use the following additional cuts to optimize the signal significance : * additional @xmath31 cuts : @xmath169 ( tevatron ) ; @xmath170 ( lhc ) . * @xmath171 cuts : @xmath172 ( tevatron ) ; @xmath173 ( lhc ) . * at least 6 jets with @xmath141 ( tevatron ) or @xmath165 ( lhc ) . as discussed above , the relevant backgrounds for the fully hadronic channel are @xmath130 , leptonically - decaying @xmath131 + jets , and @xmath132 + jets . for completeness , we also simulated @xmath174 , but this is negligible because of its small cross section . among the @xmath130 decay modes , the dominant background is from decays with at least one tau lepton , followed by the semi - leptonic decay to electron or muon ( where the lepton is either missed or non - isolated ) . @xmath31 , @xmath154 , and @xmath175 distributions for signal and backgrounds for the 10 tev lhc in the hadronic channel . the top two panels show distributions of @xmath156 and @xmath154 after the previous cuts in the precut table , with the position of the precut marked with a vertical dashed line . the lower two panels show distributions of @xmath156 and @xmath175 after all precuts . the hadronic top contribution is negligible after the @xmath176 cut and has therefore been omitted in the remaining plots . for signal , the masses @xmath157 , @xmath158 , and @xmath159 have been chosen for illustration . the @xmath128 and @xmath129 samples were simulated with a cut on @xmath177 and at least 3 jets in the parton - level generation . see text for details . , title="fig : " ] @xmath31 , @xmath154 , and @xmath175 distributions for signal and backgrounds for the 10 tev lhc in the hadronic channel . the top two panels show distributions of @xmath156 and @xmath154 after the previous cuts in the precut table , with the position of the precut marked with a vertical dashed line . the lower two panels show distributions of @xmath156 and @xmath175 after all precuts . the hadronic top contribution is negligible after the @xmath176 cut and has therefore been omitted in the remaining plots . for signal , the masses @xmath157 , @xmath158 , and @xmath159 have been chosen for illustration . the @xmath128 and @xmath129 samples were simulated with a cut on @xmath177 and at least 3 jets in the parton - level generation . see text for details . , title="fig : " ] + @xmath31 , @xmath154 , and @xmath175 distributions for signal and backgrounds for the 10 tev lhc in the hadronic channel . the top two panels show distributions of @xmath156 and @xmath154 after the previous cuts in the precut table , with the position of the precut marked with a vertical dashed line . the lower two panels show distributions of @xmath156 and @xmath175 after all precuts . the hadronic top contribution is negligible after the @xmath176 cut and has therefore been omitted in the remaining plots . for signal , the masses @xmath157 , @xmath158 , and @xmath159 have been chosen for illustration . the @xmath128 and @xmath129 samples were simulated with a cut on @xmath177 and at least 3 jets in the parton - level generation . see text for details . , title="fig : " ] @xmath31 , @xmath154 , and @xmath175 distributions for signal and backgrounds for the 10 tev lhc in the hadronic channel . the top two panels show distributions of @xmath156 and @xmath154 after the previous cuts in the precut table , with the position of the precut marked with a vertical dashed line . the lower two panels show distributions of @xmath156 and @xmath175 after all precuts . the hadronic top contribution is negligible after the @xmath176 cut and has therefore been omitted in the remaining plots . for signal , the masses @xmath157 , @xmath158 , and @xmath159 have been chosen for illustration . the @xmath128 and @xmath129 samples were simulated with a cut on @xmath177 and at least 3 jets in the parton - level generation . see text for details . , title="fig : " ] distributions for @xmath31 , @xmath154 , and @xmath175 for both signal and backgrounds in the hadronic channel at the 10 tev lhc are shown in fig . [ fig : nolep - distributions ] . the top two panels show @xmath31 and @xmath154 plotted after the cuts coming before it in the list , and the position of the precut is marked with a vertical dashed line . the bottom two panels are @xmath31 and @xmath175 distributions plotted after precuts . for clarity , we have split the @xmath130 background into components : fully hadronic decay ( negligible after @xmath31 cut ) , decays with at least one tau lepton , semi - leptonic decays ( to electron or muon ) , and purely leptonic decays ( which are negligible with these cuts ) . the corresponding distributions for the tevatron are qualitatively similar . after precuts for the hadronic channel , the combined background cross section is 21 fb for the tevatron and 1.4 pb for the 10 tev lhc . the signal efficiency of the precuts is 9 - 20% at the tevatron and 8 - 13% at the lhc . a table of cross sections after cuts for backgrounds and some signal points may be found in the appendix in tables [ table : tevhad ] and [ table : lhchad ] . the main remaining backgrounds after precuts for both the semi - leptonic and hadronic channels include tau leptons . one reason for this is that a tau lepton is often mistagged as a jet , which therefore adds significantly to the fully hadronic background with large @xmath31 ( in particular for the hadronic channel ) . it would be interesting to see an experimental study of whether an anti - tau tag could be effective in further suppressing these backgrounds , while keeping a good signal efficiency . this might be of significant importance for any new physics with signatures consisting of jets and missing energy . we now determine the discovery and exclusion reach for @xmath0 at the tevatron and the 10 tev lhc . for each parameter point @xmath178 , we use the optimum cut ( after precuts ) that gives the best signal significance , with the additional requirements that @xmath179 and more than two signal events are observed . given the small number of signal and background events after cuts , we have used poisson statistics , rather than assuming gaussian distributions , for both signal and backgrounds . figure [ fig : exclusion - tev ] shows the 95% cl tevatron exclusion contours for both the semi - leptonic and hadronic channels and integrated luminosities of 2 , 5 , 10 , and @xmath5 . even with just @xmath180 , exclusion limits of @xmath181 ( semi - leptonic mode ) and @xmath182 ( hadronic mode ) can be reached , which already extend into the interesting mass range consistent with current direct search bounds and precision electroweak data . with a combined integrated luminosity of @xmath5 at the end of tevatron running , a reach of up to 455 gev for the hadronic channel can be achieved . 95% cl tevatron exclusion contours for the semi - leptonic channel ( left ) and the hadronic channel ( right ) for integrated luminosities 2 , 5 , 10 , and @xmath5 . for each point in parameter space , the cut with the best significance has been chosen.,title="fig : " ] 95% cl tevatron exclusion contours for the semi - leptonic channel ( left ) and the hadronic channel ( right ) for integrated luminosities 2 , 5 , 10 , and @xmath5 . for each point in parameter space , the cut with the best significance has been chosen.,title="fig : " ] the reach in @xmath4 is almost independent of @xmath24 for small to medium @xmath24 . however , when @xmath24 approaches the on - shell decay threshold of @xmath183 , the reach is limited since the top and @xmath2 are produced nearly at rest in the @xmath0 rest frame , and the @xmath184 system therefore needs a transverse boost for the @xmath2 particles to produce large missing transverse momentum . this leads to the dip in the exclusion curves at @xmath185 close to @xmath183 , and indeed there is no exclusion reach at the tevatron for @xmath186 . for @xmath5 integrated luminosity and @xmath4 between 370 and 390 gev , @xmath24 could be excluded up to 160 gev at 95% cl using the hadronic mode . for smaller @xmath4 , the reach in @xmath24 is decreased due to the softness of the @xmath2 particle distributions , while for larger @xmath4 , it is decreased because of the small @xmath187 production cross section . figure [ fig : discovery - tev ] shows the 3@xmath6 ( gaussian equivalent deviation in a two - sided gaussian distribution , which is more commonly used in the literature . ] ) tevatron discovery contours for both the semi - leptonic and hadronic channels for integrated luminosities of 2 , 5 , 10 , and @xmath5 . a 3@xmath6 signal could be observed for @xmath188 and @xmath189 in the semi - leptonic channel with @xmath5 integrated luminosity . the hadronic channel is more promising . with @xmath190 integrated luminosity , a reach in @xmath4 up to 360 gev could be achieved when @xmath24 is not too large . with @xmath5 integrated luminosity , the reach is extended to 400 gev for @xmath24 up to about 80 gev . for larger @xmath24 , the reach in @xmath4 decreases . 3@xmath6 ( gaussian equivalent ) tevatron discovery contours for the semi - leptonic channel ( left ) and the hadronic channel ( right ) for integrated luminosities 2 , 5 , 10 , and @xmath5 . for each point in parameter space , the cut with the best significance has been chosen.,title="fig : " ] 3@xmath6 ( gaussian equivalent ) tevatron discovery contours for the semi - leptonic channel ( left ) and the hadronic channel ( right ) for integrated luminosities 2 , 5 , 10 , and @xmath5 . for each point in parameter space , the cut with the best significance has been chosen.,title="fig : " ] figure [ fig : exclusion - lhc10 ] shows the 95% cl exclusion contours for a 10 tev early lhc run , in the semi - leptonic and hadronic channels for integrated luminosities 100 , 200 , and @xmath9 . with just @xmath191 , the lhc exclusion reach for @xmath4 exceeds the tevatron exclusion reach with @xmath5 luminosity . exclusions of @xmath4 up to 490 , 520 , and 535 gev could be achieved with 100 , 200 , and @xmath9 integrated luminosity for the semi - leptonic channel . the exclusion region for the hadronic channel covers almost the entire interesting mass parameter space with @xmath9 luminosity . note that at the lhc , we could tolerate much smaller @xmath192 ; in particular , we start probing the off - shell decay region @xmath193 for @xmath123 . 95% cl exclusion contours for a 10 tev lhc run in the semi - leptonic channel ( left ) and the hadronic mode ( right ) , for integrated luminosities 100 , 200 , and @xmath9 . for each point in parameter space , the cut with the best significance has been chosen.,title="fig : " ] 95% cl exclusion contours for a 10 tev lhc run in the semi - leptonic channel ( left ) and the hadronic mode ( right ) , for integrated luminosities 100 , 200 , and @xmath9 . for each point in parameter space , the cut with the best significance has been chosen.,title="fig : " ] figure [ fig : discovery - lhc10 ] shows the 3@xmath6 ( gaussian equivalent ) discovery contours for a 10 tev lhc run , in the semi - leptonic and hadronic channels for integrated luminosities 100 , 200 , and @xmath9 . although the reach in both @xmath4 and @xmath24 is limited for the semi - leptonic mode , the hadronic channel could provide a 3@xmath6 signal for @xmath194 and @xmath195 with @xmath9 luminosity . we might also observe a positive signal for @xmath24 up to about 170 gev in the off - shell decay region ( @xmath196 ) for @xmath197 . 3@xmath6 ( gaussian equivalent ) discovery contours for a 10 tev lhc run , in the semi - leptonic channel ( left ) and the hadronic channel ( right ) , for integrated luminosities 100 , 200 , and @xmath9 . for each point in parameter space , the cut with the best significance has been chosen . , title="fig : " ] 3@xmath6 ( gaussian equivalent ) discovery contours for a 10 tev lhc run , in the semi - leptonic channel ( left ) and the hadronic channel ( right ) , for integrated luminosities 100 , 200 , and @xmath9 . for each point in parameter space , the cut with the best significance has been chosen . , title="fig : " ] it is clear from the discovery and exclusion contours , both for the tevatron and the lhc , that the fully hadronic channel has considerably larger reach than the semi - leptonic channel , for reasons enumerated in sec . [ sec : simulation ] . in this channel , the full , currently viable , region in parameter space can be excluded at a 10 tev lhc run . in case both channels are visible , they can be used to distinguish between different model and mass hypotheses . we have considered the prospects for hadron colliders to pair produce exotic 4th generation quarks that decay directly to a pair of dark matter particles and sm particles . although we have a particular interest in the wimpless dark matter scenario @xcite ( including a specific example @xcite that can potentially explain the dama annual modulation result ) , this scenario is motivated on quite general grounds , and , with minor modifications , our analysis applies to many other dark matter scenarios and other new physics models . we have focused on the up - type 4th generation quark @xmath0 . @xmath0 pair production leads to @xmath198 , and we have then analyzed the semi - leptonic and fully hadronic channels . the fully hadronic channel ( vetoing events with leptons ) seems to be the most efficient , because of the large branching fraction and the reduction in sm background with large @xmath31 . existing constraints require @xmath199 , where the lower bound comes from direct searches , and the upper bound is from perturbativity . we have found that there are bright prospects for probing exotic 4th generation quarks in this mass window at the tevatron and in early data from the lhc . for models with @xmath200 , the discovery of new physics is possible at the tevatron with @xmath201 of luminosity , while for @xmath195 the discovery of new physics may be possible at the lhc with @xmath202 . in particular , with @xmath203 of data , the lhc should be able to discover almost all of the relevant parameter space with @xmath204 , where wimpless models can explain the dama and cogent results . conversely , if no signal is seen in @xmath9 , the entire mass range consistent with current bounds and perturbativity will be excluded . of course , although an exclusion definitively excludes the model , a discovery will only be a discovery of a multi - jet ( + lepton ) + @xmath31 signal . considerably larger integrated luminosity would be needed to identify the signal as @xmath118 , and it is an even harder problem to determine if the new physics really is @xmath205 production , with decay to top quarks and dark matter . such a discovery analysis would require a good identification of the decaying top quarks as well as spin and mass determinations of the @xmath0 and @xmath2 particles , and would require significant amounts of data from the lhc ( and be beyond the capabilities of the tevatron ) . exotic 4th generation quark decays are not the only processes that give a signal of top quark pairs plus missing transverse energy . in particular , this is a typical signature for supersymmetry , for example , from stop pair production followed by the decay @xmath206 . however , it should be possible to distinguish these possibilities with more lhc data . since the 4th generation quark is a fermion , it has a higher production cross - section than squarks with a similar mass . moreover , squarks could also have more complicated decay chains that would be absent in @xmath0 decay . there have already been studies on how to distinguish supersymmetric signals from other new physics with similar signals @xcite , and it would be worthwhile to perform a more detailed analysis of how one would distinguish exotic 4th generation quarks from squarks . it is also worth noting that the process @xmath207 + jets would be well suited for analysis using the @xmath208 kinematic variable @xcite . this analysis has focused on pair production of the up - type 4th generation quarks @xmath0 . precision electroweak constraints imply that the down - type 4th generation quark @xmath79 must be fairly degenerate with the @xmath0 , and so @xmath209 should also be accessible . this will likely be a more difficult signal to extract from early data , since it requires a good understanding of @xmath61-tagging , and we would expect large qcd backgrounds . but in the event of a discovery , such an analysis will be immensely useful in understanding the underlying physics . there is also an interesting complementarity between the collider studies of dark matter considered here , and direct or indirect detection strategies . for example , one may consider wimpless dark matter models in the limit of small @xmath80 . in this limit the cross - sections for dark matter - nucleon scattering and for dark matter annihilation are small , and direct or indirect dark matter searches will be unsuccessful . but the production cross section for @xmath210 pairs is controlled by qcd , independent of the yukawa coupling @xmath80 . so for models in this limit of parameter space , hadron colliders may provide the only direct evidence for the nature of dark matter . interestingly , at small @xmath80 , the 4th generation quarks are long - lived . this may result in displaced decay vertices , and a sufficiently long - lived 4th generation quark may even hadronize and reach the detector . there has already been significant study of detection strategies for long - lived exotic hadrons @xcite , and these results should be directly applicable to the case when the @xmath0 travels a macroscopic distance in the detector . it would be interesting to investigate further how to determine the nature of such semi - stable color triplet particles . our results for a 10 tev lhc run can be approximately translated to the alternative of an extended 7 tev run . in this case , a coverage corresponding to the @xmath9 quoted in this study should be attainable for less than about @xmath211 . finally , and of particular interest , the annual modulation signal of dama @xcite has been recently supported by unexplained events from cogent @xcite , which , if interpreted as dark matter , also favor the same low mass @xmath212 and high cross section @xmath213 region of dark matter parameter space . these results are consistent with recent bounds and results from cdms @xcite . the consistency of several direct detection experiments is , of course , important to establish a dark matter signal . improved statistics will be essential , but given the difficulty of making definitive background determinations , independent confirmation by completely different means is also highly desirable . as discussed above , the direct detection data may naturally be explained by scalar dark matter interacting with exotic 4th generation quarks . data already taken by the super - kamiokande experiment provide a promising probe of these interpretations of dama , cdms , and cogent through indirect detection @xcite . the results derived here show that these explanations , including wimpless models , can also be tested very directly with data already taken at the tevatron and have extraordinarily promising implications for early runs of the lhc . we gratefully acknowledge h. baer , a. barr , g. kribs , b. nelson , t. tait , x. tata , d. toback and d. whiteson for useful discussions . jk is grateful to george mitchell , kias , and the kek theory center for their hospitality while this work was in preparation . the work of ja was supported by ncts , grant number nsc 982119m002001 . the work of jlf was supported in part by nsf grant phy0653656 . the work of ss was supported in part by the department of energy under grant de - fg02 - 04er-41298 . in this appendix , we present tables listing the cross sections after cuts for the @xmath214 signal and the main sm backgrounds . in the upper section of each table , each line gives the cross section after including all cuts above . in the lower section , each line gives the cross section after including the cut on that line , and all precuts . for the signal , three examples with @xmath215 and @xmath216 , 400 , and 500 gev are chosen . the @xmath128 and @xmath129 cross sections in parentheses were simulated with a cut on @xmath217 and at least 3 jets in the parton - level generation . for the lhc , we have divided the @xmath130 background according to decays : hadronic , single tau lepton , semi - leptonic ( electron or muon ) , double leptonic ( electron or muon ) , and double tau lepton . only the contributing decay modes have been included in the table . .signal and background cross sections in fb after cuts for the semi - leptonic channel at the tevatron . the signal examples are for @xmath215 and @xmath216 , 400 , and 500 gev as indicated . the @xmath128 and @xmath129 cross sections in parentheses were simulated with a cut on @xmath217 and at least 3 jets in the parton - level generation . [ cols="^,^,^,^,^,^,^",options="header " , ] 99 l. ackerman , m. r. buckley , s. m. carroll and m. kamionkowski , phys . d * 79 * , 023519 ( 2009 ) [ arxiv:0810.5126 [ hep - ph ] ] . j. l. feng , m. kaplinghat , h. tu and h. b. yu , jcap * 0907 * , 004 ( 2009 ) [ arxiv:0905.3039 [ hep - ph ] ] . h. goldberg , phys . lett . * 50 * , 1419 ( 1983 ) [ erratum - ibid . * 103 * , 099905 ( 2009 ) ] . j. r. ellis , j. s. hagelin , d. v. nanopoulos , k. a. olive and m. srednicki , nucl . b * 238 * , 453 ( 1984 ) . g. servant and t. m. p. tait , nucl . b * 650 * , 391 ( 2003 ) [ arxiv : hep - ph/0206071 ] . h. c. cheng , j. l. feng and k. t. matchev , phys . lett . * 89 * , 211301 ( 2002 ) [ arxiv : hep - ph/0207125 ] . j. l. feng and j. kumar , phys . * 101 * , 231301 ( 2008 ) [ arxiv:0803.4196 [ hep - ph ] ] . j. l. feng , j. kumar and l. e. strigari , phys . b * 670 * , 37 ( 2008 ) [ arxiv:0806.3746 [ hep - ph ] ] . r. bernabei _ et al . _ [ dama collaboration ] , eur . j. c * 56 * , 333 ( 2008 ) [ arxiv:0804.2741 [ astro - ph ] ] . r. bernabei _ et al . _ , arxiv:1002.1028 [ astro-ph.ga ] . e.g. _ , m. brhlik and l. roszkowski , phys . lett . b * 464 * , 303 ( 1999 ) [ arxiv : hep - ph/9903468 ] ; p. belli , r. bernabei , a. bottino , f. donato , n. fornengo , d. prosperi and s. scopel , phys . d * 61 * , 023512 ( 2000 ) [ arxiv : hep - ph/9903501 ] ; p. gondolo and g. gelmini , phys . d * 71 * , 123520 ( 2005 ) [ arxiv : hep - ph/0504010 ] . r. bernabei _ et al . _ , eur . j. c * 53 * , 205 ( 2008 ) [ arxiv:0710.0288 [ astro - ph ] ] ; s. chang , a. pierce and n. weiner , phys . d * 79 * , 115011 ( 2009 ) [ arxiv:0808.0196 [ hep - ph ] ] ; m. fairbairn and t. schwetz , jcap * 0901 * , 037 ( 2009 ) [ arxiv:0808.0704 [ hep - ph ] ] . f. petriello and k. m. zurek , jhep * 0809 * , 047 ( 2008 ) [ arxiv:0806.3989 [ hep - ph ] ] ; c. savage , g. gelmini , p. gondolo and k. freese , jcap * 0904 * , 010 ( 2009 ) [ arxiv:0808.3607 [ astro - ph ] ] . c. e. aalseth _ [ cogent collaboration ] , arxiv:1002.4703 [ astro-ph.co ] . z. ahmed _ et al . _ [ the cdms - ii collaboration ] , arxiv:0912.3592 [ astro-ph.co ] . h. c. cheng and i. low , jhep * 0408 * , 061 ( 2004 ) [ arxiv : hep - ph/0405243 ] . p. fileviez perez and m. b. wise , arxiv:1002.1754 [ hep - ph ] . b. zeldovich , adv . astrophys . * 3 * , 241 ( 1965 ) ; h.y . chiu , phys . * 17 * , 712 ( 1966 ) ; g. steigman , ann . rev . nucl . part . * 29 * , 313 ( 1979 ) ; r.j . scherrer and m.s . turner , phys . rev . d * 33 * , 1585 ( 1986 ) [ erratum - ibid . d * 34 * , 3263 ( 1986 ) ] . r. foot , phys . d * 78 * , 043529 ( 2008 ) [ arxiv:0804.4518 [ hep - ph ] ] ; m. y. khlopov and c. kouvaris , phys . d * 78 * , 065040 ( 2008 ) [ arxiv:0806.1191 [ astro - ph ] ] ; s. andreas , t. hambye and m. h. g. tytgat , jcap * 0810 * , 034 ( 2008 ) [ arxiv:0808.0255 [ hep - ph ] ] ; e. dudas , s. lavignac and j. parmentier , nucl . b * 808 * , 237 ( 2009 ) [ arxiv:0808.0562 [ hep - ph ] ] ; s. andreas , arxiv:0905.0785 [ hep - ph ] . t. han and r. hempfling , phys . b * 415 * , 161 ( 1997 ) [ arxiv : hep - ph/9708264 ] . l. j. hall , t. moroi and h. murayama , phys . b * 424 * , 305 ( 1998 ) [ arxiv : hep - ph/9712515 ] . d. tucker - smith and n. weiner , phys . d * 64 * , 043502 ( 2001 ) [ arxiv : hep - ph/0101138 ] ; phys . d * 72 * , 063509 ( 2005 ) [ arxiv : hep - ph/0402065 ] . s. nussinov , l. t. wang and i. yavin , jcap * 0908 * , 037 ( 2009 ) [ arxiv:0905.1333 [ hep - ph ] ] . a. menon , r. morris , a. pierce and n. weiner , arxiv:0905.1847 [ hep - ph ] . e.g. _ , v. a. kudryavtsev , m. robinson and n. j. c. spooner , j. phys . * 203 * , 012039 ( 2010 ) [ arxiv:0912.2983 [ hep - ex ] ] . a. bottino , f. donato , n. fornengo and s. scopel , phys . d * 68 * , 043506 ( 2003 ) [ arxiv : hep - ph/0304080 ] ; phys . d * 77 * , 015002 ( 2008 ) [ arxiv:0710.0553 [ hep - ph ] ] ; phys . d * 78 * , 083520 ( 2008 ) [ arxiv:0806.4099 [ hep - ph ] ] . m. a. shifman , a. i. vainshtein and v. i. zakharov , phys . b * 78 * , 443 ( 1978 ) . v. barger , j. kumar , d. marfatia , e. sessolo , in preparation . h. j. he , n. polonsky and s. f. su , phys . d * 64 * , 053004 ( 2001 ) [ arxiv : hep - ph/0102144 ] ; j. alwall _ et al . _ , eur . j. c * 49 * , 791 ( 2007 ) [ arxiv : hep - ph/0607115 ] ; g. d. kribs , t. plehn , m. spannowsky and t. m. p. tait , phys . rev . d * 76 * , 075016 ( 2007 ) [ arxiv:0706.3718 [ hep - ph ] ] ; r. fok and g. d. kribs , phys . d * 78 * , 075023 ( 2008 ) [ arxiv:0803.4207 [ hep - ph ] ] ; b. holdom , w. s. hou , t. hurth , m. l. mangano , s. sultansoy and g. unel , pmc phys . a * 3 * , 4 ( 2009 ) [ arxiv:0904.4698 [ hep - ph ] ] ; m. hashimoto , arxiv:1001.4335 [ hep - ph ] . v. m. abazov _ et al . _ [ d0 collaboration ] , phys . * 97 * , 171806 ( 2006 ) [ arxiv : hep - ex/0608013 ] . t. aaltonen _ et al . _ [ cdf collaboration ] , phys . rev . d * 76 * , 072010 ( 2007 ) [ arxiv:0707.2567 [ hep - ex ] ] . t. aaltonen _ et al . _ [ cdf collaboration ] , phys . rev . lett . * 102 * , 221801 ( 2009 ) [ arxiv:0903.2618 [ hep - ex ] ] . e.g. _ , cms collaboration , physics analysis summary sus-08 - 005 , http://cms-physics.web.cern.ch/cms-physics/public/sus-08-005-pas.pdf . a. g. ivanov [ cdf collaboration ] , arxiv:0811.0788 [ hep - ex ] . t. han , r. mahbubani , d. g. e. walker and l. t. e. wang , jhep * 0905 * , 117 ( 2009 ) [ arxiv:0803.3820 [ hep - ph ] ] . meade and m. reece , phys . d * 74 * , 015010 ( 2006 ) [ arxiv : hep - ph/0601124 ] . j. alwall _ et al . _ , jhep * 0709 * , 028 ( 2007 ) [ arxiv:0706.2334 [ hep - ph ] ] . t. sjostrand , s. mrenna and p. skands , jhep * 0605 * , 026 ( 2006 ) [ arxiv : hep - ph/0603175 ] . pgs pretty good simulator , http://www.physics.ucdavis.edu/@xmath218conway/research/software/ pgs / pgs4-general.html . a. d. martin , r. g. roberts , w. j. stirling and r. s. thorne , eur . j. c * 28 * , 455 ( 2003 ) [ arxiv : hep - ph/0211080 ] ; a. d. martin , r. g. roberts , w. j. stirling and r. s. thorne , eur . phys . j. c * 35 * , 325 ( 2004 ) [ arxiv : hep - ph/0308087 ] ; p. m. nadolsky _ et al . _ , phys . d * 78 * , 013004 ( 2008 ) [ arxiv:0802.0007 [ hep - ph ] ] . v. m. abazov _ et al . _ [ d0 collaboration ] , phys . lett . b * 638 * , 119 ( 2006 ) [ arxiv : hep - ex/0604029 ] . atlas note atl - phys - pub-2009 - 084 . a. datta , k. kong and k. t. matchev , phys . d * 72 * , 096006 ( 2005 ) [ erratum - ibid . d * 72 * , 119901 ( 2005 ) ] [ arxiv : hep - ph/0509246 ] . c. g. lester and d. j. summers , phys . b * 463 * , 99 ( 1999 ) [ arxiv : hep - ph/9906349 ] ; c. lester and a. barr , jhep * 0712 * , 102 ( 2007 ) [ arxiv:0708.1028 [ hep - ph ] ] ; w. s. cho , k. choi , y. g. kim and c. b. park , phys . lett . * 100 * , 171801 ( 2008 ) [ arxiv:0709.0288 [ hep - ph ] ] ; a. j. barr , b. gripaios and c. g. lester , jhep * 0802 * , 014 ( 2008 ) [ arxiv:0711.4008 [ hep - ph ] ] ; m. m. nojiri , y. shimizu , s. okada and k. kawagoe , jhep * 0806 * , 035 ( 2008 ) [ arxiv:0802.2412 [ hep - ph ] ] ; m. burns , k. kong , k. t. matchev and m. park , jhep * 0903 * , 143 ( 2009 ) [ arxiv:0810.5576 [ hep - ph ] ] ; a. j. barr and c. gwenlan , phys . d * 80 * , 074007 ( 2009 ) [ arxiv:0907.2713 [ hep - ph ] ] . m. drees and x. tata , phys . b * 252 * , 695 ( 1990 ) ; g. aad _ et al . _ [ the atlas collaboration ] , arxiv:0901.0512 [ hep - ex ] ; p. mermod [ atlas collaboration ] , arxiv:0909.1911 [ hep - ex ] . d. hooper , f. petriello , k. m. zurek and m. kamionkowski , phys . d * 79 * , 015010 ( 2009 ) [ arxiv:0808.2464 [ hep - ph ] ] ; j. l. feng , j. kumar , j. learned and l. e. strigari , jcap * 0901 * , 032 ( 2009 ) [ arxiv:0808.4151 [ hep - ph ] ] ; j. kumar , j. g. learned and s. smith , phys . d * 80 * , 113002 ( 2009 ) [ arxiv:0908.1768 [ hep - ph ] ] ; v. niro , a. bottino , n. fornengo and s. scopel , phys . d * 80 * , 095019 ( 2009 ) [ arxiv:0909.2348 [ hep - ph ] ] .

we study the higher derivative chiral models with four supercharges and bogomolnyi - prasad - sommerfield ( bps ) states in these models . the off - shell lagrangian generically includes higher powers of the auxiliary fields which causes distinct on - shell branches associated with the solutions to the auxiliary fields equation . we point out that the model admits a supersymmetric completion of arbitrary higher derivative bosonic models of a single complex scalar field and an arbitrary scalar potential can be introduced even without superpotentials . as an example , we present a supersymmetric extension of the faddeev - skyrme model without four time derivatives , in contrast to the previously proposed supersymmetric faddeev - skyrme - like model containing four time derivatives . in general , higher derivative terms together with a superpotential result in deformed scalar potentials . we find that higher derivative corrections to 1/2 bps domain walls and 1/2 bps lumps are exactly canceled out while the 1/4 bps lumps ( as compact baby skyrmions ) depend on a characteristic feature of the higher derivative models . we also find a new 1/4 bps condition for domain wall junctions which generically receives higher derivative corrections . june , 2014 0.5 cm * bps states in supersymmetric chiral models + 0.3 cm with higher derivative terms * 1.1 cm muneto nitta and shin sasaki 0.7 cm _ department of physics , and research and education center for natural sciences , + -0.2 cm keio university , hiyoshi 4 - 1 - 1 , yokohama , kanagawa 223 - 8521 , japan 0.1 cm department of physics , kitasato university + -0.2 cm sagamihara 252 - 0373 , japan _ 0.5 cm

1406.7647

low energy dynamics of field theories can be described by only light fields such as nambu - goldstone modes when one integrates out massive modes . the low - energy effective theories are usually expanded by derivative expansions ; thereby , they inevitably contain higher derivative terms of fields . chiral perturbation theory is such a theory describing low - energy pion dynamics in qcd with a chiral symmetry breaking @xcite . the skyrme model @xcite , which is a non - linear sigma model with fourth order derivative terms , is one of such a class . supergravity as low - energy effective theory of string theory should have higher derivative correction terms @xcite . other examples include world - volume effective actions of solitonic objects such as topological solitons in field theories and d - branes in string theories @xcite . the effective theory of a d - brane is described by the dirac - born - infeld ( dbi ) action @xcite containing an infinite number of derivatives . higher derivative field theories are also useful in other areas of physics . in the cosmological context , higher derivative theories are proposed for inflation models such as the k - inflation @xcite and the galileon inflation @xcite . these higher derivative models are known to admit characteristic soliton solutions such as k - defects @xcite , compactons @xcite and so on . on the other hand , supersymmetry is one of the most important tools in modern high energy physics . it has not only been considered as the most promising candidate to solve the naturalness problem in the standard model in the phenomenological side , but also it plays important roles to control quantum corrections in supersymmetric field theories , leading to determining exact low - energy dynamics @xcite . when one constructs low - energy effective theories in supersymmetric field theories , one is required to consider higher derivative corrections in a supersymmetric manner . it is , however , not so easy to construct supersymmetric completion of general higher derivative theories . off - shell superfield formalisms are useful to write down actions of supersymmetric higher derivative models . in particular , the four - dimensional @xmath4 superfield formalism that incorporates the chiral superfield @xmath5 is a simple starting point . it is , however , known that not all the off - shell supersymmetric higher derivative models exhibit good physical properties . off - shell formulations of higher derivative terms often encounter with an auxiliary field problem ; chiral superfields with space - time derivatives ( e.g. @xmath6 ) sometimes introduce derivative interactions of the auxiliary field @xmath0 . consequently , the auxiliary fields become dynamical . it is hard to eliminate them , and the on - shell structure of the action is not obvious . for instance , the chiral lagrangian of qcd contains the wess - zumino - witten ( wzw ) term to reproduce the quantum anomaly at low energy . however , a supersymmetric completion of the wzw term proposed in ref . @xcite suffers from this auxiliary field problem @xcite . it was proposed in ref . @xcite that a supersymmetric wzw term in superspace can be constructed without the auxiliary field problem if the number of chiral superfields is doubled . the auxiliary field problem would be more problematic if one were to introduce a superpotential , so one could not introduce a potential . nevertheless , supersymmetric higher derivative models of which the building blocks are the chiral superfields are studied in various contexts . among other things , the chiral models studied in refs . @xcite provide a good grounding for studying supersymmetric higher derivative theories . in this model , the auxiliary fields are not accompanied by the space - time derivatives and therefore they can be eliminated by their equations of motion . in principle , it is possible to write down the explicit on - shell actions of the models . in particular , the scalar potential that shows up after eliminating the auxiliary fields looks more apparent @xcite . the coupling of higher derivative chiral models to supergravity was also achieved in this type of model @xcite . a supersymmetric dbi action was constructed in ref . the other examples include a supersymmetric completion of the @xmath7 model @xcite , the supersymmetric galileon inflation models @xcite and models for the ghost condensation @xcite . the same structure appears in quantum effective actions @xcite a higher derivative supersymmetric @xmath8 model free from the auxiliary field problem was also considered previously as a supersymmetric extension @xcite of the faddeev - skyrme model @xcite and a supersymmetric baby skyrme model @xcite . the formalism in refs . @xcite has been also applied to the construction of manifestly supersymmetric higher derivative corrections to supersymmetric nonlinear realizations @xcite . in the former half of this paper , we study higher derivative chiral models developed in refs . @xcite in the superfield formalism , where higher derivative terms can be introduced as a tensor with two holomorphic and symmetric indices and two anti - holomorphic and symmetric indices . we find a surprising fact that has been overlooked in past studies on the supersymmetric completions of various higher derivative models . the model with a single chiral superfield admits a supersymmetric extension of _ arbitrary _ bosonic models that consist of a single complex scalar field . as an example , we present a supersymmetric extension of the faddeev - skyrme model @xcite . the bosonic part of this model does not contain four time derivatives . this is in contrast to the previously proposed supersymmetric extension @xcite of the faddeev - skyrme model that contains an additional four derivative term that includes four time derivatives . moreover , we point out that an arbitrary scalar potential can be introduced even without the superpotential . we further work out the higher derivative chiral models with superpotentials . the resulting on - shell lagrangians are highly non - linear . we study perturbative analysis revealing the possibility of ghost kinetic term and deformations of the scalar potential . meanwhile , bogomolnyi - prasad - sommerfield ( bps ) topological solitons play important roles in the study of non - perturbative dynamics of supersymmetric field theories since they break and preserve a fraction of supersymmetry , belong to short supermultiplets , and consequently are stable against quantum corrections @xcite . when a bps soliton preserves @xmath9 of supersymmetry , it is called a @xmath9 bps soliton . for instance , yang - mills instantons , bps monopoles , vortices , lumps and domain walls @xcite are of @xmath10 bps and composite solitons such as domain wall junctions are of 1/4 bps in theories with four supercharges @xcite and eight supercharges @xcite ( see refs . @xcite as a review for a fraction of supersymmetry for bps states ) . bps solitons remain important in supersymmetric field theories with higher derivative terms . prime examples of such solitons contain 1/2 bps lumps in supersymmetric @xmath8 models with a four - derivative term @xcite , supersymmetric baby skyrmions , which are compactons @xcite ; and bps compactons in k - field theories @xcite . the higher derivative @xmath8 model in ref . @xcite appears as the effective theory of a 1/2 bps non - abelian vortex @xcite in supersymmetric theories with eight supercharges . then , the 1/2 bps lumps in the vortex correspond to yang - mills instantons in the bulk @xcite . while a few examples of bps solitons in higher derivative supersymmetric theories have been studied thus far , a systematic study of bps solitons in such theories is needed . in the latter half of this paper , we give a general framework to examine bps states in supersymmetric higher derivative chiral models . our framework does not only reproduce , in a unified manner , a few remarkable previous studies of the bps bounds in the supersymmetric higher derivative models admitting bps baby skyrmions @xcite , bps compactons @xcite , and bps lumps @xcite , but also includes the more general cases with several new bps states ; 1/2 bps domain walls , 1/4 bps domain wall junctions , 1/2 and 1/4 bps lumps and baby skyrmions . in particular , we find bps baby skyrmions found in ref . @xcite to be 1/4 bps states . we show that 1/2 bps domain walls and 1/2 bps lumps do not receive higher derivative corrections while 1/4 bps domain wall junctions do . the organization of this paper is as follows . in sec . [ sec : hdc ] , we introduce the supersymmetric higher derivative chiral model with four supercharges . we write down the equation of motion for the auxiliary fields and analyze the structure of the on - shell lagrangians . in particular , we introduce the superpotential and the deformation of the scalar potential caused by the higher derivative terms is discussed . we then examine bps states that preserve 1/2 and 1/4 of the original supersymmetry in subsequent sections . the 1/2 bps domain wall and 1/4 bps domain wall junctions are studied in sec . [ sec : wall ] , and 1/2 bps and 1/4 bps lumps are studied in sec . [ sec : lump ] . section [ sec : conc ] is devoted to conclusion and discussions . notations and conventions of superfields are found in the appendix [ sec : notation ] . in the first subsection , we present general higher derivative chiral models with multiple chiral superfields . in the second subsection , we further work out the models with a single chiral superfield without and with a superpotential . we consider four - dimensional @xmath4 supersymmetric higher derivative chiral models that have specific properties . the lagrangian consists of chiral superfields @xmath11 @xmath12 , for which the component expansion in the chiral base @xmath13 is @xmath14 where @xmath15 is the complex scalar field , @xmath16 is the weyl fermion and @xmath17 is the complex auxiliary field . the notations and conventions of the chiral superfield are found in appendix [ sec : notation ] . the supersymmetric lagrangian with higher derivative terms is given by @xmath18 where @xmath19 is the khler potential and @xmath20 is a superpotential as usual . higher derivative terms are produced by the second term proportional to @xmath21 , which is a @xmath22 khler tensor symmetric in holomorphic and anti - holomorphic indices , of which the components are functions of @xmath11 and @xmath23 ( admitting space - time derivatives acting on them ) . as we will see , the most important feature of this model is that the auxiliary fields never become dynamical ; the equation of motion for the auxiliary fields is an algebraic equation . now we examine the component structure of the model . the fourth derivative part of the lagrangian has an essential property . this term is evaluated as @xmath24 + i_f , \label{eq:4th}\end{aligned}\ ] ] where @xmath25 stands for terms that contain fermion fields . since the bosonic part of the right hand side of saturates the grassmann coordinate @xmath26 , only the lowest component of the tensor @xmath21 contributes to the bosonic part of the lagrangian . therefore the bosonic part of the lagrangian is @xmath27 . \label{eq : comlagrangian}\end{aligned}\ ] ] this lagrangian exhibits a higher derivative model that has the following properties : ( i ) the higher derivative terms are governed by the tensor @xmath21 , and ( ii ) the model is manifestly ( off - shell ) supersymmetric and khler invariant provided that @xmath19 and @xmath20 are scalars and @xmath21 is a tensor . among other things , the auxiliary fields do not have a space - time derivative @xcite . they are irrelevant when classical configurations of fields are concerned . ] and they are eliminated by the following equation of motion : @xmath28 this is an algebraic equation and , in principle , solvable . however , the equation is a simultaneous equation of cubic power and it is hard to find explicit solutions @xmath29 . we comment that when @xmath30 at least @xmath31 is a solution . in this case , the on - shell lagrangian becomes @xmath32 in general , there are more solutions other than @xmath31 which we will show explicitly for models with one component field . now we consider the single chiral superfield @xmath5 for simplicity . the equation of motion for the auxiliary field becomes @xmath33 here @xmath34 . we solve the equation in the @xmath30 and @xmath35 cases separately . when there is no superpotential , the equation for the auxiliary field becomes @xmath37 then the solutions are found to be @xmath38 there are two different on - shell branches associated with the solutions and . for the first solution , the bosonic part of the on - shell lagrangian is @xmath39 the first term is the ordinary kinetic term and the second term contains higher derivative correction terms . we call this the canonical branch . an example of the model is the @xmath4 supersymmetric dbi action for the world - volume theory of single d3-brane . the corresponding khler metric is canonical , @xmath40 , and the function @xmath41 is given by @xcite @xmath42 the other examples include a supersymmetric completion of the @xmath7 model @xcite , the supersymmetric galileon inflation models @xcite and models for the ghost condensation @xcite . another example of @xmath41 that has been overlooked in the literature @xcite is @xmath43,\end{aligned}\ ] ] where @xmath44 is a parameter . then , with the fubini - study metric @xmath45 for the @xmath46 model , the bosonic part of the lagrangian becomes @xmath47 this is nothing but the faddeev - skyrme model @xcite . the previous trials to construct an @xmath48 supersymmetric extension of the faddeev - skyrme model concluded that one needs an extra four - derivative term containing four time derivatives @xcite while the lagrangian in eq . it was discussed in ref . @xcite that such a term destabilizes hopfions ( knot solitons ) . therefore , the lagrangian provides an @xmath49 supersymmetric extension of the faddeev - skyrme model without four time derivatives , which is expected to give stable hopfions . more generally , since the function @xmath41 is completely arbitrary , one can construct supersymmetric extension of _ any _ bosonic models that consist of a complex scalar field @xmath50 . more surprisingly , we further point out that it is also possible to introduce an arbitrary scalar potential @xmath51 even without superpotentials , by choosing @xmath41 as @xmath52 however , as we will clarify later , superpotentials play an important role when one considers bps solutions . on the other hand , for the second solution , the bosonic part of the on - shell lagrangian is @xmath53 in this branch , the canonical kinetic term disappears . is chosen as @xmath54 , the canonical kinetic term recovers . however quite non - linear higher derivative terms remain in the lagrangian due to the factor @xmath55 . this possibility was discussed in the context of higher derivative supergravity models @xcite . ] this model was first studied in ref . @xcite where supersymmetric extensions of the baby skyrme model are discussed . we note that the second branch does not have the smooth limit to the canonical theory ( @xmath56 ) . therefore we call this the non - canonical branch . since @xmath57 should be positive semi - definite , the second solution is consistent only in the region @xmath58 we comment on the last term in eq . . the term @xmath59 is considered as a scalar potential term since it remains when the function @xmath41 does not depend on fields with space - time derivatives . for a vacuum configuration , the condition implies @xmath60 for the positive definite khler metric @xmath61 . then the scalar potential at a vacuum becomes negative even for the manifestly supersymmetric construction of the model . one resolution of this puzzle is the existence of ghosts , i.e. , fields with a kinetic term of the wrong sign . however it is not obvious whether ghosts exist or not since there is no kinetic term in the lagrangian and no consistent free theory is defined . in that case , @xmath19 loses its meaning of the khler potential and what determines the sign of the potential energy is the function @xmath19 . when @xmath62 is negative , @xmath41 and the scalar potential become positive . actually , choosing the functions of @xmath19 and @xmath41 appropriately , one can construct scalar potentials that have desired properties @xcite . when @xmath35 one eliminates @xmath64 in and obtains the equation for the auxiliary field @xmath0 : @xmath65 when there are no higher derivative corrections @xmath66 , one recovers the ordinary @xmath0-term solution @xmath67 . since the equation is an algebraic equation of cubic power , the solutions are obtained by the cardano s method @xcite , @xmath68 { - \frac{q}{2 } + \sqrt{\left(\frac{q}{2}\right)^2 + \left(\frac{p}{3}\right)^3 } } + \omega^{3-k } \sqrt[3 ] { - \frac{q}{2 } - \sqrt{\left(\frac{q}{2}\right)^2 + \left(\frac{p}{3}\right)^3 } } , \notag \\ & & k = 0,1,2 , \ \omega^3 = 1 , \label{eq : aux_3sol}\end{aligned}\ ] ] where @xmath69 is a cubic root of unity and @xmath70 and @xmath71 are given by @xmath72 the on - shell lagrangian is obtained by substituting the solutions of the auxiliary field into the lagrangian , @xmath73 where @xmath74 ( @xmath75 ) is one of the solutions in . therefore , there are three different on - shell branches in this model . we note that although the model is corrected by higher derivative terms , the supersymmetry requires correction terms in the scalar potential that do not contain derivative terms . in particular , the scalar potential of the model is calculated to be @xmath76 here @xmath77 is the function @xmath41 where @xmath78 . we note that even for the manifestly supersymmetric lagrangian with the positive khler metric @xmath79 , a negative scalar potential is possible when @xmath80 . again , this fact would be an indication of ghost states in the theory . as we will see below , the on - shell lagrangian potentially includes ghost states . now we examine the structure of the on - shell lagrangians in each branch . to see the effects of superpotentials , we write down the explicit on - shell component lagrangian . in particular , we examine the relation between the positive definiteness of the scalar potential and the ghost states . a similar analysis about the scalar potential was performed in the context of the four - dimensional @xmath4 supergravity @xcite , in which negative potentials are not problematic . on the other hand , negative potentials could be problematic for the rigid supersymmetric case , on which we focus here . we note that when @xmath81 , a solution @xmath82 is not allowed . we first consider the canonical branch where the solution of the auxiliary field has the smooth limit @xmath56 @xcite . we look for a perturbative expression of the lagrangian for small @xmath41 . the solution of the auxiliary field is expanded as @xmath83 where @xmath84 is a parameter associated with the small @xmath41 expansion and @xmath85 is the solution in @xmath86 ( @xmath66 ) . this is given by @xmath87 here @xmath88 and @xmath89 is the complex conjugate of @xmath90 . the explicit forms of @xmath91 and @xmath92 are obtained iteratively . they are found to be @xmath93 , \\ f_2 = & \ - \frac{4 \lambda^2 \bar{w}'}{(k_{\varphi \bar{\varphi}})^7 } ( \bar{w } ' w ' - k_{\varphi \bar{\varphi } } \partial_m \varphi \partial^m \bar{\varphi } ) \left\ { 3 w ' \bar{w } ' - ( k_{\varphi \bar{\varphi}})^2 \partial_m \varphi \partial^m \bar{\varphi } \right\}.\end{aligned}\ ] ] then , substituting these solutions into the auxiliary field @xmath0 in the lagrangian , we obtain the on - shell lagrangian , ( we take @xmath94 for simplicity ) @xmath95 here @xmath96 is the ordinary scalar potential in the supersymmetric chiral models . we note that the scalar potential is deformed by the non - zero @xmath41 and the vacuum structure clearly depends on the structure of the function @xmath41 . the examples of the deformed scalar potentials are found in fig . [ fig : potential ] . the lagrangian contains an infinite number of the higher derivative terms that are induced by non - zero @xmath41 and @xmath20 . the structure of the derivative terms is completely determined by supersymmetry . we point out that even for the canonical kinetic term , it is deformed by @xmath41 . up to @xmath97 , it is given by @xmath98 \partial_m \varphi \partial^m \bar{\varphi } + \mathcal{o } ( \lambda^3 ) . \label{eq : lag - pot}\end{aligned}\ ] ] since @xmath41 is an arbitrary function , the sign of the kinetic term can be flipped even for the positive definite khler metric @xmath62 . if the sign of the kinetic term is changed , there appear ghost states in the theory @xcite . in that case , the model shows instability caused by the higher derivatives . this fact leads to the non - positive semi - definite potential even for supersymmetric theories . the sign of the kinetic term depends on the explicit forms of the functions @xmath41 and @xmath20 . although it is important and interesting , we do not pursue the ( non-)existence of the ghost states in this paper . we also note that the metric of the target space of the nonlinear sigma model in the lagrangian ( [ eq : lag - pot ] ) does not have to be khler anymore even though it is @xmath48 supersymmetric . next , we study the effect of the superpotential in the non - canonical branch associated with the solution . since we can not take the @xmath56 limit , we consider the small @xmath20 perturbation around @xmath36 . the solution of the auxiliary field is expanded as @xmath99 where @xmath100 is a parameter associated with the small @xmath20 expansion and @xmath101 here we choose a real solution of @xmath85 . using the @xmath102 symmetry , we make the superpotential be real and positive . then , the solutions @xmath103 and @xmath104 are found to be @xmath105 the on - shell lagrangian is @xmath106 we can observe that the scalar potential @xmath107 is deformed by the superpotential @xmath20 . finally , a comment is in order . we started from the four - dimensional theory . however , the lower dimensional models , such as three - dimensional @xmath108 and two - dimensional @xmath109 theories can be easily obtained by the dimensional reduction . actually , the @xmath30 case corresponds to the three - dimensional @xmath110 models discussed in ref . in this and the next sections , we study bps configurations in the supersymmetric higher derivative chiral models discussed in the previous section . since we consider models with scalar fields , we focus on the bps domain walls and lumps in the following . bps equations that preserve a fraction of supersymmetry are obtained from the condition that the on - shell supersymmetry transformation of the fermion vanishes @xmath111 . here @xmath112 represents the on - shell supersymmetry transformation by parameters @xmath113 , @xmath114 . the off - shell supersymmetry transformation @xmath115 of the fermion is given by @xmath116 by substituting a solution of the auxiliary field @xmath0 into @xmath117 and assuming a specific field configuration together with appropriate killing spinor conditions on @xmath113 , @xmath114 , we find corresponding on - shell bps equations . since there is the variety of branches associated with the solutions @xmath0 in our model , we study each branch separately . when a scalar field model with an ordinary canonical kinetic term has a potential with several vacua , there is a domain wall solution that interpolates between these vacua . we look for 1/2 bps domain wall solutions in the higher derivative model . we consider domain wall configurations of the complex scalar field @xmath50 . namely , the field depends on the one direction @xmath118 . we first consider the case in which the superpotential exists . in this case , the solution @xmath82 is not allowed . therefore , we generically consider the @xmath119 branch . the killing spinor condition for the 1/2 bps domain wall configuration is @xcite @xmath120 here @xmath121 is a phase factor . then the off - shell bps equation is @xmath122 by plugging a solution in into the right hand side of the equation and arranging the resulting condition by @xmath123 , we obtain the on - shell bps condition . here , instead of that , we use the equation of motion for the auxiliary field @xmath0 in order to observe the universal property of the three solutions . substituting the bps condition into the equation of motion for @xmath0 , we obtain @xmath124 the higher derivative terms including @xmath41 cancel out and we obtain the on - shell bps equation @xmath125 equation . is nothing but the ordinary ( without higher derivative terms ) bps condition for the domain wall . this result suggests that even for the existence of the three different on - shell branches in the model , the bps domain wall can not distinguish them . furthermore , the on - shell energy density of the domain wall is evaluated as @xmath126 the last expression gives the tension of the ordinary bps domain wall . therefore we conclude that all the higher derivative corrections to the solutions and energy are canceled out in the bps domain walls . this is a consequence of the fact that the configuration depends on the one direction . it is easy to confirm that the solutions to the bps condition together with the equation of motion for the auxiliary field satisfy the full equation of motion for the scalar field does not depends on the second space - time derivatives or higher of @xmath50 . ] @xmath127 \frac{\partial \lambda}{\partial \bar{\varphi } } \notag \\ & - \partial_m \left [ - k_{\varphi \bar{\varphi } } \partial^m \varphi + 2 \lambda ( ( \partial_n \varphi)^2 \partial^m \bar{\varphi } - |f|^2 \partial^m \varphi ) + \left\ { \varphi|^2 + |f|^4 \right\ } \frac{\partial \lambda}{\partial ( \partial_m \bar{\varphi } ) } \right ] = 0 . \label{eq : varphi_eom}\end{aligned}\ ] ] next , we consider the case in which @xmath36 . even for this case , there is the scalar potential @xmath59 in the lagrangian . this branch corresponds to the @xmath128 solution . substituting the off - shell bps condition into the equation of motion for the auxiliary field and assuming @xmath128 , the on - shell bps condition becomes @xmath129 this condition never provides the domain wall equation . when there is no ghost , eq . is just a vacuum condition of the scalar potential @xmath59 . therefore , although there is a scalar potential in the non - canonical branch , superpotentials are necessary for 1/2 bps domain wall solutions . we also note that the 1/2 bps domain wall solution to the equation interpolates between `` vacua '' specified by the superpotential @xmath130 as its tension stands for . we stress that the condition @xmath130 does not always imply vacua of the scalar potential especially in the non - canonical branch . the bps domain walls remain intact even for the deformation of the scalar potential . although there are other vacua that originate from the singularity of the function @xmath41 ( see fig.1 ( c ) ) , domain walls that interpolate these vacua are not bps and break all the supersymmetry . we next consider 1/4 bps domain wall junctions @xcite . the scalar field depends on the two spacial directions @xmath131 . first , we consider the @xmath81 case . we impose the killing spinor conditions on the supersymmetry parameters , @xmath132 where @xmath121 is a phase factor . then we obtain the bps condition from the supersymmetry transformation , and derivatives @xmath133 . @xmath134 is the complex conjugate of @xmath135 . ] @xmath136 here @xmath0 in the right hand side is one of the solutions in . this is the 1/4 bps condition . substituting the condition into the equation of motion for the auxiliary field , we obtain the on - shell bps equation on the scalar field : @xmath137 when @xmath66 , the on - shell bps equation becomes that of the ordinary bps domain wall junctions @xcite of which the analytic solutions are studied in ref . different from the 1/2 bps domain wall case , the higher derivative corrections do not cancel in the equation . the solutions are deformed from the ones in ref . @xcite in general and depend on the explicit form of the function @xmath41 . now we confirm that the bps solutions to satisfy the full equation of motion for the scalar field . using the bps condition , we find the following terms in vanish : @xmath138 by using the bps equation and the equation of motion for the auxiliary field , we find that the other terms in also vanish : @xmath139 = 0.\end{aligned}\ ] ] therefore we conclude that the solutions to the deformed bps equation actually satisfy the full equation of motion in eq . . the energy density of the domain wall junction is evaluated as @xmath140.\end{aligned}\ ] ] this is nothing but the expression of the ordinary ( without higher derivative terms ) domain wall junctions . after integration over the @xmath141 plane , the first term gives the junction charge and the second term gives the tension of the domain walls . they are evaluated on the boundary at the infinity of the @xmath142 plane . again , the junction charge and the domain wall tension are solely determined by the asymptotic boundary conditions of the scalar field and the superpotential , and do not depend on the function @xmath41 . although the expression of the bogomolnyi bound of the energy is not deformed by the higher derivative terms , we stress that the solutions of the 1/4 bps domain wall junction are potentially deformed in general . finally we examine 1/4 bps domain wall junctions in the @xmath36 non - canonical branch . the killing spinor and the off - shell bps conditions are given by eqs . and . the solution of the auxiliary field is given in eq . . then , the on - shell bps equation is found to be @xmath143 eq . is supplemented by the consistency condition in eq . . again , the higher derivative corrections to the on - shell 1/4 bps condition are not canceled . we will comment on this equation in the next section . next we consider lumps in @xmath144 higher derivative models . we look for the bps equation for lumps that depend on @xmath145 and @xmath146 . recall that for the @xmath36 case , the solutions of the auxiliary field are given by @xmath147 where @xmath84 is a phase factor . there are the canonical and non - canonical branches associated with the solutions in eqs . and , respectively . in the following subsections , we examine bps lump equations in each branch . we first focus on the canonical branch associated with the solution . hopfions in the supersymmetric higher derivative @xmath46 model of this type were discussed before @xcite . bps lumps in the supersymmetric higher derivative @xmath46 model were discussed in ref . bps lumps in the higher derivative @xmath148 non - linear sigma models were discussed in a different context without supersymmetry @xcite . in this branch , the bps lump equation is obtained by imposing the first condition in on the spinor @xmath114 , as can be seen in , e.g. , refs . then , the bps equation for lumps is given by @xcite @xmath149 this is nothing but the ordinary 1/2 bps lump condition . this is confirmed by the bogomolnyi bound of the energy density . for the canonical branch , we have the energy density @xmath150 where we have assumed the condition @xmath151 for the positive - semi definiteness of the energy @xmath152 . the right hand side is nothing but the topological charge density for the 1/2 bps lump . the energy bound is saturated provided the condition is satisfied . then we find that the higher derivative corrections to the solutions and the energy bound are canceled out in this branch . it is confirmed that solutions to the equation satisfy the full equation of motion for the scalar field . when we consider the fubini - study metric for the @xmath8 model and take the function @xmath41 as @xmath153 the bound becomes just the bps bound obtained in the context of the effective theory on a non - abelian vortex @xcite . in summary , although the lagrangian contains higher derivative corrections , the 1/2 bps lump solution to the equation ( which is a holomorphic function with appropriate boundary conditions ) does not receive any corrections in the canonical branch . we next consider the non - canonical branch . since this is associated with the @xmath128 solution even for @xmath30 , we need to impose both of the two conditions in in order to obtain the bps equation from the variation of the fermion . then the bps equation is @xmath154 where @xmath155 is a phase factor . this is the 1/4 bps equation . again , the bps lump does not cancel the higher derivative corrections generally . we can make the deformed bps equation into the following form , @xmath156 we confirm that the solutions to the bps equation satisfy the full on - shell equation of motion for the scalar field . in the non - canonical branch , we have the bogomolnyi completion of the energy @xmath157 ^ 2 + \frac{k_{\varphi \bar{\varphi}}}{2 } ( |\partial \varphi|^2 - |\bar{\partial } \varphi|^2 ) \notag \\ \ge & \ - i k_{\varphi \bar{\varphi } } \varepsilon_{ij } \partial_i \varphi \partial_j \bar{\varphi}. \label{eq : lump_energy_bound}\end{aligned}\ ] ] since we have @xmath158 for static configurations from the consistency condition of the solution , the energy bound is saturated by the topological charge density of lumps provided that the bps condition is satisfied . it is obvious that the expression of the topological charge is not corrected by the higher derivative terms . in the non - canonical branch , the lagrangian does not contain ordinary canonical kinetic term . an example of such a kind of non - canonical model is the extremal ( bps ) baby skyrme model . the model consists of the fourth derivative term and potential terms in ( 2 + 1 ) dimensions . more concretely , if we take the khler potential and @xmath41 as in , then the lagrangian is nothing but the fourth derivative part of the baby skyrme model with an irrelevant constant term . in ref . @xcite the authors found specific khler potentials and constructed the potentials of the baby skyrme model . actually , the condition was first found in the supersymmetric baby skyrme model @xcite . eq . is the same as the one found in the previous section , eq . . the difference of solutions is specified by boundary conditions . however , the energy bound in suggests that there are no bps domain walls junctions in the non - canonical branch . the example of solutions to eq . are the compact baby skyrmions @xcite that are solitons with compact support . the bps states in the higher derivative chiral models are summarized in table [ tb : bps_states ] . .bps states in the @xmath30 and @xmath35 higher derivative chiral models . corresponding solutions of the auxiliary field @xmath0 ( whether they vanish or not ) are also presented . [ cols="<,^,^",options="header " , ] [ tb : bps_states ] in this paper we have studied bps states in the four - dimensional @xmath4 supersymmetric higher derivative chiral model of which the lagrangian is given in eq . . the model is governed by a ( 2,2 ) khler tensor @xmath159 symmetric in holomorphic and anti - holomorphic indices , in addition to the khler potential @xmath19 and the superpotential @xmath20 . they are functions of the chiral superfields @xmath11 . in particular , the tensor @xmath159 determines the higher derivative interactions of the models . a specific feature of the model is that the auxiliary fields @xmath17 do not have space - time derivatives on them and can be eliminated by their equation of motion algebraically . one can explicitly write down the on - shell lagrangian of the model at least for a single chiral superfield . since the equation of motion for the auxiliary fields is no longer a linear equation , there are several on - shell branches in this model . this fact deserves new non - trivial bps equations that include higher derivative corrections . when there is no superpotential , there are two distinct on - shell branches . one is the canonical branch associated with the solution @xmath82 . an example of this model is the supersymmetric dbi model @xcite . we have shown that this branch , in fact , allows a supersymmetric extension of any bosonic models of complex scalar fields . we have exhibited the explicit function @xmath41 which corresponds to the supersymmetric extension of the faddeev - skyrme model without four time derivatives , which is in contrast to the previous studies @xcite concluding that such a term is necessary for supersymmetry . the other branch is the non - canonical one corresponding to the solution @xmath128 . in this branch , the ordinary canonical kinetic term disappears and the lagrangian starts from the forth order derivative terms . an example of this model is the extremal ( bps ) baby skyrme model . this branch was discussed in refs . although the @xmath30 case has been essentially discussed in the literature @xcite , things get more involved when one introduces a superpotential @xmath20 . in this case , a solution @xmath82 is not allowed . there are three on - shell branches associated with the three different solutions of the auxiliary field equation @xcite . the resulting on - shell lagrangians have highly non - linear expressions . perturbative analysis reveals the possibility of ghost kinetic terms and deformations of the scalar potential . even though the on - shell lagrangian is complicated and becomes highly non - linear in the @xmath81 case , one can derive the off - shell bps conditions from the supersymmetry transformation of fermions . these conditions are supplemented by the equation of motion for the auxiliary field giving rise to the on - shell conditions . we have analyzed the properties of bps states . for the 1/2 bps domain wall case , the higher derivative corrections are exactly canceled out in the @xmath35 case . the solution to the bps equation satisfies the full equation of motion for the scalar field . we have shown that the tension of the domain wall does not receive any higher derivative corrections . in the @xmath30 non - canonical branch , the 1/2 bps condition does not provide the domain wall equation . for the 1/4 bps domain wall junction in the @xmath35 case , the on - shell bps equation receives higher derivative corrections . this is a new 1/4 bps equation for domain wall junctions . the solution is deformed by the higher derivative effects and it is confirmed that the solution satisfies the full equation of motion . the expression of the energy bound is shown to be the same as the ordinary ( without higher derivative terms ) theory , namely , the sum of the junction charge and the tension . for lump configurations in the @xmath30 case , there are two on - shell bps equations . one is the 1/2 bps lumps associated with the @xmath160 solution , where all the derivative corrections are canceled out . the other is the 1/4 bps lumps associated with the @xmath128 solution . the on - shell bps equation is deformed by the higher derivative corrections . this is nothing but the equation studied in ref . an example of solutions to this equation is compactons in the extremal ( bps ) baby skyrme model . while we were able to solve explicitly auxiliary field equations ( [ eq : af - eom ] ) only for one chiral superfield , reducing the third order algebraic equation ( [ eq : af - eom2 ] ) , the multicomponent equation ( [ eq : af - eom ] ) has yet to be solved . when the target space has a large isometry , it should be possible to solve it . construction of more general target spaces , for instance a higher derivative @xmath161 model and its bps solitons remains as a future problem . while we have exhausted all bps states that are already known in conventional @xmath48 supersymmetric theories without higher derivatives , there may still remain unknown bps states particular for higher derivative theories . in fact , 1/4 baby bps skyrmions do not exist in conventional theories . a sine - gordon kink inside a domain wall ( corresponding to a baby skyrmion in the bulk ) @xcite , a baby skyrmion inside a domain wall ( corresponding to a three - dimensional skyrmion in the bulk ) @xcite , or a baby skyrmion string ending on a domain wall @xcite or stretched between domain walls @xcite is one of possibilities of composite bps states . it should be important to generalize our formalism to theories with extended supersymmetries such as eight supercharges . although , only four out of eight supercharges are manifestly realized in the @xmath4 superfield formalism , this is still useful to study the off - shell effective theory of bps solitons in models with eight supercharges @xcite . supersymmetric theories with eight supercharges are known to admit plenty of composite bps states @xcite . in particular , a classification of all possible bps states in supersymmetric theories with eight supercharges was given in ref . it is an interesting future problem to explore which bps states ( do not ) receive higher derivative corrections . as this problem concerns , 1/2 bps topological solitons in theories with eight supercharges preserve four supercharges on their world - volume . off - shell effective actions of the 1/2 bps domain wall and vortex were obtained in @xmath162 , @xmath48 superfield formalism at the leading order @xcite . the formulation presented in this paper should be useful to obtain the off - shell action of higher derivative corrections to these effective actions . for instance , as mentioned in eqs . ( [ eq : lump1 ] ) and ( [ eq : bps - bound - lump ] ) , the @xmath8 model with a four - derivative term appearing as the effective theory of a non - abelian vortex admits 1/2 bps lumps @xcite , corresponding to yang - mills instantons in the bulk @xcite . in the same way , an @xmath163 principal chiral model with the skyrme term appears @xcite on a non - abelian domain wall @xcite . the off - shell higher derivative corrections to the effective theories on these solitons are some of future directions . the authors would like to thank masahide yamaguchi for discussions . the work of m. n. is supported in part by grant - in - aid for scientific research ( no . 25400268 ) and by the `` topological quantum phenomena '' grant - in - aid for scientific research on innovative areas ( no . 25103720 ) from the ministry of education , culture , sports , science and technology ( mext ) of japan . the work of s. s. is supported in part by kitasato university research grant for young researchers . we use the notation of the textbook of wess and bagger @xcite . the component expansion of the @xmath4 chiral superfield in the @xmath164-basis is @xmath165 where only the bosonic components are presented . the supercovariant derivatives are defined as @xmath166 the sigma matrices are @xmath167 . here @xmath168 are pauli matrices . 0 h. leutwyler , `` on the foundations of chiral perturbation theory , '' annals phys . * 235 * , 165 ( 1994 ) [ hep - ph/9311274 ] . t. h. r. skyrme , a nonlinear field theory , " proc . r. soc . lond . a * 260 * ( 1961 ) 127 ; a unified field theory of mesons and baryons , " nucl . 31 * ( 1962 ) 556 . j. polchinski , `` string theory . 1 : an introduction to the bosonic string , '' cambridge , uk : univ . pr . ( 1998 ) 402 p , `` string theory . 2 : superstring theory and beyond , '' cambridge , uk : univ . ( 1998 ) 531 p e. s. fradkin and a. a. tseytlin , `` nonlinear electrodynamics from quantized strings , '' phys . b * 163 * ( 1985 ) 123 . m. born and l. infeld , `` foundations of the new field theory , '' proc . a * 144 * ( 1934 ) 425 ; p. a. m. dirac , `` an extensible model of the electron , '' proc . a * 268 * ( 1962 ) 57 . c. armendariz - picon , t. damour and v. f. mukhanov , `` k - inflation , '' phys . b * 458 * ( 1999 ) 209 [ hep - th/9904075 ] . a. nicolis , r. rattazzi and e. trincherini , `` the galileon as a local modification of gravity , '' phys . d * 79 * ( 2009 ) 064036 [ arxiv:0811.2197 [ hep - th ] ] . e. babichev , `` global topological k - defects , '' phys . d * 74 * ( 2006 ) 085004 [ hep - th/0608071 ] . c. adam , p. klimas , j. sanchez - guillen and a. wereszczynski , `` compact baby skyrmions , '' phys . d * 80 * , 105013 ( 2009 ) [ arxiv:0909.2505 [ hep - th ] ] . c. adam , t. romanczukiewicz , j. sanchez - guillen and a. wereszczynski , `` investigation of restricted baby skyrme models , '' phys . d * 81 * ( 2010 ) 085007 [ arxiv:1002.0851 [ hep - th ] ] . n. seiberg and e. witten , `` electric - magnetic duality , monopole condensation , and confinement in n=2 supersymmetric yang - mills theory , '' nucl . b * 426 * , 19 ( 1994 ) [ erratum - ibid . b * 430 * , 485 ( 1994 ) ] [ hep - th/9407087 ] , `` monopoles , duality and chiral symmetry breaking in n=2 supersymmetric qcd , '' nucl . phys . b * 431 * , 484 ( 1994 ) [ hep - th/9408099 ] . d. nemeschansky and r. rohm , `` anomaly constraints on supersymmetric effective lagrangians , '' nucl . phys . b * 249 * , 157 ( 1985 ) . s. j. gates , jr . , `` why auxiliary fields matter : the strange case of the 4-d , n=1 supersymmetric qcd effective action , '' phys . b * 365 * , 132 ( 1996 ) [ hep - th/9508153 ] , `` why auxiliary fields matter : the strange case of the 4-d , n=1 supersymmetric qcd effective action . 2 . , '' b * 485 * , 145 ( 1997 ) [ hep - th/9606109 ] . m. nitta , `` a note on supersymmetric wzw term in four dimensions , '' mod . a * 15 * , 2327 ( 2000 ) [ hep - th/0101166 ] . s. j. gates , jr . and s. m. kuzenko , `` the cnm hypermultiplet nexus , '' nucl . b * 543 * , 122 ( 1999 ) [ hep - th/9810137 ] , `` 4-d , n=2 supersymmetric off - shell sigma models on the cotangent bundles of kahler manifolds , '' fortsch . phys . * 48 * , 115 ( 2000 ) [ hep - th/9903013 ] . s. j. gates , jr . , m. t. grisaru , m. e. knutt and s. penati , `` the superspace wznw action for 4-d , n=1 supersymmetric qcd , '' phys . b * 503 * , 349 ( 2001 ) [ hep - ph/0012301 ] ; s. j. gates , jr . , m. t. grisaru , m. e. knutt , s. penati and h. suzuki , `` supersymmetric gauge anomaly with general homotopic paths , '' nucl . b * 596 * , 315 ( 2001 ) [ hep - th/0009192 ] ; s. j. gates , jr . , m. t. grisaru and s. penati , `` holomorphy , minimal homotopy and the 4-d , n=1 supersymmetric bardeen - gross - jackiw anomaly , '' phys . b * 481 * , 397 ( 2000 ) [ hep - th/0002045 ] . a. t. banin , i. l. buchbinder and n. g. pletnev , `` on quantum properties of the four - dimensional generic chiral superfield model , '' phys . d * 74 * , 045010 ( 2006 ) [ hep - th/0606242 ] . c. adam , j. m. queiruga , j. sanchez - guillen and a. wereszczynski , `` extended supersymmetry and bps solutions in baby skyrme models , '' jhep * 1305 * ( 2013 ) 108 [ arxiv:1304.0774 [ hep - th ] ] . j. khoury , j. -l . lehners and b. ovrut , `` supersymmetric p(x,@xmath170 ) and the ghost condensate , '' phys . rev . d * 83 * ( 2011 ) 125031 [ arxiv:1012.3748 [ hep - th ] ] . s. sasaki , m. yamaguchi and d. yokoyama , `` supersymmetric dbi inflation , '' phys . b * 718 * ( 2012 ) 1 [ arxiv:1205.1353 [ hep - th ] ] . m. koehn , j. -l . lehners and b. a. ovrut , `` higher - derivative chiral superfield actions coupled to n=1 supergravity , '' phys . rev . d * 86 * ( 2012 ) 085019 [ arxiv:1207.3798 [ hep - th ] ] . f. farakos and a. kehagias , `` emerging potentials in higher - derivative gauged chiral models coupled to n=1 supergravity , '' jhep * 1211 * ( 2012 ) 077 [ arxiv:1207.4767 [ hep - th ] ] . m. rocek and a. a. tseytlin , `` partial breaking of global d = 4 supersymmetry , constrained superfields , and three - brane actions , '' phys . rev . d * 59 * ( 1999 ) 106001 [ hep - th/9811232 ] . j. khoury , j. -l . lehners and b. a. ovrut , `` supersymmetric galileons , '' phys . d * 84 * ( 2011 ) 043521 [ arxiv:1103.0003 [ hep - th ] ] . m. koehn , j. -l . lehners and b. ovrut , `` ghost condensate in @xmath171 supergravity , '' phys . d * 87 * ( 2013 ) 6 , 065022 [ arxiv:1212.2185 [ hep - th ] ] . i. l. buchbinder , s. kuzenko and z. yarevskaya , `` supersymmetric effective potential : superfield approach , '' nucl . b * 411 * , 665 ( 1994 ) . s. m. kuzenko and s. j. tyler , `` the one - loop effective potential of the wess - zumino model revisited , '' jhep * 1409 * ( 2014 ) 135 [ arxiv:1407.5270 [ hep - th ] ] . e. a. bergshoeff , r. i. nepomechie and h. j. schnitzer , `` supersymmetric skyrmions in four - dimensions , '' nucl . b * 249 * ( 1985 ) 93 . l. freyhult , `` the supersymmetric extension of the faddeev model , '' nucl . b * 681 * ( 2004 ) 65 [ hep - th/0310261 ] . l. d. faddeev . princeton preprint ias-75-qs70 ; l. d. faddeev and a. j. niemi , `` knots and particles , '' nature * 387 * , 58 ( 1997 ) [ hep - th/9610193 ] . c. adam , j. m. queiruga , j. sanchez - guillen and a. wereszczynski , `` n=1 supersymmetric extension of the baby skyrme model , '' phys . rev . d * 84 * , 025008 ( 2011 ) [ arxiv:1105.1168 [ hep - th ] ] . m. nitta and s. sasaki , `` higher derivative corrections to manifestly supersymmetric nonlinear realizations , '' phys . d * 90 * , 105002 ( 2014 ) [ arxiv:1408.4210 [ hep - th ] ] . e. witten and d. i. olive , `` supersymmetry algebras that include topological charges , '' phys . b * 78 * , 97 ( 1978 ) . g. r. dvali and m. a. shifman , `` domain walls in strongly coupled theories , '' phys . b * 396 * , 64 ( 1997 ) [ erratum - ibid . b * 407 * , 452 ( 1997 ) ] [ hep - th/9612128 ] . g. w. gibbons and p. k. townsend , `` a bogomolny equation for intersecting domain walls , '' phys . * 83 * ( 1999 ) 1727 [ hep - th/9905196 ] ; s. m. carroll , s. hellerman and m. trodden , `` domain wall junctions are 1/4 - bps states , '' phys . d * 61 * ( 2000 ) 065001 [ hep - th/9905217 ] ; a. gorsky and m. a. shifman , `` more on the tensorial central charges in n=1 supersymmetric gauge theories ( bps wall junctions and strings ) , '' phys . d * 61 * , 085001 ( 2000 ) [ hep - th/9909015 ] . h. oda , k. ito , m. naganuma and n. sakai , `` an exact solution of bps domain wall junction , '' phys . b * 471 * , 140 ( 1999 ) [ hep - th/9910095 ] ; k. ito , m. naganuma , h. oda and n. sakai , `` nonnormalizable zero modes on bps junctions , '' nucl . b * 586 * , 231 ( 2000 ) [ hep - th/0004188 ] . m. naganuma , m. nitta and n. sakai , `` bps walls and junctions in susy nonlinear sigma models , '' phys . d * 65 * , 045016 ( 2002 ) [ hep - th/0108179 ] . m. eto , y. isozumi , m. nitta , k. ohashi and n. sakai , `` webs of walls , '' phys . d * 72 * , 085004 ( 2005 ) [ hep - th/0506135 ] , `` non - abelian webs of walls , '' phys . b * 632 * , 384 ( 2006 ) [ hep - th/0508241 ] ; m. eto , t. fujimori , t. nagashima , m. nitta , k. ohashi and n. sakai , `` dynamics of domain wall networks , '' phys . d * 76 * , 125025 ( 2007 ) [ arxiv:0707.3267 [ hep - th ] ] . m. shifman and a. yung , `` supersymmetric solitons and how they help us understand non - abelian gauge theories , '' rev . phys . * 79 * , 1139 ( 2007 ) [ hep - th/0703267 ] , `` supersymmetric solitons , '' cambridge , uk : cambridge univ . ( 2009 ) 259 p m. eto , y. isozumi , m. nitta , k. ohashi and n. sakai , `` solitons in the higgs phase : the moduli matrix approach , '' j. phys . a * 39 * , r315 ( 2006 ) [ hep - th/0602170 ] . m. eto , y. isozumi , m. nitta and k. ohashi , `` 1/2 , 1/4 and 1/8 bps equations in susy yang - mills - higgs systems : field theoretical brane configurations , '' nucl . b * 752 * , 140 ( 2006 ) [ hep - th/0506257 ] . m. eto , t. fujimori , m. nitta , k. ohashi and n. sakai , `` higher derivative corrections to non - abelian vortex effective theory , '' prog . phys . * 128 * , 67 ( 2012 ) [ arxiv:1204.0773 [ hep - th ] ] . c. adam , j. m. queiruga , j. sanchez - guillen and a. wereszczynski , `` bps bounds in supersymmetric extensions of k field theories , '' phys . d * 86 * ( 2012 ) 105009 [ arxiv:1209.6060 [ hep - th ] ] . c. adam , j. m. queiruga , j. sanchez - guillen and a. wereszczynski , `` supersymmetric k field theories and defect structures , '' phys . d * 84 * ( 2011 ) 065032 [ arxiv:1107.4370 [ hep - th ] ] . a. hanany and d. tong , `` vortices , instantons and branes , '' jhep * 0307 * , 037 ( 2003 ) [ hep - th/0306150 ] ; r. auzzi , s. bolognesi , j. evslin , k. konishi and a. yung , `` nonabelian superconductors : vortices and confinement in n=2 sqcd , '' nucl . b * 673 * , 187 ( 2003 ) [ hep - th/0307287 ] ; m. eto , y. isozumi , m. nitta , k. ohashi and n. sakai , `` moduli space of non - abelian vortices , '' phys . lett . * 96 * , 161601 ( 2006 ) [ hep - th/0511088 ] ; m. eto , k. konishi , g. marmorini , m. nitta , k. ohashi , w. vinci and n. yokoi , `` non - abelian vortices of higher winding numbers , '' phys . d * 74 * , 065021 ( 2006 ) [ hep - th/0607070 ] . m. eto , y. isozumi , m. nitta , k. ohashi and n. sakai , `` instantons in the higgs phase , '' phys . d * 72 * , 025011 ( 2005 ) [ hep - th/0412048 ] ; t. fujimori , m. nitta , k. ohta , n. sakai and m. yamazaki , `` intersecting solitons , amoeba and tropical geometry , '' phys . d * 78 * , 105004 ( 2008 ) [ arxiv:0805.1194 [ hep - th ] ] . i. antoniadis , e. dudas and d. m. ghilencea , `` supersymmetric models with higher dimensional operators , '' jhep * 0803 * ( 2008 ) 045 [ arxiv:0708.0383 [ hep - th ] ] . l. a. ferreira , `` exact vortex solutions in an extended skyrme - faddeev model , '' jhep * 0905 * , 001 ( 2009 ) [ arxiv:0809.4303 [ hep - th ] ] ; l. a. ferreira and p. klimas , `` exact vortex solutions in a @xmath172 skyrme - faddeev type model , '' jhep * 1010 * , 008 ( 2010 ) [ arxiv:1007.1667 [ hep - th ] ] ; l. a. ferreira , n. sawado and k. toda , `` static hopfions in the extended skyrme - faddeev model , '' jhep * 0911 * , 124 ( 2009 ) [ arxiv:0908.3672 [ hep - th ] ] ; l. a. ferreira , p. klimas and w. j. zakrzewski , `` some ( 3 + 1 ) dimensional vortex solutions of the cpn model , '' phys . d * 83 * , 105018 ( 2011 ) [ arxiv:1103.0559 [ hep - th ] ] , `` properties of some ( 3 + 1 ) dimensional vortex solutions of the @xmath172 model , '' phys . rev . d * 84 * , 085022 ( 2011 ) [ arxiv:1108.4401 [ hep - th ] ] , `` some properties of ( 3 + 1 ) dimensional vortex solutions in the extended @xmath172 skyrme - faddeev model , '' jhep * 1112 * , 098 ( 2011 ) [ arxiv:1111.2338 [ hep - th ] ] ; l. a. ferreira , j. jaykka , n. sawado and k. toda , `` vortices in the extended skyrme - faddeev model , '' phys . d * 85 * , 105006 ( 2012 ) [ arxiv:1112.1085 [ hep - th ] ] . a. m. polyakov and a. a. belavin , `` metastable states of two - dimensional isotropic ferromagnets , '' jetp lett . * 22 * , 245 ( 1975 ) [ pisma zh . * 22 * , 503 ( 1975 ) ] . m. nitta , `` josephson vortices and the atiyah - manton construction , '' phys . d * 86 * , 125004 ( 2012 ) [ arxiv:1207.6958 [ hep - th ] ] ; m. kobayashi and m. nitta , `` jewels on a wall ring , '' phys . rev . d * 87 * , 085003 ( 2013 ) [ arxiv:1302.0989 [ hep - th ] ] ; p. jennings and p. sutcliffe , `` the dynamics of domain wall skyrmions , '' j. phys . a * 46 * , 465401 ( 2013 ) [ arxiv:1305.2869 [ hep - th ] ] . m. nitta , `` correspondence between skyrmions in 2 + 1 and 3 + 1 dimensions , '' phys . rev . d * 87 * , 025013 ( 2013 ) [ arxiv:1210.2233 [ hep - th ] ] , `` matryoshka skyrmions , '' nucl . b * 872 * , 62 ( 2013 ) [ arxiv:1211.4916 [ hep - th ] ] ; s. b. gudnason and m. nitta , `` domain wall skyrmions , '' phys . d * 89 * , 085022 ( 2014 ) [ arxiv:1403.1245 [ hep - th ] ] , `` baryonic sphere : a spherical domain wall carrying baryon number , '' phys . rev . d * 89 * , 025012 ( 2014 ) [ arxiv:1311.4454 [ hep - th ] ] . j. p. gauntlett , r. portugues , d. tong and p. k. townsend , `` d - brane solitons in supersymmetric sigma models , '' phys . rev . d * 63 * , 085002 ( 2001 ) [ hep - th/0008221 ] ; m. shifman and a. yung , `` domain walls and flux tubes in n=2 sqcd : d - brane prototypes , '' phys . rev . d * 67 * , 125007 ( 2003 ) [ hep - th/0212293 ] . y. isozumi , m. nitta , k. ohashi and n. sakai , `` all exact solutions of a 1/4 bogomolnyi - prasad - sommerfield equation , '' phys . d * 71 * , 065018 ( 2005 ) [ hep - th/0405129 ] ; m. eto , t. fujimori , t. nagashima , m. nitta , k. ohashi and n. sakai , `` dynamics of strings between walls , '' phys . d * 79 * , 045015 ( 2009 ) [ arxiv:0810.3495 [ hep - th ] ] . m. eto , y. isozumi , m. nitta , k. ohashi and n. sakai , `` manifestly supersymmetric effective lagrangians on bps solitons , '' phys . d * 73 * , 125008 ( 2006 ) [ hep - th/0602289 ] . m. eto , m. nitta , k. ohashi and d. tong , `` skyrmions from instantons inside domain walls , '' phys . lett . * 95 * , 252003 ( 2005 ) [ hep - th/0508130 ] . m. shifman and a. yung , `` localization of nonabelian gauge fields on domain walls at weak coupling ( d - brane prototypes ii ) , '' phys . d * 70 * , 025013 ( 2004 ) [ hep - th/0312257 ] ; m. eto , t. fujimori , m. nitta , k. ohashi and n. sakai , `` domain walls with non - abelian clouds , '' phys . d * 77 * , 125008 ( 2008 ) [ arxiv:0802.3135 [ hep - th ] ] . j. wess and j. bagger , `` supersymmetry and supergravity , '' princeton , usa : univ . ( 1992 ) 259 p

the rich phenomenology of collective neutrino oscillations has been studied only in one - dimensional or spherically symmetric systems . motivated by the non - spherical example of coalescing neutron stars , presumably the central engines of short gamma - ray bursts , we use the liouville equation to formulate the problem for general source geometries . assuming the neutrino ensemble displays self - maintained coherence , the problem once more becomes effectively one - dimensional along the streamlines of the overall neutrino flux . this approach for the first time provides a formal definition of the `` single - angle approximation '' frequently used for supernova neutrinos and allows for a natural generalization to non - spherical geometries . we study the explicit example of a disk - shaped source as a proxy for coalescing neutron stars .

0805.3300

flavor transformations caused by neutrino mixing depend on the matter background and on the neutrino fluxes themselves : neutrino - neutrino interactions provide a nonlinear term in the equations of motion @xcite that gives rise to collective flavor transformation phenomena . the neutrino density needs to be so large that a typical neutrino - neutrino interaction energy @xmath0 is comparable to the vacuum oscillation frequency @xmath1 . only recently has it been fully appreciated that this condition is sufficient even if a dense background of ordinary matter provides a much larger interaction energy so that naively neutrino - neutrino interactions would seem negligible . following this crucial insight , nonlinear oscillation phenomena in the supernova ( sn ) context have been studied over the past two years in a long series of papers . one striking effect is `` self - maintained coherence '' @xcite . different neutrino modes have different vacuum oscillation frequencies @xmath1 , but with strong neutrino - neutrino interactions they `` stick together '' and oscillate as a single mode characterized by the `` synchronized oscillation frequency '' @xmath2 . this can lead to all modes going through an msw resonance together , the `` collective msw - like transition '' or `` synchronized msw effect '' @xcite . more interesting still are collective phenomena driven by the decrease of the neutrino flux with distance from the source . the adiabatic transition from a dense to a dilute neutrino gas produces step - like spectral features where the spectrum sharply splits into parts of different flavor transformation , so - called `` step - wise spectral swapping '' or `` spectral splits '' . spectral splits can result from a preceding collective msw effect ( `` msw prepared spectral split '' ) or from neutrino - neutrino interactions alone . the latter case depends on an unusual form of non - equilibrium among neutrino flavors where one has an excess of flavor pairs , say @xmath3 , over the other flavors . for neutrinos streaming off a sn core one indeed expects a hierarchy of number fluxes @xmath4 . therefore , one can have collective transformations of the form @xmath5 , where @xmath6 stands for some suitable combination of @xmath0 and @xmath7 neutrinos . these `` collective pair transformations '' do not violate any conservation law and thus can be catalyzed even by a very small mixing angle . @xmath8 can completely swap with @xmath9 whereas the larger @xmath10 converts only to the extent allowed by flavor - lepton conservation , but in a step - like spectral form . both the complete conversion of @xmath8 and the split in the @xmath11 spectrum provide signatures for the inverted neutrino hierarchy even for an extremely small 13-mixing angle @xcite . for non - isotropic enviroments , multi - angle effects may play an important role . the term `` multi - angle effects '' actually refers to two different issues . one is that the weak interaction potential between two relativistic particles is proportional to ( @xmath12 ) , where @xmath13 is their relative angle of propagation . one usually considers an isotropic background of ordinary matter so that @xmath14 averages to zero . the same is true in an isotropic neutrino gas for the neutrino - neutrino term . the second issue is the `` multi - angle instability '' . neutrinos arriving from different points on the source belong to different angular modes , which may decohere kinematically in flavor space @xcite . this effect can be self - induced in the sense that a very small initial anisotropy is enough to trigger an exponential runaway , for example in a gas consisting of equal densities of neutrinos and antineutrinos @xcite . systems consisting of very few angular modes can show a two - stream or multi - stream instability @xcite . on the other hand , numerical studies show that systems consisting of many angular modes and with a sufficient neutrino - antineutrino asymmetry do not show a multi - angle instability but rather show self - maintained coherence among different angular modes @xcite . the sn neutrino flux parameters seem to be such that the multi - angle instability plays no role in practice . based on this assumption , most of the sn studies have used the `` single - angle approximation '' , where all angular modes are assumed to have the same behavior . multi - angle effects related to the ( @xmath12 ) structure of the neutrino - neutrino term are unavoidable for an extended source radiating neutrinos into space because the emitted neutrino flux can not form an isotropic gas . however , these effects also occur , and are easier to study theoretically , in a homogeneous system evolving in time that has a non - isotropic angular distribution of neutrinos . in practice , however , one usually deals with stationary systems where one asks for the spatial variation of a neutrino ensemble as a function of distance from the source . even if the neutrino - neutrino interaction were isotropic , we still would have geometric multi - angle effects because neutrinos reaching a certain point from an extended source have traveled on different trajectories . even the simple case of an infinite radiating plane is not trivial . here the direction perpendicular to the plane is the only direction in which the overall neutrino ensemble can show any spatial variation . even if all neutrinos have the same energy and thus oscillate with the same frequency along their trajectories , the projection on the direction perpendicular to the plane yields different effective oscillation frequencies and thus kinematical decoherence . neutrino - neutrino effects can synchronize different angular modes so that a sufficiently dense neutrino gas will not show this form of multi - angle decoherence . on the contrary , all angular modes will vary with the same oscillation length as a function of distance from the plane . a similar description applies to a spherical source where one asks for the variation of all angular modes along the radial direction . the single - angle treatment of sn neutrino oscillations amounts to the assumption of self - maintained coherence among angular modes , although it has never been explicitly expressed in this form . some authors assumed that all angular modes oscillate as the radial one @xcite . however , in this case the neutrino - neutrino interaction vanishes because of the ( @xmath12 ) factor , so it was necessary to assume a certain average of the neutrino - neutrino interaction strength . other authors represented all angular modes by a single angular mode radiated at @xmath15 relative to the radial direction and then used a neutrino - neutrino interaction strength consistent with this assumption @xcite . this implementation of the single - angle approximation has the advantage that one can use the same numerical code as for multi - angle simulations , simply restricting oneself to a single angular bin . in yet other cases the system was modeled as a homogeneous and isotropic gas that evolves in time , assuming a time variation of the neutrino density that mimics the radial variation in the spherical case . one of our goals is to show that the single - angle treatment can be formulated self - consistently . the assumption that all angular modes evolve the same in flavor space provides a unique concept of what is meant by `` single - angle behavior . '' this is straightforward in the systems described so far where symmetry dictates that the spatial variation is only along a certain direction , effectively reducing the problem to one dimension . we are really motivated , however , by more general geometries where no special direction is singled out by symmetry . in particular , we are interested in the case of coalescing neutron stars that may form the inner engines of short gamma - ray bursts @xcite . the accretion torus or disk formed during neutron star coalescence is a neutrino source comparable to a sn core . however , the torus is less dense and not efficient at producing @xmath16 and @xmath17 . therefore , the torus is a source for a dominant @xmath3 pair flux which is thought to produce an @xmath18 pair plasma , thus powering short gamma - ray bursts . the annihilation cross section for @xmath19 is much larger than that for @xmath20 , so the neutrino flavor composition strongly influences the number of @xmath21 pairs produced . therefore , one may ask if collective pair conversions occur in this environment close enough to the source to modify the energy transfer to the @xmath21 plasma , and hence affect the strength of the gamma ray burst . for coalescing neutron stars one expects a flux hierarchy @xmath22 , which differs from the sn case because the matter leptonizes when neutrons convert to protons , in contrast to the deleptonization of a sn core . the asymmetry between @xmath10 and @xmath8 could be enough to prevent multi - angle decoherence so that similar collective effects as in the sn environment are conceivable . however , even granting this assumption , it is not straightforward how to implement something like a single - angle approximation in this context because it is not obvious how one should picture self - maintained coherence . the purpose of our paper is to formulate the meaning of self - maintained coherence for general source geometries and study its implications . we find that the flavor variation reduces to a quasi one - dimensional problem along the streamlines of the total neutrino flux . the main difference between the general case and the radiating plane or sphere is that the streamlines are typically curved , at least close to the source , so that self - maintained coherence applies to flavor oscillations along these curved streamlines . we begin in sec . [ sec : eoms ] with the general equations of motion for the neutrino matrices in flavor space . in sec . [ sec : synchronized oscillations ] we formulate the collective equations for neutrinos only ( no antineutrinos ) and consider only synchronized oscillations . sec . [ sec : streamlines ] shows the existence of streamlines and gives a prescription for calculating the flavor evolution along them . in sec . [ sec : geometries ] we solve the problem for several geometries . in sec . [ sec : bipolar ] we study the generalization to a mixed system of neutrinos and antineutrinos where collective pair transformations are possible . in sec . [ sec : neutron stars ] we consider an explicit example for a disk source with parameters inspired by numerical simulations of coalescing neutron stars . we summarize our conclusions in sec . [ sec : conclusions ] . a homogeneous ensemble of mixed neutrinos can be described by `` matrices of density '' @xmath23 , which really are matrices of occupation numbers , for each momentum mode @xmath24 . if @xmath25 and @xmath26 are the creation and annihilation operators of a neutrino in the mass eigenstate @xmath27 of momentum @xmath24 , we have @xmath28 so that the diagonal entries of @xmath23 are the usual occupation numbers ( expectation values of number operators ) , whereas the off - diagonal elements encode the phase relations that allow one to follow flavor oscillations . such a description assumes that higher - order correlations beyond field bilinears play no role , probably a good approximation for neutrinos produced from essentially thermal sources such as the early - universe plasma or a sn core . antineutrinos are described in an analogous way by @xmath29 . note that we always use overbars to characterize antiparticle quantities . the order of flavor indices was deliberately interchanged on the r.h.s . so that the matrices @xmath23 and @xmath30 transform in the same way in flavor space @xcite . in this way one can , for example , write the overall neutrino current in the simple form @xmath31 where here and henceforth @xmath32 stands for @xmath33 . while some authors prefer the seemingly more intuitive equal order of flavor indices for neutrinos and antineutrinos , in that convention the current would be @xmath34 and generally the equations will involve both the matrices of density and their complex conjugates . ( in a truly field - theoretic derivation there is no ambiguity about the relative structure of neutrino and antineutrino matrices . an analogous example for matrices of occupation numbers is provided by the kinetic treatment of the quark gluon plasma where the quark distribution functions are @xmath35 matrices in color space @xcite . the matrices for quarks and the ones for antiquarks transform equally under a color gauge transformation . ) the matrices @xmath23 and @xmath36 depend on time . their evolution is governed by the boltzmann collision equation @xmath37 + \dot\varrho_{\bf p}\big|_{\rm coll}\ ] ] and a similar equation for @xmath36 . the collision term will be of no further concern in our paper because we only study freely streaming neutrinos . further , @xmath38 $ ] is a commutator and @xmath39 is the matrix of oscillation frequencies . in vacuum we have @xmath40 with @xmath41 the neutrino mass matrix . in general @xmath42 also depends on the background medium and notably on the presence of other neutrinos . here and henceforth the ultrarelativistic limit for neutrinos is assumed . in other words , it is assumed that the difference between the neutrino energy @xmath43 and momentum @xmath44 is irrelevant except for oscillations : the only relevant difference between energy and momentum is captured by the matrix of oscillation frequencies . up to this point we have considered a homogeneous ensemble evolving in time , largely relevant for neutrino oscillations in the early universe for which this formalism was originally developed . however , for a realistic representation of radiating objects such as supernovae or coalescing neutron stars we need to include spatial transport phenomena . to this end one introduces space - dependent occupation numbers or wigner functions @xmath45 . the quantum - mechanical uncertainty between location and momentum implies that this construction is only valid in the limit where spatial variations of the ensemble are weak on the microscopic length scales defined by the particles typical compton wavelengths . the left hand side of the boltzmann collision equation now turns into the usual liouville operator @xcite @xmath46 the transition to the liouville operator may seem obvious , but making it conceptually precise requires a long argument @xcite . the first term represents an explicit time dependence , the second a drift term caused by the particles free streaming , and the third the effect of external macroscopic forces , for example gravitational deflection . henceforth we shall neglect external forces and only retain the drift term . in our ultrarelativistic approximation the modulus of @xmath47 is the speed of light . of course , for propagation over very large distances this may be a bad approximation when time - of - flight effects play a role . in this case the drift term would have a more complicated structure because the velocity is then also a nontrivial matrix in flavor space @xcite . we shall focus on stationary problems . even neutrino emission from a sn or coalescing neutron stars fall in this category unless there are fast variations of the source . in other words , we shall ignore a possible explicit time dependence of @xmath45 so that finally we arrive at @xmath48\,,\ ] ] where the matrix of oscillation frequencies in general also depends on space because of the influence of matter and other neutrinos . equation ( [ eq : eom1 ] ) could have been guessed starting from the usual treatment of neutrino oscillations . the neutrino density matrix describing a stationary beam in the @xmath49direction follows @xmath50 $ ] . moreover , in our ultrarelativistic approximation @xmath51 and @xmath52 . equation ( [ eq : eom1 ] ) then simply amounts to a collection of many beams in different directions originating from different source points . we finally spell out eq . ( [ eq : eom1 ] ) explicitly under the assumption that the only nontrivial ingredients consist of neutrino - neutrino refractive effects , @xmath53 + \sqrt{2}\,g_{\rm f } \int { { \rm d}}{\bf q}\,(1-{\bf v}_{\bf q}\cdot{\bf v}_{\bf p } ) \bigl[(\varrho_{\bf q , x}-\bar\varrho_{\bf q , x } ) , \varrho_{\bf p , x}\bigr]\ , , \nonumber \\ * { { \rm i}}\,{\bf v}_{\bf p}\cdot{\bf\nabla}_{\bf x}\,\bar\varrho_{\bf p , x } & = & -\frac{1}{2p}\,\bigl[{\sf m}^2,\bar\varrho_{\bf p , x}\bigr ] + \sqrt{2}\,g_{\rm f } \int { { \rm d}}{\bf q}\,(1-{\bf v}_{\bf q}\cdot{\bf v}_{\bf p } ) \bigl[(\varrho_{\bf q , x}-\bar\varrho_{\bf q , x } ) , \bar\varrho_{\bf p , x}\bigr]\ , . \end{aligned}\ ] ] the only difference between the neutrino and antineutrino equations is the sign of the vacuum oscillation term . this structure is a consequence of defining the antineutrino matrices with reversed flavor indices . the nonlinear equations of motion ( eoms ) eq . ( [ eq : eom2 ] ) simplify considerably if self - maintained coherence occurs in a dense neutrino gas and all modes can be assumed to evolve in the same way . for the moment we restrict ourselves to a source radiating only neutrinos ( no antineutrinos ) . what we mean by self - maintained coherence in this case is that at a given location all neutrino modes are aligned with each other , assuming they were aligned at the source . that is , @xmath54 here , @xmath55 is a scalar occupation number , summed over all flavors , while for @xmath56 flavors @xmath57 is a @xmath58 matrix normalized as @xmath59 . we also define the local neutrino density and flux as @xmath60 where q _ * x * is the momentum average of a quantity @xmath61 at location @xmath62 with respect to the distribution function @xmath63 . for future convenience we also define _ * x * = , a unit vector along the direction of flux @xmath64 , or equivalently , along the average velocity @xmath65 . in this language , the eom eq . ( [ eq : eom2 ] ) becomes @xmath66 \,f_{\bf p , x}\,,\ ] ] where the nonlinear term has disappeared because it involves the commutator @xmath67=0 $ ] . the only impact of the neutrino - neutrino term is to glue the different modes together , but afterwards it no longer appears in the equation . the evolution now factorizes into the free streaming of the overall neutrino flux and the oscillation of the common flavor matrix @xmath68 . the trace of eq . ( [ eq : eom4 ] ) provides @xmath69 allowing us to determine the overall neutrino density @xmath70 and flux @xmath64 at any point , assuming it is given at the source surface . with @xmath71 , integrating eq . ( [ eq : vpdelf ] ) over all modes yields @xmath72 expressing the absence of neutrino sources or sinks . the eom for @xmath57 is found by integrating eq . ( [ eq : eom4 ] ) over all modes , @xmath73 \,\int{{\rm d}}{\bf p}\,\frac{f_{\bf p , x}}{p}\,.\ ] ] the l.h.s . expands as @xmath74 where the first term disappears because of eq . ( [ eq : eom4a ] ) so that @xmath75 \,\int{{\rm d}}{\bf p}\,\frac{f_{\bf p , x}}{p}\,.\ ] ] dividing both sides by @xmath76 , we get @xmath77\ , , \ ] ] where @xmath78 equation ( [ eq : eom4cc ] ) resembles the eom in the single - angle approximation , with @xmath79 as the synchronized matrix of oscillation frequencies . note that @xmath80 is independent of location if the energy spectrum of neutrinos is assumed to be the same everywhere . equation ( [ eq : eom4cc ] ) is a partial differential equation ( pde ) for the matrix @xmath57 . it can be reduced to a set of ordinary differential equations ( odes ) @xmath81\ , , \label{ode2}\end{aligned}\ ] ] where @xmath82 is a parameter along the `` characteristic line , '' or `` streamline , '' defined by eq . ( [ ode1 ] ) . since @xmath83 is unique at each location , the streamlines do not intersect each other . along each streamline , the differential equation [ eq . ( [ ode2 ] ) ] for the matrix @xmath57 is a set of coupled , but linear odes and hence can be solved easily and uniquely , given the boundary conditions . as an illustration , we now specialize to a two - flavor system , where the coupled odes in eq . ( [ ode2 ] ) reduce to a single ode and may be interpreted as the evolution of a single phase . as uaual , we express a hermitian @xmath84 matrix @xmath85 in terms of a polarization vector @xmath86 by virtue of @xmath87 with @xmath88 the vector of pauli matrices . indicates a matrix in flavor space , a letter with an arrow such as @xmath89 indicates a vector in flavor space , and a bold - faced letter such as @xmath90 a vector in coordinate space . ] the equation for synchronized oscillations then turns into the usual precession formula @xmath91 where @xmath92 is the polarization vector corresponding to @xmath93 and @xmath94 the one corresponding to @xmath95 . here @xmath96 is a unit vector by definition , which is determined by @xmath97 and hence is independent of location . the effective synchronized oscillation frequency is _ * x * = = , [ eq : omega0 ] where @xmath98 is the synchronized oscillation frequency far away from the source , where the details of the source geometry do not matter . all the effects of the source geometry are thus captured by the quantity @xmath99 . the problem becomes even simpler if we observe that the magnitude of @xmath92 and its projection on @xmath96 are conserved . we can then write @xmath92 in terms of its components along and perpendicular to @xmath96 as _ * x * = ( p_a_*x * , p_a_*x * , p _ ) . [ pvec - param ] equation ( [ eq : eom4f ] ) then reduces to a scalar equation @xmath100 in other words , @xmath92 is fully specified by a real scalar number @xmath101 that tells us the phase of the revolution of the polarization vector around @xmath96 . equations ( [ ode1 ] ) and ( [ ode2 ] ) in this case simplify to @xmath102 as before , the streamlines are given by eq . ( [ ode1:2flav ] ) and the evolution is obtained by solving the single ode in eq . ( [ ode2:2flav ] ) . this has been possible because of a simple parametrization of @xmath92 in eq . ( [ pvec - param ] ) in two flavors . in the case of three flavors , one can express a hermitian @xmath103 matrix in terms of the gell - mann matrices @xcite . equation ( [ precession ] ) then retains the same form , except that the `` cross product '' is now understood to be @xmath104 with @xmath105 the structure constants for su(3 ) . the odes in eq . ( [ ode2 ] ) are coupled , but linear , and can be solved along the streamlines . the general strategy for calculating the neutrino flavor evolution is therefore the following . @xmath106 find the streamlines for the problem by solving eq . ( [ ode1 ] ) with the given neutrino source . @xmath107 solve for the phase @xmath101 ( or for the matrix @xmath57 ) by solving the odes [ eq . ( [ ode2 ] ) ] along each streamline . this leads to the determination of @xmath45 everywhere . an important consequence of the existence of streamlines is that the evolution along any streamline is independent of the other streamlines . moreover , each streamline is determined solely by the flux @xmath64 along it . thus the neutrino flavor content at any point in space is determined simply by identifying the streamline that passes through that point and evolving the phase @xmath101 along the streamline . this effectively reduces any multi - dimensional problem to one dimension . we now calculate the flavor evolution of neutrinos in several physical situations , ranging from the trivial to the complicated . the crucial quantity to calculate in each scenario is @xmath108 , which captures the effect of the source geometry through eq . ( [ eq : omega0 ] ) . for a homogeneous plane that radiates neutrinos isotropically , the streamlines are straight lines normal to the surface . therefore , the oscillation phase @xmath109 depends only on the distance @xmath49 from the plane and the surfaces of equal phase are planes parallel to the radiating surface . the average velocity of all neutrinos away from the plane is half the speed of light , @xmath110 everywhere . therefore , @xmath111 represents the flavor oscillations along the direction perpendicular to the plane . this gives @xmath112 , where @xmath113 may be chosen to be an arbitrary real number . this , along with the initial conditions , completely determines the flavor state of the neutrinos at any point in space . for a spherical source , symmetry dictates that the streamlines are straight lines following the radial distance from the center and that the surfaces of equal phase are spherical shells around the source . therefore spherical symmetry dictates that the average velocity @xmath114 is along the radius . all that remains to be determined is @xmath115 , the average radial velocity of all neutrinos at distance @xmath116 . to this end we note that all modes that are radiated at the neutrino sphere with the same zenith angle @xmath117 relative to the radial direction behave identically . their radial velocity at the neutrino sphere is @xmath118 , so it is useful to classify all modes by their @xmath119 . simple geometry tells us that @xmath120 , where @xmath121 is the zenith angle of a given neutrino trajectory relative to the radial direction at a distance @xmath116 . therefore , the radial velocity at distance @xmath116 of a mode that has radial velocity @xmath119 at the neutrino sphere is @xmath122 given the all - flavor phase - space density @xmath123 of a mode @xmath119 at the neutrino sphere , we can determine its value @xmath124 at another distance from flux conservation . in spherical coordinates we have @xmath125=0 $ ] so that @xmath126 . if we assume isotropic emission at the neutrino sphere , the modes are uniformly distributed in the variable @xmath127 or @xmath128 , so that @xmath129 inserting the expression eq . ( [ eq : vr ] ) for @xmath130 and performing the integral we find @xmath131\,.\ ] ] at the neutrino sphere ( @xmath132 ) one finds @xmath133 , identical to the radiating plane , whereas at large distances @xmath134 all modes move essentially in the radial direction and @xmath135 . therefore , near the neutrino sphere , the neutrino flavor evolves along the radial direction with frequency @xmath136 , whereas at large distances it evolves with @xmath137 . this case is represented in fig . [ fig : sphere ] . the decrease of @xmath138 to its asymptotic value is surprisingly fast . as usual , the phase is given by @xmath139 , and this along with initial conditions specifies the flavour content of neutrinos all over space . perhaps the simplest example where the streamlines are curved is the case of a radiating circular disk . this model approximately mimics the accretion disk formed by two coalescing neutron stars . the disk radius is taken to be @xmath140 , setting the geometric scale for the system . we assume that the disk radiates homogeneously a flux @xmath141 throughout its surface and that at each point the emission is identical to a blackbody source . very close to the disk the physical situation is identical to an infinite radiating plane . the streamlines are perpendicular to the disk with a flux @xmath141 . viewed from a large distance , on the other hand , the flux must be directed away from what looks like a point - like source , so that the streamlines are radially outward from the source . since the source is assumed to behave like a `` blackbody '' , the neutrino flux coming from the source varies with @xmath142 where @xmath143 is the zenith angle relative to the disk axis . this is simply because we are seeing a blackbody surface at an angle so that the total flux is reduced by the projected size of the source . on the other hand , the total flux passing through a spherical surface at distance @xmath116 is conserved . therefore , at a distance @xmath134 we find @xmath144 therefore , we know the flux and its direction directly at the source and at a large distance . we calculate the flux @xmath64 numerically and find that the streamlines are well represented by the hyperbolae @xmath145 here we use coordinates @xmath6 along the disk radius and @xmath49 along the disk axis . we have also calculated the magnitude of flux @xmath76 , the magnitude of average velocity @xmath146 , and the effective neutrino - neutrino interaction strength @xmath147 defined later in eq . ( [ eq : mudefinition ] ) . we show contours of these quantities in fig . [ fig : streamlines ] . moreover , in fig . [ fig : asympt ] we show the behaviour of these quantities as a function of @xmath49 for @xmath148 and their asymptotic variation as a function of @xmath143 at a large distance from the disk . details of the numerical calculations are given in appendix [ app : stream ] . we now extend the previous considerations to the more interesting situation where both neutrinos and antineutrinos stream from the source . the main difference is that we now assume that neutrinos and antineutrinos each form two separate classes of modes that oscillate together , i.e. , we assume that the behavior of all neutrino modes is well represented by a single matrix @xmath57 and that of the antineutrino modes by a matrix @xmath149 such that @xmath150 and @xmath151 . integrating the eoms of eq . ( [ eq : eom2 ] ) over all modes we arrive at @xmath152n_{\bf x } + \sqrt{2}\,g_{\rm f}\left ( [ ( n_{\bf x}{\sf p}_{\bf x}-\bar n_{\bf x}\bar{\sf p}_{\bf x } ) , n_{\bf x } { \sf p}_{\bf x } ] -[({\bf f}_{\bf x}{\sf p}_{\bf x}-\bar{\bf f}_{\bf x}\bar{\sf p}_{\bf x } ) , { \bf f}_{\bf x } { \sf p}_{\bf x}]\right)\ , , \nonumber \\ { { \rm i}}\,\bar{\bf f}_{\bf x}\cdot{\bf\nabla}_{\bf x}\,\bar{\sf p}_{\bf x } & = & -\bigl[\bar{\sf \omega}_0,\bar{\sf p}_{\bf x}\bigr]\bar n_{\bf x } + \sqrt{2}\,g_{\rm f}\left ( [ ( n_{\bf x}{\sf p}_{\bf x}-\bar n_{\bf x}\bar{\sf p}_{\bf x } ) , \bar n_{\bf x } \bar{\sf p}_{\bf x } ] -[({\bf f}_{\bf x}{\sf p}_{\bf x}-\bar{\bf f}_{\bf x}\bar{\sf p}_{\bf x } ) , \bar{\bf f}_{\bf x } \bar{\sf p}_{\bf x}]\right)\ , . \end{aligned}\ ] ] if it were not for the nonlinear term , of course , we would have synchronized oscillations separately for neutrinos and antineutrinos where @xmath153 is the oscillation matrix averaged over the neutrino spectrum whereas @xmath154 is the one averaged over the antineutrino spectrum . in order to consider the simplest nontrivial case we assume further that the energy spectra for neutrinos and antineutrinos are the same as well as their angular emission characteristics . this implies @xmath155 . moreover , we assume that the total flux of antineutrinos is @xmath156 times the flux of neutrinos . it simplifies the equations if we absorb this global factor in the normalization of the matrix @xmath149 so that @xmath157 whereas @xmath158 . with these assumptions we have @xmath159 and @xmath160 . with @xmath161 we then find @xmath162 + \mu_{\bf x } [ ( { \sf p}_{\bf x}-\bar{\sf p}_{\bf x}),{\sf p}_{\bf x}]\ , , \nonumber \\ { { \rm i}}\ , \widehat{\bf f}_{\bf x}\cdot{\bf\nabla}_{\bf x}\,\bar{\sf p}_{\bf x } & = & -\bigl[{\sf \omega}_{\bf x},\bar{\sf p}_{\bf x}\bigr ] + \mu_{\bf x } [ ( { \sf p}_{\bf x}-\bar{\sf p}_{\bf x}),\bar{\sf p}_{\bf x}]\ , . \end{aligned}\ ] ] the oscillation matrix is @xmath163 whereas @xmath164 is the effective neutrino - neutrino interaction strength to be discussed below . if neutrinos and antineutrinos have the same streamlines , we find a unique extension of the idea of self - maintained coherence to a source with general geometry . therefore , one may expect that the solutions along given streamlines depend on the source geometry primarily through the variation of @xmath0 along the streamlines . in the sn context , the evolution of the neutrino ensemble driven by the decrease of @xmath0 tends to be essentially adiabatic so that the exact variation of @xmath0 along a streamline should not matter much for the overall picture . this suggests that we may be allowed to borrow the insights gained in the sn context directly to more general cases . even if neutrinos and antineutrinos do not follow exactly the same streamlines , if the lateral variation of the solution is smooth , one may speculate that the system finds some suitable average and that the final solution could still be similar to the simple case . in the limit of self - maintained coherence discussed in the previous section , we are led to define an effective neutrino - neutrino interaction strength @xmath165 in principle , the same coefficient would have appeared in the neutrino - only case , but there is was accompanied by a vanishing commutator so that it did not appear explicitly . the definition of @xmath147 is unique except for the normalization that depends on our choice @xmath166 and @xmath167 . in our normalization , @xmath147 has the interpretation of the average neutrino - neutrino interaction energy and corresponds to the quantity @xmath168 in the ordinary matter effect . equation ( [ eq : mudefinition ] ) implies some general features of the variation of @xmath0 at large distance @xmath116 from an extended but finite source . ( we always consider sources that are not point like or else there would be no neutrino - neutrino effects , but that are finite , in contrast to the infinite plane mentioned earlier , or else there would be no systematic decrease of @xmath0 with distance . ) the flux factor @xmath76 provides a trivial @xmath169 scaling from the geometric flux dilution , as shown in fig . [ fig : asympt ] . the `` collinearity factor '' @xmath170 arises from the @xmath12 structure of the neutrino - neutrino interaction . at a large distance all neutrinos travel essentially collinear so that @xmath171 , implying an additional `` collinearity suppression . '' a typical transverse component of the neutrino velocity is of order @xmath172 if @xmath140 is the size of the source and @xmath116 the distance . therefore , a typical neutrino velocity along the radial direction is @xmath173 , implying @xmath174 . therefore , we find the general scaling @xmath175 that is well known for spherical sources , and is also confirmed numerically in fig . [ fig : asympt ] . for a spherical source we have derived an explicit expression for @xmath176 in eq . ( [ eq : vr ] ) . in this case we find explicitly @xmath177 ^ 2 -1 \nonumber\\ * & = & \frac{1}{2}\,\left(\frac{r}{r}\right)^2 \quad\hbox{for $ r\to\infty$\ , . } \label{eq : singleours } \end{aligned}\ ] ] immediately at the source we find @xmath178 . the asymptotic behavior for large @xmath116 agrees with what one finds if one uses only a single angular bin with all neutrinos being radiated at @xmath15 relative to the radial direction @xcite . in the papers by duan et al . , beginning with ref . @xcite , a somewhat different expression for @xmath179 was used because they pictured the single - angle approximation as all modes oscillating like the radial one . considering the average neutrino - neutrino interaction they find the equivalent of @xmath180 ^ 2 \nonumber\\ * & = & \frac{1}{4}\,\left(\frac{r}{r}\right)^2 \quad\hbox{for $ r\to\infty$\ , . } \label{eq : singlefull } \end{aligned}\ ] ] while the scaling at large radius is , of course , the same , their normalization is a factor of two smaller . we have checked with a numerical multi - angle example that indeed our result provides the proper approximation . in fig . [ fig : fig4 ] we show the evolution of the antineutrino total polarization vector component @xmath181 from a multi - angle simulation ( continuous curve ) . we compare it with the single - angle results corresponding to our approximation of @xmath179 of eq . ( [ eq : singleours ] ) as a dashed curve and with duan et al.s prescription of eq . ( [ eq : singlefull ] ) as a dotted curve . in our numerical example , following ref . @xcite , we have fixed the neutrino - neutrino interaction strength at the neutrino sphere ( radius @xmath182 km ) as @xmath183 km@xmath184 , the neutrino oscillation frequency @xmath185 km@xmath184 , the ratio of the number fluxes @xmath186 , and the neutrino mixing angle @xmath187 . we have assumed the inverted mass hierarchy where bipolar conversions occur . for the single - angle cases , we have used the adiabatic approximation worked out in @xcite , where @xmath181 depends on radius only through @xmath188 and where therefore there are no nutation wiggles . we discuss in more details about the adiabatic approximation below in sec . [ sec : adiab ] . of course , the difference found between duan et al.s approximation and ours is not crucial in that the neutrino fluxes and angular divergences assumed in previous studies had the character of toy examples . since @xmath0 decreases with @xmath189 , a factor 2 modification of @xmath0 translates into a 20% radial shift of @xmath188 that is not important for a toy model . the difference close to the source is much larger , but less important . close to the source we typically have strongly synchronized oscillations with a very small mixing angle , causing practically no difference in the overall solution . we finally consider the variation of @xmath0 for a disk - like source . at a large distance , the variation with distance will be proportional to @xmath189 as in the case of a spherical source where a factor of @xmath169 comes from the trivial geometric flux dilution with distance , while another factor @xmath169 arises from the increasing collinearity of the neutrino trajectories with distance . to see this more explicitly we note that a typical `` radial '' velocity at a large distance is @xmath190 where @xmath191 and @xmath192 are typical transverse velocities in two orthogonal transverse directions . at a large distance we have @xmath193 and @xmath194 so that the factor @xmath195 in eq . ( [ eq : mudefinition ] ) is @xmath196 . a typical transverse velocity component of a neutrino at a distance @xmath116 is @xmath172 if @xmath140 is the geometric size of the source , confirming the @xmath169 scaling of @xmath179 at a large distance . moreover , we can also derive the variation of @xmath147 with zenith angle @xmath143 at a large distance . if the source is a circular disk of radius @xmath140 , the azimuthal component of the velocity , @xmath191 , will typically be @xmath197 . the polar component @xmath192 , on the other hand , will vary as @xmath198 , since the projected area of the source in this direction is suppressed by @xmath142 . this leads to @xmath199 . since the flux itself varies as @xmath200 we expect @xmath201 this is numerically confirmed and agrees with the variation shown in fig . [ fig : asympt ] . the adiabatic approximation for collective neutrino oscillations in the single - angle case has been developed in a series of papers @xcite . if the neutrino flavor evolution is adiabatic , the flavor composition of neutrinos as well as antineutrinos depends only on the initial state and the value of @xmath202 at each @xmath62 , and there is no need to compute the evolution explicitly . here , we follow @xcite and consider in particular the case of a system constituted by only two polarization vectors , @xmath203 for neutrinos and @xmath204 for antinuetrinos . at extremely large matter densities , the effective mixing angles are small , as a result @xmath203 and @xmath204 are aligned with each other and with @xmath96 . in the limit of a large but slow varying neutrino interaction strength @xmath147 , the two polarization vectors move in a pure precession mode around @xmath205 . during the evolution , the conservation of `` flavor lepton number '' implies p_z - _ z = 1-is a conserved quantity , where @xmath206 is the ratio of antineutrino and neutrino fluxes . an explicit solution for the flavor evolution can be obtained simply in terms of @xmath206 as follows . the results in @xcite , when calculated in the limit of vanishing effective mixing angle , give @xmath207 where @xmath208 is the cosine of the angle between @xmath204 and @xmath96 at any location . it equals @xmath209 to begin with , and varies between @xmath209 and @xmath210 during the evolution . ( [ eq : adiab ] ) can be inverted to yield @xmath208 at any location as a function of @xmath202 . knowing @xmath208 , one may compute the directions of @xmath203 and @xmath204 through @xmath211 the values of @xmath212 and @xmath213 directly yield the survival probabilities @xmath214 and @xmath215 of @xmath11 and @xmath216 respectively . @xmath214 and @xmath215 are thus direct functions of the local value of @xmath202 . in fig . [ fig : fig4 ] we have shown how the analytical adiabatic approximation agrees very well with the multi - angle numerical simulation for the spherically symmetric sn case . in general , this result means that in the cases where neutrino ensembles exhibit a self - maintained coherent behaviour , the adiabatic approximation together with the single - angle limit described in eq . ( [ eq : mudefinition ] ) are sufficient to describe the collective flavor evolution . on the other extreme , in the case of ensembles showing kinematical decoherence , the adiabatic approximation is not appropriate , and the final outcome is a complete equalization of the neutrino fluxes . for purposes of illustration we now consider an explicit example for a disk - like source . following the numerical studies of ruffert and janka @xcite for coalescing neutron stars , we use the following parameters for a disk source as an extremely schematic model . the all - species neutrino luminosity of the models shown by ruffert and janka varies between @xmath217 and the luminosity in species other than @xmath11 and @xmath218 is always very small . typically @xmath219 , a typical average being 12 mev that we will use for both species . in contrast to the sn case , here one expects an excess of the @xmath220 number flux over @xmath11 , so that in this case we normalize the polarization vectors to the antineutrino flux . the ratio of number fluxes @xmath221 varies between about 0.65 and 0.81 if we ignore one case where the number fluxes are exactly equal and one where the @xmath11 flux practically vanishes . inspired by these numbers and the pictures shown by ruffert and janka we adopt the following parameters for our homogeneous disk - like neutrino source : radius @xmath222 km , @xmath223 , @xmath224 mev , and @xmath225 . this implies a @xmath218 flux at the disk of @xmath226 and a scale for the neutrino - neutrino interaction strength of @xmath227 . we have in mind two - flavor oscillations driven by the atmospheric neutrino mass difference for which we use @xmath228 . then the average vacuum oscillation frequency is @xmath229 . here we have used that for an average neutrino energy @xmath230 mev and assuming a maxwell - boltzmann spectrum we have @xmath231 . the transition between the synchronized and the bipolar oscillation region , where the flavor pendulum begins to nutate and neutrino transformations begin in earnest , occurs at @xcite @xmath232 which is @xmath233 for @xmath225 in this scenario . in fig . [ fig : fig5 ] we show the contours of constant @xmath234 , which are the same as those of constant @xmath214 under adiabatic approximation , for inverted hierarchy . the contour with @xmath235 is the one where the condition of eq . ( [ borderline ] ) occurs and the synchronized to bipolar transition takes place . within the region enclosed by this contour , the oscillations are strongly synchronized so that macroscopically nothing much happens . this contour delineates the area where the flavor survival probability is essentially unity for both @xmath11 and @xmath218 . outside this region , neutrino transitions begin and we show contours of the @xmath11 survival probability that indicates to which extent the initial @xmath11 and @xmath218 fluxes have been converted to other flavors . for our specific numerical example , we find that collective neutrino flavor conversions start at @xmath236 km . since the @xmath237 annihilation rate per unit volume decreases very rapidly far from the source ( for a spherical geometry one would expect @xmath238 @xcite ) we do not expect that these flavor transitions affect significantly the neutrino energy deposition rate in the plasma . however , the features of neutrino fluxes emitted from coalescing neutron stars are rather model dependent . here we speculate that if the asymmetry between the emitted neutrino species would be smaller , one could have collective flavor conversion or perhaps some sort of kinematical decoherence of the neutrino ensemble in a region close to the source , with a possible impact on the annihilation rate . we also note that the matter density in the region close to the source is expected to be @xmath239 , thus it is so strong to prevent ordinary resonant flavor conversions @xcite . at larger distance one expects that two resonant level crossings will occur . however , due to the uncertainties of the matter density profile in these regions , it is difficult to predict their effects on the neutrino burst . neutrinos streaming from powerful astrophysical sources such as sn cores or coalescing neutron stars are so dense near the source that they must show nonlinear flavor oscillations induced by the neutrino - neutrino refractive effect . numerical simulations reveal a rich variety of phenomena , some of which have been explained with simple analytic models . however , numerical simulations thus far have been restricted to homogeneous gases evolving in time or to sources with exact spherical symmetry . more general geometries are numerically much more demanding and have not yet been studied . therefore , we have studied what might be expected under the assumption that the multi - angle instability plays no role and that the neutrino ensemble is largely characterized by self - maintained coherence . in this case one is led to a unique formulation of the collective equations of motion that imply that collective flavor oscillations should be thought of as a one - dimensional phenomenon along the streamlines of the underlying neutrino flux . close to the source these streamlines are usually curved even though , of course , the underlying neutrino trajectories are straight . ( we have neglected the gravitational bending of trajectories . ) therefore , even if the neutrino stream has no global symmetries , the collective oscillation problem is relatively simple . we have used the concept of `` self - maintained coherence '' in the most restrictive sense that applies when the neutrino gas is dense , i.e. , when a typical neutrino - neutrino interaction energy @xmath0 is large compared to a typical vacuum oscillation frequency @xmath240 . the neutrino ensemble in this case evolves along a streamline as one unit that previously has been described as a gyroscopic pendulum in flavor space . all neutrino and anti - neutrino polarization vectors point essentially in the same direction in flavor space , the pendulum direction , allowing for the simplifications that lead to our collective equations . we have provided a prescription for defining the effective neutrino - neutrino interaction strength @xmath0 that works for general source geometries . our result is somewhat different from what was used in the previous literature , but our expression for @xmath0 agrees better with numerical simulations . there are more general forms of collective behavior . in a homogeneous isotropic system with a density that decreases adiabatically in time , the polarization vectors are at first aligned and stay in a single co - rotating plane even when @xmath0 becomes of order @xmath98 and smaller . however , they will align or anti - align themselves in the mass direction , leading to the phenomenon of spectral splits . this is a more general case of self - maintained coherence beyond our simple picture and is analytically fully understood . in the spherically symmetric case , numerical multi - angle simulations show a similar behavior . if the asymmetry between neutrinos and antineutrinos is large enough to prevent multi - angle decoherence , the neutrino and antineutrino modes evolve collectively essentially along the co - rotating plane , but with intriguing three - dimensional patterns that evolve collectively and are stable in flavor space . once more spectral splits develop , similar to the single - angle approximation . we speculate that these more general forms of behavior also occur for more general source geometries if we follow the neutrino stream lines . it is clear that multi - angle decoherence will occur if the asymmetry between neutrinos and anti - neutrinos is extremely small . even a few per cent asymmetry , on the other hand , implies that close to the source we have synchronized oscillations , or , in the pendulum picture , it precesses fast without nutations . therefore , close to the source where the streamlines are curved , our treatment should be most appropriate . if the effective mixing angle is small ( as it will be because of the presence of ordinary matter ) , superficially no flavor oscillations happen close to the source and the `` true action '' begins at the synchronization radius where the inverted pendulum begins to nutate and begins to lose its initial orientation . if this point is reached at some distance , the streamlines will essentially point straight away from the source and naively one should think that henceforth the evolution is similar to the spherically symmetric case . in fact in the adiabatic approximation , one only needs to know the local value of @xmath241 in order to determine the neutrino survival probabilities , for any source geometry . granting these assumptions , we have briefly studied a toy model for coalescing neutron stars where the source is taken to be a homogeneously radiating disk . for our numerical example , we find that the region where collective oscillations will convert the original @xmath11 and @xmath242 fluxes is quite far from the emission surface . therefore these flavor conversions should have a negligible effect on the neutrino energy deposition in the plasma . however , due to the uncertainties of the original neutrino fluxes , we can not exclude situations with a smaller asymmetry between the original fluxes . such tiny neutrino asymmetries may give rise to multi - angle decoherence , that could produce significant flavor conversions near the emitting disk . this question ultimately will need to be addressed numerically by means of large - scale numerical simulations of neutrino flavor evolution near a non - spherical source , in analogy to what has been done in the case of neutrinos streaming off a sn core . our study suggests that self - maintained coherence among different neutrino modes may well be a typical form of behavior even in non - spherical systems and notably in the interesting case of coalescing neutron stars . the final verdict on the role of collective neutrino oscillations in such systems can only come from numerical studies . at very least , our approach provides a simple limiting case against which one can measure the output of numerical simulations . a.d . and b.d . would like to thank k. damle , r. loganayagam , and s. raychaudhuri for useful discussions . in munich , this work was partly supported by the deutsche forschungsgemeinschaft ( grant tr-27 `` neutrinos and beyond '' ) , by the cluster of excellence `` origin and structure of the universe '' and by the european union ( contract no . rii3-ct-2004 - 506222 ) . in mumbai , partial support by a max planck india partnergroup grant is acknowledged . a.m. acknowledges support by the italian istituto nazionale di fisica nucleare ( infn ) through a post - doctoral fellowship . we here briefly describe the numerical integrations for a disk source that were used in the main text . the direction of neutrino momenta are determined by two polar angles @xmath243 and @xmath244 , by which we can express the component of the neutrino velocity in @xmath6 and @xmath49 direction respectively by @xmath245 one can also define the local neutrino density @xmath249 in the case of synchronization , from eq . ( [ eq : eom4 g ] ) we arrive at the following eom @xmath250 where @xmath251 in this case , we have obtained a first order partial differential equation ( pde ) . since @xmath252 and @xmath253 are not constant , in this case the streamlines will be curved . if the streamlines are graphs of the function @xmath254 , it follows @xmath255 ( i.e. , the tangent line to the graph of @xmath254 at @xmath256 is parallel to @xmath257 . ) the ordinary differential equation ( ode ) in eq . ( [ eq : ode ] ) is the so - called characteristic equation for the associated pde . its solutions are the streamlines for the pde . the range of the previous angular integrations are determined by the coordinates @xmath6 and @xmath49 . it could be convenient to express the previous integrals at a given point , writing @xmath258 and @xmath244 in terms of the coordinates @xmath259 and @xmath260 on the disk , by tracing back the direction of neutrino momenta on the disk , as shown in fig . [ fig : thetapphip ] . let the neutrino emitted from a point @xmath261 pass through the location @xmath262 . the direction of the momentum of this neutrino is along the unit vector @xmath263 . therefore , @xmath264 we can now write down @xmath258 and @xmath244 explicitly in terms of @xmath265 : @xmath266 so that @xmath267 for the angle @xmath244 , we get @xmath268 so that @xmath269 using as variables of integration @xmath259 and @xmath260 , we obtain @xmath270 where @xmath271 is the jacobian that relates the two coordinate systems , and @xmath272 and @xmath273 . j. pantaleone , `` neutrino oscillations at high densities , '' phys . b * 287 * , 128 ( 1992 ) . g. sigl and g. raffelt , `` general kinetic description of relativistic mixed neutrinos , '' nucl . b * 406 * , 423 ( 1993 ) . s. samuel , `` neutrino oscillations in dense neutrino gases , '' phys . d * 48 * , 1462 ( 1993 ) . v. a. kosteleck and s. samuel , `` neutrino oscillations in the early universe with an inverted neutrino mass hierarchy , '' phys . b * 318 * , 127 ( 1993 ) . v. a. kosteleck and s. samuel , `` self - maintained coherent oscillations in dense neutrino gases , '' phys . d * 52 * , 621 ( 1995 ) [ hep - ph/9506262 ] . s. samuel , `` bimodal coherence in dense selfinteracting neutrino gases , '' phys . d * 53 * , 5382 ( 1996 ) [ hep - ph/9604341 ] . s. pastor , g. g. raffelt and d. v. semikoz , `` physics of synchronized neutrino oscillations caused by self - interactions , '' phys . d * 65 * , 053011 ( 2002 ) [ hep - ph/0109035 ] . y. y. y. wong , `` analytical treatment of neutrino asymmetry equilibration from flavour oscillations in the early universe , '' phys . d * 66 * , 025015 ( 2002 ) [ hep - ph/ 0203180 ] . k. n. abazajian , j. f. beacom and n. f. bell , `` stringent constraints on cosmological neutrino antineutrino asymmetries from synchronized flavor transformation , '' phys . d * 66 * , 013008 ( 2002 ) [ astro - ph/0203442 ] . s. pastor and g. raffelt , `` flavor oscillations in the supernova hot bubble region : nonlinear effects of neutrino background , '' phys . rev . lett . * 89 * , 191101 ( 2002 ) [ astro - ph/0207281 ] . r. f. sawyer , `` classical instabilities and quantum speed - up in the evolution of neutrino clouds , '' hep - ph/0408265 . r. f. sawyer , `` speed - up of neutrino transformations in a supernova environment , '' phys . d * 72 * , 045003 ( 2005 ) [ hep - ph/0503013 ] . r. f. sawyer , `` the multi - angle instability in dense neutrino systems , '' arxiv:0803.4319 [ astro - ph ] . h. duan , g. m. fuller and y. z. qian , `` collective neutrino flavor transformation in supernovae , '' phys . d * 74 * , 123004 ( 2006 ) [ astro - ph/0511275 ] . h. duan , g. m. fuller , j. carlson and y. z. qian , `` simulation of coherent non - linear neutrino flavor transformation in the supernova environment . i : correlated neutrino trajectories , '' phys . d * 74 * , 105014 ( 2006 ) [ astro - ph/0606616 ] . s. hannestad , g. g. raffelt , g. sigl and y. y. y. wong , `` self - induced conversion in dense neutrino gases : pendulum in flavor space , '' phys . d * 74 * , 105010 ( 2006 ) ; erratum ibid . * 76 * , 029901 ( 2007 ) [ astro - ph/0608695 ] . g. g. raffelt and g. sigl , `` self - induced decoherence in dense neutrino gases , '' phys . d * 75 * , 083002 ( 2007 ) [ hep - ph/0701182 ] . h. duan , g. m. fuller , j. carlson and y. z. qian , `` analysis of collective neutrino flavor transformation in supernovae , '' phys . d * 75 * , 125005 ( 2007 ) [ astro - ph/0703776 ] . g. g. raffelt and a. yu . smirnov , `` self - induced spectral splits in supernova neutrino fluxes , '' phys . d * 76 * , 081301 ( 2007 ) ; erratum ibid . * 77 * , 029903 ( 2008 ) [ arxiv:0705.1830 ( hep - ph ) ] . g. g. raffelt and a. yu . smirnov , `` adiabaticity and spectral splits in collective neutrino transformations , '' phys . d * 76 * , 125008 ( 2007 ) [ arxiv:0709.4641 ( hep - ph ) ] . a. esteban - pretel , s. pastor , r. toms , g. g. raffelt and g. sigl , `` decoherence in supernova neutrino transformations suppressed by deleptonization , '' phys . d * 76 * , 125018 ( 2007 ) [ arxiv:0706.2498 ( astro - ph ) ] . h. duan , g. m. fuller and y. z. qian , `` a simple picture for neutrino flavor transformation in supernovae , '' phys . d * 76 * , 085013 ( 2007 ) [ arxiv:0706.4293 ( astro - ph ) ] . h. duan , g. m. fuller , j. carlson and y.z . qian , `` neutrino mass hierarchy and stepwise spectral swapping of supernova neutrino flavors , '' phys . lett . * 99 * , 241802 ( 2007 ) [ arxiv:0707.0290 ( astro - ph ) ] . g. l. fogli , e. lisi , a. marrone and a. mirizzi , `` collective neutrino flavor transitions in supernovae and the role of trajectory averaging , '' jcap * 0712 * , 010 ( 2007 ) [ arxiv:0707.1998 ( hep - ph ) ] . h. duan , g. m. fuller , j. carlson and y. z. qian , `` flavor evolution of the neutronization neutrino burst from an o - ne - mg core - collapse supernova , '' phys . lett . * 100 * , 021101 ( 2008 ) [ arxiv:0710.1271 ( astro - ph ) ] . b. dasgupta , a. dighe , a. mirizzi and g. g. raffelt , `` spectral split in prompt supernova neutrino burst : analytic three - flavor treatment , '' arxiv:0801.1660 [ hep - ph ] . a. esteban - pretel , s. pastor , r. toms , g. g. raffelt and g. sigl , `` mu - tau neutrino refraction and collective three - flavor transformations in supernovae , '' phys . rev . d * 77 * , 065024 ( 2008 ) [ arxiv:0712.1137 ( astro - ph ) ] . b. dasgupta and a. dighe `` collective three - flavor oscillations of supernova neutrinos , '' [ arxiv:0712.3798 ( hep - ph ) ] . h. duan , g. m. fuller and y. z. qian , `` stepwise spectral swapping with three neutrino flavors , '' phys . d * 77 * , 085016 ( 2008 ) [ arxiv:0801.1363 ( hep - ph ) ] . b. dasgupta , a. dighe and a. mirizzi , `` identifying neutrino mass hierarchy at extremely small theta(13 ) through earth matter effects in a supernova signal , '' arxiv:0802.1481 [ hep - ph ] . h. duan , g. m. fuller and j. carlson , `` simulating nonlinear neutrino flavor evolution , '' arxiv:0803.3650 [ astro - ph ] . m. ruffert and h .- janka , `` gamma - ray bursts from accreting black holes in neutron star mergers , '' astron . astrophys . * 344 * , 573 ( 1999 ) [ astro - ph/9809280 ] . a. d. dolgov , `` neutrinos in the early universe , '' yad . fiz . * 33 * , 1309 ( 1981 ) [ sov . . phys . * 33 * , 700 ( 1981 ) ] . m. a. rudzsky , `` kinetic equations for neutrino spin- and type - oscillations in a medium , '' astrophys . space science * 165 * , 65 ( 1990 ) . b. h. j. mckellar and m. j. thomson , `` oscillating doublet neutrinos in the early universe , '' phys . d * 49 * , 2710 ( 1994 ) . p. strack and a. burrows , `` a generalized boltzmann formalism for oscillating neutrinos , '' phys . d * 71 * , 093004 ( 2005 ) [ hep - ph/0504035 ] . c. y. cardall , `` liouville equations for neutrino distribution matrices , '' arxiv:0712.1188 [ astro - ph ] . s. mrowczynski and m. h. thoma , `` what do electromagnetic plasmas tell us about quark - gluon plasma ? , '' ann . rev . nucl . part . sci . * 57 * , 61 ( 2007 ) [ nucl - th/0701002 ] . e. c. zachmanoglou and d. w. thoe , _ introduction to partial differential equations with applications _ ( dover publications , new york , 1986 ) . r. r. volkas and y. y. y. wong , `` matter - affected neutrino oscillations in ordinary and mirror stars and their implications for gamma - ray bursts , '' astropart . * 13 * , 21 ( 2000 ) [ astro - ph/9907161 ] .

based on a dispersive approach , we apply the inverse amplitude method to unitarize one - loop and chiral perturbation theory . numerically , we find that this unitarization technique yields the correct complex analytic structure in terms of cuts and poles . indeed , using the chiral parameter estimates obtained from low energy experiments we obtain the poles associated with the and resonances . just by fixing their actual masses we obtain a parametrization of the and phase shifts in eight different channels . with this fit we have then calculated several low - energy phenomenological parameters estimating their errors . among others , we have obtained the chiral parameters and scattering lengths , which can be relevant for future experiments . pacs : 14.40.aq , 14.40.cs,11.80.et,13.75.lb 6.5 in 8.5 in epsf lbl-38645 ucm - ft 3/96 april 1996 revised nov96 to appear in : prd 1-sept-97 * the inverse amplitude method in chiral + perturbation theory * .5 cm a. dobado + departamento de fsica terica . + universidad complutense . 28040 madrid , spain + and .3 cm j.r . pelez + theoretical physics group . lawrence berkeley laboratory + university of california . berkeley , california 94720 . usa . .5 cm

hep-ph9604416

even though qcd yields a remarkably good description of the strong interaction , the low energy hadron physics has to be modelled phenomenologically . this is due to the fact that the usual perturbative approach in the coupling constant can not be applied to qcd below energies of the order of 1 gev . most of the phenomenological results were based on pcac and current algebra . however , in 1979 , weinberg @xcite showed how to reobtain many of these predictions by means of an effective lagrangian . the fields in that lagrangian are the light mesons , ( pions , kaons and etas ) which are understood as the goldstone bosons ( gb ) arising from the spontaneous breaking of chiral symmetry . the lagrangian is built as an expansion in derivatives , that respects the symmetry breaking pattern of qcd . indeed , the first term in the expansion is fixed by the symmetry requirements and accounts for the current algebra results . the next terms in the expansion produce further corrections , which depend on several phenomenological parameters but are always consistent with the qcd symmetry constraints . these techniques were later developed to one loop in a set of papers by gasser and leutwyler @xcite . they showed how to obtain amplitudes involving light mesons , as functions of their momenta , their masses and those few phenomenological parameters . by fitting these parameters from a few low energy experiments it is then possible to obtain successful predictions for other processes the whole approach is known as chiral perturbation theory ( chpt ) . very recently some partial higher order calculations @xcite have appeared in the literature as well as a complete two loop calculation of @xmath4 scattering @xcite , which will be needed in order to analyze new data to come from da@xmath8ne and brookhaven . for a general review of the available experimental data on pion physics and future prospects , we refer the reader to @xcite . nevertheless , there are some intrinsic limitations when applying chpt , namely , the fact that the amplitudes calculated within the chiral approach are only unitary in the perturbative sense , that is , up to the next order in the external momenta . such a breakdown of unitarity is most severe at high energies , where the external momenta is no longer a good expansion parameter , although it can also occur at moderate energies @xcite . as a result , it is not possible to reproduce resonant states , which are one of the most characteristic features of the strongly interacting regime . many different methods have been proposed in order to improve this behavior and thus to extend the applicability of chpt to higher energies ; among them : the use of pad approximants @xcite , the explicit introduction of resonances @xcite , the k - matrix @xcite , the large @xmath9 limit @xcite ( n being the number of gb ) or the inverse amplitude method ( iam ) @xcite . this work is devoted precisely to the last method , which can be justified within a dispersive approach and can easily reproduce the two lightest resonances : the @xmath2 in @xmath4 scattering @xcite and the @xmath10 in @xmath11 scattering @xcite . but not only that , the iam also improves considerably the fit to data even in non - resonant channels , almost up to the first two particle inelastic threshold ( the many particle inelastic thresholds can be neglected since they are suppressed by phase space factors ) . this fit provides a remarkably good parametrization that can be used for other processes . indeed , in a previous work @xcite , the authors showed how it can be used together with a simple unitarization prescription to obtain successful results on @xmath12 up to 700 mev . of course , it is also possible to obtain very good parametrizations @xcite of @xmath4 or @xmath11 elastic scattering by including all resonant states explicitly . however , our aim choosing the iam is to reproduce these phenomena just with the few phenomenological parameters present in the chpt lagrangian . in this way , even though their masses and widths will not be obtained with great accuracy , resonances can be regarded as real predictions . that is one of the relevant features of the iam since other very popular unitarization methods are not able to reproduce resonances unless they are explicitly introduced in the calculation . that is , for instance , the case with the k - matrix . the purpose of this work is , first , to study how high in energies the iam yields good results and what are its limitations . we would also like to know whether it is possible to reproduce further resonance states . it is clear that the best candidates are the lightest resonances whose dominant decay modes are @xmath4 or @xmath11 . we have listed them in table 1 . in case these resonances were not accommodated after our unitarization , it would be interesting to understand why . second , once we have a good fit to these resonances , we want to make a complete numerical analysis of several low - energy quantities of interest , like the chiral parameters or the scattering lengths , including estimations for their errors . as we will see below , we expect that the iam somehow will include effects that can not be obtained from the pure @xmath13 expansion . [ cols="^,^,^,^,^",options="header " , ] lightest resonances with @xmath4 or @xmath11 dominant decay modes . data taken from @xcite . 0.cm finally , we would like to comment on another motivation of the present work , which at first may not seem very related to the main topic . the philosophy of the chiral approach has also reached the description of the strongly interacting symmetry breaking sector ( sisbs ) of the standard model @xcite . the scalar sector of such a model displays the same symmetry breaking pattern as two flavor massless qcd . hence it is possible to build an effective lagrangian , much as it is done for chpt @xcite . although the electroweak gb are not physical , using this lagrangian it is possible to obtain predictions for the scattering of longitudinal gauge bosons @xcite at future colliders , like the lhc . indeed , there are already experimental proposals to measure the electroweak chiral parameters at cms @xcite . most of the works on the sisbs make use of the equivalence theorem @xcite , which allows us to read the observable amplitudes , in terms of longitudinal gauge bosons , directly from those with gb . this theorem has been recently proved in the chiral lagrangian formalism @xcite and seems to be severely constrained by the lack of unitarity . at this point is when the unitarization procedures come into play and it is crucial to know whether they are reliable , since what we are now looking for are real predictions and not elaborated fits to still unavailable data . in section 2 we review some basic aspects of exact and perturbative unitarity and we define the partial waves in elastic scattering . section 3 introduces the iam , first with a derivation from dispersion theory and then by studying the constraints to its applicability . section 4 and section 5 are organized in the same way , although they refer to @xmath0 and @xmath1 chpt , respectively : first we apply the iam to chpt with the chiral parameters obtained from low energy experiments in order to study the iam predictive power . next , we present an iam fit to the data . for the best @xmath1 fit we present the unitarized results for the scattering lengths and some other phenomenological parameters . then , in section 6 , we study the analytic structure on the complex plane of the iam amplitudes . in section 7 we present the conclusions . there is also an appendix where we give the elastic scattering formulae used in this work , as well as a discussion on perturbative unitarity . when dealing with strong interactions , it is usual to project the amplitudes in partial waves with definite angular momentum @xmath14 and isospin @xmath15 , as follows @xmath16 where @xmath17 or @xmath18 depending on whether the particles in the process are identical or not . the acceptable isospin values also depend on the process , namely @xmath19 for @xmath4 elastic scattering and @xmath20 for @xmath11 . for both reactions the definite isospin amplitudes @xmath21 are obtained from a single function . in the first case @xmath22 whereas for @xmath11 scattering we can write @xmath23 in order to deal with both processes on the same footing , we will label the particles in the reaction as @xmath24 and @xmath25 . thus the mandelstam variables will satisfy : @xmath26 and the threshold will be at @xmath27 . as it is well known , whenever @xmath28 , and below inelastic thresholds , the unitarity of the s - matrix implies @xmath29 where @xmath30 is the two particle phase space . explicitly : @xmath31 as a consequence of eq.[uni ] , the partial wave can be parametrized as follows : @xmath32 and @xmath33 is called the @xmath34 phase shift . we have already mentioned that the chpt amplitudes are obtained as an expansion in external momenta and masses . that is @xmath35 where , for the cases we are interested in , @xmath36 is @xmath37 , @xmath38 is @xmath39 , etc ... in practice , we can only obtain the few first terms of the series above and therefore the amplitude only satisfies the unitarity condition perturbatively @xmath40 the @xmath37 terms were given by weinberg @xcite and they are called the low energy theorems . the next order contributions to @xmath4 scattering were given in @xcite . the calculation for @xmath11 scattering can be found in @xcite , although we have found that the formulae in the literature do not satisfy eq.[punit ] . we will comment on that later . very recently it has appeared the complete calculation of the @xmath41 contribution to elastic @xmath4 scattering @xcite . although we will not use it , we will compare some of its results with those of our method . let us briefly review the standard derivation @xcite of the inverse amplitude method , since we will use it later in order to understand the applicability of the method . any partial wave obtained from a relativistic quantum field theory should present a characteristic analytic structure in the complex @xmath42 plane . indeed , the reaction threshold becomes a cut in the from @xmath43 to @xmath44 . due to crossing symmetry , there should be another left cut along the negative axis . if we now apply cauchy s theorem to our complex amplitudes we obtain integral equations known as dispersion relations . for instance , a three times subtracted dispersion relation is @xmath45 where we have not written explicitly the left cut ( lc ) contribution . the number of subtractions needed depends on how the amplitude behaves at infinity in order to ensure the vanishing of the contributions coming from closing the integral contour . we have chosen three subtractions since we are going to use @xmath39 chpt amplitudes which at high @xmath42 behave as @xmath46 . but our arguments remain valid for @xmath41 amplitudes when using four times subtracted dispersion relations , etc ... the chpt partial waves present both cuts and we can calculate both the subtraction constants @xmath47 and the integrand inside eq.[disp ] _ perturbatively _ @xmath48 where we have expanded the subtraction constants in terms of @xmath49 . the iam is based on the fact that the function @xmath50 displays the very same analytic structure of @xmath51 , apart from some possible pole contributions . for later convenience , we will make use of @xmath52 . notice that we have multiplied @xmath53 by a real function without singularities ; thus we keep the same analytic structure and we can write a very similar dispersion relation : @xmath54 where @xmath55 stands for possible pole contributions . the advantage of using @xmath56 is that , using eqs.[uni ] and [ punit ] , we can calculate exactly the integral over the right cut ( but not on the left , since those equations only hold on the elastic cut ) , as follows : @xmath57 note that we are denote by @xmath51 the exact amplitude , which is unknown , although we know its analytic properties . in contrast , the expressions for @xmath36 and @xmath38 , etc ... have been calculated explicitly . as we did before , we can also expand the @xmath58 subtraction coefficients in powers of @xmath59 , and then rewrite the dispersion relation for @xmath56 , which now reads @xmath60 where we have approximated @xmath61 on the left cut and we have neglected the pole contribution . in other words , @xmath62 this is the iam result that we are going to use in the present work . incidentally , eq.[iam ] can be understood as the formal @xmath63 $ ] pad approximant of the chpt amplitude . it is important to remark that if we expand again eq.[iam ] at low energies , we find @xmath64 that is , we recover the chpt expansion . hence , up to @xmath65 our method and chpt yield the same low energy results if the same chiral lagrangian parameters are used . let us review all the approximations made in the previous section , in order to comment how they will constraint the iam applicability : in eq.[preiam ] we have replaced the @xmath56 left cut integral by that of @xmath66 . as we have remarked in the preceeding discussion , eqs.[uni ] and [ punit ] are only exact on the right cut . on the left cut we can not write the chain of equalities that lead to eq.[img ] . nevertheless , if we use the chpt result as an approximation : @xmath67 we get @xmath68 notice that , in order to obtain the @xmath34 phase shifts , we are going to calculate @xmath69 for real @xmath70 . that means that the denominator @xmath71 inside the integrals is never going to be very small , which somehow will wash out the error on the left cut . but note also that treating differently the right and left cuts violates crossing symmetry . indeed , in @xcite it has already been pointed out that the pad approximants do not reproduce correctly the subleading logarithms that would appear at _ next order _ in the chiral expansion ( @xmath65 in this case ) . of course they would be obtained if we applied the iam to the chiral amplitudes at @xmath65 , but still the method would not yield the correct logarithms at @xmath72 and so on . at high energies chiral logarithms are not so relevant , but at low energies they are a very important feature of chpt and indeed they can give the dominant contribution in some channels . nevertheless , from eq.[recover ] we see that at low energies the iam yields the very same @xmath73 chpt expansion , _ including _ the _ dominant _ chiral logarithms . the contribution from the left cut and subleading logarithms is @xmath65 . as a consequence , if we try to make a low energy fit to the data , the parameters that we would obtain with the iam would not lie very far from those of chpt , but they will not be the same . that is the reason why , in the following sections , we will denote with a hat the parameters obtained from any iam fit . in eq.[preiam ] , we have neglected the contributions coming from zeros in the amplitude , that will appear as poles of the inverse function . there is no way to know _ a priori _ whether or not a partial wave will vanish for a given value of @xmath42 , although it is known that chiral amplitudes have zeros below threshold , which are known as adler zeros . their position is not known except for the @xmath74 channel , where the pole is located at threshold . in our derivation it is compensated by the same zero in the @xmath75 channel . that is not the case of the @xmath76 amplitudes and therefore we are neglecting the contribution of their residue . consequently , our amplitudes are not valid to obtain adler zeros and that will affect our results at low energy ( but no more than @xmath65 ) . that is another reason to differentiate the parameters obtained from our fit from those of the pure chiral expansion . this is apparently a harmless assumption in the above reasoning , although it dramatically affects the results of the iam . in fact , it can happen that @xmath77 . in the @xmath78 channels of @xmath4 scattering or in the @xmath79 in @xmath11 , this only occurs for isolated values of @xmath42 , at or below threshold . in particular , that means that the iam amplitudes will have the same zeros as the lowest order chiral amplitudes . however , every other partial wave vanishes at @xmath37 , for any @xmath42 . as a consequence , the formula in eq.[iam ] is no longer valid . nevertheless , we can generalize our previous derivation , in order to include those channels whose leading order is @xmath39 . we only have to go through the very same steps , although now we would write a dispersion relation for @xmath80 . but let us remember that the main improvement of the approach is that we are calculating exactly the integral of @xmath81 over the right cut . however , for that purpose we need an imaginary part , and by looking at eq.[punit ] we can see that @xmath77 implies that @xmath82 . therefore , unless we have a calculation up to @xmath83 , the corresponding imaginary part will vanish . hence when following the derivation of the iam if @xmath77 the best we can get is plain chpt again . at present , only @xmath84 calculations are available and we can only expect to obtain a real improvement with our approach in the six channels listed above . thus , we will not be able to reproduce the @xmath85 resonance . in order to obtain @xmath86 on the right cut , eq.[img ] , we have just made use of the elastic unitarity condition of eq.[uni ] . however , the right cut is composed of many superimposed cuts , each one corresponding to a different inelastic intermediate channel . actually , eq.[uni ] is only true below the first inelastic threshold , and the real unitarity condition reads @xmath87 the sum is over all the physically accessible intermediate states @xmath88 , whose phase space is @xmath89 . as far as we are neglecting electromagnetic interactions , the first inelastic channel in @xmath4 is the four pion intermediate state , at 550 mev . similarly , for @xmath11 is @xmath90 , whose threshold is @xmath91 910 mev . strictly speaking , only for lower energies the elastic approximation is exact . nevertheless , the contribution of these intermediate states is strongly suppressed by the four particle phase space and we expect the iam to provide a good approximation . unfortunately within the range of energies we are interested in , there are intermediate channels which are not suppressed by phase space . indeed , at approximately @xmath92 mev the inelastic @xmath93 threshold opens up . its phase space factor is the @xmath89 in eq.[sigma ] , with @xmath94 . therefore , above the two kaon threshold we have to reconsider the derivation of the iam . let us illustrate with @xmath4 scattering how inelastic effects modify our result . as the starting point , for @xmath95 , we have a new unitarity relation : @xmath96 where we have denoted by @xmath97 the generic @xmath51 pion elastic scattering amplitude and by @xmath98 the @xmath34 partial wave of the process @xmath99 . thus we now have , for @xmath100 , that @xmath101 which differs from eq.[img ] in the term coming from two kaon intermediate production . if we follow the very same steps of our previous derivation , we arrive at @xmath102 notice that , using chpt , @xmath103 . but at these high energies that is not negligible . besides , we are interested in the above integral for physical values of @xmath42 and therefore the denominator will be almost divergent for some @xmath104 . for these reasons we can not neglect this integral and then we should not trust the iam since it could miss some relevant physical features . that is indeed the case in pion scattering since , as it can be seen in table 1 , there is one resonance , the @xmath105 , whose nature is closely related to the @xmath93 threshold . nowadays , the interpretation of that resonance is still controversial : different authors propose different poles ( not always just one ) in the vicinity of the @xmath93 inelastic cut @xcite . as we will see later , our approach is not able to reproduce any of these poles , consistent with the fact that the iam makes use just of elastic unitarity . at this point we want to remark the importance of understanding why and when the method does no longer yield the right results . let us remember that we are also thinking in possible applications of this unitarization procedures to the electroweak chiral effective lagrangian , whose reference model is the standard model with a heavy higgs . in such case , one would expect to see a broad resonance in the scalar channel and we want to have a unitarization procedure whose predictions we can trust . throughout the derivation of the iam we have been using the chiral amplitudes up to @xmath39 . nevertheless , it is possible to extend the argument to include higher order terms , as for instance the @xmath41 contributions . in that case we would have started from a four times subtracted dispersion relation for the two - loop calculation . once more , the integral over the right cut would be related to the one for @xmath52 . working out the expansion of the subtraction constants , we would then arrive to @xmath106 again that is the formal [ 1,2 ] pad approximant , and it satisfies the elastic unitarity condition . as we have already mentioned two recent papers have appeared with @xmath41 calculations of @xmath4 scattering within @xmath0 chpt @xcite . we have not used these results , since , as we have just seen , they will not help us to overcome any of the preceeding objections to the iam . however it is quite likely that , have we used them , the parameters of the fits that we will present in the next sections would had been slightly modified . the inverse amplitude method was first applied @xcite to @xmath4 scattering without the strange quark . in that case , the massless limit displays spontaneous symmetry breaking from @xmath107 to @xmath108 , which is nothing but the usual isospin . the @xmath39 expression for @xmath4 scattering was obtained in @xcite , and it is written in terms of four phenomenological parameters @xmath109 as well as the mass and pion decay constants , @xmath110 and @xmath111 . in this section we will review how the method is able to reproduce the @xmath112 resonance . we will show some results for recently proposed new parameters in order to test the iam predictive power , but we will also present a unitarized fit to the data . as a novelty we will use not only the @xmath76 phase shifts , but also those with @xmath113 , in order to obtain the best fit with the iam . in this new calculation , we have also estimated the error bars of the unitarized parameters . let us now illustrate what happens if we apply the iam on the chpt amplitudes using the chiral parameters obtained from low energy experiments . we want to see quantitatively to what extent the main physical features are reproduced . in order to simplify the comparison with previous works , we have chosen @xmath114 mev and @xmath115 mev . the values of the chiral parameters are not so clear , since they have considerable error bars . in table 2 we have listed the different sets of parameters that we have taken from the literature to obtain fig.1 . ' '' '' method & @xmath116 & @xmath117 & @xmath118 + chpt & [email protected] & [email protected] & no resonances + inverse & [email protected] & [email protected] & 715 mev + amplitude & [email protected] & [email protected] & 675 mev + .2 cm sets of parameters and methods used in the text . those in the first two lines come from @xmath120 decays @xcite . those in the third , from data on @xmath120 and @xmath4 together with some unitarization procedure ref.@xcite . @xmath118 is calculated with the central values . let us remark at this point that for the chpt phase shifts we are using the definition @xmath121 suggested in @xcite . of course , chpt is just a low energy approach , but incidentally , these phase shifts coincide with those obtained from the k - matrix unitarization defined as @xmath122 it can be easily verified that @xmath123 satisfies elastic unitarity , eq.[uni ] , exactly . consequently , the dotted lines in fig.1 not only give the chpt predictions , but also the results of the k - matrix unitarization . we will thus confirm that such a method is not able to reproduce resonances by itself . they have to be added by hand . in fig.1 it can be clearly seen , in the @xmath74 channel , that the iam yields a @xmath112-like resonance . the value of its mass is obtained from the point where @xmath124 and it lies 10% to 15% away from its real value . in this way , the _ existence _ of the @xmath112 resonance can be regarded as a prediction of the iam with chpt and the parameters obtained from some low energy data . it is also evident that the fit of the @xmath125 channel is correct up to much higher energies . in table 2 we have also included the values of @xmath118 corresponding to each choice of parameters . for all the cases we have set @xmath126 and @xmath127 following reference @xcite . the only feature of @xmath4 scattering that is evidently missing from the unitarized results is the @xmath105 resonance in the @xmath128 channel . in the previous section we saw that this fact is connected with the failure of the whole approach to reproduce the kaon inelastic cut . but let us first obtain a better fit to the data . now that we have an amplitude that describes the right cut , while keeping at the same time the correct polynomial form from chpt , it seems natural to use @xmath118 @xcite to fit the data . note that fixing the correct mass does not imply a good fit . -2 cm 0.cm 0.cm for instance , we could get a wrong width . in order to differentiate the parameters thus obtained from those coming from plain chpt we will call them @xmath129 . the ( 1,1 ) channel is almost only sensible to @xmath130 . with @xmath131 we get the @xmath118 listed in table 1 and in fig . 2 it can be seen that the results are remarkably successful . later we will show that we also get the right width . once that difference is fixed , we just have to determine one parameter , say @xmath132 . in previous studies @xcite the unitarized fit to the other phase shifts was used in order to estimate the values of @xmath133 and @xmath132 . but , as we commented above , the data in the ( 0,0 ) channel is not as good as that of ( 1,1 ) . the same happens for the ( 2,0 ) channel , where the curves are not very sensible to small variations in the @xmath134 parameters . therefore , in the present work , we have also used the @xmath113 channels ( mainly that with @xmath135 ) to further constrain the parameter range . let us remember now that in these channels @xmath136 and , as we have already discussed in section 2 , the iam leads again to plain chpt . that is why we will only use for them data up to @xmath137mev , although in other channels we are using data at higher energies . thus , the values given in table 3 are just a conservative estimate of the range where we obtain a reasonable fit in the @xmath138 and @xmath139 channels , when @xmath118 fixed to its actual value . the results are shown in fig.2 , where the continuous line corresponds to the central @xmath134 values and the shaded area to their uncertainties . notice that the shaded area has always been obtained by varying @xmath132 within its estimated error . in fig.2 it can be seen how it is not possible to fit the @xmath105 resonance with @xmath0 chpt and the iam . it is clear that , even though the actual value of the @xmath140 phase shift may not lie very far from the unitarized prediction , the qualitative behavior of the curves in this channel is not correct above @xmath141mev . with the @xmath134 fit we can obtain the total breit - wigner width of the @xmath112 resonance from : @xmath142 indeed we have computed it for different values of @xmath42 around @xmath143 . the result is given in table.3 and it is quite close to the experimental result ( see table 1 ) although slightly high . we will see that it is possible to obtain the right value when using @xmath1 chpt . as we have already commented , this result is not at all trivial , since fitting the right mass does not ensure a correct description of the resonance . therefore , even though we are now using the @xmath118 experimental value , the @xmath144 width is again a prediction of the iam . in contrast , in a unitarization scheme where one introduces the resonances by hand , one has to give both the masses and the widths . ' '' '' method & @xmath133 & @xmath132 & @xmath118 ( input ) & @xmath144 + inverse & & & & + amplitude & [ 1ex]-0.5 @xmath119 0.6 & [ 1ex]5.4@xmath119 0.6 & [ 1ex]768.8@xmath119 1.1 mev & [ 1ex][email protected] mev + .2 cm parameters and results of the one - loop iam when @xmath118 is fixed to its actual value . 0.cm -2 cm 0.cm 0.cm the extension of the chpt approach to include the strange quark was done , once more , by gasser and leutwyler @xcite . in this case there are eight goldstone bosons , which are identified with the three pions , the four kaons and the eta . in principle it is possible to calculate the amplitudes of any process involving any combination of these particles . but the thresholds for these reactions are much higher than in pion scattering , which in practice restricts severely the effectiveness of the approach . nevertheless , the lowest two particle threshold apart from @xmath4 scattering is that of @xmath11 elastic scattering at @xmath145mev , which is still within the applicability range of chpt . the calculation of this amplitude to @xmath39 was performed by bernard , kaiser and meiner @xcite who also gave the @xmath39 result for @xmath4 within @xmath1 chpt . in the literature , these formulae have sometimes appeared with some minor errata which have been corrected in the da@xmath8ne physics handbook @xcite . however , even those formulae do not satisfy perturbative unitarity ( see appendix ) . following the work in @xcite we have rederived an expression which does satisfy that requirement , and we have included it in the appendix , together with a discussion on how it is obtained and its unitarity properties . in the @xmath1 case there are more phenomenological parameters that we have set to : @xmath146 there are also twelve one - loop parameters , denoted by @xmath147 . however only @xmath148 and @xmath149 appear in @xmath11 in scattering , whereas in pion scattering only the following combinations are present : @xmath150 again , and in order to simplify the comparison with previous works , we have fixed the following values @xcite : @xmath151 a precise value of these parameters is not very important since they are related to the different masses and decay constants that we had already fixed . hence , in practice , the only relevant parameters for @xmath4 and @xmath11 scattering in @xmath1 are @xmath152 and @xmath153 . the iam was first applied to @xmath1 chpt by the authors in @xcite , were we showed that it reproduces not only the @xmath2 resonance but also the @xmath10 . our aim in this section is first to study the predictive power of the method , whether it can accommodate further resonant states , or why it can not . then we will present a simultaneous fit to @xmath4 and @xmath11 scattering to the @xmath112 and @xmath154 masses . the new features of this analysis is that it uses the corrected chpt expressions for @xmath11 scattering which now satisfy perturbative unitarity ( see the appendix ) and the fact that we also use the data on the @xmath113 @xmath4 scattering channels . we will also estimate the error bars on the best fit that will be used to obtain numerical values for some interesting phenomenological quantities . this fit will also allow us , in section 6 , to perform a numerical study of the analytic structure of the iam amplitudes in the complex @xmath42 plane . let us then start with the iam using parameters obtained from low energy data . in table 4 we list different choices of parameters and methods together with their results for the @xmath112 and @xmath154 masses . as in the case of @xmath0 chpt the iam is able to predict from low energy data the existence of both resonant states . remarkably , the masses thus obtained lie again 10% to 15% away from their actual values . in fig.3 we show the result of applying the iam to @xmath11 scattering , with the parameters given in table 4 . in contrast with plain @xmath39 chpt ( or the k - matrix unitarization method , since they yield the same phase shifts ) , it is evident that the iam not only accommodates the @xmath154 resonance , but it also reproduces the @xmath155 channel . we do not display the results for @xmath4 scattering in @xmath1 because they will almost superimpose with those in fig.1 . indeed , the @xmath156 parameters in lines 2 and 3 of table 2 were obtained , respectively , from the @xmath157 in lines 2 and 3 of table 4 @xcite , by means of @xmath158 as a matter of fact , we have calculated independently the @xmath4 elastic scattering in @xmath0 and @xmath1 . using the above equations to relate the parameters in both cases , and below kaon threshold , we have obtained the same results up to numerical differences ( @xmath911% ) , which would be unobservable in the figures . that is a nice check of our programs . therefore , fig.1 is also the result for @xmath4 scattering in the @xmath1 formalism , but now with the parameters in table 4 . -2 cm 0.cm 0.cm ' '' '' method & @xmath159 & @xmath160 & @xmath161 & @xmath118 & @xmath162 + chpt & [email protected] & [email protected] & [email protected] & - & - + inverse & [email protected] & [email protected] & [email protected] & 717 mev & 847 mev + amplitude & [email protected] & [email protected] & [email protected] & 680 mev & 804 mev + .2 cm different sets of parameters and methods used in the text . those of the first two lines come from @xmath120 decays @xcite . those of the third line come from data on @xmath120 and @xmath4 together with some unitarization procedure ( for details see ref.@xcite ) . the quoted values of @xmath118 and @xmath162 are calculated with the central values . again we have an expression for the amplitude that behaves correctly with respect to unitarity and that presents the right form in the low energy limit . therefore , we can try to use the actual @xmath2 and @xmath10 masses in order to fit the @xmath4 and @xmath11 phase shifts . we remark once more that nothing ensures that fitting the right masses will give us the right description , since , among other things , the widths of the resonances could be wrong . when dealing with the @xmath1 chiral lagrangian we have more parameters , and the way they appear in the amplitudes is more complicated . let us first start with the @xmath4 scattering partial waves in @xmath1 . as we have commented in section 3.2.1 , in order to avoid confusions with the chpt low - energy parameters , we will denote the parameters of our fit by @xmath163 . the ( 1,1 ) channel only depends on @xmath164 , and will be fixed with @xmath118 . in so doing we get @xmath165 as a consistency check we see that it is within a 1% of @xmath166 which is obtained from the @xmath134 parameters of the @xmath0 case , with the help of eq.[ll ] . once again we use the channels @xmath138 and @xmath139 to determine the best @xmath167 value , which indeed is the same that we would have obtained from the @xmath132 @xmath0 parameter by means of eq.[ll ] . it can be found in table 5 . hence , the best @xmath1 fit of the @xmath4 phase shifts , yields almost the same results as those obtained with @xmath0 and the very same fig.2 remains valid for @xmath1 . nevertheless , when computing the @xmath144 within the @xmath1 formalism , we obtain a much better value than in @xmath0 , which was about @xmath168 mev too high . it is also listed in table 5 finally we will use @xmath169 to fix the correct @xmath10 mass . however , the @xmath10 has an added subtlety , namely , that the mass splitting between different charge states is of the order of 5 mev . this is a small isospin breaking effect that we have not included in our approach . therefore , we have used an average mass @xmath170 mev with an error bar that includes the mass of any @xmath10 state , no matter what its charge may be . that uncertainty has also been taken into account in the @xmath163 error estimates . once we have @xmath169 , we use @xmath171 and eq.[dif ] to obtain @xmath172 . the parameters of this fit have been collected in table 5 , together with @xmath144 and @xmath173 , which can be considered as predictions of the approach . notice , however , that in this case the width of the @xmath10 resonance lies 20% away from its actual value , which nevertheless is a reasonably good result in view of the whole fit in that channel . ' '' '' method & @xmath174 & @xmath175 & @xmath176 & @xmath144 & @xmath173 + inverse & & & & & + amplitude & [ 1ex][email protected] & [ 1ex]1.48 @xmath119 0.33 & [ 1ex]-2.44@xmath119 0.21 & [ 1ex]149.9@xmath119 1.2 mev & [ 1ex]41.2 @xmath119 1.9 mev + .2 cm parameters and results of the @xmath1 iam when @xmath177mev and @xmath178 mev are fixed to their actual values . notice that for @xmath10 we have chosen an average mass between its different charge states . 0.cm concerning the @xmath179 parameters , they are compatible with those in table 4 , which were obtained from low energy data . even more , they are also consistent with other parameters obtained from the iam applied to the form factors of the @xmath180 decays @xcite , which are very well know experimentally : @xmath181 ( notice that in that reference they are using @xmath182 , so that the parameters do necessarily differ ) . nevertheless , it would not make any sense to try to reduce the error bars of these parameters . we consider that the approach that we have been following here can only be consistent within a few percent error level . in order to have a better accuracy it would be necessary to take into account higher order chpt corrections , isospin - breaking effects and the whole approach should be modified following the comments that we made in previous sections . in fig.4 we show the results of the @xmath1 iam fit to the resonance masses , in terms of elastic scattering phase shifts , which we think deserve some comments : * first notice that we are not showing the curves for @xmath4 scattering because they are exactly those in fig.2 . the differences only appear above the two kaon threshold , since in the @xmath1 formulae we are also considering internal loops of kaons and etas . * in the @xmath183 case we can extend the graphs up to 1100 mev , or even more . the reason is that the first two body inelastic threshold is @xmath184 production at 1040 mev and , in contrast to the @xmath4 case , there is no nearby resonance . indeed , the next resonant state in @xmath11 elastic scattering is @xmath185 , very high to affect dramatically our results at @xmath186 mev , but also to be correctly reproduced by the iam method . nevertheless , the existence of the @xmath184 threshold can be noticed in the @xmath187 channel , as a small bump in the curves at precisely 1040 mev . * the shaded area in the @xmath10 channel is not only due to the averaged mass for @xmath10 with @xmath188mev error , but also to the fact that we have to determine several parameters to get the right mass , in contrast with the @xmath2 case , when we only had to fix one . -2 cm 0.cm 0.cm we have explicitly checked that our chpt amplitudes satisfy perturbative unitarity . as it is explained in the appendix , previous calculations @xcite , including ours , did not respect this condition , although by a very small amount . that is why the values of the best parameters for this fit are slightly different from those of our previous work @xcite . .5 cm * phenomenological parameters * .5 cm once we have a good parametrization of @xmath4 and @xmath11 elastic amplitudes , we can use it to obtain the values of some relevant phenomenological parameters . first we can calculate the scattering lengths , which determine the strength of the interactions at low energy . despite our iam fit makes use of high energy data , we expect that it will reproduce the low energy behavior since in the low energy limit it reduces to the chiral expansion , which at @xmath73 already yields quite good values ( see tables 6 and 7 ) . however , as far as the iam is non - perturbative we are also taking into account higher order effects , that will modify the results . indeed , some of these lengths have already been calculated with the iam and it yields slightly better results than plain chpt @xcite . we have made again the calculation with our fit , but as far as we have an estimate of the error bars in the @xmath179 parameters , we will also give the error estimates coming _ only _ from the uncertainties in @xmath179 ( mostly dominated by that of @xmath189 ) . before giving the results , it is convenient to recall that the scattering lengths have two different normalizations . namely : @xmath190 for @xmath4 scattering , where @xmath191 is the c.m . momentum @xmath192 , and @xmath193 for @xmath11 scattering , where now @xmath194[s-(m_k - m_\pi)^2]/4s$ ] . the predictions of our fit for the @xmath4 and @xmath11 scattering lengths are given in tables 6 and 7 ( in @xmath110 units ) . notice that all the values are compatible with the experimental data , and in general they only differ very slightly from the @xmath39 chpt results , usually in the right direction toward the central value . however , the experimental error bars are still too big to arrive at any conclusion . also the error bars in the iam have to be interpreted cautiously , since they are obtained _ only _ from the uncertainties in the @xmath179 parameters . as we have already commented , very recently there has appeared a two loop calculation of @xmath4 scattering within su(2 ) chpt . it estimates @xmath195 or @xmath196 and @xmath197 or @xmath198 , which are precisely the values obtained with our iam fit . this fact gives support to the idea that the iam somehow takes into account higher order terms even at low energies . notice that we do not compare with the two - loop calculation of the scattering lengths and slopes in @xcite because they have used them as an _ input _ in a @xmath199-fit to determine their additional @xmath24 and @xmath25 parameters . therefore , their values are almost exactly those of the experimental data . however , as far as there is no data for the @xmath200 @xmath4 slope parameter , their value can be regarded as a prediction . they give @xmath201 , which is consistent with our result and with @xmath202 , that was obtained from sum rules in @xcite . we have also calculated the phase of the @xmath203 parameter , which measures direct cp violation in @xmath204 decays @xcite . it is related to the @xmath42-wave phase shifts as follows : @xmath205 our result is : @xmath206 very close to @xmath207 which is obtained in plain chpt @xcite . in contrast with the case of the scattering lengths , the value of this angle is not used as an input in @xcite and is therefore a prediction of their best fit . the value they quote is @xmath208 . lccc ' '' '' & & & + @xmath209 & 0.201 & 0.216 @xmath119 0.008 & 0.26 @xmath119 0.05 + @xmath210 & 0.26 & 0.289 @xmath119 0.025 & 0.25 @xmath119 0.03 + @xmath211 & -0.041 & -0.0417 @xmath119 0.0014 & -0.028 @xmath119 0.012 + @xmath212 & -0.070 & -0.075 @xmath119 0.003 & -0.082 @xmath119 0.008 + @xmath213 & 3.6 @xmath214 10@xmath215 & ( 3.744 @xmath119 0.002)@xmath21410@xmath215 & ( 3.8 @xmath119 0.2)@xmath21410@xmath215 + @xmath200 & 0.43@xmath21410@xmath215 & ( 0.515 @xmath119 0.001)@xmath21410@xmath215 & - + @xmath216 & 20 @xmath214 10@xmath217 & ( 17.1 @xmath119 3.5)@xmath21410@xmath217 & ( 17 @xmath119 3)@xmath21410@xmath217 + @xmath218 & 3.5 @xmath214 10@xmath217 & ( 2.8 @xmath119 1.5)@xmath21410@xmath217 & ( 1.3 @xmath119 3.1)@xmath21410@xmath217 + .1 cm @xmath4 scattering lengths . the one - loop chpt results are taken from @xcite . the experimental data come from @xcite . the errors in the iam fit come only from the uncertainties in the parameters . they do not include other theoretical uncertainties . lccc ' '' '' & & & + @xmath219 & -0.043 & -0.049 @xmath119 0.004 & -0.13 ... -0.05 + @xmath220 & - & -0.026 @xmath119 0.003 & - + @xmath221 & 0.148 & 0.155 @xmath119 0.012 & 0.13 ... 0.24 + @xmath222 & - & 0.087 @xmath119 0.016 & - + @xmath223 & 0.012 & 0.0146 @xmath119 0.0012 & 0.017 ... 0.018 + .1 cm @xmath11 scattering lengths . note that the chpt results have been obtained using the corrected formulae in the appendix . the experimental data come from @xcite 0.cm 0.cm finally , in fig.5 we show the phase difference @xmath224 , compared with the hitherto available experimental data @xcite . the difference between the iam and plain chpt at high energies is due to the presence of the @xmath112 resonance . nevertheless , there are also some differences at low energies , since the dispersive approach is somehow taking into account higher order contributions . -2 cm 0.cm 0.cm the main objection to unitarization procedures is the apparent arbitrariness in their predictions , which may differ from one another . in most cases , these methods are nothing but a small modification of the amplitudes so that they can satisfy the unitarity constraint in eq.[uni ] , while keeping at the same time the good low energy behavior . but that constraint is not enough to determine the amplitude completely . thus there are as many unitarization techniques as algebraic tricks to implement such a constraint exactly or to get a better approximation . however , we have already seen in section 3.1 that , below any other inelastic threshold , the inverse amplitude method can be derived directly from the analytic structure of the general two body elastic scattering amplitude . our purpose in this section is to show that , apart from satisfying elastic unitarity , it provides the correct analytic structure required from relativistic quantum field theory . such an structure is not trivial at all and can not be reproduced by other unitarization procedures . both the left and right unitarity cuts are already present in plain chpt , therefore , we will mainly focus on the poles in the second riemann sheet . in the previous section we used the most naive criteria to identify resonances , i.e. , that the phase crosses the @xmath124 value . however , that is only true for the simplest cases . the rigorous characterization of resonances is made in terms of poles in the second riemann sheet of the amplitudes in the @xmath42 complex plane . indeed , when a resonance is produced by just one of these poles , both its mass and width are related to the pole position by @xmath225 provided the width is small enough . in this work we have extended to the @xmath42 complex plane both the @xmath4 and the @xmath11 elastic scattering iam amplitudes obtained in the previous section . notice that the cuts in chpt come from logarithmic functions , so that we have infinite sheets in the complex plane . however , only two of them correspond to the first and second riemann sheets . once we have identified these sheets we can check whether the resonances that we found in previous sections are produced by a pole in the second riemann sheet and thus whether they have a real sound basis . -2 cm 0.cm 0.cm we will first analyze the @xmath226 process . in figure 6 we represent the imaginary part of the amplitude in the complex @xmath42 plane for the three channels @xmath227 and @xmath228 . notice that when we say complex @xmath42 plane , we mean that we have parametrized @xmath42 as @xmath229 , where @xmath230 is the cm energy and is represented in the real axis whereas @xmath231 provides the complex part . on the left column we have displayed the results in the first riemann sheet , whereas in the right column we have continued through the cut to the lower half of the second riemann sheet . in all cases it can be clearly noticed the existence of a cut on the real axis on the first riemann sheet . as we had commented before , a right cut is not anything completely new , since it is already present in one - loop chpt , although in that case , the values that the amplitudes take on it are different . in contrast , the most striking new feature in the iam amplitudes is the appearance of poles in the second riemann sheet and how they determine the amplitude shape for the physical values of @xmath42 . indeed , we have found two poles with @xmath232 in the second riemann sheet , one in the @xmath233 partial wave and another one in @xmath234 . let us start with the second , which clearly corresponds to the @xmath112 resonance . the position of this pole can be obtained from the contour plots in fig.7 , and it is found at around @xmath235 . using eq.[respar ] we see that it is in a good agreement with the @xmath2 mass and width parameters given in table 1 . therefore , we can conclude that this pole is completely consistent with the @xmath2 resonance . the other pole that can be seen in @xmath226 is on the @xmath233 channel . using the parameters of the best @xmath1 iam fit of the previous section , we find that it is located at @xmath236 . it is not responsible for the appearance of any resonance , since it is very far away from the real axis . however , from purely phenomenological fits to pion scattering data it had already been pointed out the existence of such a pole around @xmath237 mev @xcite . this pole is responsible for the strong interaction in that dominates the at low energy the two pion @xmath233 channel . we can now see that even in the channel where there is not an apparent improvement , the iam yields the correct analytic structure . -2 cm 0.cm 0.cm much as it happened in previous sections , the method is not able to reproduce the @xmath105 resonance . as we already commented , the interpretation in terms of poles of this resonance is still controversial . following the same steps as before , we have also identified the four riemann sheets that now appear due to the superposition of two cuts . indeed , we have even implemented the iam derived with the inelastic unitarity condition in section 3.2.4 . we have not found any pole that could hint at the existence of such a resonance . -2 cm 0.cm 0.cm as we have already explained , we should not expect to find anything since the approach is not able to reproduce properly the two kaon unitarity cut and consequently neither its associated sheet structure . let us now address to the @xmath11 elastic scattering case . again , in fig.8 we have displayed the imaginary part of the amplitudes for the @xmath238 and @xmath239 . those pictures on the left represent the first riemann sheet and those on the right , the second . once more the existence of a unitarity cut is clear , but there is also the appearance of a pole in the appropriate channel . in particular , using the third contour plot of fig.7 , we have found a pole in @xmath240 mev , which using eq.[respar ] yields again the mass and width for the @xmath10 resonance that we gave in table 5 . in this work we have shown how the iam provides a consistent technique to accommodate resonances . indeed , based on its derivation from dispersion theory , we have made a systematic analysis of its applicability , which is mainly limited by the existence of two body inelastic thresholds and by the fact that the tree level approximation vanishes in some channels . we have found that it is able to predict the most relevant features of strong elastic scattering , once the chiral parameters are determined from low energy data . quantitatively , the errors are hard to estimate , but we have found in all cases that the mass of the predicted resonances fall within approximately 15% of their actual values . we think this fact gives a sound basis for its application in order to obtain at least a qualitative description of resonances in the strongly interacting symmetry breaking sector . moreover , once we force the iam results to fit the actual resonance mass values , we get a remarkably good fit which is able to reproduce the experimental data up to the next relevant two body inelastic threshold . following that procedure , we have given the unitarized @xmath0 chpt fit to @xmath241 as well as that of @xmath1 to @xmath4 and @xmath11 elastic scattering . for the first time we have estimated the values of the unitarized chiral parameters together with their error bars . these values do not lie very far from those obtained without the iam , and therefore do not spoil the low energy expansion , as can be noticed from the scattering lengths that we have given . with this fit , we have calculated several low - energy phenomenological parameters , like the scattering lengths . our values differ from those obtained at @xmath39 due to the unitarization . however we expect that they include other corrections due to unitarity and resonant effects . we consider that it would not make any sense to try to reduce the error bars in the unitarized parameters within this approach . one has to keep in mind that we have neglected higher order chpt corrections , isospin - breaking contributions , and that we have used high energy data which is very sensible to such effects . it is quite likely that , in order to obtain results consistent to a higher degree of accuracy , the iam in the simple version that has been used here , will not be enough . finally we have also shown how the iam yields the proper analytic structure in the complex @xmath42 plane , in contrast with other unitarization techniques . indeed , we have found that the apparent resonant behavior that is observed on phase shifts , is produced by the corresponding poles in the second riemann sheet , meeting the strict requirements imposed by general relativistic quantum field theory . therefore , we think that the iam and unitarization by means of dispersion theory is the most natural and economic way to extend the applicability of chiral lagrangians . we have seen however , that its main limitations come from the existence of two body inelastic thresholds . nevertheless , work is still in progress in the subject , the iam has been recently applied to other processes and higher order chpt calculations will be soon available . as far as some other physically relevant features do not lie very far from the present applicability limits , it seems very likely that they can be reproduced in the near future . j.r.p would like to thank the theory group at berkeley for their kind hospitality and the jaime del amo foundation for a fellowship . he also wishes to thank m.s.chanowitz for calling his attention on the @xmath242 problem . we have also profited from discussions with u.g . meiner on the @xmath11 elastic scattering amplitudes and m.r.pennington on the adler zero problem and the left cut contribution . we thank d.toublan too , for pointing an error in our calculations as well as c.carone for reading the manuscript . this work has been partially supported by the ministerio de educacin y ciencia ( spain ) ( cicyt aen93 - 0776 ) . partial support by us doe under contract de - ac03 - 76sf00098 is gratefully acknowledged . the first calculation of elastic @xmath11 scattering was later performed in @xcite . these amplitudes were given in terms of physical as well as _ lowest order _ masses and decay constants , which are usually denoted by @xmath243 and @xmath244 , respectively ( @xmath245 being either @xmath246 or @xmath247 ) . of course , the only measurable parameters are the first , and when comparing with experimental observations , one has to eliminate those from _ lowest order _ in terms of the physical constants . indeed , it is possible to find @xcite the relation between @xmath248 and @xmath110 as well as that between @xmath249 and @xmath250 . unfortunately , at _ lowest order _ there is only one @xmath251 , which is related both to @xmath111 and @xmath252 . hence , whenever one finds @xmath251 in an expression there are two choices : either relate it to @xmath111 or to @xmath252 . the difference between the two choices will be one order higher in the chiral expansion . for instance , if one has an @xmath37 expression with @xmath253 , in principle one can substitute it by @xmath254 or @xmath255 . all these choices are equally acceptable . when one is working only with pions , the natural choice is the one that leaves all the expressions in terms of @xmath111 . when one is dealing both with pions and kaons , it is not so obvious . however , once one choice is done for the @xmath256 term , we have to keep it for the @xmath73 contribution , otherwise one would violate perturbative unitarity , eq.[uni ] . surprisingly , in the amplitude in the literature @xcite , which is the one we had also followed in our previous work @xcite , the choice for the @xmath256 term is different from that of the @xmath73 contribution that yields the imaginary part . indeed , the @xmath256 term is written just in terms of @xmath111 whereas @xmath257 is written in terms of @xmath111 and @xmath252 . as a consequence , there is a @xmath258 factor of difference between @xmath259 and @xmath260 . numerically that amounts to a @xmath261 factor . thus , we have rederived from the original work @xcite the amplitudes in terms of physical quantities , ( also correcting some small errata ) so that they satisfy perturbative unitarity . we have chosen to write the formulae symmetrically with respect to @xmath111 and @xmath252 . but the other choices are equally acceptable . the result is : @xmath262\right . \\ \nonumber & + & l_3^r\left [ ( u - m_\pi^2-m_k^2)^2 + ( t-2m_\pi^2 ) ( t-2m_k^2)\right ] + 4l_4^r\left[t(m_\pi^2+m_k^2 ) - 4m_\pi^2m_k^2\right]\\ \nonumber & + & \left . 2l_5^rm_\pi^2(m_\pi^2-m_k^2-s)+8(2l_6^r+l_8^r)m_\pi^2m_k^2 \right\ } \\ \nonumber t^u_4(s , t , u)&= & \frac{1}{4f^2_\pi f^2_k}\left\ { \frac{3}{2}\left [ ( s - t)\left(l_{\pi k}(u)+l_{k\eta}(u ) -u\left(m_{\pi k}^r(u)+m_{k\eta}^r(u)\right)\right ) \right.\right . \\ \nonumber & + & \left.\left(m_k^2 - m_\pi^2)^2(m_{\pi k}^r(u)+m_{k \eta}^r(u ) \right)\right ] + t(u - s)[2m_{\pi\pi}^r(t)+m_{kk}^r(t ) ] \\ \nonumber & + & \frac{1}{2 } ( m_k^2 - m_\pi^2 ) [ k_{\pi k}(u)(5u-2m_k^2 - 2m_\pi^2)+ k_{k \eta}(u)(3u-2m_k^2 - 2m_\pi^2 ) ] \\ \nonumber & + & \frac{1}{8}j_{\pi k}^r(u)\left [ 11u^2 - 12u(m_k^2+m_\pi^2)+4(m_k^2+m_\pi^2)^2 ] + + j_{\pi k}^r(s)(s - m_k^2-m_\pi^2\right)^2 \\ \nonumber & + & \frac{3}{8}j_{k \eta}^r(u ) \left(u - \frac{3}{2}(m_k^2+m_\pi^2)\right)^2 + \frac{1}{2}j_{\pi\pi}^r(t)t(2t - m_\pi^2 ) + \frac{3}{4}j_{kk}^r(t)t^2 \\ \nonumber & + & \left . \frac{1}{2}j_{\eta\eta}^r(t)m_\pi^2 \left(t-\frac{8}{9}m_k^2\right)\right\}\\ \nonumber\end{aligned}\ ] ] the functions @xmath263 , with @xmath264 , can be found in @xcite although they should be written in terms of physical quantities . we want to remark again that this way to write the @xmath11 amplitude is one of several possible choices , since we could have chosen to write everything just in terms of @xmath111 , for example . the important point is to keep the same choice _ both _ for the @xmath256 and the @xmath73 . for completeness , we will also give the @xmath1 formulae used in this work for @xmath4 scattering , because they have also appeared with some minor errata in the literature : @xmath265 + ( t \leftrightarrow u ) \right\}\\ \nonumber c(s , t , u)&=&\frac{4}{f_{\pi}^4 } \left\ { ( 2l_1^r + l_3 ) ( s - 2m_{\pi}^2)^2 + l_2^r [ ( t-2m_{\pi}^2)^2 + ( u - 2m_{\pi}^2)^2 ] + \right . \\ \nonumber & + & \left . ( 4l_4^r+2l_5^r)m_{\pi}^2(s-2m_{\pi}^2 ) + ( 8l_6^r+4l_8^r)m_{\pi}^4 \right\ } \\ \nonumber\end{aligned}\ ] ] g. ecker , j. gasser , a. pich and e. de rafael , _ nuc . * b321 * ( 1989)311 . + g. ecker , j. gasser , h. leutwyler , a. pich and e. de rafael , _ phys . * b223 * ( 1989)425 . donoghue , c. ramirez and g. valencia , _ phys . _ * d39*(1989)1947 . + v. bernard , n. kaiser and u.g . meiner , _ nuc . * b364 * ( 1991)283 . cornwall , d.n . levin and g. tiktopoulos , _ phys * d10 * ( 1974 ) 1145 . + b.w . lee , c. quigg and h. thacker , _ phys . rev . _ * d16 * ( 1977 ) 1519 . + m. veltman , _ acta phys . _ * b8 * ( 1977 ) 475 . chanowitz and m.k . gaillard , _ nucl . * b261 * ( 1985 ) 379 . a. dobado and m.j . herrero , _ phys . * b228 * ( 1989 ) 495 and * b233 * ( 1989 ) 505 . + j. donoghue and c. ramirez , phys . a. dobado , m.j . herrero and t.n . truong , _ phys . _ * b235 * ( 1989 ) 129 . + a. dobado , m.j . herrero and j. terrn , _ z. phys . _ * c50 * ( 1991 ) 205 and _ z. phys . _ * c50 * ( 1991 ) 465 . + s. dawson and g. valencia , _ nucl . * b352 * ( 1991)27 . + a. dobado , d. espriu and m.j . herrero , _ phys . lett . _ * b255*(1991)405 . + d. espriu and m.j . _ * b373 * ( 1992)117 . k.l.au , d.morgan and m.r.pennington , _ phys . * d35 * ( 1987)1633 . + d. morgan and pennington , _ phys . rev . _ * d48 * ( 1993 ) 1185 . + g.jansen et al . , _ phys . _ * d52 * ( 1995)2690 . + n.tornqvist and m.roos _ phys . * 76 * ( 1996)1575 . basdevant , c.d . froggat and j.l.petersen , _ nucl . phys . _ * b72*(1974)413 . + j.l.basdevant , p.chapelle , c.lpez and m.sigelle,_nucl . _ * b98*(1975)285 . + c.d.froggat and j.l . petersen , _ nucl . phys . _ * b129*(1977)89 . + j.l.petersen , _ the @xmath4 interaction_. cern yellow report 77 - 04 ( 1977 ) .

since tassiulas and ephremides proposed the maximum weight scheduling algorithm of throughput - optimality for constrained queueing networks in 1992 , extensive research efforts have been made for resolving its high complexity issue under various directions . in this paper , we resolve the issue by developing a generic framework for designing throughput - optimal and low - complexity scheduling algorithms . under the framework , an algorithm updates current schedules via an interaction with a given oracle system that can generate a solution of a certain discrete optimization problem in a finite number of interactive queries . therefore , one can design a variety of scheduling under this framework by choosing different oracles , e.g. , the exhaustive search ( es ) , the markov chain monte carlo ( mcmc ) , the belief propagation ( bp ) and the cutting - plane ( cp ) algorithms . the complexity of the resulting algorithm is decided by the number of operations required for an oracle processing a single query , which is typically very small . somewhat surprisingly , we prove that an algorithm using any such oracle is throughput - optimal for general constrained queueing network models that arise in the context of emerging large - scale communication networks . in particular , the ` pick - and - compare ' algorithms developed by tassiulas in 1998 and recently developed queue - based csma algorithms can be also understood as special cases of such algorithms using es and mcmc oracles , respectively . to our best knowledge , our result is the first that establishes a rigorous connection between iterative optimization methods and low - complexity scheduling algorithms , which we believe provides various future directions and new insights in both areas .

1407.3694

the problem of dynamic resource allocation in constrained queueing networks is fundamental to modern communication networks . this has long been the subject of significant research , where the @xcite of tassiulas and ephremides proposed the mw ( maximum weight ) scheduling algorithm of throughput - optimality . it has been also a powerful paradigm in cross - layer control and resource allocation @xcite . however , the algorithm can not be implemented in practice due to its high algorithmic complexity . been made since then for developing algorithms of lower complexity while maintaining the same performance guarantee . examples of such algorithms include greedy algorithms @xcite and random access algorithms @xcite . recently , there have been exciting progresses for the topic that even fully distributed random access algorithms based on csma ( carrier sense multiple access ) can achieve the desired high performance , i.e. , throughput - optimality , in wireless interference network models , e.g. , see @xcite . the main intuition underlying these results is that nodes in a network can adjust their csma parameters dynamically using local information such as queue - length so that they can simulate the mw algorithm asymptotically for throughput - optimality . however , it is known that the queue - based csma algorithms suffer from very poor delay performance . there have been also significant efforts formally addressing the limitation in the delay performance of the queue - based csma algorithms @xcite as well as designing better alternatives in delay under certain scenarios @xcite . in addition , another issue is that the throughput - optimality proof in @xcite is very sensitive to the underlying wireless network model . namely , it is not an easy task to design similar csma - like throughput - optimal algorithms in other network models since one has to reproduce painful , long and routine proof - arguments for throughput - optimality in each different model . the main goal of this paper is to resolve these issues of the recent queue - based csma algorithms by developing a general framework that can produce a variety of throughput - optimal and low - complexity scheduling algorithms , not to the underlying network models . design and analysis of resource allocation algorithms in constrained queueing networks have been of great interest for last four decades across various research communities , where the contention resolution protocols in multiple - access broadcast channels have been studied since the mid-1970s @xcite . the broadcast channels induce a special constraint in resources usages among queues ( i.e. , every queue contend with all other queues ) , and simple distributed contention resolution protocols gained much attention in the ages . it is the year 1992 that the seminal work by tassiulas and ephremides proposed a throughput - optimal algorithm , referred to as mw , for general resource constraints . its main appeal is that it does require only information on the current queue lengths ( and instantaneous service rates ) and does not rely on the knowledge of underlying system parameters . however , as we mentioned earlier , the mw algorithm suffers from the high implementation complexity , and a huge array of subsequent research has been made to develop algorithms with high performance guarantees and low - complexity . maximal scheduling or longest - queue - first algorithm are low - complexity alternatives to mw , but they achieve only some fraction of the maximal throughput region . parallel iterative matching @xcite and islip @xcite were shown to be 50% throughput - optimal @xcite . on the other hand , kumar et al . @xcite and dimakis and walrand @xcite identified sufficient conditions on the network topology for throughput - optimality . those conditions were further analyzed to obtain fractional throughput results about a class of wireless networks by joo et al . @xcite and leconte et al . these algorithms are generally not throughput - optimal and require multiple rounds of message exchanges between nodes . simpler or distributed implementations of mw has been also extensively studied in the literature . tassiulas @xcite provides linear - complexity versions of mw , called ` pick - and - compare ' , but it is known that they suffer from the bad delay performance . the works on this line also include a variant of mw by giaccone , prabhakar and shah @xcite and a gossip - based algorithm by modiano , shah and zussman @xcite . however , these algorithms are specific to certain network models and still require high information ( or message ) exchanges for each new scheduling decision . a few years ago , it was proven that even fully distributed queue - based random access algorithms based on csma with very little message passing can simulate the mw algorithm asymptotically and hence achieve throughput - optimality in wireless interference network models @xcite . however , the csma algorithms are known to suffer from the following two issues : the bad delay performance and the limited applicability to other network models . for the delay issue , bouman et al . @xcite formally established explicit lower bounds on the delay of queue - based random access algorithms . on the other hand , significant efforts have also been made to design csma algorithms with better delay performance , e.g. , see @xcite . the limited applicability issue is rather technical : it is a technically tedious task to design similar csma - like throughput - optimal algorithms in other network models since one has to reproduce painful , long and routine proof - arguments for analyzing ` time - varying ' markov chains in each different model , e.g. , see the buffered circuit switched network model @xcite . the main motivation of this paper is for resolving these issues and goes beyond csma - like algorithms for achieving high performance guarantees and low implementation complexity , not sensitive to the underlying network models . we develop a generic framework for designing throughput - optimal and low - complexity scheduling algorithms , where queue - based csma algorithms belong to it as special cases . under the framework , an algorithm updates current schedules via an interaction with a given oracle system that can generate a solution of a certain discrete optimization problem in a finite number of interactive queries . specifically , an algorithm maintains three parameters , ` schedule ' , ` weight ' and ` advice ' , at each time instance . the schedule and advice are kept being updated under interactions with the oracle system and the weight is decided by some function of current queue - size , which we call ` queue - length - function ' . therefore , the scheduling algorithm design is determined by ` which oracle is used ' and ` which queue - length - function is chosen ' . the oracle system can be designed through iterative optimization methods including the naive exhaustive search algorithm , the markov chain monte carlo ( e.g. , the glauber dynamics @xcite ) , the message - passing inference algorithms ( e.g. , the belief propagation @xcite ) and the linear / integer programming ( e.g. , the cutting - plane method @xcite ) . in this paper , we describe three oracle examples , the exhaustive search ( es ) , the markov chain monte carlo ( mcmc ) and the belief propagation ( bp ) in section [ sec : exam ] . in particular , the ` pick - and - compare ' algorithms @xcite and queue - based csma algorithms can be understood as such algorithms ( with minor modifications ) using es and mcmc oracles , respectively ( see section [ sec : exam ] ) . the scheduling complexity is decided by the number of operations required for an oracle processing a single query , which is typically very small , e.g. , es , mcmc and bp . we prove that given any oracle system , there exists a queue - length - function such that an algorithm using the oracle is throughput - optimal under general constrained queueing network models , which includes wireless network models , input - queued switch models , the buffered circuit switched network model , etc . in particular , we first show there exists a queue - length - function such that its value does not change too often ( i.e. , remains fixed for a long time ) and hence the mw optimization problem can be solved through ( enough ) interactions to the oracle , i.e. , the algorithm simulates mw and is throughput - optimal . this proof strategy is similar to those of @xcite in spirit . the main difference is that we force the queue - length - function value being ` rarely - varying ' in integers , while the authors in @xcite do it being ` slowly - varying ' in real numbers . due to such a difference , now we do not need to analyze ` time - varying ' algorithms ( e.g. , markov chains ) , i.e. , analysis when algorithm parameters are fixed is enough . this makes the proof much simpler and not sensitive to the given oracle and underlying network model . our strong theoretical statement provides a quite powerful paradigm for developing a variety of low - complexity scheduling algorithms of high performance . for example , our experimental results show that the delay performance of such an algorithm using the bp oracle is close to that of the mw algorithm , and it significantly outperforms csma algorithms ( see section [ sec : exp ] ) . moreover , one can also easily design other types of throughput - optimal algorithms , via designing various oracle systems , under a quite broad class of constrained queueing network models , without suffering from repeating tedious proofs sensitive to the underlying assumptions and setups . at a high level , we use the interactive oracle concept to provide a ` black - box ' tool for designing scheduling algorithms of high performance and low - complexity . this establishes a rigorous connection between iterative optimization methods and low - complexity scheduling algorithms , which we believe provides various future directions and new insights in both areas . in section [ sec : pre ] , we describe the constrained queueing network model of our interest and the performance metric ( i.e. , throughput - optimality ) for scheduling algorithms . section [ sec : main ] provides the main result of this paper , i.e. , our generic framework for designing a throughput - optimal and low - complexity scheduling algorithm which decides its current schedule via an interaction with an oracle system . in section [ sec : exam ] , we provide several examples of scheduling algorithms under the framework . section [ sec : pfmainthm ] provides the formal proof of the main result , and our experimental results are presented in section [ sec : exp ] . we consider a stochastic network of @xmath0 queues represented by @xmath1 with @xmath2 , and time is discrete indexed by @xmath3 . external packets ( or jobs ) arrive at each queue as per an exogenous stochastic process , where we let @xmath4 denote the number of packets arriving at queue @xmath5 during the time interval @xmath6 . we assume that @xmath7 are independent bernoulli random variables with @xmath8=\lambda_i\ ] ] for some @xmath9 $ ] , called the _ arrival rate _ for queue @xmath5 . packets in queues are served at unit rate subject to scheduling constraints . formally , a schedule is denoted by @xmath10\in \{0,1\}^n$ ] , where @xmath11 means queue @xmath5 is transmitting ( or active ) and @xmath12 otherwise . we let @xmath13 be a set of all available schedules and let @xmath14 be the schedule during @xmath6 for @xmath15 . to describe the dynamics in the network , we let @xmath16 denote the number of packets in @xmath5-th queue at time @xmath17 ( i.e. , _ length of queue _ ) and let @xmath18 $ ] be the vector of queue lengths . then the queueing dynamics for queue @xmath5 can be described as @xmath19 where @xmath15 and @xmath20 for given @xmath1 and @xmath21 , our goal is to develop low - complexity and high - performance scheduling algorithms which determine @xmath22 for each time instance @xmath15 . a scheduling algorithm is required to choose @xmath14 at each time instance @xmath15 . we say an algorithm is throughput - optimal if it keeps queues finite for as large a set of arrival rates @xmath23 $ ] as possible . to describe it formally , we define the capacity region as follows : @xmath24 where it is the convex hull of the feasible scheduling set @xmath21 . since @xmath14 , the effective ` service ' rate induced by any algorithm over time is essentially in @xmath25 . therefore , if arrival rates @xmath26 can be served by any algorithm , then there exists @xmath27 such that @xmath28 component - wise . we say @xmath26 is _ ( strictly ) admissible _ if @xmath29 is ( strictly ) less than @xmath30 for some @xmath31 . let @xmath32 and @xmath33 be the set of all admissible and strictly admissible rates , respectively : @xmath34 a throughput - optimal algorithm keeps queues finite for any @xmath35 , where there can be many different ways to argue the finiteness of queues . in this paper , we use the following popular definition of throughput - optimality . [ def : throughput - optimality ] a scheduling algorithm is called _ throughput - optimal _ if for any @xmath35 the ( appropriately defined ) network markov chain is positive recurrent . if the state of the network markov chain includes queue - sizes , then its positive recurrence implies that the queues remain finite with probability 1 , which motivates the above definition . we recall the definition of the positive recurrence of a markov chain @xmath36 on a countable state space @xmath37 . a state @xmath38 is said to be _ recurrent _ if @xmath39=1 $ ] , where @xmath40 . if @xmath41<\infty$ ] , a recurrent state @xmath42 is called _ positive recurrent_. if a markov chain is irreducible and it has a positive recurrent state , all states are positive recurrent . in this case , we call such a markov chain positive recurrent . one way to show the positive recurrence of a markov chain is using the following lyapunov - foster criteria . a function @xmath43 is called a ` lyapunov function ' if @xmath44 . the following theorem provides a criteria about the positive recurrence using a lyapunov function , e.g. , see @xcite . [ thm : lyapunov ] consider a lyapunov function @xmath45 such that @xmath46 is finite for any @xmath47 . suppose there exist functions @xmath48 such that @xmath49\leq -\kappa({\mathsf{x}}),~\forall x\in\omega,\ ] ] and they satisfy the following conditions : 1 . 2 . @xmath51 . @xmath52 for all @xmath47 . 4 . @xmath53 . then , @xmath54 is positive recurrent . this section presents our main result , throughput - optimal and low - complexity scheduling algorithms . we first describe the generic concept of our algorithms utilizing certain oracle systems . the oracle systems are motivated by ( randomized or deterministic ) iterative algorithms which find an approximate optimal solution to the following optimization problem : @xmath55 typically , in an iterative algorithm , the output at each iteration depends only on information from the previous iteration ( and weight @xmath56 ) . we call such information ` advice ' . one can understand an iteration of the algorithm as an interaction to an oracle system , where a query consisting of advice and weight is sent to an interactive oracle system , and the oracle returns a schedule together with an updated advice for the next iteration . this naturally suggests the following scheduling algorithm : at each time instance @xmath17 , ( a ) it sends a query consisting of the advice @xmath57 produced in the previous step and weight @xmath58 decided by current queue lengths to the oracle system , and then ( b ) the oracle returns a schedule to use at the next time instance @xmath59 . hence , the complexity of the scheduling algorithm is that of the oracle system processing a query ( equivalently , that of one iteration of an iterative algorithm ) , which is typically very small as we provide examples in section [ sec : exam ] . in the following sections , we provide the formal descriptions of oracle systems of interest and scheduling algorithms utilizing them . and weight @xmath58 is sent to the oracle . the oracle returns the schedule @xmath60 to use at the next time instance and updated advice @xmath61 . , width=288 ] we consider a network model introduced in section [ subsec : network_model ] with set of queues @xmath1 with @xmath2 and available schedules @xmath62 . then the oracle system we are interested in has an advice space @xmath63 with the following input and output : * it takes inputs @xmath64 and @xmath65 $ ] , * it outputs ( or returns ) @xmath66 and @xmath67 , where @xmath56 , @xmath68 , and @xmath69 are called ` weights ' on queues , ` advice ' , and ` updated advice ' , respectively . formally , let the outputs @xmath70 and @xmath71 , where the oracle can generate random outputs in general . we assume that the oracle system satisfies the following properties : 1 . for any @xmath72 , if @xmath73 is large enough , then there exists @xmath74 such that for any @xmath75 and advice @xmath64 , @xmath76 where @xmath77 is the function composing @xmath78 ` @xmath17 times ' , i.e. , @xmath79 . 2 . for any @xmath56 , @xmath80 the first assumption * a1 * implies that the schedule @xmath30 generated by the oracle system after @xmath81 interactions is an approximate solution to the optimization problem . the second assumption * a2 * is merely for a technical purpose : it provides the countability of the underlying space of the network markov chain which we define in the following section . in this section , we describe our scheduling algorithm given an oracle system . to this end , we let @xmath82 be positive - real - valued functions and @xmath83 where @xmath84 . the algorithm maintains parameters @xmath85 . it updates the schedule @xmath22 and parameters @xmath86 under an interaction with the oracle system as follows : ' '' '' * @xmath87 , * @xmath88 , * @xmath89 is the closest integer to @xmath90 if @xmath91 and @xmath92 otherwise . ' '' '' figure [ fig : oracle ] illustrates the above scheduling algorithm using an interactive oracle system . for each time @xmath93 , we let @xmath94 then , @xmath95 becomes a markov chain on the following state space @xmath37 : @xmath96 then , @xmath37 is countable due to the assumption * a2 * of the oracle system . now we are ready to state the main result in this paper , where its proof is presented in section [ sec : pfmainthm ] . [ thm : mainresult ] the above scheduling algorithm is throughput - optimal , i.e. , @xmath54 is positive recurrent for any @xmath35 , if functions @xmath97 satisfy the assumptions * a1-a2 * and the following conditions : 1 . @xmath98 and @xmath99 are increasing , differentiable and concave . 2 . @xmath100 and @xmath101 . 3 . @xmath102 . 5 . for any fixed @xmath104 , @xmath105 6 . there exists @xmath106 such that , for any fixed @xmath104 , @xmath107 in the following section , we provide several specific examples of throughput - optimal and low - complexity scheduling algorithms under theorem [ thm : mainresult ] . in this section , we show the wide - applicability of our framework by designing several throughput - optimal and low - complexity scheduling algorithms under it . in particular , we provide examples of low - complexity oracle systems with corresponding explicit functions @xmath97 so that they satisfy the assumptions * a1-a2 * and the conditions * c1-c6 * of theorem [ thm : mainresult ] . the first oracle system we introduce utilizes an exhaustive search ( es ) algorithm to find a maximum weight schedule @xmath66 , i.e. , it maximizes @xmath108 over @xmath66 for given @xmath1 with @xmath2 , @xmath62 , and @xmath109 . an es algorithm keeps a search - index @xmath110 which may not be in @xmath21 , and current maximum schedule @xmath66 . at each iteration , if @xmath111 is in @xmath21 and its weight is greater than the weight of @xmath30 , replace @xmath30 by @xmath111 . in addition , it updates the search index @xmath111 by @xmath112 where @xmath113 can be considered as a non - negative integer less than @xmath114 and @xmath112 is naturally defined under the interpretation . * es oracle system . * formally , the es oracle system has an advice space @xmath115 . for an advice @xmath116 and weights @xmath117 , the oracle outputs @xmath118 and @xmath119 are decided as follows : ' '' '' * set @xmath120 * set @xmath121 . ' '' '' since it compares all schedules in @xmath21 exhaustively , after @xmath114 interactions with the oracle system , it returns a schedule @xmath66 whose weight is maximum . thus we have , for all @xmath64 and @xmath122 , @xmath123 which implies that we can take @xmath124 for the the assumption * a1 * , i.e. , @xmath81 is a constant function with respect to @xmath125 . in addition , the assumption * a2 * holds obviously . using these facts and theorem [ thm : mainresult ] , the following corollary can be derived . [ cor : es ] the scheduling algorithm described in section [ sec : algorithm_description ] using the es oracle system is throughput - optimal if @xmath126 it is elementary to check the conditions * c1 * -*c5 * of theorem [ thm : mainresult ] , where @xmath124 . the condition * c6 * of theorem [ thm : mainresult ] can be derived as follows : for @xmath127 , @xmath128 this completes the proof of corollary [ cor : es ] . the second oracle system we introduce is the markov chain monte carlo ( mcmc ) algorithm . for a detailed description of the oracle , we consider a special setup of available schedules @xmath21 . [ fig : wireless ] queues in grid interference topology . available schedules are @xmath129 , @xmath130 , @xmath131 , @xmath132 , @xmath133 , @xmath134 , etc.,title="fig:",width=240 ] * wireless network model . * we consider an undirected graph @xmath135 and the set of available schedules @xmath21 is defined as : @xmath136 the above scheduling constraint has been popularly used for modeling interference in wireless networks @xcite , i.e. , two queues @xmath5 and @xmath137 can not transmit simultaneously if they share an edge in the ` interference ' graph @xmath138 . we let @xmath139 denote the set of interfering neighbors of queue @xmath5 . * mcmc oracle system . * in the mcmc oracle system , the advice space is @xmath140 and for an advice @xmath141 , the oracle outputs @xmath118 and @xmath142 are decided randomly as follows : ' '' '' 1 . choose a queue @xmath5 uniformly at random , and set @xmath143 2 . if @xmath144 for some @xmath145 , then set @xmath146 . otherwise , set @xmath147 @xmath148 where @xmath149 . 4 . for each @xmath150 , * if @xmath11 , then set @xmath151 * else if @xmath152 for all @xmath145 , then set @xmath153 * otherwise , set @xmath154 . ' '' '' then , using known results on the mixing time of mcmc , one can prove that the assumption * a1 * holds with @xmath155 where @xmath156 are some ( ` @xmath0-dependent ' ) constants independent of @xmath157 . the proof of is a direct consequence of lemma 3 and lemma 7 in @xcite , and we omit the details due to space constraints . the assumption * a2 * also holds obviously again . using these facts and theorem [ thm : mainresult ] , the following corollary can be derived . [ cor : mcmc ] the scheduling algorithm described in section [ sec : algorithm_description ] using the mcmc oracle system is throughput - optimal if @xmath158 it is elementary to check the conditions * c1 * -*c4 * of theorem [ thm : mainresult ] . the condition * c5 * is from and @xmath159 : @xmath160 furthermore , the condition * c6 * can be derived as follows : this completes the proof of corollary [ cor : mcmc ] . [ cor : mcmc ] the scheduling algorithm described in section [ sec : algorithm_description ] using the mcmc oracle system is throughput - optimal if @xmath161 where @xmath162 . it is elementary to check the conditions * c1 * -*c4 * of theorem [ thm : mainresult ] . for the condition * c5 * , @xmath163 which comes from and @xmath164 . then @xmath165 and it implies the condition * c5*. the condition * c6 * can be derived as follows : for large enough @xmath166 , @xmath167 . combining this with , we have that @xmath168 it completes the proof of corollary [ cor : mcmc ] . we note that the scheduling algorithm described in section [ sec : algorithm_description ] using the mcmc oracle system is the discrete - time version of the csma algorithm in @xcite . the third oracle system we introduce is the belief propagation ( bp ) algorithm , where bp is a popular heuristic iterative algorithm for solving inference problems arising in graphical models @xcite . for the provable throughput - optimality of the scheduling algorithm using the bp oracle system , we consider a special setup of available schedules @xmath21 which is described in what follows . * input - queued switch model . * we consider the undirected complete bipartite graph of left - vertices @xmath169 , right - vertices @xmath170 , and edges @xmath171 , where @xmath172 then , the set of queues is @xmath173 of @xmath174 queues and @xmath175 where this is a popular model for input - queued switches @xcite , and also a special case of the wireless network model described in the previous section . input ports and @xmath176 output ports , there are @xmath177 queues . available schedules are @xmath178 , @xmath179 , @xmath132 , @xmath133 , @xmath180 , and @xmath181.,width=240 ] * bp oracle system . * in the bp oracle system , the advice space is @xmath182 and for an advice @xmath183 , where @xmath184 $ ] , the oracle outputs @xmath118 and @xmath185 are decided as follows : ' '' '' * for each @xmath186 , set @xmath187 where @xmath188 * if @xmath189 , reset @xmath190 . ' '' '' in the above , one can set @xmath191\in [ 0,1]^{|\mathcal e|}$ ] such that @xmath192 is unique , and @xmath193 for some constant @xmath194 , e.g. , let @xmath195 be a random number in the interval @xmath196 $ ] . then , from the works by bayati et al . @xcite or sanghavi et al . @xcite , it follows that the assumption * a1 * holds with @xmath197 for the assumption * a2 * , one can observe that @xmath198 which is obviously a finite set given @xmath56 . using these facts and theorem [ thm : mainresult ] , the following corollary can be derived . in the bp oracle system , the advice space is @xmath199 and for an advice @xmath200\in \mathbb a$ ] , the oracle outputs @xmath118 and @xmath201 are decided as follows : * for each @xmath202 , set @xmath203 where @xmath204 * if @xmath189 , reset @xmath205 by an arbitrary schedule in @xmath21 . in the above , we carefully design @xmath206 so that ( a ) @xmath192 is unique , ( b ) @xmath207 and ( c ) @xmath208 . then , from the work by bayati et al . @xcite , it follows that the assumption * a1 * holds with @xmath209 for the assumption * a2 * , one can observe that @xmath210 which is obviously a finite set given @xmath56 . using these facts and theorem [ thm : mainresult ] , the following corollary follows . the ( max - product ) bp is an iterative , message - passing algorithm for finding @xmath66 which maximizes @xmath108 . the algorithm maintains messages @xmath211 at the @xmath17-th iteration , and they are updated as : @xmath212 where @xmath213 is @xmath166 if @xmath214 and @xmath215 otherwise . after enough iterations , the algorithm outputs @xmath216 if @xmath217 , and @xmath218 otherwise . bayati et al . @xcite establish the following result . * if the desired maximizer @xmath219 is unique , the bp algorithm outputs it for all @xmath220 , where @xmath221 is the difference between the weights of the maximum weight schedule and the second maximum weight schedule , i.e. , @xmath222 . the unique condition in the above theorem can be removed by adding small noises to weights . for example , for integer weight @xmath56 , by adding @xmath223 for the @xmath5-th weight , we guarantee unique maximizer . in our model , for @xmath224 and @xmath225 , we add @xmath226 to @xmath227 weight and we get the following conclusion : * for integer weight @xmath56 , the bp algorithm outputs @xmath219 for all @xmath228 . now we explain how the bp algorithm can be understood as an oracle system . an advice is represented by @xmath229 and , with input @xmath68 , the oracle system outputs following @xmath230 and @xmath231 : @xmath232 namely , the iterative bp mechanism can be thought as an interactive oracle system , where messages corresponds to advices . due to * bp2 * , we have @xmath233 and by theorem [ thm : mainresult ] , the following corollary follows . [ cor : bp ] the scheduling algorithm described in section [ sec : algorithm_description ] using the bp oracle system is throughput - optimal if @xmath234 it is elementary to check the conditions * c1 * -*c5 * of theorem [ thm : mainresult ] , where @xmath235 . the condition * c6 * of theorem [ thm : mainresult ] can be derived as follows : for @xmath236 , @xmath237 where @xmath238 is some constant depending on @xmath239 and the last equality is from @xmath240 and @xmath241 . this completes the proof of corollary [ cor : bp ] . we note that there are many different ways to design the bp oracle , for example , one can consider the following variant . ' '' '' * for each @xmath186 , set @xmath242 * choose @xmath186 so that @xmath243 is the largest one among those which @xmath244 after resetting @xmath245 . * reset @xmath245 and keep this procedure until no more edge is found in the step 2 . ' '' '' the scheduling algorithm described in section [ sec : algorithm_description ] using the above second bp oracle system is also throughput - optimal for the same set of functions @xmath246 in theorem [ cor : bp ] , where its proof is identical to that of the first bp oracle system . the main difference in the second bp oracle is that it decides the schedule @xmath247 greedily based on the ` belief ' @xmath248 $ ] , while the first one simply checks whether @xmath243 is positive or not . a similar version of the scheduling algorithm using the second oracle system was actually studied in @xcite heuristically , where by carefully choosing queue - length functions , theorem [ thm : mainresult ] provides its formal throughput - optimality proof which is missing in @xcite . in this section , we introduce the fourth oracle system combining two mechanisms , bp and es , in the wireless interference network model described in section [ sec : mcmc_wireless ] . we call it the bp - es oracle system . * bp - es oracle system . * in the oracle system , the advice space is @xmath249 , and for an advice @xmath250 where @xmath251 $ ] , the oracle outputs @xmath118 and @xmath252 are decided as follows : ' '' '' * for each edge @xmath186 , set @xmath253 * choose a queue @xmath5 uniformly at random , and set @xmath254 * set @xmath255 where @xmath256 $ ] is defined as @xmath257 ' '' '' as in section [ sec : mcmc_wireless ] , we consider an undirecteted graph @xmath258 and let @xmath259 the first algorithm we utilize for this model is exhaustive search ( es ) which is introduced in section [ sec : es ] and for fixed weight @xmath56 , it find a maximum weight scheduling in @xmath114 iterations . another algorithm is belief propagation ( bp ) for maximum weight independent set . in this algorithm , at every iteration @xmath17 , each node @xmath5 sends a message @xmath260 to each neighbor @xmath145 . each node also maintains a belief @xmath261 . the message and belief updates are as follows : * initially , set @xmath262 for all @xmath263 . * update messages : @xmath264 * for each @xmath150 , @xmath265 * output @xmath22 as @xmath266 bp algorithm for wireless networks sometimes finds a maximum weight schedule but it does not always converge . so , we combine two algorithms and call it es+bp . in the above , the second and third steps can be implementable in distributed fashions using poisson clocks @xcite and gossip algorithms for distributed averaging @xcite . the bp - es oracle system outputs schedule @xmath205 as the maximum weight one among three schedules @xmath30 , @xmath111 and @xmath267 , where @xmath30 is the current maximum schedule , @xmath111 is the es search index and @xmath267 is decided by the bp mechanism . since the search index @xmath111 is decided randomly , the bp - es oracle finds an maximum weight schedule with probability @xmath268 after @xmath269 interactions for the the assumption * a1*. the assumption * a2 * holds with similar reasons as before . using these facts and theorem [ thm : mainresult ] , the following corollary can be derived . [ cor : esbp ] the scheduling algorithm described in section [ sec : algorithm_description ] with the bp - es oracle system is throughput - optimal if @xmath270 the proof of corollary [ cor : esbp ] is almost identical to that of corollary [ cor : es ] . this section provides the proof of theorem [ thm : mainresult ] , i.e. , we show that @xmath36 described in section [ sec : algorithm_description ] is positive recurrent , as long as @xmath271 and @xmath97 satisfy the assumptions * a1-a2 * and the conditions * c1 * -*c6 * stated in theorem [ thm : mainresult ] . towards this , we shall construct a lyapunov function @xmath169 and auxiliary functions @xmath272 , @xmath273 so that they satisfy the conditions of theorem [ thm : lyapunov ] . first , note that the state space @xmath37 is countably infinite from the assumption * a2 * of the oracle system . for @xmath274 , we consider the following function : @xmath275 where @xmath276 . since @xmath277 from the condition * c2 * , we have @xmath278 as @xmath279 . thus , @xmath280 and @xmath43 is a lyapunov function . with an abuse of notation , we use @xmath281 to denote @xmath282 . now , given @xmath271 , there exists @xmath283 and @xmath284\in [ 0,1]^{|s|}$ ] so that @xmath285 for a state @xmath274 , we consider the following functions @xmath272 and @xmath273 : @xmath286 where @xmath287 is a constant in the condition * c6 * , @xmath288 is the largest integer not greater than @xmath166 , and @xmath289 are some constants satisfying @xmath290 for example , one can choose @xmath291 . using the above functions , we shall establish the following lemma . [ lemma : negative_drift ] for every initial state @xmath38 with large enough @xmath292 and every arrival rate @xmath35 , it follows that @xmath293\,|\,{\boldsymbol{x}}(0)={\mathsf{x}}]\leq -\kappa({\mathsf{x}}).\label{eq : negative_drift}\ ] ] the proof of the above lemma is presented in section [ sec : pf : lemma : negative_drift ] . in the remainder of this section , we check that @xmath272 and @xmath273 satisfy the desired conditions * l1-l4 * of theorem [ thm : lyapunov ] , which leads to the positive recurrence of the markov chain @xmath95 . first , it is easy to check that @xmath294 is finite . without loss of generality , we can also assume that holds for every @xmath38 by letting @xmath295 if @xmath292 is not large enough . from the conditions * c2 * and * c4 * , we have @xmath296 as @xmath297 . since @xmath98 is an unbounded increasing function due to * c1 * and * c2 * , we have @xmath298 as @xmath297 . thus , the first two conditions * l1 * and * l2 * are satisfied . by the definition of function @xmath272 , it is not hard to check that @xmath272 increases as @xmath299 does . therefore , it follows that @xmath300 which implies the condition * l3*. finally , as @xmath278 ( i.e. , @xmath297 ) , @xmath301^{-1}\to 0,\end{aligned}\ ] ] which implies the condition * l4*. this completes the proof of theorem [ thm : mainresult ] . to begin with , we describe the main steps to prove lemma [ lemma : negative_drift ] at a high level : 1 . for given initial state @xmath42 with large enough @xmath299 , @xmath58 does not change many times in @xmath302 $ ] , which is formally stated in lemma [ lemma : stopping_time ] . the schedule @xmath22 is close to a max - weight one with respect to @xmath303 ` mostly ' in the time interval @xmath304 $ ] , which is formally stated in lemma [ lemma : bernoulli_main ] . the first step plays a key role for establishing the second step : since @xmath58 does not change often , the interactive oracle system can find a max - weight schedule with respect to @xmath305 . the max - weight property in the second step will lead to the negative drift property in lemma [ lemma : negative_drift ] . now we proceed toward proving lemma [ lemma : negative_drift ] formally . let @xmath306 is the time at which the @xmath307-th change of @xmath58 occurs , i.e. , @xmath58 remains fixed during the time interval @xmath308 . formally , let @xmath309 and for @xmath310 , iteratively define @xmath311 in the following lemma , for initial state @xmath42 with large enough @xmath299 , we show that @xmath58 can be changed at most @xmath0 times in time interval @xmath302 $ ] , i.e. , @xmath312 . [ lemma : stopping_time ] for any initial state @xmath313 with large enough @xmath299 , the following event occurs with probability @xmath314 : @xmath312 and @xmath315,\label{eq : bound_of_q_max}\end{aligned}\ ] ] where @xmath287 is the constant in the condition * c6 * of theorem [ thm : mainresult ] . since @xmath58 remains fixed for @xmath316 , the assumption * a1 * implies that that with high probability , @xmath22 is close to the max - weight schedule with respect to @xmath58 for @xmath317 using this observation with lemma [ lemma : stopping_time ] , we obtain the following lemma that with high probability , the schedule @xmath22 is closed to a max - weight one with respect to @xmath303 ` mostly ' in the time interval @xmath304 $ ] . [ lemma : bernoulli_main ] for any @xmath318 and initial state @xmath274 with large enough @xmath299 , it follows that @xmath319~\geq~1-\beta,\ ] ] where @xmath320\,:\\ & { \boldsymbol{\sigma}}(t)\cdot f({\boldsymbol{q}}(t))\geq ( 1-\eta)\max_{{\boldsymbol{\rho}}\in s}{\boldsymbol{\rho}}\cdot f({\boldsymbol{q}}(t))\bigg\}. \end{aligned}\ ] ] in the remainder of this section , we shall establish the proof of lemma [ lemma : negative_drift ] using lemma [ lemma : stopping_time ] and lemma [ lemma : bernoulli_main ] . to this end , we first recall that @xmath321 from the conditions * c1 * and * c4 * , we have @xmath322 where @xmath323 $ ] . in addition , from , we have an upper bound on the expectation of the first term in : @xmath324 } \nonumber \\ & = & { \mathbb e}[{\boldsymbol{a}}(t)]\cdot { \mathbb e}[f({\boldsymbol{q}}(t))\;|\ ; { \boldsymbol{x}}(0)={\mathsf{x } } ] \nonumber \\ & \leq & \sum_{{\boldsymbol{\rho}}\in s } \alpha_{{\boldsymbol{\rho}}}{\boldsymbol{\rho}}\cdot { \mathbb e}[f({\boldsymbol{q}}(t))\;|\ ; { \boldsymbol{x}}(0)={\mathsf{x } } ] \nonumber \\ & \leq & ( 1-\varepsilon ) { \mathbb e}[\max_{{\boldsymbol{\rho}}\in s}{\boldsymbol{\rho}}\cdot f({\boldsymbol{q}}(t))\;|\ ; { \boldsymbol{x}}(0)={\mathsf{x } } ] . \label{eq : arrival_bound}\end{aligned}\ ] ] by combining and , it follows that @xmath325\,|\,{\boldsymbol{x}}(0)={\mathsf{x}}]}\\ & = & { \mathbb e}\left[\sum_{t=0}^{\tau({\mathsf{x}})-1}l(t+1)-l(t)\,\bigg|\,{\boldsymbol{x}}(0)={\mathsf{x}}\right]\\ & \leq & \sum_{t=0}^{\tau({\mathsf{x}})-1}{\mathbb e}\left [ { \boldsymbol{a}}(t)\cdot f({\boldsymbol{q}}(t))\,|\,{\boldsymbol{x}}(0)={\mathsf{x}}\right]\\ & & \mbox { } -\sum_{t=0}^{\tau({\mathsf{x}})-1}{\mathbb e}\left[{\boldsymbol{\sigma}}(t)\cdot f({\boldsymbol{q}}(t))\,|\,{\boldsymbol{x}}(0)={\mathsf{x}}\right ] + n f'(0)\tau({\mathsf{x}})\\ & = & { \mathbb e}\left[\sum_{t=0}^{\tau({\mathsf{x}})-1}\delta(t)\;\bigg|\ ; { \boldsymbol{x}}(0)={\mathsf{x}}\right],\end{aligned}\ ] ] where @xmath326 now we will bound @xmath327 . from lemma [ lemma : bernoulli_main ] with @xmath328 , where @xmath329 is defined in , it follows that for @xmath330 , @xmath331 where the last inequality is from lemma [ lemma : stopping_time ] . on the other hand , for any @xmath332 $ ] , we know that @xmath333 using these facts with lemma [ lemma : bernoulli_main ] , we further have that @xmath325\,|\,{\boldsymbol{x}}(0)={\mathsf{x}}]}\\ & \leq & { \mathbb e}\left[\sum_{t=0}^{\tau({\mathsf{x}})-1}\delta(t)\;\bigg|\ ; { \boldsymbol{x}}(0)={\mathsf{x}}\right]\\ & = & \pr [ \mathcal a^c]\cdot { \mathbb e}\left[\sum_{t=0}^{\tau({\mathsf{x}})-1}\delta(t)\;\bigg|\ ; { \boldsymbol{x}}(0)={\mathsf{x}},\mathcal a^c\right]\\ & & \qquad + ~\pr[\mathcal a]\cdot { \mathbb e}\left[\sum_{t=0}^{\tau({\mathsf{x}})-1}\delta(t)\;\bigg|\ ; { \boldsymbol{x}}(0)={\mathsf{x}},\mathcal a\right]\\ & \leq & \beta\,n\,f((1+c)q_{\max})\tau({\mathsf{x } } ) + n\,f'(0)\tau({\mathsf{x}})\\ & & \qquad + ( 1-\beta)\,n\,f((1+c)q_{\max})\alpha\,\tau({\mathsf{x}})\\ & & \qquad + \left(1-\beta\right)\left(-\frac{\varepsilon}{2 } f((1-c)q_{\max})\right)(1-\alpha)\tau({\mathsf{x}})\\ & \leq & \left ( -\frac{\varepsilon}{2}(1-\beta)(1-\alpha)+\frac{2n(\beta+(1-\beta)\alpha)}{1-c}\right)\times\\ & & \qquad f((1-c)q_{\max})\tau({\mathsf{x } } ) + n\,f'(0)\tau({\mathsf{x}})\\ & = & -\kappa({\mathsf{x}}),\end{aligned}\ ] ] where @xmath334 denotes the event that @xmath335 and in the last inequality , we use the following property for concave function @xmath98 : @xmath336 this completes the proof of lemma [ lemma : negative_drift ] . we first proceed toward proving in lemma [ lemma : stopping_time ] . it follows from the following observations : @xmath337 where the first inequality is from our modeling assumption that @xmath338 and @xmath339 , and the second inequality is from the definition of @xmath272 . hence , for @xmath340 $ ] , @xmath341 does not change more than @xmath342 from its initial value @xmath343 , which completes the proof of . now , we aim for proving @xmath312 . it suffices to show that , for given initial state @xmath42 with large enough @xmath299 , @xmath344,\ ] ] i.e. , @xmath345 does not change more than @xmath314 , and hence @xmath346 can be updated at most once for each @xmath150 . to prove , it is enough to show the following inequality : @xmath347 since using the definition of @xmath272 , @xmath348 to this end , for @xmath340 $ ] , consider following cases : 1 . if @xmath349 , then from the definition @xmath350 the amount of change @xmath351 is bounded by @xmath352 where we use and the condition * c1 * , i.e. , @xmath353 is decreasing . 2 . otherwise , it is bounded by @xmath354 where we use and the conditions * c1-c2 * , i.e. , @xmath353 is decreasing , @xmath355 and @xmath356 for large enough @xmath166 . this completes the proof of and hence that of lemma [ lemma : stopping_time ] . without loss of generality , assume that @xmath357 and let @xmath358 to simplify notation , we will use @xmath359 instead of @xmath360 . from conditions * c1 * -*c6 * , it is elementary to check that for large enough @xmath166,@xmath361 where their detailed proofs are given in appendix [ sec : appenda ] . due to lemma [ lemma : stopping_time ] , we can assume that @xmath362 and @xmath363 the sufficient conditions to prove lemma [ lemma : bernoulli_main ] are the following : 1 . for all @xmath332$],@xmath364 \leq \max_{{\boldsymbol{\rho}}\in s } { \boldsymbol{\rho}}\cdot { \boldsymbol{w}}(t).\ ] ] 2 . with probability @xmath365 , at least @xmath366 number of time instance @xmath367 $ ] satisfy @xmath368\leq { \boldsymbol{\sigma}}(t)\cdot{\boldsymbol{w}}(t).\label{eq : claimb}\ ] ] 3 . for all @xmath340 $ ] at which is satisfied , @xmath369 the proof of lemma [ lemma : bernoulli_main ] comes immediately from ( a ) , ( b ) and ( c ) as follows : @xmath370 & = & ( 1-\eta')^3\left[\max_{{\boldsymbol{\rho}}\in s } { \boldsymbol{\rho}}\cdot f({\boldsymbol{q}}(t))\right ] \\ & \leq & ( 1-\eta')^2\left[\max_{{\boldsymbol{\rho}}\in s } { \boldsymbol{\rho}}\cdot { \boldsymbol{w}}(t ) \right]\\ & \leq & ( 1-\eta ' ) \left({\boldsymbol{\sigma}}(t)\cdot{\boldsymbol{w}}(t)\right ) \\ & \leq & { \boldsymbol{\sigma}}(t)\cdot f({\boldsymbol{q}}(t)),\end{aligned}\ ] ] where with probability @xmath365 , at least @xmath366 number of time instance @xmath367 $ ] satisfy the second last and last inequalities . hence , we proceed toward proving ( a ) , ( b ) and ( c ) . * proof of ( a ) . * recall that our scheduling algorithm in section [ sec : algorithm_description ] maintains @xmath371 , where @xmath372 . thus , for @xmath15 and @xmath150 , we have @xmath373 and @xmath374 for large enough @xmath299 , combining above inequality with , we conclude that @xmath375}\\ & \leq & \left(1-\frac{2n}{f((1-c)q_{\max})}\right ) \left[\max_{{\boldsymbol{\rho}}\in s } { \boldsymbol{\rho}}\cdot f({\boldsymbol{q}}(t))\right]\\ & \leq & \max_{{\boldsymbol{\rho}}\in s } { \boldsymbol{\rho}}\cdot f({\boldsymbol{q}}(t))-2n\\ & \leq & \max_{{\boldsymbol{\rho}}\in s } { \boldsymbol{\rho}}\cdot { \boldsymbol{w}}(t),\end{aligned}\ ] ] where we use @xmath376 for @xmath332 $ ] from . * proof of ( b ) . * we first state the following key proposition for the proof of ( b ) . [ lem : assum1 ] suppose the assumption * a1 * holds . then , for any advice @xmath64 and @xmath377 , the following event occurs with probability @xmath378 : @xmath379 for @xmath75 , define a binary random variable @xmath380 such that @xmath381 if @xmath382 and @xmath383 otherwise . then , from the assumption * a1 * , @xmath384 < \delta$ ] and @xmath385 < \delta(l - h)$ ] . the conclusion of proposition [ lem : assum1 ] directly follows from the markov inequality applying to the random variable @xmath386 . from the above proposition , it follows that with probability @xmath378 , @xmath387 where from lemma [ lemma : stopping_time ] , we set @xmath388 due to @xmath389 . since @xmath58 changes at most @xmath0 times in @xmath302 $ ] from lemma [ lemma : stopping_time ] , one can use the union bound and conclude that with probability @xmath390 , @xmath391 for at least @xmath392 fraction of times in @xmath393 . furthermore , from , , and the definition of @xmath272 , we have @xmath394 thus , it follows that @xmath395 therefore , with probability @xmath365 , for at least @xmath396 fraction of times in the interval @xmath302 $ ] , holds . * proof of ( c ) . * from , we first have that @xmath397 for all @xmath340 $ ] at which is satisfied , @xmath398\\ & \leq & { \boldsymbol{\sigma}}(t)\cdot{\boldsymbol{w}}(t),\end{aligned}\ ] ] where we use , and @xmath399 . for large enough @xmath299 , combining the above inequality with leads to ( c ) : @xmath400 this completes the proof of lemma [ lemma : bernoulli_main ] . this section presents experimental results for the performance of our scheduling algorithm using various interactive oracle systems . we consider two network models : input - queued switch networks and wireless networks described in section [ sec : mcmc_wireless ] and section [ sec : bp ] , respectively . we compare the average queue - length in our scheduling algorithm using the mcmc , es and bp oracle systems with that of the mw algorithm @xcite . * input - queued switch network . * we first consider an input - queued switch network on @xmath401 complete bipartite graph topology in section [ sec : bp ] . we report the average queue lengths in the switch network for @xmath402 with load 90% , i.e. , arrival rates @xmath403 for all @xmath186 and any algorithm can not stabilize the queues if @xmath404 for all @xmath405 . we run the mw algorithm and our low - complexity scheduling algorithms using the mcmc , es , and bp oracle systems , where we consider the second bp oracle described in section [ sec : bp ] . here we use @xmath406 for es and bp oracles and @xmath407 for mcmc oracle . also , in bp oracle system , instead of using arbitrary schedule in @xmath21 when @xmath408 , we use a greedy algorithm with belief - weight to find a schedule , where belief at each @xmath202 is defined by @xmath409 : * find @xmath410 with maximum belief @xmath411 . * when @xmath412 are given , choose @xmath413 such that @xmath414 is a matching and @xmath415 is maximum among such @xmath416 . figure [ fig : switch_performance ] summarizes our experimental results and we observe that the performance of our scheduling algorithm using bp oracle system is as good as that of the mw algorithm , although it is much simpler to code and faster to run . grid topology.,title="fig:",width=288 ] grid topology.,title="fig:",width=288 ] * wireless networks in grid topology . * we also consider a wireless network with two dimensional grid interference topology . due to the high complexity of the mw algorithm , we chose the @xmath417 grid , i.e. , @xmath418 queues , where figure [ fig : wireless ] represents the @xmath419 grid . we again compare the evolution of the average queue lengths of the mw algorithm and our scheduling algorithms using es , mcmc and bp - es oracle systems . we note that we consider a minor variant of bp - es oracle so that it chooses a schedule greedily based on bp beliefs @xmath420 $ ] similarly as the second bp oracle in the input - queued switch network . we use the same weight functions as in the input - queued switch case and our experimental results are in figure [ fig : wireless_performance ] . we again observe that the performance of our scheduling algorithm using bp - es oracle system is as good as that of the mw algorithm , although it is much simpler to code and faster to run . the problem of dynamic resource allocation among network users contending resources has long been the subject of significant research for last four decades . in this paper , we develop a generic framework for designing resource allocation algorithms of low - complexity and high - performance via connecting iterative optimization and low - complexity scheduling . our work shows sufficient conditions on queue - length functions so that a queue - based scheduling algorithms is throughput - optimal . to our best knowledge , our result is the first that establishes a rigorous connection between iterative optimization methods and low - complexity scheduling algorithms , which we believe provides various future directions and new insights in both areas . 10 l. tassiulas and a. ephremides . stability properties of constrained queueing systems and scheduling policies for maximum throughput in multihop radio networks . _ ieee transactions on automatic control _ , 37:1936 - 1948 , 1992 . l. georgiadis , m. j. neely , and l. tassiulas . resource allocation and cross - layer control in wireless networks . _ foundations and trends in networking _ , 1:1 - 144 , 2006 . t. e. anderson , s. s. owicki , j. b. saxe , and c. p. thacker . high - speed switch scheduling for local - area networks . _ acm transactions on computer systems _ , 11(4):319 - 352 , 1993 . n. mckeown . islip : a scheduling algorithm for input - queued switches . _ ieee transaction on networking _ , 7(2):188 - 201 , 1999 . c. joo , x. lin , and n. b. shroff . understanding the capacity region of the greedy maximal scheduling algorithm in multi - hop wireless networks . _ ieee infocom _ , 2008 . m. leconte , j. ni , and r. srikant . improved bounds on the throughput efficiency of greedy maximal scheduling in wireless networks . _ acm mobihoc _ , 2009 . l. a. goldberg and p. d. mackenzie . analysis of practical backoff protocols for contention resolution with multiple servers . _ acm soda _ , 1996 . p. gupta and a. l. stolyar . optimal throughput allocation in general randomaccess networks . , 2006 . p. marbach , a. eryilmaz , and a. ozdaglar . achievable rate region of csma schedulers in wireless networks with primary interference constraints . _ ieee conference on decision and control _ , 2007 . s. rajagopalan , d. shah , and j. shin . network adiabatic theorem : an efficient randomized protocol for contention resolution . _ acm sigmetrics _ , 2009 . l. jiang , d. shah , j. shin , and j. walrand . distributed random access algorithm : scheduling and congestion control . _ ieee transactions on information theory _ , 56(12):6182 - 6207 , 2010 . d. shah and j. shin . randomized scheduling algorithm for queueing networks . _ annals of applied probability _ , 22:128 - 171 , 2012 . d. shah , d. n. c. tse , and j. n. tsitsiklis . hardness of low delay network scheduling . _ ieee transactions on information theory _ , 57(12):7810 - 7817 , 2011 . n. bouman , s. borst , and j. leeuwaarden . delays and mixing times in random - access networks . acm sigmetrics _ , 2013 . d. shah and j. shin , delay optimal queue - based csma . _ acm sigmetrics performance evaluation review _ , 38(1):373 - 374 , 2010 . m. lotfinezhad and p. marbach . throughput - optimal random access with order - optimal delay . _ ieee infocom _ , 2011 . l. jiang , m. leconte , j. ni , r. srikant , and j. walrand . fast mixing of parallel glauber dynamics and low - delay csma scheduling . _ ieee infocom _ , 2011 . p. k. huang and x. lin , improving the delay performance of csma algorithms : a virtual multi - channel approach . _ ieee infocom _ , 2013 . n. abramson and f. kuo ( editors ) . the aloha system . _ computer- communication networks _ , 1973 . j. g. dai and b. prabhakar . the throughput of data switches with and without speedup . _ ieee infocom _ , 2000 . s. kumar , p. giaccone , and e. leonardi . rate stability of stable - marriage scheduling algorithms in input - queued switches . , 2002 . l. tassiulas . linear complexity algorithms for maximum throughput in radio networks and input queued switches . _ infocom _ , 1998 . p. giaccone , b. prabhakar , and d. shah . randomized scheduling algorithms for high - aggregate bandwidth switches . _ ieee journal on selected areas in communications high - performance electronic switches / routers for high - speed internet _ , 21(4):546 - 559 , 2003 . e. modiano , d. shah , and g. zussman . maximizing throughput in wireless network via gossiping . _ acm sigmetrics _ , 2006 . a. dimakis and j. walrand . sufficient conditions for stability of longest - queue- first scheduling : second - order properties using fluid limits . _ advances in applied probability _ , 38(2):505 - 521 , 2006 . d. a. levin , y. peres , and e. l. wilmer . markov chains and mixing times . _ american mathematical society _ , 2008 . j. yedidia , w. freeman , and y. weiss . constructing free energy approximations and generalized belief propagation algorithms . _ ieee transactions on information theory _ , 51:2282 - 2312 , 2004 . g. cornuejols . valid inequalities for mixed integer linear programs . mathematical programming series b , 112:3 - 44 , 2008 . s. foss and t. konstantopoulos . an overview of some stochastic stability methods . _ journal of operations research , society of japan _ , 47(4 ) , 2004 . m. i. jordan . graphical models . _ statistical science _ , 19:140 - 155 , 2004 . m. bayati , d. shah , and m. sharma . max - product for maximum weight matching : convergence , correctness , and lp duality . _ ieee transactions on information theory _ , 54(3 ) : 1241 - 1251 , 2008 . s. sanghavi , d. malioutov , and a. s. willsky . linear programming analysis of loopy belief propagation for weighted matching . , 2007 . s. atalla , d. cuda , p. giaccone , m. pretti . belief - propagation assisted scheduling in input - queued switches . _ ieee 18th annual symposium on high performance interconnects ( hoti ) _ , 2010 d. shah . gossip algorithms . _ now publishers inc _ , 2009 . here , we provide the proofs of - for large enough @xmath166 . due to conditions * c2 * and * c5 * , @xmath421 which proves and . now , follows from the condition * c1 * and * c2 * : @xmath422 where we use the following property for concave function @xmath98 : @xmath423 finally , for , let @xmath424 and we observe that @xmath425 as @xmath426 due to the condition * c6*. this completes the proof of .

we investigate classical thermal activation ( ta ) and macroscopic quantum tunneling ( mqt ) for a ybacuo(ybco ) josephson junction coupled to an lc circuit theoretically . due to the coupling between the junction and the lc circuit , the macroscopic phase dynamics can be described as the escape process of a fictitious particle with an mass moving in a _ two - dimensional _ potential . we analytically calculate the escape rate including both the ta and mqt regime by taking into account the peculiar dynamical nature of the system . in addtion to large suppression of the mqt rate at zero temperature , we study details of the temperature dependece of the escape rate across a crossover region . these results are in an excellent agreement with recent experimental data for the mqt and ta rate in a ybco biepitaxial josephson junction . therefore the coupling to the lc circuit is essential in understanding the macroscopic quantum dynamics and the qubit operation based on the ybco biepitaxial josephson junctions .

0901.1723

macroscopic quantum tunneling ( mqt ) has become a focus of interest in physics and chemistry because it can provide a signature of quantum behavior in a macroscopic system . @xcite among several works on mqt , josephson junctions have been intensively studied . @xcite heretofore experimental tests of mqt were focused on low-@xmath4 superconductor josephson junctions . , the stray capacitance @xmath5 and kinetic inductance @xmath6 . the josephson junction with the critical current @xmath7 and the capacitance @xmath8 is formed at the boundary between ( 001 ) and ( 103 ) ybco electrodes . @xmath9 and @xmath10 are the phase difference across the josephson junction and the stray capacitance @xmath5 , respectively . , width=226 ] [ fig:1 ] renewed interest in mqt occurred after the recent experimental observations of mqt @xcite and energy level quantization ( elq ) @xcite in high-@xmath4 superconductor josephson junctions , e.g. , yba@xmath0cu@xmath1o@xmath2 ( ybco ) grain - boundary biepitaxial junctions [ see fig . 1 ( a ) ] @xcite and bi@xmath0sr@xmath0cacu@xmath0o@xmath11 intrinsic junctions . @xcite high-@xmath4 josephson junctions , characterized by a high josephson plasma frequency @xmath12 ( up to several thz ) , exhibit a crossover from the thermal activation ( ta ) to mqt at relatively high temperatures in comparison with low-@xmath4 systems . these intriguing findings have definitively opened up the way to quantum systems based on the @xmath13-wave symmetry of high-@xmath4 superconductors . the @xmath13-wave order parameter can be used to create naturally degenerate two - level systems which can offer significant advantages for quantum computation . @xcite however , one of the main arguments against high-@xmath4 qubits based on systems with @xmath13-wave order parameter symmetry was the presence of low energy excitations , i.e. , nodal quasiparticles and zero - energy andreev bound states ( zess ) , destroying quantum coherence . recent theoretical works suggest that nodal quasiparticles and zess , respectively , give super - ohmic and ohmic dissipation on mqt . @xcite the ybco biepitaxial grain - boundary junctions which were used in mqt and elq experiments , @xcite have quite novel structure and several fascinating advantages for the coherent property over other types of high-@xmath4 junctions . first , the relative orientation of the @xmath13-wave order parameter between two ybco electrodes can be designed artificially . @xcite therefore it is possible to make a junction in which a lobe of the @xmath13-wave order parameter of one electrodes is facing a node in the other electrodes as shown in fig . 1(a ) . in such configuration , super - ohmic dissipation resulting from the nodal quasiparticles can be suppressed . @xcite moreover the junction has an additional tilting of one electrode [ ( 103 ) ybco ] with respect to ( 001 ) ybco electrodes [ see fig . 1(a ) ] . in such tilt - junctions , the formation of zess causing ohmic dissipation can be inhibited as was predicted by golubov and tafuri . @xcite recently , it was found from elq experiment that the measured bias - current @xmath14 dependence of the resonant frequency deviates significantly from what one expects from the josephson plasma frequency @xmath12 in a conventional single josephson junction . @xcite the resonant frequency is a factor of five less than what one would expect from the estimated values of the josephson critical current @xmath7 and the junction capacitance @xmath8 . this fact could be explained by taking into account the existence of a large kinetic inductance and stray capacitance coupled to the junction . @xcite in the biepitaxial junctions , the film of the ( 103 ) ybco electrode is oriented in such a way that the electric transport has a large component in the @xmath15-axis direction [ see fig . so the kinetic inductance @xmath6 of the ( 103 ) electrode can be significant and is much larger than the josephson inductance of the junction @xmath16 . @xcite another important effect which was observed in elq experiments is the influence of the stray capacitance @xmath5 of the srtio@xmath1 ( sto ) substrate with a huge dielectric constant [ see fig . hence a large part of the measured capacitance is not due to the junction interface but to a distributed stray capacitance @xmath5 which in effect shunts the junction . @xcite the influence of @xmath6 and @xmath5 on the macroscopic dynamics is inevitable in the biepitaxial junctions and can be taken into account by an extended circuit model [ see fig.1(b ) ] . @xcite in fig . 1 ( b ) , @xmath9 is the phase difference across the josephson junction and @xmath17 is the phase difference across the capacitor @xmath5 , where @xmath18 is the current through the inductor @xmath6 . as will be mentioned later , the addition of the lc circuit results in a two - dimensional ( 2d ) potential @xmath19 and an anisotropic mass which make the dynamics much more complex than for an ordinary single junction . in refs . [ ] we have theoretically investigated the @xmath20 resonant activation based on the extended circuit model , and found that the bias - current @xmath14 dependence of the resonant frequencies can be quantitatively explained by the normal modes ( upper and lower resonant mode ) of the system . those normal modes , under certain conditions , can be completely different from the bare josephson plasma frequency @xmath12 . therefore the extended circuit model well describes the elq experiment in the ybco biepitaxial junctions . in contrast to elq experiment , it was experimentally confirmed that the thermal escape process above the crossover temperature @xmath21 can be quantitatively explained by the single junction model @xmath22 the lc circuit , @xmath23 , one - dimensional ( 1d ) model . @xcite therefore a natural question to ask is _ can the extended circuit model explain both the ta and mqt escape dynamics in the ybco biepitaxial junction consistently ? _ in this paper , in order to answer the important open question , we explore the validity of the extended circuit model for both the ta and mqt escape processes by extending the previous classical theory . @xcite then we will show that the presence of the lc circuit has negligible influence on the ta escape rate , but the mqt escape rate is suppressed considerably due to the coupling to the lc circuit . this behavior is nicely consistent with a recent experimental result of the temperature dependence of the ta and mqt escape rate . @xcite therefore , _ the presence of the lc circuit is fundamental and essential in understanding the macroscopic phase dynamics in the ybco biepitaxial junctions . _ this paper is organized as follows . in sec . ii , we show the lagrangian describing the ybco josephson junction coupled to the lc circuit . we also discuss the anisotropy of the mass and the two - dimensional potential profile for this model . in sec . iii , we derive an effective action from the lagrangian and show that the system can be mapped into a one - dimensional model . the crossover temperature @xmath21 , the ta , and the mqt escape rate are calculated in secs . iv , v , and vi , respectively . we try to compare theoretical results with recent experimental data of the ybco biepitaxial josephson junction in sec . finally , we summarize our results and draw future directions in sec . viii . in this section we derive the lagrangian for the josephson junction coupled to the lc circuit and discuss the anisotropy of the mass and two - dimensional potential structure of this model . the hamiltonian of the circuit [ fig . 1(b ) ] can be written as @xmath24 where @xmath25 and @xmath26 @xmath27 are the charge on the junction and the stray capacitor @xmath5 , respectively , @xmath28 is the josephson coupling energy , and @xmath29 is the flux quantum . the lagrangian is then given by @xmath30 @xmath31 , \label{eq:1 - 3}\end{aligned}\ ] ] where @xmath32 and @xmath33 with @xmath34 being the zero - bias josephson inductance . the lagrangian @xmath35 describes the quantum dynamics of a fictitious particle with an anisotropic mass moving in a two - dimensional tilted washboard potential @xmath19 . therefore the escape paths traverse a two - dimensional landscape , so the macroscopic dynamics in this model becomes more complicated than that in the simple one - dimensional model . of a ybco josephson junction coupled to a lc circuit for @xmath36 and @xmath37 . initially the junction oscillate around the local minima at @xmath38 . if @xmath39 is increased , the junction eventually switches to the finite voltage state by escaping either thermally or via mqt through the saddle point at @xmath40 . , width=302 ] [ fig:2 ] the two - dimensional potential profile @xmath19 is shown in fig . the mean slope along the @xmath10 direction is proportional to the bias current @xmath14 , and the mean curvature perpendicular to the diagonal direction ( @xmath41 is due to the inductive coupling between the josephson junction and capacitance @xmath5 characterized by @xmath42 . the local minimum point @xmath38 and the saddle point @xmath40 is given from eq . ( [ eq:1 - 3 ] ) as @xmath43 then the potential barrier height @xmath44 is given by @xmath45 and is a decreasing function of @xmath39 . importantly , the potential barrier height @xmath44 does not depend on the lc circuit parameters @xmath6 and @xmath5 and the expression of @xmath44 is identical with the barrier height for usual single josephson junctions , i.e. , one - dimensional model . @xcite on the other hand , the overall potential shape can be controlled drastically by changing @xmath6 . in the case of small @xmath6 or equivalently large coupling strength @xmath46 , the side well becomes shallower and the potential is almost confined to the diagonal direction ( @xmath47 ) . on the other hand , when the coupling strength @xmath46is decreased , corresponding to a large @xmath6 , the confinement to the diagonal direction tends to be weak . this makes the escape from a metastable well much easier . in secs . v and vi , we quantitatively investigate the @xmath5 and @xmath6 dependence of the thermal and quantum escape rate . next we derive an approximate expression of @xmath19 for @xmath48 . by introducing the new coordinate @xmath49 and assuming @xmath48 , we can rewrite the lagrangian as @xmath50 \nonumber\\ & = & u_\mathrm{1d } ( x ) + e_j \frac{(x - y)^2}{2 \eta } , \label{eq:1 - 8}\end{aligned}\ ] ] where @xmath51 and @xmath52 . in this equation , @xmath53 \approx ( 1/2 ) m \omega_p^2 ( x^2 -x^3/x_1)$ ] is the potential of the josephson junction without the lc circuit , where @xmath54 is the second zero of @xmath55 and @xmath56 is the josephson plasma frequency with @xmath57 being the zero - bias plasma frequency . here we derive an effective action from the lagrangian eq . ( [ eq:1 - 7 ] ) of the extended circuit model . note that , in the elq experiment , the quality factor @xmath58 of the junction was found to be sufficiently large ( @xmath59 ) , @xcite so we can safely neglect the damping effect resulting from external environments other than the lc circuit . by using the functional integral method , @xcite the partition function @xmath60 of the system can be written as @xmath61 \right ) , \label{eq:2 - 1}\end{aligned}\ ] ] where the functional integral is performed over all the periodic paths with the period @xmath62 . in this equation , @xmath63 = ( m/2 ) \dot{x}^2 + ( m/2 ) \dot{y}^2 + u(x , y ) $ ] is the euclidean lagrangian and @xmath64 . the lagrangian is a quadratic function of @xmath65 and the coupling term between @xmath66 and @xmath65 is linear , so the functional integral over variable @xmath65 can be performed explicitly by use of the feynman - vernon influence functional technique . @xcite then the partition function is reduced to a single functional integral over @xmath66 , i.e. , @xmath67 /\hbar)$ ] , where the effective action is given by @xmath68={\cal s}_\mathrm{eff}^\mathrm{1d}[x]+{\cal s}_\mathrm{eff}^\mathrm{ret}[x]$ ] with @xcite @xmath69 & = & \int_0^{\hbar \beta } d \tau \left [ \frac{1}{2 } m \dot{x}^2 + u_\mathrm{1d } ( x ) \right ] , \label{eq:2 - 2 } \\ { \cal s}_\mathrm{eff}^\mathrm{ret}[x ] & = & \frac{1}{4 } \int_{0}^{\hbar \beta } d \tau \int_0^{\hbar \beta } d \tau ' \left [ x(\tau ) - x(\tau ' ) \right]^2 k ( \tau-\tau ' ) \ ! . \nonumber\\ \label{eq:2 - 3}\end{aligned}\ ] ] thus the dynamics of the phase difference with an anisotropic mass moving in a two - dimensional potential @xmath70 can be mapped into a simple one - dimensional model . note that due to the coupling between the junction and the lc circuit , the effective action @xmath68 $ ] contains a kind of @xmath71 action @xmath72 $ ] in a sense that a retardation ( or nonlocal ) effect exists . the nonlocal kernel @xmath73 in eq . ( [ eq:2 - 3 ] ) is defined by @xmath74 } { \sinh \left [ \frac { \hbar \beta \omega_{lc } } { 2 } \right ] } \nonumber\\ & = & \frac{m}{\hbar \beta } \sum_{n=-\infty}^\infty \zeta_n e^ { - i \omega_n \tau } , \label{eq:2 - 4}\end{aligned}\ ] ] where @xmath75 is the lc resonant frequency , @xmath76 is the matsubara frequency , and @xmath77 is the fourier coefficient of @xmath73 . the nonlocal kernel @xmath73 is related to the spectral density @xmath78 in the language of the caldeira - leggett theory @xcite via @xmath79 where @xmath80 is the matsubara green s function of a free boson , @xmath81 @xmath82 in the caldeira - leggett theory , the external bath is modeled by a sum of an @xmath83 number of harmonic oscillators which is capable of destroying mqt . @xcite on the other hand , in our case , the bath can be described by a @xmath84 harmonic oscillator and then @xmath78 is given as a delta - function , i.e. , @xmath85 for later convenience , we rewrite the effective action @xmath72 $ ] in terms of the memory kernel @xmath86 , @xcite i.e. , @xmath87 & = & \frac{1}{2 } \int_{0}^{\hbar \beta } d \tau \int_0^{\hbar \beta } d \tau ' x(\tau ) k ( \tau-\tau ' ) x(\tau ' ) , \label{eq:2 - 6 } \\ k(\tau ) & = & m \omega_{lc}^2 \sum_{n=-\infty}^\infty \delta(\tau- n \hbar \beta ) -k(\tau ) \nonumber\\ & = & \frac{m}{\hbar \beta } \sum_{n=-\infty}^\infty \xi_n e^ { - i \omega_n \tau } , \label{eq:2 - 7}\end{aligned}\ ] ] where the fourier coefficient of the memory kernel @xmath88 is related to the fourier transform of the memory - friction kernel @xmath89 , @xcite @xmath90 @xmath91 , via @xmath92 in below , by using the derived effective action @xmath93 , we calculate the crossover temperature and the escape rate , and then discuss the influence of the anisotropic mass and the two - dimensional nature of the potential profile on the macroscopic phase dynamics in the ybco junction coupled to the lc circuit . based on the normal mode analysis at the barrier top of the potential @xmath70 , the crossover temperature between the mqt and ta escape process is given by @xcite @xmath94 where @xmath95 is the renormalized trial frequency due to the coupling to the lc circuit and is a positive root of the equation , @xmath96 by using eq . ( [ eq:2 - 8 ] ) , we get @xmath97 for the adiabatic ( @xmath98 and @xmath99 ) and the nonadiabatic ( @xmath100 and @xmath101 ) limit , where @xmath102 , and @xmath103 is the josephson inductance . in the adiabatic ( nonadiabatic ) limit , the lc resonant frequency @xmath104 is much larger ( smaller ) than the josephson plasma frequency @xmath12 . as a result , the lc resonant mode can ( can not ) adiabatically follow the dynamics of the josephson junctions . note that , in the ybco biepitaxial junction used in the mqt experiment , @xcite the nonadiabatic case is approximately realized ( see sec . vii ) . by substituting eq . ( [ eq:3 - 2 ] ) into eq . ( [ eq:3 - 1 ] ) , we get a simple expression for @xmath21 as @xmath105 where @xmath106 is the crossover temperature without coupling to the lc circuit . therefore , due to the influence of the lc circuit , @xmath107 becomes smaller than @xmath108 . in the nonadiabatic ( adiabatic ) case , @xmath107 reduces with decreasing the kinetic inductance @xmath6 ( increasing the stray capacitance @xmath5 ) . the ta escape rate well above the crossover temperature @xmath107 is given by @xcite @xmath109 where @xmath110 is the quantum - mechanical enhancement factor resulting form stable fluctuation modes and @xmath111 is the fourier transform of the memory - friction kernel . as was shown in sec . ii , the potential barrier height is not changed even in the presence of the lc circuit , i.e. , @xmath112 . therefore the coupling to the lc circuit only modifies the prefactor of @xmath113 . note that the exponent in @xmath113 for the josephson junction coupled to the lc circuit was discussed by fistul . @xcite here we calculate the ta escape rate @xmath114 including the prefactor in order to discuss the influence of the lc circuit explicitly and compare with experimental results . the quantum correction @xmath115 in the prefactor can be calculated analytically as @xmath116 with @xmath117 where @xmath118 and @xmath119 the effective trial frequency @xmath120 in eq . ( [ eq:4 - 1 ] ) is given by eq . ( [ eq:3 - 2 - 1 ] ) . in the adiabatic limit ( @xmath98 and @xmath99 ) , the quantum enhancement factor @xmath110 ( [ eq:4 - 3 ] ) can be simplified to @xmath121 on the other hand , in the non - adiabatic limit ( @xmath100 and @xmath101 ) , we get @xmath122 note that the nonadiabatic limit is almost realized in the actual ybco junction @xcite ( see sec . the quantum enhancement factor @xmath110 almost coincides with the result without retardation effects , @xcite i.e. , @xmath123 for both the adiabatic and non - adiabatic limit ( see also secs . c and vii . b ) . from eqs . ( [ eq:3 - 2 ] ) , ( [ eq:4 - 1 ] ) , ( [ eq:4 - 4 ] ) , and ( [ eq:4 - 5 ] ) , therefore , we can conclude that the influence of the coupling to the lc circuit on the thermal activation process is quite weak , so the system behaves as a one - dimensional system well above the crossover temperature @xmath107 . this result is qualitatively consistent with the experimental observation in the ybco biepitaxial junction . @xcite in sec . vii we will numerically compare theoretical results with experimental data in the ta regime . to obtain the mqt escape rate @xmath124 for the josephson junction coupled to the lc circuit below @xmath107 , the usual procedure is to apply the so - called im-@xmath125 method . @xcite it is based on the calculation of the free energy @xmath126 and thus of the partition function @xmath60 of an unstable system . according to the metastable decay theory , @xcite the mqt escape rate has the form @xmath127 . in below we derive a weak retardation condition in which @xmath128 in the effective action @xmath93 can be treated as a small perturbation . then we calculate @xmath124 for zero temperature and finite temperature ( @xmath129 in a weak retardation limit . here we derive a condition for the weak retardation or nonlocal effect . the bounce trajectory @xmath130 satisfies the euler - lagrange equation obtained from the variation principle @xmath131=0 $ ] . taking the variation of the effective action eqs . ( [ eq:2 - 2 ] ) and ( [ eq:2 - 3 ] ) , we then get @xmath132 introducing the fourier expansion @xmath133 @xmath134 into eq . ( [ eq:5 - 1 ] ) and using eq . ( [ eq:2 - 7 ] ) , we obtain the following equation for the fourier coefficient @xmath135 @xmath136 \nonumber\\ & + & m \xi_n \tilde{x}_n^{(b ) } = 0 . \label{eq:5 - 2 } \end{aligned}\ ] ] for @xmath137 , we get @xmath138 the criterion for weak retardation can be obtained from eq . ( [ eq:5 - 3 ] ) by comparing the retardation term ( @xmath139 ) with the first term ( @xmath140 ) . by defining the parameter @xcite @xmath141 the weak retardation limit corresponds to the case of @xmath142 . in this case @xmath128 in the effective action can be treated as a small perturbation . figure 3 shows the @xmath6 and @xmath5 dependence of the retardation parameter @xmath143 . in the ybco junction used in the mqt experiment , @xcite @xmath144 and @xmath145 as will be estimated in sec . vii , so @xmath146 . therefore , the weak retardation condition is almost satisfied in this case . and @xmath5 dependence of the parameter @xmath143 which characterizes the retardation effect due to the junction - lc coupling . in the ybco biepitaxial junction which was used in actual mqt experiment , @xmath146 ( red circle ) . , width=302 ] [ fig:3 ] the mqt escape rate at zero temperature is given by @xmath147 . @xcite by use of the bounce techniques , the mqt escape rate @xmath124 in the weak retardation limit @xmath142 is perturbatively determined by @xcite @xmath148 where @xmath149/\hbar=36 v_0^\mathrm{1d } / 5 \hbar \omega_p$ ] and @xmath150/\hbar$ ] are the bounce exponents , that are the value of the actions evaluated along the bounce trajectory @xmath151 . the bounce action @xmath152 $ ] can be written as @xmath153 & = & 8 \pi^2 \frac{m \omega_{lc}^2 x_1 ^ 2}{\omega_p^4 } \frac{1}{\hbar \beta } \nonumber\\ & \times & \sum_{n=-\infty}^\infty \frac{\omega_n^4 } { \left ( \omega_n^2 + \omega_{lc}^2 \right ) \sinh^2 \left ( \frac{\pi \omega_n}{\omega_p } \right ) } . \label{eq:5 - 6}\end{aligned}\ ] ] in the limit of zero temperature , we have @xmath154 \ ! \ ! \ ! & = & \ ! \ ! \ ! \frac { 8 \pi m x_1 ^ 2 \omega_{lc}^2 } { \omega_p } \int_0^\infty d z \frac{z^4 } { \left\ { z^2 + \left ( \frac{\omega_{lc}}{\omega_p } \right)^2 \right\ } \sinh^2 \left ( \pi z \right ) } \nonumber\\ \ ! & \approx & \ ! \left\ { \begin{array}{cl } \displaystyle{\frac{4}{15 } m \omega_p x_1 ^ 2 } & \quad l_j \gg l_s \ \ & \ c_j \gg c_s \\ \displaystyle{\frac{4}{3 } m \left ( \frac{\omega_{lc}}{\omega_p } \right)^2 \omega_p x_1 ^ 2 } & \quad l_j \ll l_s \ \ & \ c_j \ll c_s \end{array } \right . \ ! \ ! . \nonumber\\ \label{eq:5 - 7}\end{aligned}\ ] ] therefore the total bounce exponent is given by @xmath155 where @xmath156 is the retardation correction to the mass @xmath157 . thus , due to the influence of the lc circuit , the bounce exponent is increased with respect to the one - dimensional case . by substituting eq . ( [ eq:5 - 8 ] ) into eq . ( [ eq:5 - 5 ] ) , we finally get the zero - temperature mqt escape rate for @xmath142 as @xmath158 . \label{eq:5 - 10 } \end{aligned}\ ] ] therefore the coupling to the lc circuit effectively increases the barrier height , @xmath159 @xmath160 . this behavior is consistent with the result for low-@xmath4 junctions coupled to an lc circuit which was derived in a different context . @xcite in contrast to the ta regime , the coupling to the lc circuit reduces the mqt escape rate @xmath124 considerably . therefore the anisotropic mass and the two - dimensional nature of the potential profile have large influence on the mqt escape process in the low - temperature regime . in this section we calculate the mqt escape rate at finite temperature ( @xmath161 ) for the weak retardation limit ( @xmath142 ) and show that the @xmath162 dependence of the mqt escape is drastically influenced by the coupling to the lc circuit . at finite - temperature the mqt escape rate is enhanced by thermal fluctuations . if the temperature is much lower than @xmath163 , the finite temperature enhancement to the @xmath164 is negligible in contrast to the system with an ohmic dissipative environment ( see appendix a ) . as will be shown later , however , at sufficiently high temperature but still smaller than @xmath107 , the coupling to the lc circuit gives a large enhancement to the @xmath164 . the finite - temperature bounce trajectory @xmath165 is given by a periodic solution in the inverted potential @xmath166 with energy @xmath167 ( @xmath168 ) . the corresponding solution is given by @xmath169 , where @xmath170 are the solutions of the equation @xmath171 . @xcite in this equation @xmath172 denotes the jacobi elliptic function with the parameter @xmath173 and the coefficient @xmath174 . the period @xmath62 of the finite - temperature bounce @xmath165 is given by the complete elliptic integral of the first kind @xmath175^{-1/2}$ ] as @xmath176 . @xcite the energy @xmath177 is related to the temperature @xmath178 via this equation . in the limit of zero - temperature ( @xmath179 ) , we recover the zero temperature bounce solution @xmath180 . by substituting @xmath165 into the effective action , a finite - temperature bounce action @xmath181 $ ] can be evaluated as @xmath182 & = & { \cal s}_t^\mathrm{1d } [ x_b]+{\cal s}_t^\mathrm{ret}[x_b ] , \label{eq:5 - 11 } \\ { \cal s}_t^\mathrm{1d } [ x_b ] & = & m \int_0^{\hbar \beta } d \tau { \dot x}_b^t(\tau)^2 + e \hbar \beta \nonumber\\ & = & \frac{\lambda \omega_p m ( q_1-q_2)^2}{15 k^2 } f(\lambda \omega_p \hbar \beta , k ) + e \hbar \beta , \label{eq:5 - 12 } \\ { \cal s}_t^\mathrm{ret}[x_b ] & = & \frac{1}{2 } \int_0^{\hbar \beta } d \tau d \tau ' k(\tau-\tau ' ) \left [ x_b^t(\tau ) - x_b^t(\tau ' ) \right]^2 \nonumber\\ & = & \frac{1}{2 } m \omega_\mathrm{lc}^2 \hbar \beta \sum_{n=-\infty}^\infty \frac{\omega_n^2}{\omega_n^2 + \omega_\mathrm{lc}^2 } \left| x_b^t ( n)\right|^2 , \label{eq:5 - 13}\end{aligned}\ ] ] where @xmath183 is the fourier transform of the finite temperature bounce @xmath165 and the function @xmath184 is defined by @xmath185 , \label{eq:5 - 14 } \end{aligned}\ ] ] with @xmath186 and @xmath187 the jacobi elliptic functions , @xmath188 is the jacobi amplitude , @xmath189 is the elliptic integral of the second kind , @xmath190 , @xmath191 , @xmath192 , and @xmath193 . the prefactor of the mqt escape rate @xmath194 is determined from the fluctuation modes around the bounce solution @xmath165 and can be calculated from the gelfand - yaglom formula @xcite as @xmath195 where @xmath196 , @xmath197 , and @xmath198 is the elliptic integral of the first kind . @xcite for @xmath179 , @xmath199 is reduced to well - known zero - temperature result , i.e. , @xmath200 . thus we analytically obtain the finite - temperature mqt escaper rate for the weak retardation limit : @xmath201}{\hbar } \right ) . \label{eq:5 - 16 } \end{aligned}\ ] ] this result is smoothly matched to the zero - temperature expression ( [ eq:5 - 10 ] ) when @xmath202 . for the ybco junction coupled to the lc circuit . ( a ) @xmath6 dependence of @xmath203 for @xmath204 pf . ( b ) @xmath5 dependence of @xmath203 for @xmath205 nh . we choose @xmath206 , @xmath207 @xmath208a ( @xmath209 0.54 nh ) and @xmath210 pf for both graphs . in the above parameter range , the weak retardation condition ( @xmath211 ) is satisfied . solid and dotted arrows indicate the crossover temperature @xmath107 and @xmath212 , respectively . , width=264 ] [ fig:4 ] the arrhenius plot of the escape rate @xmath194 and @xmath213 are shown in fig . 4 . note that we have used an interpolate crossover - formula @xcite near the crossover temperature @xmath107 because @xmath194 and @xmath213 diverges as @xmath214 due to the breakdown of the gaussian approximation ( see appendix b ) . in the numerical calculation , we have used @xmath207 @xmath208a and @xmath210 pf . these parameters agree with those directly measured or estimated from experiment , @xcite allowing a comparison between numerical and experimental data ( see sec . the ta escape rate @xmath213 above @xmath107 is almost same as in the case without the lc circuit as clearly seen from fig . on the other hand , @xmath194 is considerably reduced due to the coupling to the lc circuit . as was explained in sec . ii , the mqt escape rate is significantly reduced with decreasing @xmath6 . this is due to the drastic change in the two - dimensional potential profile . similarly the mqt escape rate is decreasing with increasing @xmath5 . this behavior can be explained as follows . in the case of small @xmath5 , the kinetic energy to the @xmath65 ( or @xmath215 direction becomes small . therefore , due to the strong anisotropy of the mass , mqt along the external escape direction which connects the minimum @xmath38 and the saddle point @xmath40 ( see fig . 2 ) is inhibited . interestingly , the temperature dependence of the mqt escape rate @xmath194 for the two - dimensional system is quite different from that of the one - dimensional system . in the case without a lc circuit , @xmath216 is almost constant below the crossover temperature @xmath108 . on the other hand , in the case of small @xmath6 and large @xmath5 , @xmath194 significantly depends on the temperature in the range of @xmath217 . therefore , from the temperature dependence of the mqt escape rate , we can distinguish qualitatively the anisotropy of the mass and the dimensionality of the quantum phase dynamics . in this section , we first summarize the experimental device parameters relevant to the ta and mqt phenomena of a ybco grain - boundary biepitaxial josephson junction ( sec . vii . then , in order to check the validity of the extended circuit model @xcite for the ta and mqt escape process , we try to compare our result with the experimental data @xcite of the switching current distribution at the high - temperature ta and the low temperature mqt regimes ( sec . vii . .experimental junction parameters used for the estimation of the ta and mqt escape rate . [ cols="<,^",options="header " , ] experimental parameters used in numerical calculation are given in table i. the values given in the table are indeed typical ones in actual experiments . by the fitting to the kramers formula with experimental data of the ta escape rate well above the crossover temperature , @xmath7 was experimentally estimated as @xmath207 @xmath208a which corresponds to @xmath218 nh . @xcite the values of @xmath205 nh and @xmath204 pf have been directly determined from the elq experiments . @xcite remaining unknown junction parameter is @xmath8 . as will be shown in the next section , this value is estimated as @xmath210 pf . this allows for a qualitative comparison between numerical and experimental data of the zero temperature mqt escape rate . in this case @xmath144 and @xmath219 ( @xmath220 for @xmath206 ) . therefore , in the actual junction , the nonadiabatic limit is approximately realized . in this section , we numerically calculate the switching current distribution @xmath221 which is related to the escape rate @xmath222 as @xcite @xmath223 , \label{eq:6 - 1}\end{aligned}\ ] ] where @xmath224 is the sweep rate of the external bias current . in the actual experiment , @xcite the temperature dependence of the full width at half maximum ( hmfw ) @xmath225 of @xmath221 is measured as shown in fig . 5 . of the switching current distribution @xmath221 . both the one- ( dotted black line ) and two - dimensional models ( red solid line ) give almost same result above the crossover temperature . the calculated @xmath225 from the zero - temperature mqt escape rate for the one - dimensional model and the finite - temperature mqt escape rate for two - dimensional model [ eq . ( 44 ) ] are shown by dashed - dotted ( blue ) line and ( red ) squares , respectively . experimental data of @xmath225 ( black circles ) for a ybco biepitaxial josephson junction @xcite is also plotted . we choose @xmath207 @xmath208a ( @xmath218 nh ) , @xmath210 pf , @xmath205 nh , @xmath204 pf and @xmath226 ma / s . arrows indicate the several characteristic temperatures , i.e. , @xmath212 , @xmath108 , and @xmath107 for @xmath206 . , width=264 ] [ fig:5 ] first we investigate the ta regime . in fig . 5 we show the temperature dependence of @xmath225 in the ta escape regime ( red solid and black dotted lines ) . in calculation we have substituted eqs . ( [ eq:4 - 1 ] ) and ( [ eq:4 - 3 ] ) into eq . ( [ eq:6 - 1 ] ) . both the one- and two - dimensional model give good agreement with the experimental data ( black circles ) well above the crossover temperature ( @xmath227k ) . therefore , in the ta regime , the system can be treated as a one - dimensional model without the lc circuit . in the mqt regime , the measured saturated value of @xmath225 at @xmath228 k is found to be 11.9 na . @xcite from the numerical estimation of @xmath225 , we found that @xmath229 pf gives good agreement with the experimental value of @xmath225 as shown in fig . 5 . the obtained value of @xmath8 is consistent with the estimated value @xmath230 pf@xmath231 based on the geometry of the junction . @xcite therefore , we can conclude that the extended circuit model quantitatively explain the mqt experiment @xcite in the ybco biepitaxial junction near zero temperature . in order to test the validity of the extended circuit model more systematically , experimental measurements of hmfw @xmath225 by changing @xmath6 and @xmath5 are needed . @xcite in fig . 6 we show @xmath6 and @xmath5 dependence of @xmath225 . if we use the substrates with low dielectric constant @xmath232 , e.g. , mgo ( @xmath233 ) and laalo@xmath1 ( @xmath234 ) , the reduction in @xmath225 with respect to the one - dimensional model ( blue dotted line in fig . 6 ) becomes small . and ( b ) @xmath235 dependence of the full width at half maximum @xmath225 of the switching current distribution @xmath221 at @xmath236 k for the case with ( red solid ) and without ( blue dotted ) the lc circuit . we choose the parameters as @xmath207 @xmath208a ( @xmath218 nh ) , @xmath210 pf and @xmath226 ma / s . , width=264 ] [ fig:6 ] in the present work , the ta and the mqt escape process of the ybco josephson junction coupled to the lc circuit has been analyzed by taking into account the anisotropy of the mass and the two - dimensional nature of the phase dynamics . based on the feynman - vernon approach , the effective one - dimensional action is derived by integrating out the degree of freedom of the lc circuit . we found that the coupling to the lc circuit gives negligible reduction for ta escape rate . on the other hand , we also found that the mqt escape rate is considerably reduced due to the coupling between the junction and the lc circuit . more importantly , the temperature dependence of the mqt escape rate for the ybco junction coupled with the lc circuit is quite different from that without the lc circuit . these theoretical results are in an excellent agreement with experimental data of the ybco biepitaxial josephson junction . @xcite therefore we can conclude that _ the anisotropy of the mass and the two - dimensional nature of the potential profile due to the coupling to the lc circuit are quite important and essential to understand macroscopic quantum phenomena and qubit operation in such systems . _ there , however , remain a question , which can not be treated by the present approach . as seen from fig . 5 , the extended circuit model can not explain a hump structure of @xmath225 near 0.1 k. @xcite we note that such a characteristic behavior has been observed in a dc - squid ( superconducting quantum interference device ) system composed only by low-@xmath4 superconductors . @xcite while the deviation between our theory and the experimental data is left for a future problem , it might be attributed to the fact that our model neglects thermal / dynamical population from the quasi - ground state to the excited states in the metastable well . we note that in the nonadiabatic cases ( @xmath237 ) corresponding to the mqt experiment for the ybco biepitaxial junction @xcite more quantum levels due to excitation of the lc circuit are relevant to the decay process than in a simple one - dimensional system . in order to investigate the thermal / dynamical population effect , we have to solve a master equation to obtain population probabilities of each level by the larkin and ovchinnikov theory . @xcite this consideration may explain the anomaly of the escape rate @xmath203 near the crossover temperature . finally we would like to comment advantages of the junction - lc coupling to the qubit and quantum optics applications . the system that we considered in this paper can be regarded as an artificial atom ( the josephson junction ) coupled to the quantized electromagnetic field ( the lc circuit ) . therefore the appearance of several interesting phenomena relating to quantum optics , @xcite e.g. , the vacuum rabi oscillation , @xcite generation of a non - classical state of the lc system , @xcite and laser oscillation @xcite is expected also in the ybco biepitaxial junctions . additionally the lc circuit will act as a quantum information bus . @xcite therefore the entanglement or the coupling between separated high-@xmath4 qubits and eventually a high-@xmath4 version of the circuit - qed system @xcite will be realized in such biepitaxial junctions . these studies will open up the possibility of future applications for high-@xmath4 superconductor materials . we would like to thank j. ankerhold , a. barone , m. fogelstrm , a. a. golubov , d. r. gulevich , g. johansson , j. r. kirtley , t. lfwander , f. lombardi , j. p. pekola , g. rotoli , v. s. shumeiko , and f. tafuri for useful discussions . one of the authors ( s. k. ) would like to thank the applied quantum physics laboratory at the chalmers university of technology for its hospitality during the course of this work . this work was supported by nano - ned program under project no . tcs.7029 , crest ( jst ) , eu nanoxide , eu strep project midas , the swedish research council ( vr ) under the linnaeus center on engineered quantum systems , the swedish research council ( vr ) under the project macroscopic quantum tunneling and coherence in superconductive d - wave junctions , the knut and alice wallenberg foundation , the swedish foundation for strategic research ( ssf ) under the project oxide , and the jsps - rsas scientist exchange program . if the temperature @xmath162 is much smaller than the crossover temperature @xmath107 , the finite - temperature mqt escape rate ( in the weak retardation limit ) is given by the product of @xmath164 and a finite - temperature correction as @xcite @xmath238 , \end{aligned}\ ] ] where @xmath239 is the finite - temperature correction to the bounce action and is given in term of the spectral density @xmath78 as @xmath240 . \end{aligned}\ ] ] in this equation @xmath241 , where @xmath242 is the zero temperature bounce . for the spectral density eq . ( 14 ) , the enhancement function takes the form @xmath243 \nonumber\\ & \approx & \frac{72 m \omega_\mathrm{lc}^3}{\hbar \omega_p^2 } ( 1- \gamma^2 ) e^{-\hbar \beta \omega_\mathrm{lc } } , \end{aligned}\ ] ] for @xmath244 . thus the thermal enhancement shows exponentially weak temperature dependence as long as @xmath245 . on the other hand , for a damped system with @xmath246 , we get algebraic large enhancement @xmath247 . @xcite thus the coupling to the lc circuit gives the weak finite - temperature correction to the zero - temperature mqt escape rate @xmath164 for @xmath248 . note that , in the actual ybco junction , @xmath249 . in sec . c , we show the finite - temperature mqt escape rate @xmath194 for @xmath161 . in this appendix , we show an expression for the escape rate near @xmath107 . the quantum enhancement factor @xmath110 in the ta escape rate @xmath213 increases with decreasing the temperature , and diverges as @xmath214 . this unphysical divergence is due to the appearance of a bounce trajectory for temperature below @xmath107 and hence the breakdown of the gaussian approximation . @xcite a crossover region is characterized by the condition @xmath250 in which the dimensionless parameter @xmath251 is given by @xcite @xmath252 \sqrt { \frac{m \beta}{2 b_4 } } . \end{aligned}\ ] ] here the coefficient @xmath253 measures the strength of the anharmonicity of the potential @xmath254 and is given by @xmath255 . \end{aligned}\ ] ] in the crossover region , functional integral can not be done by steepest descents , but requires a more careful treatment , leading to a escape rate of the form @xcite @xmath256 \nonumber\\ & \times & \exp \left [ -\beta v_0^\mathrm{1d } + \frac{\hbar \left ( \beta - \beta_\mathrm{co } \right)^2}{2 | \tau'| } \right ] , \end{aligned}\ ] ] with @xmath257 . in this equation , @xmath258 , @xmath259 is the dimensionless prefactor , and @xmath260 is the energy derivative of the bounce period . this expression applies for the temperature @xmath162 slightly above and below @xmath107 and smoothly interpolates between the high - temperature escape rate @xmath213 and the low - temperature escape rate @xmath194 as demonstrated in fig . the ta escape rate @xmath213 is represented by a dotted ( red ) line and the mqt escape rate @xmath194 as a dashed ( blue ) line . the crossover formula @xmath261 ( b3 ) is represented by a solid ( black ) line and smoothly matches onto these functions below and above the crossover temperature @xmath107 ( arrow ) . parameters are @xmath207 @xmath208a , @xmath210 pf , @xmath2621.7 nh , and @xmath204 pf.,width=264 ] 99 s. takagi , _ macroscopic quantum tunneling _ ( cambridge univ . press , cambridge , 2002 ) . j. ankerhold , _ quantum tunneling in complex systems : the semiclassical approach _ ( springer - verlag , berlin , 2007 ) . u. weiss , _ quantum dissipative systems _ ( world scientific , singapore , 2008 ) . e. simanek , _ inhomogeneous superconductors : granular and quantum effects _ ( oxford univ . press , 1994 ) . t. bauch , f. lombardi , f. tafuri , a. barone , g. rotoli , p. delsing , and t. claeson , phys . lett . * 94 * , 087003 ( 2005 ) . k. inomata , s. sato , k. nakajima , a. tanaka , y. takano , h. b. wang , m. nagao , h. hatano , and s. kawabata , phys . . lett . * 95 * , 107005 ( 2005 ) . x. y. jin , j. lisenfeld , y. koval , a. lukashenko , a. v. ustinov , and p. mller , phys . lett . * 96 * , 177003 ( 2006 ) . t. matsumoto , h. kashiwaya , h. shibata , s. kashiwaya , s. kawabata , h. eisaki , y. yoshida , and y. tanaka , supercond . technol . * 20 * , s10 ( 2007 ) . s. x. li , w. qiu , s. han , y. f. wei , x. b. zhu , c. z. gu , s. p. zhao , and h. b. wang , phys . * 99 * , 037002 ( 2007 ) . h. kashiwaya , t. matsumoto , h. shibata , s. kashiwaya , h. eisaki , y. yoshida , s. kawabata , and y. tanaka , j. phys . . jpn . * 77 * , 104708 ( 2008 ) . a. yurgens , m. torstensson , l. x. you , t. bauch , d. winkler , i. kakeya , and k. kadowaki , physica c * 468 * , 674 ( 2008 ) . t. bauch , t. lindstrm , f. tafuri , g. rotoli , p. delsing , t. claeson , and f. lombardi , science * 311 * , 57 ( 2006 ) . k. inomata , s. sato , m. kinjo , n. kitabatake , h. b. wang , t. hatano , and k. nakajima , supercond . technol . * 20 * , s105 ( 2007 ) . h. kashiwaya , t. matsumoto , s. kashiwaya , h. shibata , h. eisaki , y. yoshida , s. kawabata , and y. tanaka , physica c * 468 * , 1919 ( 2008 ) . f. tafuri and j. r. kirtley , rep . 68 * , 2573 ( 2005 ) . a. a. yurgens , supercond . * 13 * , r85 ( 2000 ) . l b. ioffe , v. b. geshkenbein , m. v. feigelman , a. l. fauchre , and g. blatter , nature * 398 * , 679 ( 1999 ) . a. blais and a. m. zagoskin , phys . a * 61 * , 042308 ( 2000 ) . g. blatter , v. b. geshkenbein , and l. b. ioffe , phys . b * 63 * , 174511 ( 2001 ) . y. v. fominov , a. a. golubov , and m. kupriyanov , jetp lett . * 77 * , 587 ( 2003 ) . m. h. s. amin , a. y. smirnov , a. m. zagoskin , t. lindstrm , s. a. charlebois , t. claeson , and a. y. tzalenchuk , phys . b * 71 * , 064516 ( 2005 ) . t. kato , a. a. golubov , and y. nakamura , phys . b * 76 * , 172502 ( 2007 ) . s. kawabata , s. kashiwaya , y. asano , and y. tanaka , phys . b * 70 * , 132505 ( 2004 ) . s. kawabata , s. kashiwaya , y. asano , and y. tanaka , phys . b * 72 * , 052506 ( 2005 ) . s. kawabata , s. kashiwaya , y. asano , and y. tanaka , physica e ( amsterdam ) * 29 * , 669 ( 2005 ) . s. kawabata , s. kashiwaya , y. asano , y. tanaka , t. kato , and a. a. golubov , supercond . technol . * 20 * , s6 ( 2007 ) . s. kawabata , a. a. golubov , ariando , c. j. m. verwijs , h. hilgenkamp , and j. r. kirtley , phys . b * 76 * , 064505 ( 2007 ) . t. yokoyama , s. kawabata , t. kato , and y. tanaka , phys . b * 76 * , 134501 ( 2007 ) . t. umeki , t. kato , t. yokoyama , y. tanaka , s. kawabata , and s. kashiwaya , physica c * 463 - 465 * , 157 ( 2007 ) . f. tafuri , f. carillo , f. lombardi , f. miletto granozio , f. ricci , u. scotti di uccio , a. barone , g. testa , e. sarnelli , and j. r. kirtley , phys . rev . b * 62 * , 14431 ( 2000 ) . f. lombardi , f. tafuri , f. ricci , f. miletto granozio , a. barone , g. testa , e . sarnelli , j. r. kirtley , and c. c. tsuei , phys . * 89 * , 207001 ( 2002 ) . a. golubov and f. tafuri , phys . b * 62 * , 15200 ( 2000 ) . g. rotoli , t. bauch , t. lindstrm , d. stornaiuolo , f. tafuri , and f. lombardi , phys . rev . b * 75 * , 144501 ( 2007 ) . f. lombardi , t. bauch , g. rotoli , t. lindstrm , j. johansson , k. cedergren , f. tafuri , and t. claeson , ieee trans . appl . supercond . * 17 * 653 ( 2007 ) . g. schn and a. d. zaikin , phys . rep . * 198 * , 237 ( 1990 ) . y. c. chen , j. low temp . phys . * 65 * , 133 ( 1986 ) . b. r. trees , y. h. helal , j. s. schiffrin , and b. m. siller , phys . b * 76 * , 224513 ( 2007 ) . a. o. caldeira and a. j. leggett , ann . ( ny ) * 149 * , 374 ( 1983 ) . h. grabert and u. weiss , phys . lett . * 53 * , 1787 ( 1984 ) . m. v. fistul , phys . b * 75 * , 014502 ( 2007 ) . p. g. wolynes , phys . lett . * 47 * , 968 ( 1981 ) . d. esteve , m. h. devoret , and j. m. martinis , phys . b * 34 * , 158 ( 1986 ) . j. m. martinis and h. grabert , phys . b * 38 * , 2371 ( 1988 ) . w. zwerger , z. phys . b * 51 * , 301 ( 1983 ) . y. yasui , t. takaai , and t. ootsuka , j. phys . 34 * , 2643 ( 2001 ) . i. m. gelfand and a. m. yaglom , j. math . phys . * 1 * , 48 ( 1960 ) . the retardation correction in the prefactor @xmath199 gives negligible contribution to @xmath124 in compared with that in the exponent @xmath263 $ ] . therefore , in the calculation of @xmath124 , we have ignored the retardation correction in @xmath199 . r. f. voss and r. a. webb , phys . lett . * 47 * , 265 ( 1981 ) . a. garg , phys . rev . b*51 * , 15592 ( 1995 ) . the junction capacitance @xmath8 was estimated by numerically calculating the electrostatic charge distribution in our biepitaxial junction geometry . in the previous study , @xcite the remarkable temperature dependence of @xmath264 near @xmath265 k has been attributed to the anomalous temperature dependence of the higher harmonic component [ @xmath266 of the josephson current , i.e. , @xmath267 . however the measured temperature dependence of the total josephson critical current @xmath268 is almost constant up to 1 k ( see inset of fig . 3 in ref . [ ] ) . in order to consistently explain these observations , the anomalous temperature dependence of @xmath269 should be completely compensated by @xmath270 . although opposite temperature - dependence between @xmath270 and @xmath271 has been observed experimentally in an in - plane @xmath13-wave junction and interpreted theoretically [ e. ilichev et al . 86 * , 5369 ( 2001 ) ] , the complete compensation between @xmath270 and @xmath269 for the @xmath272 @xmath13-wave junction is an open problem to be explained . we conjecture that the lc model adopted in this paper may provide more reasonable explanation for the anomalous hump structure . detailed study is an interesting future problem . f. balestro , j. claudon , j. p. pekola , and o. buisson , phys . lett . * 91 * , 158301 ( 2003 ) . a. i. larkin and y. n. ovchinnikov , sov . jetp * 64 * , 185 ( 1986 ) . l. mandel and e. wolf , _ optical coherence and quantum optics _ ( cambridge univ . press , cambridge , 1995 ) . j. johansson , s. saito , t. meno , h. nakano , m. ueda , k. semba , and h. takayanagi , phys . lett . * 96 * , 127006 ( 2006 ) . d. i. schuster , a. a. houck , j. a. schreier , a. wallraff , j. m. gambetta , a. blais , l. frunzio , j. majer , b. r. johnson , m. h. devoret , s. m. girvin , and r. j. schoelkopf , nature * 445 * , 515 ( 2007 ) . o. astafiev , k. inomata , a. o. niskanen , t. yamamoto , y. a. pashkin , y. nakamura , and j. s. tsai , nature * 449 * , 588 ( 2007 ) . t. p. spiller , k. nemoto , s. l. braunstein , w. j. munro , p. van loock , and g. j. milburn , new j. phys . * 8 * , 30 ( 2006 ) . j. majer , j. m. chow , j. m. gambetta , j. koch , b. r. johnson , j. a. schreier , l. frunzio , d. i. schuster , a. a. houck , a. wallraff , a. blais , m. h. devoret , s. m. girvin , and r. j. schoelkopf , nature * 449 * , 443 ( 2007 ) . j. q. you and f. nori , phys . b * 68 * , 064509 ( 2003 ) . a. blais , r. -s . huang , a. wallraff , s. m. girvin , and r. j. schoelkopf , phys . a * 69 * , 062320 ( 2004 ) . a. wallraff , d. i. schuster , a. blais , l. frunzio , r. -s . huang , j. majer , s. kumar , s. m. girvin , and r. j. schoelkopf , nature * 431 * , 162 ( 2004 ) . h. grabert and u. weiss , z. phys . b * 56 * , 171 ( 1984 ) . h. grabert , u. weiss , and p. hnggi , phys . * 52 * , 2193 ( 1984 ) . p. hnggi and w. hontscha , j. chem . phys . * 88 * , 4094 ( 1988 ) .

a foundation is laid for a theory of _ combinatorial groupoids _ , allowing us to use concepts like `` holonomy '' , `` parallel transport '' , `` bundles '' , `` combinatorial curvature '' etc . in the context of simplicial ( polyhedral ) complexes , posets , graphs , polytopes and other combinatorial objects . a new , holonomy - type invariant for cubical complexes is introduced , leading to a combinatorial `` theorema egregium '' for cubical complexes non - embeddable into cubical lattices . parallel transport of-complexes and maps is used as a tool for extending babson - kozlov - lovsz graph coloring results to more general statements about non - degenerate maps ( colorings ) of simplicial complexes and graphs .

math0510204

there has been new and encouraging evidence that the language and methods of the theory of groupoids , after being successfully applied in other major mathematical fields , offer new insights and perspectives for applications in combinatorics and discrete and computational geometry . the groupoids ( groups of projectivities ) have recently appeared in geometric combinatorics in the work of m. joswig @xcite , see also a related paper with izmestiev @xcite and the references to these papers , where they have been applied to toric manifolds , branched coverings over @xmath1 , colorings of simple polytopes etc . the purpose of this paper is to show that these developments should not be seen as isolated examples . quite the opposite , they serve as a motivation for further extensions and generalizations . as a first application we show how a cubical extension of joswig s groupoid provides new insight about cubical complexes non - embeddable into cubical lattices ( a question related to a problem of s.p . novikov which arose in connection with the @xmath2-dimensional ising model ) @xcite @xcite . the second application leads to a generalization , from graphs to simplicial complexes , of a recent resolution of the lov ' asz conjecture by babson and kozlov @xcite @xcite . combinatorial groupoids also appear ( implicitly ) in other contemporary combinatorial constructions and applications , see babson et al . @xcite , recent work of barcelo et al . on atkin - homotopy @xcite @xcite one of our central objectives is to advocate their systematic use , and to propose their recognition as a valuable tool for geometric and algebraic combinatorics . our point of departure is an observation that different problems , including the question of existence of an embedding @xmath3 of riemann manifolds , the existence and classification of embeddings ( immersions ) of cubical surfaces into hypercubes or cubical lattices , the existence of a coloring of a graph @xmath4 with a prescribed number of colors etc . , can all be approached from a similar point of view . the unifying theme and a single point of view is provided by the concept of a groupoid . in each of the listed cases there is a groupoid naturally associated to an object of the given category . for each of these groupoids there is an associated parallel transport , holonomy groups and other related invariants which serve as obstructions for the existence of morphisms indicated in figure [ fig : sphere7 ] . if @xmath5 is a riemannian manifold , the associated groupoid @xmath6 has @xmath7 as the set of objects while the morphism set @xmath8 consists of all linear isomorphisms @xmath9 arising from a parallel transport along piece - wise smooth curves from @xmath10 to @xmath11 . one of the manifestations of gauss `` theorema egregium '' is that the associated holonomy group is a metric invariant , consequently an isometry depicted in figure [ fig : sphere7 ] ( a ) is not possible . in a similar fashion a cubical `` theorema egregium '' ( theorem [ thm : cub - thm - egr ] ) provides an obstruction for an embedding of a cubical complex into a hypercube or a cubical lattice . as a consequence the cubulation ( quadrangulation ) of an annulus depicted in figure [ fig : sphere7 ] ( b ) does not admit a cubical embedding into a hypercube for the same formal reason ( incompatible holonomy ) a spherical cap can not be isometrically represented in the plane . perhaps it comes as a surprise that the graph coloring problem , figure [ fig : sphere7 ] ( c ) , can be also approached from a similar point of view . an analysis of holonomies ( `` parallel transport '' ) of diagrams of @xmath0-complexes of graphs ( simplicial complexes ) over the associated joswig groupoid eventually leads to a general result ( theorem [ thm : main ] ) which includes the `` odd '' case of babson - kozlov - lovsz coloring theorem as a special case . our objective is to lay a foundation for a program of associating groupoids to posets , graphs , complexes , arrangements , configurations and other combinatorial structures . this `` geometrization of combinatorics '' should provide useful guiding principles and transparent geometric language introducing concepts such as combinatorial holonomy , discrete parallel transport , combinatorial curvature , combinatorial bundles etc . the first steps in this direction are made in section [ sec : groupoids ] . section [ sec : cubical ] , devoted to cubical complexes , is an elaboration of the theme depicted in figure [ fig : sphere7 ] ( b ) . a holonomy @xmath12-valued invariant @xmath13 of a cubical complex is introduced . this leads to a `` combinatorial curvature '' @xmath14 and associated `` cubical theorema egregium '' ( theorem [ thm : cub - thm - egr ] ) which provides necessary conditions for the existence of an embedding / immersion @xmath15 of cubical complexes . section [ sec : gen - lov - conj ] , formally linked to figure [ fig : sphere7 ] ( c ) , introduces groupoids to the problem of coloring graphs and complexes . the focus is on the lovsz @xmath0-conjecture and its ramifications . parallel transport of graph and more general @xmath0-complexes over graphs / complexes is introduced and fundamental invariance of homotopy types of maps is established in propositions [ prop : transport ] and [ prop : prenos ] . this ultimately leads to general results about coloring of graphs and simplicial complexes ( theorems [ thm : main ] and [ thm : paradigm ] , corollary [ cor : lbk - gen ] ) . it is , or should be , well known that the concept of a _ group _ is sometimes not sufficient to deal with the concept of symmetry in general . groupoids , understood as groups with `` space - like properties '' , often alow us to handle objects which exhibit what is clearly recognized as symmetry although they admit no global automorphism whatsoever . unlike groups , groupoids are capable of describing reversible processes which can pass through a number of states . for example according to connes @xcite , heisenberg discovered quantum mechanics by considering the groupoid of quantum transitions rather than the group of symmetry . by definition a groupoid is a small category @xmath16 such that each morphism @xmath17 is an isomorphism . we follow the usual terminology and notation @xcite @xcite @xcite @xcite . perhaps the only exception is that we call the _ vertex _ ( isotropy ) group @xmath18 the _ holonomy _ group of @xmath19 at @xmath20 . if the groupoid @xmath19 is connected all holonomy groups @xmath21 are isomorphic and their isomorphism type is denoted by @xmath22 . the notion of groupoid is a common generalization of the concepts of space and group , i.e. the theory of groupoids allows us to treat spaces , groups and objects associated to them from the same point of view . a common generalization for the concept of a bundle @xmath23 over @xmath24 and a @xmath25-space @xmath23 is a @xmath19-space or more formally a diagram over the groupoid @xmath19 defined as a functor @xmath26 . in order to preserve the original intuition we use a geometric language which resembles the usual concepts as much as possible . in particular we clarify what is in this paper meant by a @xmath19-parallel transport on a `` bundle '' of spaces over a set @xmath27 . a ( naive ) `` bundle '' is a map @xmath28 . we assume that @xmath27 is a set and that @xmath29 is a topological space , so a bundle is just a collection of spaces ( fibres ) @xmath30 parameterized by @xmath27 . if all spaces @xmath30 are homeomorphic to a fixed `` model '' space , this space is referred to as the fiber of the bundle @xmath31 . suppose that @xmath19 is a groupoid on @xmath27 as the set of objects . in other words @xmath32 is a small category where @xmath33 , such that all morphisms @xmath17 are invertible . a `` connection '' or `` parallel transport '' on the bundle @xmath34 is a functor ( diagram ) @xmath35 such @xmath36 for each @xmath37 . informally speaking , the groupoid @xmath19 provides a `` road map '' on @xmath27 , while the functor @xmath38 defines the associated transport from one fibre to another . sometimes it is convenient to view the bundle @xmath34 as a map @xmath39 . then to define a `` connection '' on this bundle is equivalent to enriching the map @xmath40 to a functor @xmath35 . the following definition introduces the first in a series of _ combinatorial groupoids_. [ def : comb - groupod ] suppose that @xmath41 is a ( not necessarily finite ) poset . suppose that @xmath42 and @xmath43 two families of subposets of @xmath44 . choose @xmath45 . if for some @xmath46 both @xmath47 and @xmath48 , then the posets @xmath49 are called @xmath50-adjacent , or just adjacent if @xmath50 is not specified . define @xmath51 as a small category over @xmath52 as the set of objects as follows . for two @xmath50-adjacent objects @xmath53 and @xmath54 , an _ elementary morphism _ @xmath55 is an isomorphism @xmath56 of posets which leaves @xmath50 point - wise fixed . a morphism @xmath57 from @xmath58 to @xmath59 is an isomorphism of posets @xmath58 and @xmath59 which can be expressed as a composition of elementary morphisms . given two adjacent objects @xmath53 and @xmath54 , an elementary morphism @xmath60 may not exist at all , or if it exists it may not be unique . in case it exists and is unique it will be frequently denoted by @xmath61 and sometimes referred to as a `` flip '' from @xmath53 to @xmath54 . in this case a morphism @xmath62 is by definition a composition of flips @xmath63 * caveat : * here we adopt a useful convention that @xmath64 for each two composable maps @xmath65 and @xmath66 . the notation @xmath67 is often given priority over the usual @xmath68 if we want to emphasize that the functions act on the points from the right , that is if the arrows in the associated formulas point from left to the right . definition [ def : comb - groupod ] can be restated in a much more general form . for example the ambient poset @xmath44 can be replaced by a small category @xmath69 , @xmath42 and @xmath43 by families of subobjects of @xmath69 , elementary morphisms are defined as commutative diagrams etc . however our objective in this paper is not to explore all the possibilities . instead we create an `` ecological niche '' for combinatorial groupoids which may be populated by new examples and variations as the theory develops . suppose that @xmath44 is a ranked poset of depth @xmath70 with the associated rank function @xmath71 $ ] . let @xmath72 be the @xmath19-groupoid described in definition [ def : comb - groupod ] associated to the families @xmath73 and @xmath74 . it is clear that other `` rank selected '' groupoids can be similarly defined . the definitions of groupoids @xmath19 and @xmath75 are easily extended from posets to simplicial , polyhedral , or other classes of cell complexes . if @xmath76 is a complex and @xmath77 the associated face poset , then @xmath78 and @xmath79 are groupoids associated to the poset @xmath80 . we will usually drop the subscript whenever it is clear from the context what is the ambient poset @xmath44 or complex @xmath76 . suppose that @xmath76 is a pure , @xmath81-dimensional simplicial complex . let @xmath82 be the associated @xmath75-groupoid corresponding to ranks @xmath81 and @xmath83 . then the _ groups of projectivities _ @xmath84 , introduced by joswig in @xcite , are nothing but the holonomy groups of the groupoid @xmath82 . for this reason the groupoid @xmath82 is in the sequel often referred to as joswig s groupoid and denoted by @xmath85 . @xmath85 is _ connected _ as a groupoid if and only if @xmath76 is `` strongly connected '' in the sense of @xcite . a simplicial map of simplicial complexes is non - degenerate if it is 11 on simplices . the following definition extends this concept to the case of posets . [ def : nondeg ] a monotone map of posets @xmath86 is non - degenerate if the restriction of @xmath65 on @xmath87 induces an isomorphism of posets @xmath87 and @xmath88 for each element @xmath89 . similarly , a map of simplicial , cubical or more general cell complexes is non - degenerate if the associated map of posets is non - degenerate . in this case we say that @xmath44 is _ mappable _ to @xmath90 while a non - degenerate map @xmath86 is often referred to as a _ combinatorial immersion _ from @xmath44 to @xmath90 . a graph homomorphism @xcite @xmath91 can be defined as a non - degenerate map of associated @xmath92-dimensional cell complexes . a @xmath70-coloring of a graph @xmath25 is a non - degenerate map ( graph homomorphism ) @xmath93 where @xmath94 is a complete graph on @xmath70 vertices . [ prop : functor ] suppose that @xmath44 and @xmath90 are ranked posets of depth @xmath70 and let @xmath95 be a non - degenerate map . then there is an induced map ( functor ) @xmath96 of the associated @xmath75-groupoids . moreover , @xmath97 induces an inclusion map @xmath98 of the associated holonomy groups . the following two problems serve as a motivation for studying non - degenerate morphisms between cubical complexes , the case of embeddings being of special importance . 1 . novikov ) characterize @xmath99-dimensional complexes that admit a ( cubical ) embedding , or more generally a `` combinatorial immersion '' , into the standard cubical lattice of @xmath100 for some @xmath81 , @xcite @xcite . 2 . ( n. habegger ) suppose we have two cubulations of the same manifold . are they related by the bubble moves , @xcite ? the first question was according to @xcite motivated by a problem from statistical physics ( @xmath2-dimensional ising model ) . the second initiated the study of `` bubble moves '' on cubical subdivisions ( cubulations ) of manifolds and complexes which are analogs of stellar operations in simplicial category . for these and related questions about cubical complexes the reader is referred to @xcite @xcite @xcite @xcite @xcite @xcite @xcite @xcite @xcite @xcite and the references in these papers . recall that a cell complex @xmath76 is cubical if it is a regular @xmath101-complex such that the associated face poset @xmath80 is cubical in the sense of the following definition . [ def : cubical ] @xmath44 is a cubical poset if : 1 . for each @xmath102 , the subposet @xmath87 is isomorphic to the face poset of some cube @xmath103 2 . @xmath44 is a semilattice in the sense that if a pair @xmath104 is bounded from above then it has the least upper bound . if a space @xmath24 comes equipped with a standard cubulation , clear from the context , this cubical complex is denoted by @xmath105 , the associated @xmath99-skeleton is denoted by @xmath106 etc . for example @xmath107 is the @xmath99-skeleton of the standard cubulation of the @xmath81-cube . the group @xmath108 of all symmetries of a @xmath99-cube is isomorphic to the group of all signed , permutation @xmath109-matrices . its subgroup of all matrices with even number of @xmath110-entries is denoted by @xmath111 . [ thm : cubical - lattice ] suppose that @xmath76 is a @xmath99-dimensional cubical complex which is embeddable / mappable to @xmath112 , the @xmath99-dimensional skeleton of the standard cubical decomposition @xmath113 of @xmath114 . then @xmath115 for each cube @xmath116 . moreover , if @xmath117 where @xmath118 is a subcomplex of @xmath76 isomorphic to the @xmath99-skeleton of the @xmath119-dimensional cube @xmath120 , then @xmath121 * proof : * the finiteness of the holonomy group @xmath84 allows us to assume that the complex @xmath76 is also finite , hence embeddable / mappable to @xmath122^n\}_{(k)}$ ] for some integer @xmath123 . the case @xmath124 of the theorem is a consequence of the simple fact that each closed edge - path in @xmath125 is of even length , so let us assume that @xmath126 . each finite cubical subcomplex of @xmath112 is embeddable in a cube @xmath107 for sufficiently large @xmath81 . indeed , the @xmath92-complex @xmath127 is a hamiltonian graph with @xmath128-vertices which implies that the `` chain '' @xmath129\}\subset \{\mathbb{r}^1\}$ ] is embeddable in the @xmath92-skeleton of the cube @xmath130 if @xmath131 . since @xmath132 , we observe that @xmath129^n\}$ ] is embeddable in @xmath130 for some @xmath81 , hence the same holds for their @xmath99-skeletons . in light of proposition [ prop : functor ] , the holonomy group @xmath84 is a subgroup of @xmath133 for some @xmath134 so it is sufficient to establish the result for the complex @xmath135 where @xmath136 . as a preliminary step let us recall some basic facts about the @xmath81-cube @xmath137 . assume that @xmath138 and let @xmath139 $ ] be the standard orthonormal frame in @xmath100 . let @xmath140 , respectively @xmath141 , be the _ front _ , respectively _ back _ face of @xmath137 in the direction of the basic vector @xmath142 . given a face @xmath143 , let @xmath144 \mid \sigma\subset i^d_{x_i=0}\ } \quad \mbox{\rm and } \quad i_1(\sigma):= \{i\in [ d ] \mid \sigma\subset i^d_{x_i=1}\}.\ ] ] by definition @xmath145 . let @xmath146\setminus(i_0(\sigma)\cup i_1(\sigma))$ ] . a cube @xmath147 is a front face , respectively back face of @xmath148 , if for some @xmath149 , respectively @xmath150 . suppose that @xmath58 and @xmath53 are two adjacent @xmath99-dimensional faces in @xmath130 such that @xmath151 is their common @xmath152-dimensional face . let @xmath153 be the associated element in the combinatorial groupoid of @xmath107 , section [ subsec : comb - group ] . recall that @xmath154 is the isomorphism of cell complexes ( posets ) which keeps the subcomplex @xmath155 fixed . since @xmath58 and @xmath53 are faces of a regular cube , there is a unique isometry @xmath156 of faces @xmath58 and @xmath53 which keeps the common face @xmath157 point - wise fixed . let @xmath158 and @xmath159 be the @xmath99-dimensional linear subspaces in @xmath100 tangent ( that is parallel ) to @xmath58 and @xmath53 respectively , and let @xmath160 be the corresponding isometry associated to @xmath161 . the introduction of these maps allows us to pass from the combinatorial groupoid @xmath162 to the isomorphic groupoid @xmath163 of isometries of @xmath99-faces in @xmath137 . the following lemma is the essential step in describing the action of the groupoid @xmath164 on the set of _ admissible _ frames on @xmath99-faces of the cube @xmath137 . by definition a frame @xmath165 $ ] is admissible for a face @xmath148 if @xmath166 for each @xmath167 where @xmath168 and @xmath169 is some permutation of the set @xmath170 . the _ sign characteristic _ @xmath171 of an admissible frame @xmath172 is the number of negative signs in the sequence @xmath173 . * lemma : * the linear isometry @xmath174 maps a frame @xmath175 $ ] , admissible for @xmath58 , to a frame @xmath176 $ ] admissible for the face @xmath53 where @xmath177 for each @xmath178 , such that @xmath179 and @xmath180 is a permutation of @xmath181 . moreover , @xmath182 if @xmath157 is either a front face for both @xmath58 and @xmath53 or a back face for both @xmath58 and @xmath53 . in the opposite case the parity of the sign characteristic changes , more precisely @xmath183 . the proof of the lemma is by inspection with figure [ fig : flip ] illustrating the case when the sign characteristic of a frame is affected by the flip from @xmath58 to @xmath53 . suppose that @xmath184 is a sequence of adjacent @xmath99-faces in @xmath107 and let @xmath185 be the associated element in @xmath186 . let @xmath187 be the isometry associated to @xmath188 and @xmath189 the corresponding linear isometry . assume that @xmath190 which implies that @xmath191 $ ] is the canonical admissible frame for @xmath58 . by a successive application the lemma on pairs @xmath192 , we observe that @xmath193= [ \epsilon_1e_{\nu_1},\epsilon_2e_{\nu_2},\ldots , \epsilon_ke_{\nu_k}]$ ] where @xmath169 is a permutation of the set @xmath194 such that the number of negative signs in the sequence @xmath195 must be even . from here we deduce that @xmath196 . one easily proves by inspection that @xmath197 . this in turn provides a key step for the proof that @xmath198 which completes the proof of the second part of the theorem . [ cor : ribica ] the complex @xmath76 depicted in figure [ fig : ribica ] is not embeddable ( mappable ) to a cubical lattice ( hypercube ) of any dimension . indeed , @xmath200 \in \pi(k,\sigma)\ ] ] while by theorem [ thm : cubical - lattice ] only signed permutation matrices with even number of @xmath110-entries can arise as holonomies of subcomplexes of cubical lattices ! theorem [ thm : cubical - lattice ] and corollary [ cor : ribica ] serve as a motivation for introducing an invariant @xmath201 of a cubical complex @xmath76 and an associated combinatorial curvature @xmath14 of @xmath76 . both invariants , especially the invariant @xmath14 , can be used for testing if there exists an embedding ( non - degenerate mapping ) @xmath202 of cubical complexes @xmath76 and @xmath203 . the @xmath12-characteristics @xmath13 is cruder than @xmath14 , however it possesses an additional property that it is invariant with respect to bubble moves , proposition [ thm : bubble ] . suppose that @xmath76 is a @xmath99-dimensional cubical complex and let @xmath84 be its combinatorial holonomy group based at @xmath116 . by definition let @xmath204 if @xmath115 for all @xmath148 , and @xmath205 in the opposite case . given a @xmath99-dimensional cubical complex , let let @xmath206 where @xmath207 is the number of @xmath178-dimensional cells ( cubes ) in @xmath76 . in particular @xmath208 if @xmath209 . the _ combinatorial curvature _ of @xmath76 is the number @xmath210 which is by definition @xmath211 if @xmath209 . both @xmath13 and @xmath14 are global invariants of the cubical @xmath99-complex @xmath76 . one can introduce a local invariant @xmath212 as the minimum length @xmath213 of a closed chain @xmath214 of adjacent cubes in @xmath76 such that @xmath215 then @xmath216 is the combinatorial curvature of @xmath76 at @xmath148 and @xmath217 . this not only provides a answer in the case of hypercubes ( cubical lattices ) but resembles one of the usual definitions of the curvature as the ( limit ) quotient of locally defined quantities . the following theorem summarizes the monotonicity properties of invariants @xmath218 and @xmath14 in the form suitable for immediate applications . ( cubical `` theorema egregium'')[thm : cub - thm - egr ] if there exists an embedding or more generally a non - degenerate mapping ( cubical immersion ) @xmath202 of @xmath99-dimensional cubical complexes @xmath76 and @xmath203 then @xmath219 in other words the `` curvature '' of @xmath76 must not exceed the `` curvature '' of @xmath203 if @xmath76 is to be embedded in @xmath203 . the following result shows the relevance of the invariant @xmath13 for the problem of habegger ( problem @xmath220 ) . it says that @xmath13 is invariant with respect to cubical modifications known as `` bubble moves '' , @xcite @xcite . . ] [ thm : bubble ] suppose that @xmath76 is a @xmath99-dimensional cubical complex . then @xmath13 is invariant with respect to bubble moves , that is @xmath221 where @xmath222 is the complex obtained from @xmath76 by a sequence of bubble moves . * proof : * suppose that @xmath222 is obtained from @xmath76 by a bubble move . by definition this means that @xmath222 can be obtained from @xmath76 by excising @xmath223 from @xmath76 and replacing it by @xmath224 where @xmath223 and @xmath224 are complementary balls in the boundary of the @xmath119-cube , see figure [ fig : bubble ] . each cubical path @xmath188 in @xmath76 which enters the ball @xmath225 can be modified to a cubical path @xmath226 by replacing each cubical fragment of @xmath188 in @xmath223 by a corresponding fragment in @xmath224 . then the equality @xmath227 follows from @xmath228 . @xmath229 if @xmath230 and @xmath231 then there does not exists an embedding ( combinatorial immersion ) from @xmath76 to @xmath203 , even after @xmath76 and @xmath203 are modified by some bubble moves . in this section we demonstrate how the groupoid of projectivities introduced by joswig can be used as a basis for a construction of parallel transport of graph and more general @xmath0-complexes . in this framework we develop a general conceptual approach to the lovsz @xmath0-conjecture , recently resolved by e. babson and d. kozlov @xcite @xcite , and extend their result both from graphs to simplicial complexes ( theorem [ thm : main ] , corollary [ cor : lbk - gen ] ) and to other test graphs ( theorem [ thm : paradigm ] ) . one of central themes in topological combinatorics , after the landmark paper of laszlo lov ' asz @xcite where he proved the classical kneser conjecture , has been the study and applications of graph complexes . the underlying theme is to explore how the topological complexity of a graph complex @xmath232 reflects in the combinatorial complexity of the graph @xmath25 itself . the results one is usually interested in come in the form of inequalities @xmath233 , or equivalently in the form of implications @xmath234 where @xmath235 is a topological invariant of @xmath232 , while @xmath236 is a combinatorial invariant of the graph @xmath25 . the most interesting candidate for the invariant @xmath237 has been the chromatic number @xmath238 of @xmath25 , while the role of the invariant @xmath239 was played by the `` connectedness '' of @xmath232 , its equivariant index , the height of an associated characteristic cohomology class etc . , see @xcite @xcite @xcite @xcite for recent accounts . the famous result of lov ' asz quoted above is today usually formulated in the form of an implication @xmath240 where @xmath241 is ( together with the neighborhood complex , `` box complex '' etc . ) one of avatars of the homotopically unique @xmath12-graph complex of @xmath25 , @xcite @xcite . this complex is a special case of a general graph complex @xmath242 ( also introduced by l. lov ' asz ) , a cell complex which functorially depends on the input graphs @xmath243 and @xmath25 . an outstanding conjecture in this area , referred to as the `` lov ' asz conjecture '' , was that one obtains a better bound if the graph @xmath244 in ( 1 ) is replaced by an odd cycle @xmath245 . more precisely lov ' asz conjectured that @xmath246 this conjecture was confirmed by babson and kozlov in @xcite , see also @xcite for a more detailed exposition and @xcite @xcite . ] for subsequent developments . our objective is to develop methods which both offer a simplified approach to the proof of implication ( [ eq : lovasz - conj ] ) , at least in the case when @xmath99 is odd , and providing new insight , open a possibility of proving similar results for other classes of ( hyper)graphs and simplicial complexes . an example of such a result is theorem [ thm : main ] . one of its corollaries is the following implication , @xmath247 which , under a suitable assumption on the complex @xmath248 and the assumption that integer @xmath99 is odd , extends ( [ eq : lovasz - conj ] ) to the case of pure @xmath81-dimensional simplicial complexes . a consequence of ( [ eq : complex - conn ] ) is theorem [ thm : paradigm ] which provides a `` receipt '' how to for odd @xmath99 generate new examples of `` homotopy test graphs '' @xcite , that is graphs @xmath249 which satisfy the formula @xmath250 examples include the lower three complexes depicted in figure [ fig : piramida ] which all have chromatic number @xmath251 and clique number @xmath2 . for the definition and basic properties of graph complexes @xmath252 the reader is referred to @xcite . more general @xmath0-complexes associated to simplicial complexes @xmath76 and @xmath203 are introduced in section [ sec : ramifications ] . section [ sec : generalities ] can be used as a glossary for basic concepts like `` bundles '' , `` parallel transport '' , `` connection '' etc . suppose that @xmath76 and @xmath203 are finite simplicial complexes and let @xmath99 be an integer such that @xmath253 . denote by @xmath254 the set of all vertices of @xmath76 . let @xmath255 be the set of all @xmath99-dimensional simplices in @xmath76 . define a bundle @xmath256 by the formula @xmath257},l)\ ] ] where @xmath258 is one of the @xmath0-complexes introduced in section [ sec : hom ] and @xmath259}$ ] is the simplex with @xmath260=\{1,\ldots , k+1\}$ ] as the set of vertices . by definition a typical cell in @xmath261},l)$ ] is of the form @xmath262 where @xmath263 is a collection of non - empty simplices in @xmath203 such that @xmath264 for @xmath265 and @xmath266 the corresponding cell in @xmath258 is described by a function @xmath267 such that @xmath268 for @xmath269 and @xmath270 * example : * the complex @xmath271},\delta^{[m+1]})$ ] is well known in combinatorics as the @xmath119-fold `` deleted product '' @xmath272})^{k+1}_\delta$ ] of @xmath273}$ ] , ( * ? ? ? * section 6.3 ) . it is well known that for @xmath124 the associated deleted square @xmath272})^{2}_\delta$ ] is homeomorphic to a @xmath274-dimensional sphere . in other words , @xmath275 is a spherical bundle naturally associated to the simplicial complex @xmath76 . our next goal , in the spirit of section [ sec : generalities ] , is to identify a groupoid on the set @xmath276 which acts on the bundle @xmath277 , i.e. a groupoid which provides a parallel transport of fibres of the bundle @xmath277 . it turns out ( proposition [ prop : connection ] ) that this groupoid is precisely the @xmath75-groupoid @xmath278 , associated to the @xmath99-skeleton of @xmath76 , introduced in section [ sec : groupoids ] . it came as a pleasant surprise that this groupoid has already appeared in geometric combinatorics @xcite @xcite . indeed , the _ groups of projectivities _ m. joswig introduced and studied in these papers are just the holonomy groups of a groupoid which we call the @xmath99-th _ groupoid of projectivities _ of @xmath76 and denote by @xmath279 . in these and subsequent papers @xcite @xcite @xcite , the groups of projectivities found interesting applications to toric manifolds , branched coverings over @xmath1 , colorings of simple polytopes , etc . the following excerpt from @xcite reveals the role this construction played as a key motivating example for more general concepts introduced in section [ sec : groupoids ] . 1 . `` ... for each ridge @xmath280 contained in two facets @xmath281 , there is a unique vertex @xmath282 which is contained in @xmath148 but not in @xmath157 . we define the _ perspectivity _ @xmath283 by setting @xmath284 ... the _ projectivity _ @xmath285 from @xmath53 to @xmath286 along @xmath66 is a concatenation @xmath287 of perspectivities . the map @xmath285 is a bijection from @xmath58 to @xmath286 . ... '' [ prop : connection ] for each simplicial complex @xmath76 and an auxiliary `` coefficient '' complex @xmath203 , there exists a canonical @xmath279-connection @xmath288 on the bundle @xmath277 . in other words the function @xmath289 can be enriched ( extended ) to a functor @xmath290 where @xmath291 is the @xmath99-th joswig s groupoid of projectivities of @xmath76 . * proof : * if @xmath292 is a perspectivity from @xmath58 to @xmath53 ( a `` flip '' in the language of section [ subsec : comb - group ] ) and if @xmath293 is a cell in @xmath294 , then @xmath295 is the map defined by @xmath296 . more generally , if @xmath297 is a `` projectivity '' between @xmath58 and @xmath286 , then @xmath298 or in other words @xmath299 which demonstrates in passing that the map @xmath300 depends only on the morphism @xmath301 alone , and not on the associated path @xmath302 . @xmath229 the main motivation for introducing the parallel transport of @xmath0-complexes is the lovsz conjecture and its ramifications . this is the reason why the case of graphs and the graph complexes deserves a special attention . additional justification for emphasizing graphs comes from the fact that graph complexes @xmath252 have been studied in numerous papers and today form a well established part of graph theory and topological combinatorics . the situation with simplicial complexes is quite the opposite . in order to extend the theory of @xmath0-complexes from graphs to the category of simplicial complexes , many concepts should be generalized and the corresponding facts established in a more general setting . one is supposed to recognize the main driving forces and to isolate the most desirable features of the theory . a result should be a dictionary / glossary of associated concepts , cf . table 1 . consequently , section [ sec : functor ] should be viewed as an important preliminary step , leading to the more general theory developed in sections [ sec : ramifications ] and [ sec : main ] . in order to simplify the exposition we assume , without a serious loss of generality , that all graphs @xmath303 are without loops and multiple edges . in short , graphs are @xmath92-dimensional simplicial complexes . let @xmath304 be the restriction of @xmath25 on the edge @xmath305 . following the definitions from section [ sec : natural ] the map @xmath306 where @xmath307 , can be thought of as a `` bundle '' over the graph @xmath25 , with @xmath308 in the role of the `` fibre '' over the edge @xmath309 . more generally , given a class @xmath19 of subgraphs of @xmath25 , say the subtrees , the chains , the @xmath99-cliques etc . , one can define an associated `` bundle '' @xmath310 by a similar formula @xmath311 , where @xmath312 . the parallel transport @xmath313 , for a given graph ( @xmath92-dimensional , simplicial complex ) @xmath243 , is a specialization of the parallel transport @xmath314 introduced in section [ sec : natural ] . for example if @xmath315 is the flip ( perspectivity ) between adjacent edges @xmath316 and @xmath317 in @xmath25 , and if @xmath318 is a cell in @xmath319 , then @xmath320 is defined by @xmath321 * fundamental observation : * the constructions of the connections @xmath314 , respectively @xmath313 , are quite natural and elementary but it is proposition [ prop : transport ] , respectively its more general relative proposition [ prop : prenos ] , that serve as an actual justification for the introduction of these objects . proposition [ prop : transport ] allows us to analyze the parallel transport of homotopy types of maps from the complex @xmath252 to complexes @xmath322 , where @xmath323 , providing a key for a resolution of the lovsz conjecture in the case when @xmath99 is an odd integer . implicit in the proof of proposition [ prop : transport ] is the theory of _ folds _ of graphs and the analysis of natural morphisms between graph complexes @xmath324 , where @xmath249 is a tree , as developed in @xcite @xcite @xcite . this theory is one of essential ingredients in the babson and kozlov spectral sequence approach to the solution of lovsz conjecture . some of these results are summarized in proposition [ prop : help ] , in the form suitable for application to proposition [ prop : transport ] . as usual @xmath325 is the graph - chain of vertex - length @xmath213 , while @xmath326 is the graph isomorphic to @xmath325 defined on a linearly ordered set of vertices @xmath327 . in this context the `` chain - flip '' is a generic name for the automorphism @xmath328 of the graph - chain such that @xmath329 for each @xmath178 . [ prop : help ] suppose that @xmath316 and @xmath317 are two distinct , adjacent edges in the graph @xmath25 . let @xmath330 be the chain - flip automorphism of @xmath331 and @xmath332 the associated auto - homeomorphism of @xmath333 . suppose that @xmath334 is an obvious embedding and @xmath335 the associated maps of graph complexes . then , * proof : * both statements are corollaries of babson and kozlov analysis of complexes @xmath324 , where @xmath249 is a tree , and morphisms @xmath339 , where @xmath340 is a subtree of @xmath249 and @xmath341 the associated embedding . our starting point is an observation that both @xmath342 and @xmath343 are retracts of the graph @xmath331 in the category of graphs and graph homomorphisms . the retraction homomorphisms @xmath344 , where @xmath345 and @xmath346 are examples of _ foldings _ of graphs . by the general theory @xcite @xcite , the maps @xmath347 and @xmath348 are homotopy equivalences . actually @xmath335 is a deformation retraction and @xmath349 is the associated embedding such that @xmath350 is the identity map . the part ( a ) of the proposition is an immediate consequence of the fact that @xmath351 is an identity map . it follows that @xmath352 , and in light of the fact that @xmath353 and @xmath354 are homotopy inverses to each other , we conclude that @xmath355 . [ prop : transport ] suppose that @xmath361 are distinct vertices in @xmath25 such that @xmath362 . let @xmath363 be the inclusion map of graphs and @xmath364 the associated map of @xmath365 complexes . then the following diagram commutes up to a homotopy , * proof : * the diagram ( [ eqn : diagram1 ] ) can be factored as @xmath367 where @xmath368 and @xmath369 are obvious inclusions of indicated graphs such that @xmath370 . then the result is a direct consequence of proposition [ prop : help ] , part ( b ) . @xmath229 the proof @xcite of lov ' asz conjecture splits into two main branches , corresponding to the parity of a parameter @xmath70 , where @xmath70 is an integer which enters the stage as the size of the vertex set of the complete graph @xmath94 . the first branch relies on theorem 2.3 . cit . ) , more precisely on part ( b ) of this result , while the second branch is founded on theorem 2.6 . both theorems are about the topology of the graph complex @xmath371 . theorem 2.3 . ( b ) is a statement about the height of the first stiefel - whitney class , equivalently the conner - floyd index @xcite of the @xmath12-space @xmath371 . theorem 2.6 . claims that for @xmath70 even , @xmath372 is a zero homomorphism where @xmath373 is the homomorphism associated to the continuous map @xmath374 which in turn comes from the inclusion @xmath375 . [ thm : glavna ] suppose that @xmath376 is an inclusion map , @xmath377 a nontrivial automorphism of @xmath244 , and @xmath378 the associated maps of graph complexes . then the following diagram is commutative up to a homotopy @xmath379 * proof : * assume that the consecutive vertices of @xmath380 are @xmath381 and let @xmath382 be the associated sequence of edges where by convention @xmath383 . identify the graph @xmath244 to the subgraph @xmath384 of @xmath380 . by iterating proposition [ prop : transport ] we observe that the diagram @xmath385 is commutative up to a homotopy , where @xmath386 . the proof is completed by the observation that @xmath387 in the groupoid @xmath388 . @xmath229 * proof : * it is sufficient to observe that for @xmath391 , @xmath392 is an even dimensional sphere and that the automorphism @xmath393 from the diagram ( [ eqn : diagram3 ] ) is in this case essentially an antipodal map . it follows that @xmath393 changes the orientation of @xmath394 and as a consequence @xmath395 . @xmath229 graphs are viewed as @xmath92-dimensional simplicial complexes . graph homomorphisms are special cases of _ non - degenerate _ simplicial maps of simplicial complexes , see @xcite @xcite and definition [ def : nondeg ] . the definition of @xmath252 is extended to the case of @xmath0-complexes @xmath396 of simplicial complexes @xmath76 and @xmath203 . the groupoids needed for the definition of the parallel transport of @xmath0-complexes are described in section [ sec : natural ] . theory of folds for graph complexes @xcite @xcite is extended in section [ sec : tree - like ] to the case of @xmath0-complexes in sufficient generality to allow `` parallel transport '' of homotopy types of maps between graph complexes . this development eventually leads to theorem [ thm : main ] which extends theorem [ thm : glavna ] to the case of @xmath0-complexes @xmath396 and represents the currently final stage in the evolution of theorem 2.6 . from @xcite . graphs & simplicial complexes + trees & tree - like complexes + foldings of graphs & vertex collapsing of complexes + graph homomorphisms & non - degenerate simplicial maps + @xmath252 & @xmath396 + chromatic number @xmath238 & chromatic number @xmath397 + 1 . for each two vertices @xmath405 , if @xmath406 then @xmath407 , 2 . for each simplex @xmath116 , the join @xmath408 of all sets ( or @xmath211-dimensional complexes ) @xmath409 is a subcomplex of @xmath203 . [ exam : companion ] the definition of the complex @xmath396 is a natural extension of @xmath252 and reduces to it if @xmath76 and @xmath203 are @xmath92-dimensional complexes . moreover , @xmath414 where @xmath415 is the simplicial complex of all cliques in a graph @xmath248 . the set @xmath402 is easily identified as the @xmath211-dimensional skeleton of the cell - complex @xmath396 . moreover , the reader familiar with @xcite can easily check that @xmath396 is determined by the family @xmath416 in the sense of definition 2.2.1 . from that paper . the construction of @xmath396 is functorial in the sense that if @xmath417 is a non - degenerate simplicial map of complexes @xmath76 and @xmath222 , then there is an associated continuous map @xmath418 of @xmath0-complexes . indeed , if @xmath419 is a multi - valued function indexing a cell in @xmath420 , then it is not difficult to check that @xmath421 is a cell in @xmath396 . * proof : * assume that @xmath425 is a cell in @xmath396 . then @xmath426 is a cell in @xmath427 . suppose @xmath428 and @xmath429 are distinct vertices in @xmath400 . by assumption @xmath430 . we deduce from here that @xmath431 , otherwise @xmath66 would be a degenerated simplicial map . the second condition from definition [ def : hom(k , l ) ] is checked by a similar argument . @xmath229 the chromatic number @xmath397 of a simplicial complex @xmath76 is @xmath432})\neq\emptyset\}.\ ] ] in other words @xmath397 is the minimum number @xmath213 such that there exists a non - degenerate simplicial map @xmath433}$ ] . it is not difficult to check that @xmath434 where @xmath435 is the vertex - edge graph of the complex @xmath76 . in particular @xmath397 reduces to the usual chromatic number if @xmath76 is a graph , that is if @xmath76 is a @xmath92-dimensional simplicial complex . aside from the usual chromatic number @xmath238 , there are many related colorful graph invariants @xcite @xcite . among the best known are the fractional chromatic number @xmath436 and the circular chromatic number @xmath437 of @xmath25 . these and other related invariants are conveniently defined in terms of graph homomorphisms into graphs chosen from a suitable family @xmath438 of test graphs . motivated by this we offer an extension of the chromatic number @xmath397 in hope that some genuine , new invariants of simplicial complexes may arise this way . suppose that @xmath439 is a family of `` test '' simplicial complexes and let @xmath440 is a real - valued function . a @xmath441-coloring of @xmath76 is just a non - degenerate simplicial map @xmath442 and @xmath443 , the @xmath444-chromatic number of @xmath76 , is defined as the infimum of all weights @xmath445 over all @xmath441-colorings , a pure , @xmath81-dimensional simplicial complex @xmath76 is _ shellable _ @xcite @xcite , if there is a linear order @xmath447 on the set of its facets , such that for each @xmath448 , the complex @xmath449 is a pure @xmath450-dimensional subcomplex of the simplex @xmath451 . the _ restriction _ @xmath452 of the facet @xmath451 is the minimal new face added to the complex @xmath453 by the addition of the facet @xmath451 . let @xmath454 be the type of the facet @xmath451 . if @xmath455 for each @xmath178 then the complex @xmath76 is collapsible . the collapsing process is just the shelling order read in the opposite direction . from this point of view , @xmath452 can be described as a free face in the complex @xmath456 , and the process of removing all faces @xmath97 such that @xmath457 is called an elementary @xmath458-collapse . a pure @xmath81-dimensional simplicial complex @xmath76 is called _ tree - like _ or vertex collapsible if it is collapsible to a @xmath81-simplex with the use of elementary @xmath211-collapses alone . in other words @xmath76 is shellable and for each @xmath448 , the intersection @xmath449 is a proper face of @xmath451 . in order to establish analogs of propositions [ prop : help ] and [ prop : transport ] for complexes @xmath396 , we prove a result which shows that elementary vertex collapsing provides a good substitute and a partial generalization for the concept of `` foldings '' of graphs used in @xcite @xcite in the theory of graph complexes @xmath252 . [ prop : vertex - collapse ] suppose that the simplicial complex @xmath222 is obtained from @xmath76 by an elementary vertex collapse . in other words we assume that @xmath459 , where @xmath148 is a simplex in @xmath76 and @xmath460 a facet of @xmath148 . assume that @xmath460 is not maximal in @xmath222 , i.e. that for some simplex @xmath461 and a vertex @xmath462 . then for any simplicial complex @xmath203 , the inclusion map @xmath463 induces a homotopy equivalence @xmath464 * proof : * let @xmath465 . aside from the inclusion map @xmath466 , there is a retraction ( folding ) map @xmath467 , where @xmath468 and @xmath469 . since @xmath470 , we observe that @xmath471 is the identity map on @xmath420 , i.e. the complex @xmath420 is a retract of the complex @xmath396 . it remains to be shown that @xmath472 is homotopic to the identity map on @xmath396 . note that if @xmath473 then @xmath474 is the function defined by @xmath475 let @xmath476 be the map defined by @xmath477 note that @xmath478 is well defined since if a vertex @xmath10 is adjacent to @xmath429 it is also adjacent to @xmath428 , hence the condition @xmath479 is a consequence of @xmath480 . since for each @xmath473 and each vertex @xmath481 , @xmath482 by the order homotopy theorem @xcite @xcite @xcite all three maps @xmath483 and @xmath484 are homotopic . this completes the proof of the proposition . @xmath229 as in the case of graph complexes , the real justification for the introduction of the parallel transport of @xmath0-complexes comes from the fact that it preserves the homotopy type of the maps @xmath487 . as a preliminary step we prove an analogue of proposition [ prop : help ] . [ prop : helphelp ] suppose that @xmath53 and @xmath54 are two distinct , adjacent @xmath99-dimensional simplices in a finite simplicial complex @xmath76 which share a common @xmath152-dimensional simplex @xmath157 . let @xmath488 . let @xmath489 be the automorphism of @xmath42 which interchanges simplices @xmath53 and @xmath54 keeping the common face @xmath157 point - wise fixed . * proof : * by proposition [ prop : vertex - collapse ] , both maps @xmath495 for @xmath496 are homotopy equivalences . let @xmath497 and @xmath498 be the folding maps . then @xmath499 @xmath500 and we conclude that @xmath501 is also a homotopy equivalence . part ( a ) of the proposition follows from the fact that @xmath502 is an identity map . indeed , an immediate consequence is that @xmath503 is also an identity map and , in light of the fact that @xmath504 and @xmath505 are homotopy inverses to each other , we deduce that @xmath506 . [ prop : prenos ] suppose that @xmath76 and @xmath203 are finite simplicial complexes and @xmath49 a pair of adjacent ( distinct ) , @xmath99-dimensional simplices in @xmath76 . let @xmath512 be the embedding of @xmath513 in @xmath76 and @xmath514 the associated map of @xmath0-complexes . then the following diagram commutes up to a homotopy . [ cor : posledica ] suppose that @xmath76 and @xmath203 are finite simplicial complexes , @xmath148 a @xmath99-dimensional simplex in @xmath76 and @xmath521 the associated embedding . let @xmath522 . then the following diagram commutes up to a homotopy . in this section we prove the promised extension of the babson - kozlov - lovsz theorem . the graphs are replaced by pure @xmath81-dimensional simplicial complexes , while the role of the odd cycle @xmath245 is played by a complex @xmath248 which has some special symmetry properties in the sense of the following definition . as usual , an involution @xmath524 is the same as a @xmath12-action on @xmath24 . an involution on a simplicial complex @xmath248 induces an involution on the complex @xmath525 for each simplicial complex @xmath203 . for all other standard facts and definitions related to @xmath12-complexes the reader is referred to @xcite . [ def : phi - complex ] a pure @xmath81-dimensional simplicial complex @xmath248 is a @xmath526-complex if it is a @xmath12-complex with an invariant @xmath81-simplex @xmath527 such that the restriction @xmath528 of the involution @xmath529 on @xmath148 is a non - trivial element of the group @xmath530 . the graph @xmath245 is obviously an example of a @xmath533-complex . figure [ fig : piramida ] displays four examples of @xmath534-complexes , initial elements of two infinite series @xmath535 and @xmath536 . the complexes @xmath537 and @xmath538 etc . are obtained from two triangulated annuli , glued together along a common triangle @xmath148 . similarly , the complexes @xmath539 , are obtained by gluing together two triangulated mbius strips . the associated group of projectivities are @xmath540 and @xmath541 . [ thm : main ] suppose that @xmath248 is a @xmath526-complex in the sense of definition [ def : phi - complex ] , with an associated invariant simplex @xmath527 . suppose that @xmath76 is a pure @xmath81-dimensional simplicial complex . than for @xmath213 even , @xmath542 * proof : * by definition @xmath543 means that there exists a @xmath12-equivariant map @xmath544 . assume that @xmath545 which means that there exists a non - degenerate simplicial map @xmath546}$ ] . by functoriality of the construction of @xmath0-complexes , section [ sec : naturality ] , there is an induced @xmath12-equivariant map @xmath547})$ ] and similarly a map @xmath548})\rightarrow hom(\sigma , \delta^{[m+d+1]})$ ] . by @xcite theorem 3.3.3 . , the complex @xmath549})\cong hom(k_{d+1},k_{m+d+1})\ ] ] is a wedge of @xmath213-dimensional spheres . since @xmath550})$ ] is a free @xmath12-complex , we deduce that there exists a @xmath12-map @xmath550})\rightarrow s^m$ ] . all these maps can be arranged in the following sequence of @xmath12-equivariant maps @xmath551 } ) \stackrel{\widehat{\alpha}}\longrightarrow hom(\sigma , \delta^{[m+d+1 ] } ) \stackrel{\nu}\longrightarrow s^m.\ ] ] by corollary [ cor : posledica ] , there is a homotopy equivalence @xmath552 . this contradicts proposition [ prop : not - homotopic ] , which completes the proof of the theorem . [ prop : not - homotopic ] suppose that @xmath553 is a @xmath12-equivariant map of free @xmath12-complexes @xmath24 and @xmath23 where @xmath554 . assume that @xmath555 , where @xmath213 is an even integer . in other words our assumption is that there exist @xmath12-equivariant maps @xmath556 and @xmath557 such that @xmath558 then the maps @xmath65 and @xmath559 are not homotopic . * proof : * if @xmath560 then @xmath561 and by the equivariance of @xmath562 , @xmath563 where @xmath564 . it follows that @xmath565 i.e. @xmath566 , which is in contradiction with a well known fact @xcite that a @xmath12-equivariant map @xmath567 of spheres must have an odd degree . @xmath229 [ cor : lbk - gen ] suppose that @xmath248 is a @xmath526-complex with an associated invariant simplex @xmath527 . suppose that @xmath76 is a pure @xmath81-dimensional simplicial complex . than for @xmath99 odd , @xmath568 a simplicial complex is a flag - complex if @xmath116 if and only if the associated @xmath92-skeleton @xmath571 of @xmath148 is a subcomplex of @xmath76 . examples of flag - complexes include the clique - complex @xmath572 of a graph and the order complex @xmath573 of all chains ( flags ) in a poset @xmath44 . if @xmath76 is a flag - complex and @xmath574 is the associated vertex - edge graph of @xmath76 , then @xmath575 . from here we deduce , relying on the isomorphism given in example [ exam : companion ] , that if @xmath76 and @xmath203 are flag complexes then [ thm : paradigm ] suppose that @xmath577 is @xmath526-complex ( definition [ def : phi - complex ] ) which is also a flag complex . assume that @xmath99 is an odd integer . let @xmath578 . then for each graph @xmath25 @xmath579 examples of @xmath534-complexes @xmath577 satisfying conditions of theorem [ thm : paradigm ] are the complexes @xmath580 depicted in figure [ fig : piramida ] . moreover , for all associated graphs @xmath581 which in light of ( [ eq : paradigm ] ) means that they are all candidates for `` homotopy test graphs '' in the sense of ( * ? ? ? * chapter 6 ) . dolbilin , m.a . shtanko , and m.i.shtogrin . _ cubic manifolds in lattices _ , izv . , ser . mat . 58(1994 ) , 93107 , english translation : _ russ.acad . _ , 44 ( 1995 ) , 301313 . i. izmestiev , _ free actions of a torus on a manifold @xmath583 and the projectivity group of a polytope @xmath44 _ , uspekhi mat . nauk , * 56 * ( 2001 ) pp . 169170 , english translation : russian math . survays * 56 * ( 2001 ) . r. kirby , _ problems in low - dimensional topology _ , in `` geometric topology '' , georgia international topology conference , ( w.h . kazez , editor ) , ams - ip studies in advanced math . , @xmath584 , part @xmath585 , 35473 , 1995 . kozlov , _ chromatic numbers , morphism complexes , and stiefel - whitney characteristic classes _ , `` geometric combinatorics '' , ias / park city mathematics series 14 , american mathematical society , providence , ri ; institute for advanced study ( ias ) , princeton , nj , in press .

we present 0.8-mm band molecular images and spectra obtained with the atacama large millimeter / submillimeter array ( alma ) toward one of the nearest galaxies with an active galactic nucleus ( agn ) , ngc 1068 . distributions of co isotopic species (co and co ) = 32 , cn = 32 and cs = 76 are observed toward the circumnuclear disk ( cnd ) and a part of the starburst ring with an angular resolution of. 1. . the physical properties of these molecules and shock - related molecules such as hnco , chcn , so , and choh detected in the 3-mm band were estimated using rotation diagrams under the assumption of local thermodynamic equilibrium . the rotational temperatures of the co isotopic species and the shock - related molecules in the cnd are , respectively , 1422 k and upper limits of 2040 k. although the column densities of the co isotopic species in the cnd are only from one - fifth to one - third of that in the starburst ring , those of the shock - related molecules are enhanced by a factor of 310 in the cnd . we also discuss the chemistry of each species , and compare the fractional abundances in the cnd and starburst ring with those of galactic sources such as cold cores , hot cores , and shocked molecular clouds in order to study the overall characteristics . we find that the abundances of shock - related molecules are more similar to abundances in hot cores and/or shocked clouds than to cold cores . the cnd hosts relatively complex molecules , which are often associated with shocked molecular clouds or hot cores . because a high x - ray flux can dissociate these molecules , they must also reside in regions shielded from x - rays .

1410.5912

it is important to investigate the relationships between the power sources of galaxies ( i.e. , active galactic nuclei ( agn ) and/or starbursts ) and the chemical properties of the surrounding dense interstellar medium in order to study the effects on molecular abundances and to probe the power source with molecular line observations . chemical properties , especially , have been expected to be powerful astrophysical tools for the study of galaxies , because the molecular line observations of different galaxies allow us to study the effects of these different physical properties / activities on the molecular medium . in fact , some groups have suggested that it is possible to diagnose power sources in dusty galaxies using molecular line ratios in nearby galaxies ( e.g. , kohno et al . 2001 , 2008 ; usero et al . 2004 ; kohno 2005 ; imanishi et al . 2007 ; krips et al . 2008 ; izumi et al . 2013 ) . for example , elevated hcn emission with respect to co and/or hco@xmath8 has often been detected toward agns ( e.g. , kohno et al . 1996 , 2003 ) , where it is expected to be the imprint of either strong x - ray irradiation / ionization ( e.g. garca - burillo et al . 2010 ) and/or a high - temperature environment caused by agn activity ( e.g. , harada et al . 2010 ; izumi et al . 2013 ) . on the other hand , we reported no significant differences in the relative abundances of the carbon containing molecules c@xmath9h and cyclic - c@xmath7h@xmath9 between one of the nearest galaxies with an agn , ngc 1068 , and the prototypical starburst galaxy , ngc 253 . it was concluded that these basic carbon - containing molecules are insensitive to agns and/or these molecules exist in a cold gas away from an agn ( nakajima et al . 2011 ) . systematic unbiased scans ( i.e. molecular line survey observations ) are the most effective method not only for complete understanding of chemical compositions in representative sources , but also for probing interstellar medium and star formation . over the last 10 years or so , millimeter / sub - millimeter observing systems with very high sensitivity , wide frequency range , and high velocity resolution have been put to use in a number of telescopes ( e.g. carter et al . 2012 ; nakajima et al . 2008 ; 2013a ) . with these , we can also obtain higher quality line survey observations of millimeter / sub - millimeter molecular lines toward nearby galaxies as well as galactic sources . to date , several line surveys have been reported towards the center of ngc 1068 using single dish telescopes ( snell et al . 2011 ; costagliola et al . 2011 ; kamenetzky et al . 2011 ; spinoglio et al . 2012 ; aladro et al . 2013 ; nakajima et al . for example , aladro et al . ( 2013 ) compared their results with ngc 253 , and they suggested that ngc 1068 has a different chemical composition from those of starburst galaxies . they discussed that ch@xmath7cch is not detected in ngc 1068 , yet cn , sio , hco@xmath8 , and hcn are detected and enhanced . although the chemical environment of ngc 1068 has become clearer thanks to these line survey observations , contamination from starbursts associated with inner spiral arms / ring ( @xmath1030@xmath11 ) could be a problem , if we consider the sizes of the observing beams ( 15@xmath1170@xmath11 ) in these single dish telescopes . on the other hand , interferometric imaging of the circumnuclear disk ( cnd ) gives clean measurements of spectral lines , but the observed lines are limited to major species such as @xmath12co ( @xmath2 = 10 , 21 and 32 ) , @xmath0co ( @xmath2 = 10 , 21 and 32 ) , c@xmath1o ( @xmath2 = 10 and 21 ) , hcn ( @xmath2 = 10 and 32 ) , hco@xmath8 ( @xmath2 = 10 , 32 and 43 ) , cs ( @xmath2 = 21 ) , cn ( @xmath3 = 21 ) , and sio ( @xmath2 = 21 ) ( e.g. planesas et al . 1991 ; kaneko et al . 1992 ; jackson et al . 1993 ; tacconi et al . 1994 , 1997 ; helfer and blitz 1995 ; papadopoulos et al . 1996 ; schinnerer et al . 2000 ; kohno et al . 2008 ; garca - burillo et al . 2010 ; krips et al . 2011 ; tsai et al . 2012 ) due to the limitation in sensitivity of the existing pre - alma ( atacama large millimeter / sub - millimeter array ) interferometers . although it is important to estimate the physical properties , such as temperature and density and so on , only one transition has been observed except for co , hcn and hco@xmath8 . in this situation , we proposed to observe several interesting molecules sensitively with alma in the 96100 ghz and 108111 ghz ( 3-mm ) regions for lower excitation lines ( takano et al . 2014 , hereafter paper-1 ) and also 327.5330.5 ghz and 338.5342.5 ghz ( 0.8-mm ) regions for higher excitation lines . these frequency regions are rich in molecules , including typical shock / dust related species and the co isotopologues . our aim has been to focus on the effects of strong x - ray and starburst on shock and/or dust related molecules , ch@xmath7oh , so , hnco , and ch@xmath7cn , which are well observed in galactic sources to study chemical and physical conditions . our observational results with alma will be compared with observations in galactic sources , model calculations , and laboratory experiments to study formation and destruction mechanisms of the above molecules . the complex organic molecules ch@xmath7oh and ch@xmath7cn are thought to be efficiently produced on dust ( e.g. watanabe and kouchi 2002 , takano et al . 1995 ) and subsequently desorbed into gas - phase ( garrod et al . 2008 ) , while hnco is also thought to be produced at least partially on dust ( quan et al . 2010 ) . the radical so is thought to have an enhanced abundance in shocked regions ( leen and graff 1988 ) . however , the abundances and chemical processes forming and destroying these molecules are not well understood in galaxies with active galactic nuclei . from our observations in the two frequency bands with alma , rotational temperatures and column densities can be obtained using intensities of two transitions . the field of view of the 0.8-mm observations is about 18@xmath11 , which means that only the cnd region can be covered with one pointing in ngc 1068 . therefore , we additionally observe a part of the starburst ring at the offset position from the center , so that distributions and abundances of molecules in the starburst environment as well as the cnd can be studied . this comparison will enable us to further study the impact of agns on the surrounding dense interstellar medium . in paper-1 , we have presented the distributions of detected molecules in the 3-mm band , such as @xmath0co , c@xmath1o , @xmath0cn , cs , so , hnco , hc@xmath7n , ch@xmath7oh , ch@xmath7cn , and discussed the implications of diversity . the main results of paper-1 are that the molecular distributions are reflections of both physical and chemical properties . in this paper , we report a high - resolution imaging study of molecular lines in the 0.8-mm band toward ngc 1068 observed with alma , and also the calculation of the fractional abundances of shock / dust related molecules based on not only the observations in 0.8-mm band but also on the earlier 3-mm band results from paper-1 . even in its early science operation phase , alma was already powerful enough to simultaneously observe four high - excitation molecular lines ( @xmath0co and c@xmath1o @xmath2 = 32 , cn @xmath3 = 32 and cs @xmath2 = 76 ) in the 0.8-mm band , uncovering a wide variety of molecular line distributions . we describe our observations and data reduction in section 2 , and present the images and spectra of the observed molecular lines and also estimated physical properties of individual molecules in section 3 . in section 4 , we discuss relationships among the molecular abundances toward the cnd and starburst ring in ngc 1068 , and galactic sources such as hot cores , cold cores , and shocked clouds . throughout the paper , we assume that the distance of ngc 1068 is 14.4 mpc ( tully 1988 ; bland - hawthorn et al . 1997 ) ; at this distance , 1@xmath11 corresponds to 72 pc . we observed ngc 1068 using alma in cycle 0 ( early science program ) with band-3 ( @xmath133 mm ) and band-7 ( @xmath130.8 mm ) receivers in january of 2012 and november of 2011 , respectively ( i d = 2011.0.00061.s ; pi = s. takano ) . the details of the 3-mm band observations are described in paper-1 , and the 0.8-mm band observations are explained below . table 1 summarizes the observational parameters of the 0.8-mm band . the centers of four spectral windows were tuned to 329.50 ghz ( spw0 ) , 330.85 ghz ( spw1 ) , 340.90 ghz ( spw2 ) , and 342.80 ghz ( spw3 ) , each with a 1.875 ghz bandwidth and 3840 channels of 488.28 khz width . the final results were presented with a velocity resolution of @xmath420 km s@xmath14 to improve the signal - to - noise ratio . the 14 operational 12-m antennas in the compact array configuration were arranged to span baseline lengths of 20196 m , which gave a beam size of 1.@xmath53 @xmath6 1.@xmath52 ( @xmath490 @xmath6 86 pc ) for the lowest frequency , spw0 . a two - pointing mosaic was observed in order to image the cnd of @xmath15280 pc ( @xmath154@xmath11 ) in diameter and the surrounding starburst ring of @xmath42 kpc ( @xmath430@xmath11 ) in diameter ( schinnerer et al . 2000 ) . the phase reference center has been set to @xmath16 = 2@xmath1742@xmath1840.@xmath19798 and @xmath20 = -00@xmath2100@xmath2247@xmath5938 ( schinnerer et al . 2000 ) , which corresponds to the radio position of the active nucleus ( muxlow et al . the position of a part of the starburst ring is at the offset position to the southwest from the agn of ( @xmath23 , @xmath24 ) = ( -10@xmath11 , -10@xmath11 ) , which is stronger in co , hcn ( schinnerer et al . 2000 , kohno et al . the systemic velocity employed was 1150 km s@xmath14 , which was taken from schinnerer et al . the total integration times on the cnd and starburst ring were approximately 30.3 minutes in each position and the total observation time was 71.4 minutes . the system noise temperatures including the atmosphere were @xmath4120250 k and @xmath4150700 k in the upper sideband ( usb ) and the lower sideband ( lsb ) , respectively , depending on frequency and antennas . the delivered data were already bandpass , flux , and gain ( phase and amplitude ) calibrated by the alma regional center . all of the above calibration and imaging were carried out using the common astronomy software applications ( casa ; mcmullin et al . 2007 ) . we have successfully detected c@xmath1o ( @xmath2 = 32 ) , @xmath0co ( @xmath2 = 32 ) , cn ( @xmath3 = 32 ) and cs ( @xmath2 = 76 ) in the cnd , and the same molecules except for cs were also detected in the starburst ring . figure 1 shows the integrated intensity maps of four clearly detected molecules . these are two - point mosaic images toward the cnd and a part of the starburst ring . the c@xmath1o emission was not significantly detected toward the cnd , but relatively strong in the starburst ring . we found a similar feature for c@xmath1o ( @xmath2 = 10 ) in the 3-mm band observations ( paper-1 ) . on the other hand , @xmath0co ( @xmath2 = 32 ) is very strong in the cnd and also detected towards the starburst ring . the cn and cs radicals are very weak and not detected towards the starburst ring , respectively , but both are clearly detected towards the cnd ( see also the spectra in fig.2 ) . moreover , the @xmath0co , and cn emission features from the cnd are spatially separated into two conspicuous knots , and the cs emission is also separated into two knots , albeit faintly . the difference in the c@xmath1o and @xmath0co distributions can not be explained by different excitation conditions , because the abundance ratio of these species is not sensitive to the molecular gas temperature and density . therefore , this feature is possibly the result of differences in isotopic ratios . the [ @xmath12c]/[@xmath0c ] isotopic ratios in the galaxy can be quite diverse . for example , the ratio is 2025 in the galactic center clouds ( gsten et al . 1985 ) ; @xmath489 in the solar system ( wilson & matteucci 1992 ) and outer galaxies ( e.g. @xmath440 ( henkel et al . 1993 ) and @xmath480 ( martn et al . 2010 ) for ngc 253 ; and @xmath25100 for the clover leaf quasar ( henkel et al . 2010 ) ) . given this diversity , differences in the isotopic ratios [ @xmath12c]/[@xmath0c ] and [ @xmath26o]/[@xmath1o ] between the cnd and the starburst ring of ngc 1068 are possible . another possibility is the effect of isotope - selective photodissociation , in which the abundance of naturally poor isotopic species are decreased because of their lower optical depth ( e.g. harrison et al . details concerning these discussions for ngc 1068 will appear in a separate paper ( taniguchi et al . , in preparation ) . figure 2 shows the spectra of all spectral windows toward the cnd and starburst ring at the southwest position , which is the @xmath0co ( @xmath2 = 32 ) intensity peak , as mentioned in the caption of figure 2 . table 2 and 3 list the gaussian fitted line parameters toward the cnd and starburst ring , respectively . the spectra of the lsb consist of spw0 ( 327.29330.16 ghz ) and spw1 ( 328.49330.49 ghz ) , those of the usb consist of spw2 ( 338.68340.54 ghz ) and spw3 ( 340.56342.43 ghz ) . to obtain the spectra and measure the gaussian fitted line parameters , we convolved all 0.8-mm band images to the beam size of the 3-mm band data ( 4.@xmath521@xmath62.@xmath536 , p.a . = 176@xmath21 ) , and then the correction of the primary beam attenuation was applied . @xmath0co , cn and cs were clearly detected toward the cnd although c@xmath1o was not clearly detected ( @xmath44 @xmath27 ) . on the other hand , c@xmath1o , @xmath0co and cn were clearly detected toward the starburst ring . unfortunately , expected lines of shock and/or dust related molecules , ch@xmath7oh , so , hnco , and ch@xmath7cn , were not detected . we calculated the upper limits of the integrated flux ( 1@xmath27 ) for these molecules from the rms noise level . the line widths are broad in the cnd with approximately 170180 km s@xmath14 except for cn . for cn , two partially separated fine structure components of @xmath2 = 5/23/2 and @xmath2 = 7/25/2 were identified at the band edge . therefore the line parameters of cn may not be very precise . moreover , it is not possible to resolve each hyperfine component . unlike the case with the cnd , the line widths of all detected lines in the starburst ring are narrow , approximately 3040 km s@xmath14 . we compared the flux obtained with alma with those obtained with the single dish james clerk maxwell telescope ( jcmt ) to estimate the recovery of the flux with the interferometer . for @xmath0co @xmath2 = 32 , the image obtained with alma ( primary beam corrected ) was convolved with the jcmt beam of 14@xmath11 , and the flux obtained was converted to the brightness temperature , which is [email protected] k km s@xmath14 . the corresponding value obtained with the jcmt is [email protected] k km s@xmath14 , which is observed by israel ( 2009 ) . therefore , the alma observations recover 32@xmath285 % of the single dish flux . for cs @xmath2 = 76 , the flux obtained with alma was converted to brightness temperature , which is @xmath40.4 k km s@xmath14 . although the detection of cs @xmath2 = 76 with the jcmt is marginal , the corresponding value is @xmath41.4 k km s@xmath14 ( bayet et al . therefore , the recovered flux is about 26 % . the recovered flux of @xmath0co @xmath2 = 10 in the 3-mm band is about 8090 % , which was obtained by comparing with the single dish telescopes nro 45-m and iram 30-m ( see the section 3.4 in paper-1 ) . the probably reason for the increase in the missing flux in the 0.8-mm band is that the effective minimum baseline of the observations in this band is longer than that for the observations in the 3-mm band we constructed the rotation diagrams of observed molecules in the 3-mm band ( paper-1 ) and the 0.8-mm band ( this work ) toward each position of the cnd and starburst ring ( figure 3 ) . all 0.8-mm band images were convolved to the beam size of the 3-mm band data . thus , the data points of the 3-mm and 0.8-mm bands were obtained from the emission in the same regions . the rotational temperatures ( @xmath29 ) and column densities ( @xmath30 ) are calculated from the slope and intercept at @xmath31 = 0 of the diagram under the assumptions that all lines are optically thin , and that a single excitation temperature ( @xmath32 ) characterizes all transitions ( the local thermodynamic equilibrium ( lte ) assumption ) ( turner 1991 ) . the non - detected lines are plotted with their rms noise level ( 3@xmath27 ) , and the upper or lower limits of physical parameters are calculated . for the calculation of @xmath30 for molecules with only one detected transition , we assumed some temperatures of @xmath29 = 5 k , 10 k , 15 k , 30 k and 45 k. the results of all molecules are shown in tables 4 and 5 . the calculated rotation temperatures and column densities of the observed molecules , c@xmath1o in the starburst ring , cs in the cnd , and @xmath0co and hc@xmath7n in both regions are listed in table 4 . the column densities for the assumed some rotation temperatures of c@xmath1o and ch@xmath7cn ( cnd ) , hnco , cn , so , and ch@xmath7oh ( cnd and starburst ring ) , and cs ( starburst ring ) are listed in table 5 , because these molecules were detected in only one transition . note that these rotation diagrams are fitted with only two points , so that the possibility of multiple components at different temperatures can not be ruled out ( bayet et al . 2009 ) . the rotation diagrams of c@xmath1o and @xmath0co are shown in figure 3 ( a ) and ( c ) , respectively . the c@xmath1o @xmath2 = 10 line was not detected toward the cnd , thus we obtained a lower limit of @xmath29 @xmath25 14.2 k. these co isotopic molecules trace cold gas ( 8.0 k and 8.5 k ) in the starburst ring and some warm gas ( @xmath25 14.2 k , 21.5 k ) in the cnd . aladro et al . ( 2013 ) suggested that @xmath0co has two different components in the rotation diagram , 4.3 k and 24.1 k. in that work , it is possible that the reported emissions from the cnd and starburst ring are blended , because they used a single dish telescope . in addition , they obtained that the @xmath29 of c@xmath1o is the lowest in all observed molecules , 3.3 k. therefore , as compared with our result , previous work with the single dish telescope overestimated the intensity of a vertical axis in the rotation diagram for the purpose of examination of property in the cnd . or , as we saw in section 3.3 , the flux of the @xmath2 = 32 line in the starburst ring with alma is not sufficiently recovered for this study . if the previous results with the single dish telescope are accurate enough , the flux in the starburst ring with alma should be increased by a factor of 4.5 . then the @xmath29 and column density of @xmath0co will be 16.3 k and 7.1@xmath610@xmath26 @xmath33 , respectively . we leave the effect of missing flux out of consideration , because these values make little impact on the following discussion . unlike other molecules , the column densities of both co isotopologs in the starburst ring are larger than those in the cnd . the column density of @xmath0co is a factor of 3.5 , and c@xmath1o a factor of 2.4 larger than those in the cnd under the assumption of @xmath29 = 10 k. this tendency is the same with @xmath12co . tsai et al . ( 2012 ) calculated the @xmath34 in the cnd and the starburst ring from the integrated intensity of @xmath12co @xmath2 = 10 line . @xmath34 in the starburst ring at the area r15 , which includes the area of our observations at the southwest region in the starburst ring , is a factor of about 20 larger than those in the cnd . on the other hand , although the @xmath0co @xmath2 = 10 line is rather weak and c@xmath1o @xmath2 = 10 line is not detected toward the cnd ( paper-1 ) , emission of high-@xmath35 transition lines such as @xmath0co @xmath2 = 32 in our observation and @xmath12co @xmath2 = 65 based on the data by garca - burillo et al . ( 2014 ) are strong and clearly detected . thus , the kinetic temperature in the cnd is very high and presumably in a non - lte environment . the details of the physical properties based on a large velocity gradiant ( lvg ) analysis of the co lines will be described in taniguchi et al . ( in preparation ) . the rotation diagrams of hnco and ch@xmath7oh are shown in figure 3 ( b ) and ( g ) , respectively . the hnco @xmath36 = 15@xmath3714@xmath38 and ch@xmath7oh @xmath36 = 13@xmath3913@xmath40-+ lines were not detected toward both the cnd and the starburst ring , thus we could obtain only upper limits of @xmath29 . the ch@xmath7oh @xmath41 = 2@xmath421@xmath42 lines are composed of four transitions ( 2@xmath431@xmath44 , 2@xmath451@xmath46 + , 2@xmath451@xmath47 , 2@xmath481@xmath49 ) , but the quartet of lines is blended in the spectra from ngc 1068 . therefore , we calculated @xmath29 using the method for blended lines by martn et al . @xmath29 in the cnd and the starburst ring is less than 28.6 k and 36.0 k , respectively , for hnco , and 40.4 k and 38.9 k , respectively , for ch@xmath7oh . it is possible that these lines trace warm gas ( @xmath15 3040 k ) . this value of hnco is consistent with the value of @xmath430 k obtained by aladro et al . however , they obtained 5.6 k for ch@xmath7oh , which is almost six times lower than our result . the column densities of hnco in the cnd are a factor of @xmath45 larger than those in the starburst ring assuming a @xmath29 of 1030 k. on the other hand , the column density of ch@xmath7oh is not much different between the cnd and the starburst ring under the assumption of a similar @xmath29 . if the @xmath29 in the cnd is higher than the starburst ring ( e.g. 30 k in the cnd and 10 k in the starburst ring ) , the column density of ch@xmath7oh in the cnd is a factor of @xmath43 larger than that in the starburst ring . the rotation diagram is shown in figure 3 ( d ) . the @xmath41=6@xmath425@xmath42 line toward the starburst ring and the @xmath41=18@xmath4217@xmath42 lines toward both positions are not detected . we obtained an upper limit of @xmath29 , @xmath50 26.1 k in the cnd . the column density in the cnd ( @xmath41@xmath610@xmath0 @xmath33 ) is the smallest among the observed shock / dust related molecules . ch@xmath7cn is a tracer of star formation in galactic sources , such as hot cores and outflows from young stellar objects . therefore , the non - detection of this molecule in the starburst ring , but the detection in the cnd are interesting results . we discuss the abundance of this molecule in section 4.2.2 . [ [ cn ] ] cn ^^ the rotation diagram is shown in figure 3 ( e ) . we did not observe cn in the 3-mm band , thus we plotted only the @xmath3 = 32 transition in the 0.8-mm band . in line survey observations with the nro 45-m telescope , we found that the cn/@xmath0co intensity ratios are significantly higher in ngc 1068 ( @xmath42.3 ) than in ngc 253 ( @xmath41.1 ) and ic 342 ( @xmath40.6 ) ( nakajima et al . 2013b ) . the cn/@xmath0co intensity ratios are @xmath44.3 in the cnd and @xmath40.1 in the starburst ring with this alma observation . moreover , the column density of cn in the cnd is a factor of 2050 larger than that in the starburst ring . cn is one of the key molecules to trace x - ray dominated regions ( xdr ) , as mentioned in theoretical work ( e.g. meijerink et al . therefore , the enhancement of intensity and column density of cn in the cnd may be the effects of extreme physical conditions such as those that pertain to xdrs . [ [ so ] ] so ^^ the rotation diagram is shown in figure 3 ( f ) . the @xmath51=7@xmath526@xmath53 line is not detected toward both the cnd and the starburst ring . we obtained upper limits of @xmath29 in the cnd and the starburst ring of 2026 k. this temperature is consistent with the value of 22.8 k of aladro et al . the previous work of so , with a single dish telescope , traced mainly the cnd , because the distribution of so is concentrated in the cnd and the contribution is less from the starburst ring , as can be seen in figure 2 in paper-1 . the column density in the cnd is a factor of 56 larger than that in the starburst ring assuming the same @xmath29 . if the rotational temperature in the cnd is higher than in the starburst ring , the ratio of column densities changes . for example , if the rotational temperature in the cnd is 30 k and that in the starburst ring 10 k , the column densities will be 2.4 @xmath6 10@xmath54 @xmath33 and 0.2 @xmath6 10@xmath54 @xmath33 , respectively , leading to a column density of approximately 12 . [ [ cs ] ] cs ^^ the rotation diagram is shown in figure 3 ( h ) . the @xmath2=76 line is not detected toward the starburst ring . we obtained an upper limit of @xmath29 , @xmath50 12.1 k , in this region . @xmath29 in the cnd is 13.1 k , which is consistent with the result of 13.9 k by bayet et al . since the @xmath2 = 76 line was only marginally detected by bayet et al.(2009 ) , they obtained @xmath29 = 7.1 k from a rotation diagram excluding this line . however , our results support the value including the @xmath2 = 76 line . moreover , bayet et al . ( 2009 ) suggested two fitting components , but the cs molecule is distributed not only in the cnd , but also in the starburst ring ( see figure 1 in paper-1 ) . thus , it is possible to have a blending of emission from the starburst ring . the rotation diagram is shown in figure 3 ( i ) . we did not observe hc@xmath7n in the 0.8-mm band , and therefore we obtained only the @xmath2 = 1110 and 1211 lines in the 3-mm band . using these lines , @xmath29 in the cnd and the starburst ring are 22.1 k and 17.7 k , respectively . these values are not consistent with that of aladro et al . ( 2013 ) , where they obtained 7.3 k. the reason for the discrepancy is not clear , but , as can be seen , the value for @xmath2 = 109 is afield from the fitting line of the values for @xmath2 = 1110 and 1211 in their rotation diagram . therefore , hc@xmath7n has possibly more than one component in the cnd . the column density of hc@xmath7n in the cnd is about an order of magnitude larger than that in the starburst ring . the asymmetrical distributions of @xmath0co , cn and cs in the cnd can be seen in figure 1 . usero et al . ( 2004 ) suggested that two velocity components , which are velocity @xmath50 systemic velocity and velocity @xmath25 systemic velocity , correspond to the eastern knot ( e - knot ) and western knot ( w - knot ) , respectively , in the spectra of co , cs , hcn , sio , h@xmath0co@xmath8 , hco@xmath8 , and hoc@xmath8 . the e - knot and w - knot are clearly separated in the integrated intensity maps of @xmath0co , cn , and cs ( figure 1 ) in our observations . so far , @xmath12co @xmath2=10 and 21 ( schinnerrer et al . 2000 ) , cn @xmath3=21 ( garca - burillo et al . 2010 ) , hcn @xmath2=32 , hco@xmath8 @xmath2=32 , @xmath12co @xmath2=32 , and @xmath0co @xmath2=10 , and 21 ( krips et al . 2011 ) have been found to be associated with both the e - knot and the w - knot with high resolution interferometric observations . we find similar distributions for @xmath0co ( @xmath2=32 ) , cn ( @xmath3=32 ) and cs ( @xmath2=76 ) for the first time with our alma observations . we have measured the peak positions for the e - knot and w - knot seen in these lines , as well as the flux ratios of the e - knot / w - knot for each molecule , both of which are listed in table 6 . all measured peak positions of the e - knot and w - knot seen in these molecules are consistent with each other . the flux ratios between the e - knot and w - knot are [email protected] for @xmath55co , [email protected] for cn and [email protected] for cs ; i.e. , the e - knot / w - knot flux ratio of cs is higher than those of @xmath0co and cn . krips et al . ( 2011 ) obtained the e - knot / w - knot flux ratios of @xmath12co ( @xmath2=10 , 21 and 32 ) , @xmath0co ( @xmath2=10 and 21 ) , hcn ( @xmath2=32 ) and hco@xmath8 ( @xmath2=32 ) to be 1.02.3 with sma and pdbi observations , and usero et al . ( 2004 ) shows the ratios of @xmath12co ( @xmath2=10 and 21 ) , hcn ( @xmath2=10 ) , sio ( @xmath2=21 and 32 ) , h@xmath0co@xmath8 ( @xmath2=10 ) , hco@xmath8 ( @xmath2=10 ) , and hoc@xmath8 ( @xmath2=10 ) with iram 30 -m observations to be 0.61.8 and cs ( @xmath2=21 ) to be 2.2 . therefore , the measured e - knot / w - knot flux ratio of cs ( @xmath2=76 ) is higher than any other molecular lines . these results suggest that there is a chemical differentiation between the e and w knots in the cnd ( e.g. usero et al . 2004 ) . however , the details of the chemical differentiation and/or the physical structure in the cnd are still not clear based on our observations with the spatial resolution available with alma during cycle 0 . in order to understand which mechanism , such as x - rays , shock waves , or ultraviolet photons , dominates the observed chemical features , it is helpful to obtain fractional abundances of each species with respect to molecular hydrogen . there are several different possible driving forces of chemistry in the observed regions of ngc 1068 . for example , since the agn emits intense x - rays , it is likely to have xdrs ( e.g. , garca - burillo et al . , 2010 ) . elevated massive star formation can create photon - dominated regions ( pdrs ; e.g. , hollenbach & tielens 1999 ) due to ultraviolet radiation from ob stars . in the site of embedded on - going star formation , hot cores can also contribute to molecular emission . furthermore , the shock waves affect the chemistry by heating up the gas as well as by sputtering dust grains . with our beam size ( @xmath490 pc ) , a mixture of molecular clouds governed by these different mechanisms may be observed . less dense gas than the mean density of the molecular cloud is likely to have a higher volume - filling factor , while emission from compact sources such as hot cores tends to have much smaller volume - filling factors unless the region hosts an extremely large number of hot cores . with this complexity in mind , we derived the fractional abundances and compared our results with various types of galactic sources to understand the nature of molecular material in the cnd and starburst ring in ngc 1068 . to compute fractional abundances , we first need to estimate the column density of molecular hydrogen @xmath56 , which is often derived from observations of co molecules . here we adopt @xmath56 values of @xmath57 @xmath33 at the cnd and @xmath58 @xmath33 at the starburst ring , based on the estimated column densities of @xmath0co , which is an optically thin tracer of molecular gas mass . we computed these by the rotation diagram analysis of our alma data ( summarized in table 4 ) with an assumption of a [ @xmath0co]/[h@xmath9 ] fractional abundance of @xmath60 ( dickman 1978 ) . these values can be compared with @xmath56 values based on @xmath12co line measurements adopting a co - to - h@xmath9 conversion factor ( @xmath61 factor ) to check the consistency with previous studies . this approach was taken by tsai et al . ( 2013 ) , who derived @xmath56 values of @xmath62 @xmath33 at the cnd and @xmath63 @xmath33 at the starburst ring , where a similar angular resolution co @xmath2 = 10 image ( produced from visibilities with similar minimum @xmath64 distances to our 0.8-mm band data ) is available . we find that the @xmath56 values at the cnd agree well with each other , whereas we find a factor of @xmath44 disagreement in the starburst ring between the @xmath56 values from tsai et al . ( 2013 ) and our @xmath0co multi - transition analysis . this disagreement in the starburst region may be caused by uncertainties in the adopted @xmath61 factors ( see bolatto et al . 2013 for a recent review on this issue ) and/or in the [ @xmath0co]/[h@xmath9 ] fractional abundance . therefore , we should bear in mind that the estimated fractional abundances in each molecule may suffer from uncertainties in the @xmath56 values by a factor of a few at least at the starburst ring . table 7 lists the derived fractional abundances of molecules in the cnd and starburst ring of ngc 1068 . large differences in the fractional abundances are derived between the two sources for hnco , ch@xmath65cn , cn , so , ch@xmath65oh , cs , and hc@xmath65n . there are a number of possible reasons for this difference . first , because the cnd is denser than the starburst ring , species with higher critical densities may have higher emission in the cnd , as discussed in paper-1 . also , in the starburst ring , relatively more complex molecules such as hnco , ch@xmath7cn , ch@xmath7oh , and hc@xmath7n can be more easily photo - dissociated because of the lower density . for granular species , lower abundances in the starburst ring can also result from relatively fewer shocked clouds or hot cores compared with the cnd where molecular clouds are heated up so that desorption can occur efficiently , either by gravitational contraction leading to a hot core or by shock waves . finally , the large uncertainties in the h@xmath9 column densities can lead to large error bars in the fractional abundances with respect to h@xmath9 . below we discuss the chemistry of each species , and compare the observed abundances in the cnd and starburst ring with those of galactic sources such as tmc-1 ( cold core ) , sgr b2(n ) and afgl2591 ( hot cores ) , and l1157 ( shocked molecular cloud ) . comparisons are mostly done for the cnd , because the lower fractional abundances of molecules in the starburst ring can be caused by any reason stated above . quan et al.(2010 ) show that hnco can be formed both in the gas phase and on grain surfaces from ocn either through the gas - phase hydrogenation of protonated species or hydrogenation on grains with neutral atomic hydrogen . in cold or lukewarm cores , gas - phase reactions can produce hnco with fractional abundances with respect to total hydrogen density of about @xmath66 , whereas the abundance can rise to @xmath67 when hnco on the grain surface sublimates in a hot core or its warm envelope . shock waves are also likely to cause sublimation of hnco into the gas - phase , and raise the fractional abundance of gaseous hnco . the average fractional abundance of hnco in the cnd is comparable to that of a hot core or shocked region ( table 7 ) , so that sublimation during warm up or via shock waves in the cnd may be causing the high abundance of hnco there . the peak emission of ch@xmath65cn in paper-1 shows no obvious offset from the central radio continuum position as far as we could tell from the current angular resolution of the map . the formation of ch@xmath65cn occurs at lower temperatures through a radiative association reaction in the gas - phase @xmath68 followed by @xmath69 as the prestellar core condenses isothermally , ch@xmath65cn can freeze onto the dust grains . additional formation routes of ch@xmath7cn are through surface reactions on dust grains . one of them is a series of hydrogenation reactions starting from c@xmath9n : @xmath70 @xmath71 @xmath72 where ( s ) stands for solid phase . another surface reaction to form ch@xmath7cn is @xmath73 although the route via reactions ( [ eq : c2n_hyd])([eq : ch2cn_hyd ] ) is much more efficient in the estimate of the gas - grain model based on the reaction network by garrod et al . ( 2008 ) with some added reactions from harada et al . ch@xmath7cn formed through the above mentioned routes can again desorb into the gas - phase when a protostar is formed and the collapsing gas and dust heats up , forming a hot core . ch@xmath65cn is observed with high abundances in hot cores , especially in regions with higher rotational temperatures ( nummelin et al . although the overall abundance in the cnd is only slightly higher than that of a cold core and equivalent to that of the shocked cloud , the fractional abundance of ch@xmath65cn at its peak location might be much higher . strong emission of ch@xmath65cn near the galactic nucleus in the cnd might be coming from star formation regions . [ [ cn-1 ] ] cn ^^ the distribution of cn emission seems to be clearly separated into the e and w - knots . cn is an unstable radical , and both chemical models and observations suggest that the cn fractional abundance is high when the chemistry is at an early stage of evolution of molecular clouds or when there is high flux of uv - photon , x - rays , and/or cosmic - rays to ionize and dissociate precursor molecules such as hcn . therefore , its fractional abundance in diffuse clouds is high ( @xmath74 , in a model by le petit et al . 2006 ) , but its fractional abundance decreases as the chemistry evolves to a dense cloud and to hot core / corino . especially in hot cores , a high - temperature reaction @xmath75 efficiently converts cn into hcn ( harada et al . although this reaction can reduce the fractional abundance of cn at an elevated temperature caused by shock waves , a young shock still has a high fractional abundance of cn ( mitchell & deveau , 1983 ) . even in dense regions , the cn fractional abundance can be abundant in pdrs ( jansen et al . 1995 ) or xdrs with a fractional abundance of @xmath76 ( meijerink et al . 2007 ; harada et al . the value derived in the cnd is closer to the value in xdrs and in the shocked region l1157(b1 ) , but higher than that of cold cores or , especially , hot cores . since a high x - ray luminosity from the agn is seen , the high abundance of cn seen in the cnd can be explained by xdrs . although the emissions of cn peak away from agns , a model by harada et al . ( 2013 ) shows that xdrs can exist in relatively less dense regions @xmath4 100 pc away from the agn depending on the density structure of the disk . another possible scenario can be a mixing of ionized / atomic gas with molecular gas caused by turbulence , which is expected to increase the amount of radicals . however , this scenario should be tested by a model . as mentioned earlier , the emission of ch@xmath65oh is separated into the e - knot and w - knot with weaker emission near the agn core . the most efficient production mechanism of ch@xmath65oh is through grain - surface reactions . a series of hydrogenations starting with co can occur while the dust is cold ( @xmath77k ) . in order for methanol to be observed in the gas - phase , methanol on grains must sublime into the gas - phase . in the cold cores , a small amount of methanol can escape into the gas - phase through non - thermal desorption to make a fractional abundance of about 10@xmath78 ( vasyunin & herbst 2013 ) . when the dust grains are heated due to either warm - up of the hot core ( garrod et al . 2008 ) or shock waves ( viti et al . 2011 ) , most of the methanol on grains can sublime into the gas - phase to give a fractional abundance of @xmath79 . the regions that have low abundances of gaseous ch@xmath65oh may be free of a mechanism to heat up the dust grains , or the dust can be initially warm so that co , the precursor species of methanol , desorbs into the gas - phase before it can lead to the formation of methanol . our observed abundance of ch@xmath65oh in the cnd is higher than that of cold cores , which suggests that the emission is likely , once again , to be coming from either hot cores or shocked regions . a high abundance of methanol can lead via photolysis and surface chemistry to more complex organic species such as methyl formate ( hcooch@xmath7 ) and dimethyl ether ( ch@xmath7och@xmath7 ) ( garrod et al . 2008 ) . the emission of hc@xmath65n comes from a compact area near the agn core ( paper-1 ) , probably corresponding to the e - knot . hc@xmath65n can be synthesized in the gas - phase , and the peak fractional abundance in cold cores is 2 @xmath80 , a small factor of 1.7 higher than in the cnd . the fractional abundance can increase with an elevation of temperature ( harada et al . 2010 , 2013 ) , but too high a value for the x - ray ionization rate can dissociate hc@xmath65n , which is therefore not abundant in xdrs . the emission of hc@xmath65n in the cnd must be coming from regions shielded from x - rays . since the peak of hc@xmath7n , which may be compact , does not seem to be resolved , the fractional abundance of hc@xmath7n can only be precisely determined by higher angular resolution data . therefore , its emission may be coming from hot cores in compact star forming regions , or more cold and yet dense cores ( @xmath8110@xmath82 @xmath83 ) in a more spread out distribution . the existence of hc@xmath7n suggests that other carbon - chain molecules might also be observable . the model for the cnd by harada et al . ( 2013 ) shows that the cs / so ratio increases with decreasing density both at the free - fall time of a molecular cloud and at steady - state . for both cs and so , the values in the shocked cloud are more than an order of magnitude higher than those in the cold core . considering the uncertainty , it is not possible to identify the source of the emission in ngc 1068 by a comparison with these galactic sources . high abundances of the sulfur - containing species so and cs may result from high elemental sulfur abundance in the gas phase due to shock waves ( wakelam et al . 2005 ) . overall , the cnd hosts relatively complex molecules , which are often associated with shocked molecular clouds or hot cores . because a high x - ray flux can dissociate these molecules , they must also reside in regions shielded from x - rays . although we have derived the column densities of these molecules , there are some difficulties in discerning which exact source the emission is coming from . in this paper , we took the column density ratios over co column densities . since the critical density of co is lower than the rest of the molecules , co emission can also come from translucent clouds of @xmath84@xmath83 , which are likely to have a higher volume filling factor . even if most of the emission of a molecule such as ch@xmath7cn comes from hot cores , the calculated column density ratio of ch@xmath7cn over co will turn out to be lower than the value in table 7 if the mean density is low . therefore , it is hard to determine whether the ch@xmath7cn or hc@xmath7n emission is coming from an extended volume of molecular clouds with @xmath85@xmath83 with a low fractional abundance or from hot cores of @xmath86@xmath83 and @xmath87k with a small volume filling factor . besides the difference in the density , the difference in abundances of each type of source may also explain the observed emission . for example , low abundances of ch@xmath7cn and hc@xmath7n are seen in shocked clouds , and their abundances are high in hot cores , so that the differences between the peak locations of ch@xmath7oh and hnco and those of ch@xmath7cn and hc@xmath7n may come from different environments ; viz . , hot cores versus shocked clouds . these differences indicate a higher star formation activity within tens of parsecs from the agn core , where ch@xmath7cn and hc@xmath7n have peak abundances , albeit considerably lower than expected from hot core material . it should be also noted that the shock chemistry might also be different around the agn . the shocked molecular cloud value used in table 7 is from an outflow of the young stellar object l1157 . larger scale shocks in the cnd of ngc 1068 may be different in both the pre - shock density and evolutionary timescale of the chemistry . we observed the seyfert 2 galaxy ngc 1068 with the 0.8-mm band in the cnd and starburst ring during the alma early science program . the analyses were carried out with our data obtained at 3-mm band ( paper-1 ) . the main results of this work are summarized in the following list . we successfully observed images of @xmath0co ( @xmath2=32 ) , c@xmath1o ( @xmath2=32 ) , cn ( @xmath3=32 ) , and cs ( @xmath2=76 ) with an angular resolution of @xmath41.@xmath53 @xmath6 1.@xmath52 . @xmath0co and cn were detected in both of the cnd and starburst ring , while cs was only detected in the cnd . c@xmath1o was not significantly detected toward the cnd , but is relatively strong in the starburst ring . 2 . at present , some molecules have been found to be separated into eastern and western knots in the cnd with interferometric observations , and similar distributions of @xmath0co ( @xmath2=32 ) , cn ( @xmath3=32 ) and cs ( @xmath2=76 ) are found in our observations for the first time . we have determined rotation diagrams of observed molecules in the 3-mm band ( paper-1 ) and 0.8-mm band ( this work ) toward the cnd and the southwest position in the starburst ring . the rotational temperatures and column densities have been calculated from these diagrams . @xmath0co and c@xmath1o molecules trace cold gas ( @xmath50 10 k ) in the starburst ring and a little warmer gas ( 14.221.5 k ) in the cnd . although the column densities of the co isotopic species in the cnd are only one - third of that in the starburst ring , those of the shock / dust related molecules are enhanced by a factor of 34 in the cnd under the assumption of the same @xmath29 in the cnd and starburst ring . if the @xmath29 in the cnd is higher than the starburst ring , the column density of the shock related molecules , such as hnco , ch@xmath7cn , so , and ch@xmath7oh in the cnd are much larger than that in the starburst ring . the difference of the distributions between @xmath0co and c@xmath1o is unexplained by only a difference of excitation conditions , because the abundance ratio of these species is not sensitive to the molecular gas temperature and density . the column density ratios of @xmath0co / c@xmath1o in the cnd and the starburst ring are @xmath46 under the assumption that @xmath29 = 30 k for c@xmath1o and @xmath43 , respectively . the amount of c@xmath1o in the cnd is smaller than that in the starburst ring as compared with that of @xmath0co . this feature is possibly the result of differences in isotopic ratios and/or the effect of isotope - selective photodissociation . we found an especially large enhancement of cn in the cnd . specifically , the column density in the cnd is a factor of 2050 larger than that in the starburst ring . cn is one of the key molecules to study an x - ray dominated region ( xdr ) , as mentioned in theoretical studies ( e.g. meijerink et al . 2007 ; harada et al . 2013 ) . moreover the fractional abundance derived in the cnd is closer to the value in xdrs and in the shocked region l1157(b1 ) , but higher than that of cold cores or , especially , hot cores . since a high x - ray luminosity from the agn is seen , the high abundance of cn seen in the cnd can be explained by xdrs . therefore , the enhancement of column density and abundance of cn in the cnd may be the effects of extreme physical conditions such as those in xdrs . we discuss the chemistry of each species , and compare the observed abundances in the cnd and starburst ring with those of galactic sources such as cold cores , hot cores , and shocked molecular clouds in order to study the overall characteristics . as a result , the cnd of ngc 1068 seems to be chemically very rich . the cnd hosts relatively complex molecules , which are often associated with shocked molecular clouds or hot cores . because a high x - ray flux can dissociate these molecules , they must also reside in regions shielded from x - rays . this paper makes use of the following alma data : ads / jao.alma#2011.0.00061.s . alma is a partnership of eso ( representing its member states ) , nsf ( usa ) and nins ( japan ) , together with nrc ( canada ) and nsc and asiaa ( taiwan ) , in cooperation with the republic of chile . the joint alma observatory is operated by eso , aui / nrao and naoj . we thank the alma staff , especially akiko kawamura , for their supports . e. h. wishes to acknowledge the support of the national science foundation ( us ) for his astrochemistry program . he also acknowledges support from the nasa exobiology and evolutionary biology program through a subcontract from rensselaer polytechnic institute . kohno , k. , matsushita , s. , vila - vilar , b. , okumura , s. k. , shibatsuka , t. , okiura , m. , ishizuki , s. , & kawabe , r. 2001 , in asp conf . ser . , 249 , the central kiloparsec of starburst and agn : the la palma connection , ed . j. h. knapen et al . ( san francisco : asp ) , 672 mcmullin , j. p. , waters , b. , schiebel , d. , young , w. , & golap , k. 2007 , in asp conf . 376 , astronomical data analysis software and systems xvi , ed . r. a. shaw , f. hill , & d. j. bell ( san francisco , ca : asp ) , 127 nakajima , t. , takano , s. , kohno , k. , & line survey team 2013b , in asp conf . 476 , new trends in radio astronomy in the alma era : the 30th anniversary of nobeyama radio observatory , ed . r. kawabe , n. kuno , & s. yamamoto ( san francisco , ca : asp ) , 299 lc parameter & value + observation date & 28 nov 2011 + number of antennas & 14 + observing time ( min ) & 71.4 + bandpass calibrator & j0423 - 013 + flux calibrator & callisto + phase calibrator & j0339 - 017 + rms noise ( mjy beam@xmath14 ) & ( cnd / starburst ring ) + spw0 & 8.614.3/9.013.6 + spw1 & 5.46.6/5.47.6 + spw2 & 3.24.5/3.33.7 + spw3 & 3.54.0/3.64.5 + synthesized beam & ( major @xmath6 minor , p.a . ) + & 1.@xmath529@xmath61.@xmath523 , 2.@xmath893 + llllllll frequency & molecule & transition & @xmath90/k & peak flux & @xmath91 & line width & integrated flux + ( ghz ) & & & ( k ) & ( jy beam@xmath14 ) & ( km s@xmath14 ) & ( km s@xmath14 ) & ( jy beam@xmath14 km s@xmath14 ) + 329.330546 & c@xmath1o & @xmath2 = 32 & 31.76 & [email protected] & 1106@xmath2820 & 173@xmath2847 & [email protected] + 329.66437 & hnco & @xmath36 = 15@xmath3714@xmath38 & 126.69 & & & & @xmath500.97 ( 1@xmath27 ) + 330.587960 & @xmath0co & @xmath2 = 32 & 31.64 & [email protected] & 1102@xmath284 & 178@xmath289 & [email protected] + 331.071548 & ch@xmath7cn & @xmath41 = 18@xmath4217@xmath42 & 151.09 & & & & @xmath500.83 ( 1@xmath27 ) + 340.031567 & cn & @xmath3 = 32 , @xmath2 = 5/2 - 3/2 & 32.66 & [email protected] & 1049@xmath288 & 272@xmath2835 & [email protected] + 340.247874 & cn & @xmath3 = 32 , @xmath2 = 7/2 - 5/2 & 32.69 & [email protected] & 1171@xmath287 & 308@xmath2815 & [email protected] + 340.714350 & so & @xmath51 = 7@xmath526@xmath53 & 81.31 & & & & @xmath500.76 ( 1@xmath27 ) + 342.729781 & ch@xmath7oh & @xmath36 = 13@xmath3913@xmath40-+ & 232.49 & & & & @xmath500.49 ( 1@xmath27 ) + 342.882866 & cs & @xmath2 = 76 & 65.88 & [email protected] & 1109@xmath287 & 183@xmath2817 & [email protected] + llllllll frequency & molecule & transition & @xmath90/k & peak flux & @xmath91 & line width & integrated flux + ( ghz ) & & & ( k ) & ( jy beam@xmath14 ) & ( km s@xmath14 ) & ( km s@xmath14 ) & ( jy beam@xmath14 km s@xmath14 ) + 329.330546 & c@xmath1o & @xmath2 = 32 & 31.76 & [email protected] & 1216@xmath281 & 28@xmath283 & [email protected] + 329.66437 & hnco & @xmath36 = 15@xmath3714@xmath38 & 126.69 & & & & @xmath500.75 ( 1@xmath27 ) + 330.587960 & @xmath0co & @xmath2 = 32 & 31.64 & [email protected] & 1216@xmath281 & 40@xmath282 & [email protected] + 331.071548 & ch@xmath7cn & @xmath41 = 18@xmath4217@xmath42 & 151.09 & & & & @xmath500.41 ( 1@xmath27 ) + 340.031567 & cn & @xmath3 = 32 , @xmath2 = 5/2 - 3/2 & 32.66 & & & & @xmath500.27 ( 1@xmath27 ) + 340.247874 & cn & @xmath3 = 32 , @xmath2 = 7/2 - 5/2 & 32.69 & [email protected] & 1212@xmath283 & 42@xmath286 & [email protected] + 340.714350 & so & @xmath51 = 7@xmath526@xmath53 & 81.31 & & & & @xmath500.27 ( 1@xmath27 ) + 342.729781 & ch@xmath7oh & @xmath36 = 13@xmath3913@xmath40-+ & 232.49 & & & & @xmath500.31 ( 1@xmath27 ) + 342.882866 & cs & @xmath2 = 76 & 65.88 & & & & @xmath500.31 ( 1@xmath27 ) + lccccc & & & + molecule & @xmath29 & @xmath30 & & @xmath29 & @xmath30 + & ( k ) & ( @xmath610@xmath54 @xmath33 ) & & ( k ) & ( @xmath610@xmath92@xmath33 ) + c@xmath1o & & & & [email protected] & 167@xmath93 + @xmath0co & 21.5@xmath94 & 147@xmath95 & & [email protected] & 521@xmath96 + cs & [email protected] & 2.7@xmath97 & & & + hc@xmath7n & 22 & 0.9 & & 18 & 0.1 + lccccccccccc & & & + molecule & 5(k ) & 10(k ) & 15(k ) & 30(k ) & 45(k ) & & 5(k ) & 10(k ) & 15(k ) & 30(k ) & 45(k ) + c@xmath1o & 901.9 & 69.8 & 35.4 & 24.0 & 25.1 & & & & & & + hnco & 1.2 & 0.5 & 0.4 & 0.5 & 0.6 & & 0.3 & 0.1 & 0.1 & 0.1 & 0.2 + ch@xmath7cn & 0.2 & 0.08 & 0.07 & 0.09 & 0.1 & & & & & & + cn & 38.4 & 2.9 & 1.5 & 1.0 & 1.0 & & 0.8 & 0.06 & 0.03 & 0.02 & 0.02 + so & 1.2 & 1.1 & 1.4 & 2.4 & 3.6 & & 0.2 & 0.2 & 0.2 & 0.4 & 0.6 + ch@xmath7oh & 2.4 & 2.5 & 3.0 & 5.4 & 8.5 & & 1.8 & 1.9 & 2.3 & 3.5 & 5.7 + cs & & & & & & & 0.4 & 0.4 & 0.5 & 0.7 & 1.0 + lccccc molecular line & component & @xmath23 & @xmath24 & integrated flux & flux ratio + & & ( @xmath11 ) & ( @xmath11 ) & ( jy beam@xmath14 km s@xmath14 ) & ( e - knot / w - knot ) + @xmath0co ( @xmath2=32 ) & e - knot & + 0.7 & 0.0 & [email protected] & + & w - knot & -1.8 & + 0.5 & [email protected] & [email protected] + cn ( @xmath3=32 ) & e - knot & + 0.6 & -0.1 & [email protected] & + & w - knot & -1.7 & + 0.8 & [email protected] & [email protected] + cs ( @xmath2=76 ) & e - knot & + 0.6 & -0.2 & [email protected] & + & w - knot & -1.8 & + 0.4 & [email protected] & [email protected] + lcccccc species & cnd & starburst ring & cold core & & shocked cloud + & ( ngc 1068 ) & ( ngc 1068 ) & ( tmc-1 ) & ( sgr b2(n ) ) & ( afgl2591 ) & ( l1157(b1 ) ) + hnco & 0.7@xmath98(-8 ) & 0.4@xmath99(-9 ) & [email protected](-10 ) & 5.6(-10)(halo ) & & [email protected](-8 ) + & & & & 1.2(-8)(core ) & & + ch@xmath65cn & 1.1@xmath100(-9 ) & & 6(-10 ) & 2.2(-7)(core ) & & [email protected](-9 ) + cn & 0.5@xmath101(-7 ) & 0.2@xmath102(-9 ) & 5(-9 ) & & [email protected](-12 ) & 1.6(-7 ) + so & 1.6@xmath103(-8 ) & [email protected](-10 ) & 2(-9 ) & 2(-8)(halo ) & 1(-8 ) & 2.5(-7 ) + ch@xmath65oh & 3.4@xmath104(-8 ) & 7.3@xmath105(-9 ) & 3(-9 ) & 3.4(-9)(halo ) & @xmath1062(-8 ) & [email protected](-5 ) + & & & & 1.7(-6)(core ) & & + cs & 3.7@xmath103(-8 ) & 1.5@xmath107(-9 ) & 4(-9 ) & & 2(-8 ) & 1.9(-7 ) + hc@xmath65n & 1.2(-8 ) & 3.8(-10 ) & 2(-8 ) & 5(-9)(halo ) & 7(-9 ) & 1.0(-8 ) + & & & & 1(-7)(core ) & & +

we study lorentzian spacetimes for which all scalar invariants constructed from the riemann tensor and its covariant derivatives are constant ( spacetimes ) . we obtain a number of general results in arbitrary dimensions . we study and construct warped product spacetimes and higher - dimensional kundt spacetimes . we show how these spacetimes can be constructed from locally homogeneous spaces and spacetimes . the results suggest a number of conjectures . in particular , it is plausible that for spacetimes that are not locally homogeneous the weyl type is , , or , with any boost weight zero components being constant . we then consider the four - dimensional spacetimes in more detail . we show that there are severe constraints on these spacetimes , and we argue that it is plausible that they are either locally homogeneous or that the spacetime necessarily belongs to the kundt class of spacetimes , all of which are constructed . the four - dimensional results lend support to the conjectures in higher dimensions .

gr-qc0509113

let @xmath6 denote a differentiable manifold , either of riemannian or lorentzian signature . we are interested in all spacetimes for which all scalar invariants constructed from the riemann tensor and its covariant derivatives are constant ( denoted as @xmath0 spacetimes ) . in the case of a riemannian manifold , if it is @xmath0 then it is locally homogeneous @xcite . this theorem is false for lorentzian manifolds ( there are a number of counterexamples ; e.g. , the non - homogeneous @xmath1 spacetimes @xcite ) . there are a number of @xmath0 spacetimes that arise from homogeneous spacetimes and simple warped products ( see section 3 ) . some examples of @xmath0 spacetimes that arise as simple _ tensor sum _ generalizations of @xmath1 spacetimes with a cosmological constant are given in @xcite ( some higher dimensional versions are given in @xcite ) . simple examples of _ fibered products _ were given in @xcite and @xcite ( see also @xcite ) . it is also straightforward to construct @xmath0 spacetimes as _ warped products_. all of these spacetimes belong to the class @xmath7 described below . we begin with some definitions , and a brief discussion of the four - dimensional case . let us denote by @xmath8 the set of all scalar invariants constructed from the curvature tensor and its covariant derivatives up to order @xmath9 . @xmath10 is called @xmath1@xmath11 if for any invariant @xmath12 , @xmath13 over @xmath10 . @xmath10 is called @xmath0@xmath11 if for any invariant @xmath12 , @xmath14 over @xmath10 . moreover , if a spacetime is @xmath1@xmath11 or @xmath0@xmath11 for all @xmath9 , we will simply call the spacetime @xmath1 or @xmath0 , respectively . the set of all locally homogeneous spacetimes will be denoted by @xmath15 . clearly @xmath16 and @xmath17 . let us denote by @xmath7 all reducible @xmath0 spacetimes that can be built from @xmath1 and @xmath15 by ( i ) warped products ( ii ) fibered products , and ( iii ) tensor sums ( defined more precisely later ) . let us denote by @xmath18 those spacetimes for which there exists a frame with a null vector @xmath19 such that all components of the riemann tensor and its covariants derivatives in this frame have the property that ( i ) all positive boost weight components ( with respect to @xmath19 ) are zero and ( ii ) all zero boost weight components are constant . note that @xmath20 and @xmath21 . ( there are similar definitions for @xmath22 etc . @xcite ) . finally , let us denote by @xmath23 , those @xmath0 spacetimes that belong to the ( higher - dimensional ) kundt class ( defined later ) ; the so - called kundt @xmath0 spacetimes . in particular , we shall study the relationship between @xmath7 , @xmath18 , @xmath23 and especially with @xmath24 . we note that by construction @xmath7 is at least of weyl type @xmath2 ( i.e. , of type @xmath2 , @xmath3 , @xmath4 or @xmath5 @xcite ) , and by definition @xmath18 and @xmath23 are at least of weyl type @xmath2 ( more precisely , at least of riemann type @xmath2 ) . in 4d , @xmath7 , @xmath18 and @xmath23 are closely related , and it is plausible that @xmath24 is at least of weyl type @xmath2 ( see section 7 ) . in four dimensions the weyl classification and the petrov classification are closely related @xcite . however , due to the fact that the weyl classification contains many more possibilities in dimensions higher than four we will _ restrict the term petrov classification to four dimensions only_. we are particularly interested in the four - dimensional @xmath0 spacetimes . for a riemannian manifold every @xmath0 is homogeneous @xmath25 . this is not true for lorentzian manifolds . however , for every @xmath0 with particular constant invariants there is a homogeneous spacetime ( not necessarily unique ) with precisely the same constant invariants . this suggests that @xmath0 can be `` built '' from @xmath15 and @xmath1 ( e.g. , @xmath7 ) . in addition , there is a relationship between the @xmath0 conditions and ( i ) the rank of the riemann tensor and its holonomy class @xcite ( ii ) the existence of curvature collineations @xcite ( iii ) the condition of non - complete backsolvability ( ncb ) @xcite , in addition to ( iv ) curvature homogeneity is curvature homogeneous ( of order zero ) if there exists a frame with respect to which the riemann tensor has constant components . ] and ( v ) sectional curvature @xcite . the relationship between @xmath0 and curvature homogeneity in @xmath26 is studied in @xcite . in locally homogeneous spacetimes all of the sectional curvatures ( gaussian curvatures ) are constant . with the exception of certain special plane wave and constant curvature spacetimes ( and all vacuum spacetimes ) , the sectional curvature uniquely determines the spacetime metric @xcite . it is clear that @xmath0 consists of @xmath7 and , possibly , some other very special spacetimes . let us present a _ heuristic argument _ for this in @xmath26 . let us suppose that the spacetime is @xmath0 . for spacetimes that are completely backsolvable ( cb ) there is a special invariant frame such that all components of the riemann tensor are constant . with respect to this invariant frame there exist smooth vector fields @xmath27 that act transitively on the manifold whose directional derivatives leave the riemann tensor invariant ( i.e. , @xmath28 form a lie algebra of curvature collineations that span @xmath10 ) . in general , every curvature collineation is a homothetic vector , so as a result there is a homothetic group acting transitively on @xmath10 . therefore , this then implies that in general the @xmath0 spacetime is locally homogeneous . the exceptions are those spacetimes that are not cb ( ncb @xcite ) and those spacetimes for which a curvature collineation is not a homothetic vector @xcite , and these spacetimes are related to the @xmath1 spacetimes . in this paper we shall study @xmath0 spacetimes . in the next section we shall begin by summarizing some important results . in section 3 we construct a subclass of @xmath7 spacetimes that arise as warped products of a homogeneous space and a @xmath1 spacetime . in section 4 we consider the higher - dimensional kundt class . in section 5 examples of @xmath23 spacetimes in higher dimensions are given . we discuss the results in section 6 , and present a number of conjectures . in section 7 we summarize the results in 4d in detail . there are three appendices . in appendix a , we present all of the 3-dimensional @xmath1 metrics , which is necessary for the explicit determination of the metrics in the 4-dimensional @xmath7 spacetimes . in appendix b , 4-dimensional @xmath0 spacetimes are given explicitly , and the relationship between @xmath7 , @xmath18 , and @xmath23 in 4d is discussed . finally , in appendix c , we write down the metric for higher dimensional @xmath1 spacetimes in a canonical form . let us first consider the riemannian case where a great deal about these spacetimes is known . the reason the riemannian case is easier to deal with is because the orthogonal group @xmath29 is compact , and hence , the orbits of its group action are also compact . if @xmath10 is a riemannian spacetime , then : 1 . @xmath1@xmath30 @xmath31 @xmath1 . 2 . there exists a @xmath32 such that @xmath0@xmath11 @xmath31 @xmath0 . ( 1 ) : see proof below . ( 2 ) : see @xcite a riemannian space is @xmath1 if and only if it is flat . consider the invariant @xmath33 . using an orthonormal frame , this invariant is a sum of squares ; hence , @xmath34 @xmath35 @xmath36 . a riemannian space is @xmath0 if and only if it is locally homogeneous . moreover , the set of curvature invariants , @xmath37 , uniquely determines the metric up to isometries . see prfer - tricerri - vanhecke @xcite . [ singer ] if @xmath10 is curvature homogeneous of order @xmath38 , then @xmath10 is locally homogeneous . see singer @xcite . the lorentzian case is more difficult to deal with and only partial results are known . however , in the @xmath1 case , it was shown in @xcite that @xmath1@xmath39 implies @xmath1 . a similar result in the @xmath0 case is not known ; however , in this case we believe that an @xmath40 exists such that @xmath0@xmath41 implies @xmath0 . in the riemannian case @xmath10 is locally homogeneous if and only if @xmath10 is curvature homogeneous to all orders ( theorem 2.4 ) . in the lorentzian case , we have that there is an integer @xmath42 such that if @xmath43 is a complete , simply connected pseudo riemannian manifold of type @xmath44 which is @xmath42-curvature homogeneous , then @xmath10 is locally homogeneous . see podesta and spiro @xcite . we argued above that in 4d lorentzian manifolds for every @xmath0 with particular constant invariants there is a locally homogeneous spacetime with the same constant invariants . it is plausible that this is true in higher dimensions . assume that a lorentzian spacetime @xmath10 is a @xmath0 spacetime with curvature invariants @xmath37 . then there exists a locally homogeneous space @xmath45 with curvature invariants @xmath46 . consider an arbitrary covariant tensor @xmath47 and a null frame @xmath48 ; i.e. , @xmath49 consider now a boost in the @xmath50-plane : \{_a,_a , ^ _ a}= \{e^_a , e^-n_a , m^ _ a}. we can consider the vector - space decomposition of the tensor @xmath47 in terms of the boost weight with respect to the above boost ( see @xcite ) : t=_b(t)_b , where @xmath51 denotes the projection of the tensor @xmath47 onto the vector space of boost - weight @xmath52 . the components of the tensor @xmath51 with respect to the frame will transform according to : ( t)_b ab ... =e^-b(t)_b .... furthermore , we note that ( ts)_b=_b+b=b(t)_b(s)_b . moreover , the connection @xmath53 can similarly be decomposed according to whether it raises , lowers or preserves the boost weight : = ( ) _ -1+()_0+()_1 , where @xmath54 is defined by @xmath55 for all vectors @xmath56 . we note that for the metric , @xmath57 ; hence , raising or lowering tensor indices preserves the boost weight . an invariant of a tensor , @xmath47 , is necessarily @xmath58-invariant ; in particular , for any full contraction of @xmath47 : [ t]=[(t)_0 ] . this property can now be utilized to construct spaces with constant curvature invariants . in the @xmath1 case , there exists a null - frame such that the riemann tensor , @xmath59 , has the property : ( r)_b=0,b0 . this property alone implies that @xmath60 , regardless of @xmath61 ( @xmath62 ) . if the curvature tensor is of the form : ( r)_b=0,b > 0 , ( r)_b abcd , e=0,b= 0 , in a frame for which @xmath63 , it is necessarily a @xmath0@xmath30 spacetime . in general , further restrictions must be put on @xmath53 in order to make it @xmath0 . we note that spacetimes can can be classified according to their boost - weight components @xmath61 . in particular , a classification using the weyl tensor @xmath64 , which generalises the petrov classification in 4d , has been employed based on the components @xmath65 @xcite . in the 4d @xmath1 spacetimes , all of the positive boost weight spin coefficients ( e.g. , @xmath66 and @xmath67 ) are zero , and hence it follows that @xmath19 is geodesic , non - expanding , shear - free and non - twisting and the spacetime is kundt @xmath68 @xcite . this is also true for @xmath0 spacetimes , in the sense that if the @xmath0 spacetime is not locally homogeneous , then in general it belongs to @xmath68 . that is , it is plausible that in 4d , @xmath69 . in higher dimensions it was also shown that @xmath19 is geodesic , non - expanding , shear - free and non - twisting ( i.e. , @xmath70 ) in @xmath1 spacetimes @xcite . in locally homogeneous spacetimes , in general there exists a null frame in which the @xmath71 are constants . we therefore anticipate that in @xmath0 spacetimes that are not locally homogeneous , @xmath70 . for higher dimensional @xmath0 spacetimes with @xmath70 , the ricci and bianchi identities appear to be identically satisfied @xcite . a higher dimensional spacetime which admits a null vector @xmath19 which is geodesic , non - expanding , shear - free and non - twisting , will be denoted as a higher - dimensional kundt spacetime . let us assume that the spacetime admits a geodesic , non - twisting , non - expanding , shear - free null vector @xmath72 . it can be then shown that the existence of a twist - free and geodesic null vector implies that there exists a local coordinate system @xmath73 such that s^2=2u(v+hu+w _ ix^i)+_ij(v , u , x^k)x^ix^j . in essence , the twist - free condition implies that the vector field is surface forming so that there exists locally an exact null one - form @xmath74 . using a coordinate transformation and a null - rotation we can now bring the metric into the above form . in this coordinate system , @xmath75 . in addition , requiring that this vector field is non - expanding and shear - free implies that @xmath76 . we will therefore consider metrics of the from s^2=2u(v+hu+w _ ix^i)+_ij(u , x^k)x^ix^j , [ kundt ] where @xmath77 and @xmath78 . the metric ( [ kundt ] ) possesses a null vector field @xmath19 obeying ^a_b;a=^a _ ; a=^a;b_(a;b)=^a;b_[a;b]=0 ; i.e. , it is geodesic , non - expanding , shear - free and non - twisting . we will refer to the metrics ( [ kundt ] ) as higher - dimensional _ kundt metrics _ ( or simply kundt metrics ) , since they generalise the 4-dimensional kundt metrics . all examples to the authors knowledge of lorentzian @xmath0 spacetimes are of the following two forms : 1 . homogeneous spaces 2 . a subclass of the kundt spacetimes for the homogeneous spaces , their decomposition can be of the most general type , @xmath79 since a frame can be chosen such that all components are constants . however , for all known examples of non - homogeneous @xmath0 metrics , a frame can be found such that we have the decomposition @xmath80 with constant boost - weight zero components . since a homogeneous space is automatically a @xmath0 space , we will in this paper consider metrics which are not necessarily homogeneous . since all known examples of non - homogeneous metrics are of kundt form , we will henceforth only consider kundt metrics . it is known that @xmath7 spacetimes can be constructed in all dimensions via warped products . let us first determine necessary and sufficient conditions so that the warped product of a homogeneous space and a @xmath1 is @xmath0 . we consider the warped product metric @xmath81 where @xmath82 is a riemannian homogenous space and @xmath83 is a @xmath9-dimensional @xmath1 lorentzian manifold , the warping function @xmath84 only depends on points of the homogenous space . using the left - invariant frame of the homogeneous space , @xmath85 we complete it to a basis of the warped product by appending the null frame , @xmath86 where the primed vectors represent the canonical @xmath1 frame in which @xmath87 ( and hence @xmath19 ) is the aligned geodesic null congruence that is expansion , shear , and twist free . the non - vanishing inner products are such that @xmath88 defining @xmath89 , we shall show that if @xmath90 and @xmath91 are constant then this implies @xmath0 . these two conditions imply that @xmath92 is an eigenvector of @xmath91 with constant eigenvalue @xmath90 , and thus @xmath93 must be constant . following @xcite , we perform a decomposition of the riemann tensor for the warped product to obtain @xmath94 evidently the riemann tensor is of boost order zero with boost weight zero components : @xmath95 \label{rbw0.4 } \\ r_{\hat a\hat\mu\hat b\hat\lambda } & = & - t_{\hat a\hat b}\stackrel{2}{\delta}_{\hat\mu\hat\lambda } \label{rbw0.5 } \\ r_{\hat a\hat b\hat c\hat d } & = & \stackrel{1}{r}_{\hat a\hat b\hat c\hat d}. \label{rbw0.6}\end{aligned}\ ] ] the negative boost weight components , arising from the @xmath1 spacetime , are given by @xmath96 the condition that @xmath93 is constant ( and hence @xmath90 and @xmath97 are constant ) implies that the boost weight zero components of the riemann tensor are constant ; the absence of positive boost weight components then gives @xmath0@xmath98 . the covariant derivative of the null frame has the form , @xmath99 where @xmath100 denotes terms of lower order boost weight with respect to previous terms in an expression , and the unspecified @xmath101 s correspond to @xmath1 ricci - rotation coefficients @xcite [ for example , @xmath102 , except for the @xmath103 which are the rotation coefficients associated with the homogeneous space . equations ( [ dl])-([dma ] ) have leading order boost weight terms of @xmath104 and @xmath105 , respectively , which is the same boost weight as the corresponding null frame vector before taking a covariant derivative . it follows that the covariant derivative of the @xmath106 quantities appearing in ( [ riem ] ) will produce terms of equal or lesser boost weight . furthermore , in @xcite it was proven that the negative boost weight curvature components appearing in ( [ riem ] ) will remain negative upon covariant differentiation . therefore , if the boost weight zero components of the riemann tensor and its derivatives are constant then any number of covariant derivatives of the riemann tensor will be of boost order zero , which implies that the warped product will be @xmath0 . supposing that the boost weight zero components of @xmath107 are constant , then for @xmath108 the boost weight zero components will have contributions from @xmath109 , which is constant , and the non - constant @xmath101 s arising from the @xmath1 spacetime , however these boost weight zero @xmath1 @xmath101 s do not occur as a result of theorem 5.1 . therefore we have that if @xmath93 is constant , then the warped product is @xmath0 . for reference , we include the covariant derivative of the riemann tensor . by using the ricci identity we have , @xmath110}=t_{\hat b\hat c;\hat d}-t_{\hat b\hat d;\hat c } , \label{riciden}\ ] ] which can be used to express the non - vanishing boost weight zero components of the covariant derivative of the riemann tensor as @xmath111\delta_{\hat\mu\hat\lambda } \label{drbw0.2 } \\ r_{\hat 0\hat b\hat 1\hat c;\hat e } & = & -t_{\hat b\hat c;\hat e } \label{drbw0.3 } \\ r_{\hat b\hat \mu\hat c\hat \lambda;\hat e } & = & -t_{\hat b\hat c;\hat e}\delta_{\hat \mu\hat \lambda}. \label{drbw0.4}\end{aligned}\ ] ] as expected all components in ( [ drbw0.1])-([drbw0.4 ] ) are constant ( as a result of requiring @xmath93 to be constant ) . we shall now show the converse ; that is , if the ricci invariants are constant then @xmath93 is constant . in local coordinates , the ricci tensor and ricci scalar are @xmath112\delta_{\mu}^{\rho } \label{ric2 } \\ r_{m}^{~ p } & = & \stackrel{1}{r}^{~ p}_{m}-kt_{m}^{~ p } \label{ric1 } \\ r & = & \stackrel{1}{r}-k[2t+(k-1)\tau_{,m}\tau^{,m}],\label{ricsc}\end{aligned}\ ] ] where in ( [ ricsc ] ) we have used that @xmath113 and set @xmath114 . choosing new coordinates @xmath115 so that @xmath116 , then @xmath117 now , with normal coordinates along @xmath118 we have @xmath119 . therefore from ( [ ricsc ] ) , since @xmath120 is constant we have that @xmath121 , and hence @xmath47 , is constant . in this coordinate system the constraint @xmath121 constant implies that @xmath122 is constant . constant ricci scalar implies that @xmath47 and @xmath90 are constant . further conditions are then provided by the constancy of higher degree ricci invariants . from ( [ ric2 ] ) and ( [ ric1 ] ) , the ricci tensor of a warped product is @xmath123 , thus a degree @xmath124 ricci invariant always has the form @xmath125 the second term of ( [ ricinv ] ) is constant since @xmath47 and @xmath90 are constant ; hence invariants constructed from ( [ ric1 ] ) must be constant . assuming the determinant of @xmath126 or @xmath127 is nonzero , then there exists a frame in which @xmath128 and @xmath129 can simultaneously be expressed in block diagonal form with constant eigenvalues @xcite . therefore , ( [ ric1 ] ) implies that in this frame @xmath130 must also have constant components and thus constant invariants . in this case , the constancy of the ricci invariants completely determines the conditions on @xmath130 and specifies a frame in which the components of each term in ( [ ric1 ] ) are constant . the frame in which @xmath129 and @xmath130 are constant is determined by the left - invariant 1-forms . as before , the conditions that @xmath131 and @xmath90 are constant imply @xmath109 is constant , and consequently constant ricci invariants implies @xmath0 . many examples of @xmath0 spacetimes can be found among the cases where the homogeneous space is a solvable lie group : assume that @xmath132 is a solvable lie group equipped with a left - invariant metric . assume also that @xmath132 is connected and simply connected ( i.e. , @xmath133 ) . then there exists a non - constant @xmath84 such that the warped product of @xmath132 and a @xmath1 , as constructed above , is a @xmath0 . we identify @xmath134 , where @xmath135 is a solvable lie group , with the lie algebra @xmath136 . since @xmath137 is solvable the set @xmath138 $ ] will be a proper vector subspace of @xmath137 . thus for a solvable lie algebra , there exists a non - zero @xmath139^{\perp}$ ] where @xmath138^{\perp}$ ] is the complement of the derived algebra @xmath138 $ ] . this further implies that there is a left - invariant one - form @xmath140 such that @xmath141 . since @xmath133 , there exists a non - constant function @xmath142 such that @xmath143 . we can now choose @xmath144 for a constant @xmath145 . in 2-dimensions flat space is the only @xmath1 metric , and in the appendix we list all of the 3-dimensional @xmath1 metrics . examples of @xmath0 warped products where the homogeneous space is either @xmath146 or @xmath147 yield a linear function for @xmath84 ( since both of these can be considered as solvmanifolds ) ; however , for @xmath148 , since the coordinates are cyclic , we find that @xmath84 is a constant . another example , in 5-dimensions , is where the homogeneous space is a bianchi type viii lie group ( @xmath149 ) warped with a 2-dimensional @xmath1 , with coordinates @xmath150 the metric is s^2 & = & e^2 + & & + ( zx+zy)^2+(zy - zx)^2+c^2(z+)^2 . in this case @xmath151 must be linear in @xmath152 to give @xmath0 , and it turns out that for the warped product to be @xmath0 we have to consider two separate cases : 1 . the maximal isometry group of the homogeneous space is 3-dimensional ( @xmath153 ) : @xmath84 is a constant . the maximal isometry group of the homogeneous space is 4-dimensional ( @xmath154 ) : @xmath84 can be non - constant , @xmath155 . the reason why we have to distinguish these to cases is that in the first case , the isometry group is the semi - simple group , @xmath156 . however , in the second case , the isometry group allows for a transitive group which is solvable ( bianchi type iii ) . let us next consider the higher - dimensional kundt metrics ( [ kundt ] ) in the form s^2=2u(v+hu+w_i*m*^i)+_ij*m*^i*m*^j . [ kundtmetric ] it is convenient to introduce the null frame & = & u , + * n*&= & v+hu+w_i*m*^i , + * m*^i&= & e^i _ j(u ; x^k)x^j , + ( _ ab ) & = & 0 & 1 & 0 + 1 & 0 & 0 + 0 & 0 & _ ij such that s^2_h_ij*m*^i*m*^j=_ijx^ix^j,_ij , v=0 . there is a class of coordinate transformations that preserve the form of the metric ( [ kundtmetric ] ) . in particular , we can define new variables , ( v,u,x^i)=(v , u , f^i(u;x^k ) ) . for our purposes we can always use this coordinate transformation to simplify the spatial metric @xmath157 : consider the metric ( [ kundt ] ) and assume that there exists a null frame @xmath158 such that all scalars @xmath159 and @xmath160 are constants . then there exists ( locally ) a coordinate transformation @xmath161 such that _ ij_kl , _ij , u=0 . moreover , @xmath162 is a locally homogeneous space . [ thm : kundt ] by calculating the curvature tensor of the metric ( [ kundt ] ) , and its covariant derivatives , we note that r_= _ , r_;_1_n=_;_n , where the tilde refers to curvature tensors with respect to the spatial metric @xmath163 . since this metric is riemannian with all its components constant we can use the results of @xcite which state that the metrics @xmath163 , for different @xmath164 , are equivalent up to isometries . thus , given @xmath165 , there exists for every @xmath164 sufficiently close to @xmath165 an isometry @xmath166 such that @xmath167 . since , we assume that @xmath168 is sufficiently smooth in a neighbourhood of @xmath165 , we can find a sufficently smooth @xmath169 in @xmath164 . the map @xmath166 provides us with the functions @xmath170 in the theorem by composition with the coordinate charts . hence , we have @xmath171 . moreover , @xmath172 is a locally homogeneous space @xcite . henceforth , we shall assume that we consider solutions in the set @xmath173 . consequently , from theorem 4.1 , there is no loss of generality in assuming that the metric ( [ kundt ] ) has @xmath174 . the remaining coordinate freedom preserving the kundt form is then : 1 . @xmath175 and @xmath176 . h , w_i = w_j(j^-1)^j _ i , _ij=_kl(j^-1)^k _ i(j^-1)^l _ @xmath177 h=h - h_,u , w_i = w_i - h_,i , _ ij=_ij . @xmath178 h=1g_,u^2(h+v ) , w_i=1g_,uw_i , _ij=_ij . the linearly independent components of the riemann tensor with boost weight 1 and 0 are : r_010i&=&-12 w_i , vv , + r_0101&= & -h_,vv+14(w_,v)(w^i , v ) , + r_01ij&= & w_[iw_j],vv+w_[i;j],v , + r_0i1j&= & 12 , + r_ijnm&=&_ijnm . hence , the spacetime is a @xmath0@xmath30 spacetime if there exists a frame @xmath179 , a constant @xmath67 , anti - symmetric matrix @xmath180 , and symmetric matrix @xmath181 such that w_i , vv & = & 0 , [ wcsi1 ] + h_,vv-14(w_i , v)(w^i , v ) & = & , + w_[i;j],v & = & a_ij , + w_(i;j),v-12 ( w_i , v)(w_j , v ) & = & s_ij , [ wcsi4]and the components @xmath182 are all constants ( i.e. , @xmath183 is curvature homogeneous ) . we note that the first equation implies that w_i(v , u , x^k)=vw_i^(1)(u , x^k)+w_i^(0)(u , x^k ) , while the second implies h(v , u , x^k)=+vh^(1)(u , x^k)+h^(0)(u , x^k ) . if @xmath184 is a ( locally ) homogeneous riemannian space , then there exists a frame where @xmath182 are all constants . in general , curvature homogeneous does not imply homogeneous ; however , since we are mostly interested in the @xmath0 case , in light of theorem [ thm : kundt ] , we will henceforth take @xmath183 as a ( locally ) homogeneous metric . for a @xmath0@xmath30 spacetime , we have r_010i;1&=&-12 , + r_0ijk;1&=&-12[w^n , v_nijk - w_i , va_jk+(s_i[j+a_i[j)w_k],v ] . for the spacetime to be @xmath0@xmath185 , it is sufficient to require that the above components are constants ; i.e. w_i , v-12(s_ji+a_ji)w^j , v & = & _ i , + ( s_ij+a_ij)_;k-(s_ik+a_ik)_;j&= & _ ijk , where the ricci identity has been used to rewrite the latter condition . various classes of @xmath0 spacetimes that arise as members of @xmath186 can now be found . the solutions come in classes according to the properties of @xmath187 . in the following , @xmath132 is a locally homogeneous riemannian space . assuming @xmath188 immediately makes the metric ( [ kundtmetric ] ) a @xmath0 space . a special subcase of this class which is worth mentioning is the brinkmann metrics for which @xmath15 and @xmath189 are all independent of @xmath152 ; thus @xmath190 . in this case , the expressions for the curvature tensors simplify drastically . in particular , if @xmath191}$ ] , the ricci tensor is r_=h-14 ^ 2 , r_=^f _ , r_=_. this case is @xmath0 if and only if the simpler metric , @xmath45 : = 2u(v+u+vw_i^(1)*m*^i)+_ij*m*^i*m*^j , is a homogeneous space . the curvature invariants of the kundt metrics in this class will have the same invariants as @xmath45 ; i.e. @xmath193 . the homogeneous space @xmath132 can be considered as a quotient space @xmath194 , where @xmath135 is the isometry group and @xmath15 is the isotropy group . the corresponding lie algebra decomposition is @xmath195 , and we let this correspond to the algebra of left - invariant vectors . we note that if there exists a subalgebra @xmath196 such that @xmath197=0 $ ] , where @xmath198 is the complement of @xmath199 in @xmath137 , then we can choose an @xmath200-invariant @xmath201 : @xmath202 by letting @xmath203 be its corresponding left - invariant one - form , the associated kundt spacetime will be a @xmath0 . note that some homogeneous spaces may be represented by inequivalent lie algebra decompositions . for example , hyperbolic space , @xmath204 , can both be considered as the quotient @xmath205 and as a solvable lie group @xmath135 @xcite . in the latter case the decomposition gives @xmath206 and @xmath207 ; hence , @xmath208=0 $ ] . @xmath211 : s^2&=&^2(x ) + & & + x^2+^2(x)s^2_s^n + s^2_h , [ eq : csi2]where & & w^(1)_=2(x ) , + & & _ i = _ i^(0)(u , x^j ) , = + v^(1)(u , x^j)+^(0)(u , x^j ) . the riemann tensor is of type : @xmath212 special cases : 1 . if @xmath213 then @xmath214 . 2 . if @xmath215 then @xmath216 and @xmath217 . @xmath219 : s^2_i&=&^2(x ) + & & + x^2+^2(x)s^2_s^n + s^2_h , [ eq : csi3]where & & w^(1)_=-2 ( x ) , + & & _ i = _ i^(0)(u , x^j ) , = -||+v^(1)(u , x^j)+^(0)(u , x^j ) . the riemann tensor is of type : @xmath212 special cases : 1 . if @xmath220 then @xmath214 . 2 . if @xmath215 then @xmath216 and @xmath221 . let us present another example : s^2_ii&=&^2(x ) + & & + x^2+^2(x)s^2_^n + s^2_h , [ eq : csi4]where & & w^(1)_=2 ( x ) , + & & _ i = _ i^(0)(u , x^j ) , = ||+v^(1)(u , x^j)+^(0)(u , x^j ) . the riemann tensor is of type : @xmath212 special cases : 1 . if @xmath222 then @xmath214 . 2 . if @xmath215 then @xmath216 . there are more @xmath0 spacetimes with @xmath223 . they are constructed similarly to the ones above . they are all related to the fact that de sitter space or anti - de sitter space can be written in many ways as fibered spaces over the spheres , @xmath224 , or the hyperbolic space , @xmath147 , respectively ( see appendix [ app:4d ] for the 4d case ) . this is a generalisation of the warped @xmath0 spacetimes considered earlier . the metric can be written s^2&=&e^-2(x^ ) + & & + _ ^^ , [ eq : csi]where _ a & = & v_a^(1)(u , x^a)+_a^(0)(u , x^a , x^ ) , + _ & = & _ ^(0)(u , x^a , x^ ) , + & = & ( x^ ) , _ , = , + & = & ( ^(1)_a)(^(1)a)+v^(1)(u , x^a , x^)+^(0)(u , x^a , x^ ) . furthermore , we require that @xmath226 is a riemannian homogeneous space , @xmath132 . the allowed forms of the function @xmath142 depend on the homogeneous space @xmath132 . topologically , this space is a fibered manifold with base manifold @xmath132 and we will require that the fiber is a @xmath1 space ; in particular , this means that @xmath227 satisfy the @xmath1 equations ( [ wcsi1])-([wcsi4 ] ) with @xmath228 . special cases : 1 . if @xmath229 then @xmath214 . 2 . if @xmath230 then @xmath216 . the connection @xmath53 for the above metric can be decomposed as = - , where @xmath231 is a @xmath232 tensor ( or operator acting in the obvious way ) with only non - positive boost - weight components , while @xmath233 has the following property : _ there exists a null frame such that the connection coefficients of @xmath233 has the following properties : _ 1 . positive boost weight connection coefficients are all zero . 2 . zero boost weight connection coefficients are either all constants or are connection coefficients of @xmath132 . henceforth we will assume that this null frame is chosen . moreover , @xmath231 and @xmath233 can be chosen such that & & _ * e*_x=_x*e*_=0 , + & & _ xg=0 , g(_xy , z)+g(y,_xz)=0 , x , y , ztm . note that the torsion of @xmath234 is @xmath235 . let @xmath59 and @xmath236 be the curvature tensors to the connections @xmath53 and @xmath233 respectively . then r=+ , where @xmath237 for @xmath238 while @xmath239 for @xmath240 . moreover , all boost weight 0 components are constants . we also have for all @xmath9 , @xmath241 has maximal boost order 0 where all boost weight 0 components are constants . we will prove the theorem by induction . let us in the following denote the projection of a tensor @xmath47 onto the vector space spanned by the components of boost weight @xmath52 by @xmath51 . the key observation is that the @xmath242 is of constant curvature over the @xmath1 fibers . as a result , @xmath243 . thus the theorem is true for @xmath244 . moreover , by applying the bianchi identity for @xmath245 , we can show that it is true for @xmath246 . hence , the theorem is true for @xmath247 . assume the theorem is true for @xmath9 . then consider @xmath248 . immediately , we have that it is true for @xmath249 . moreover , the curvature tensor is of constant curvature over the @xmath1 fibers , so @xmath250 ; hence , the theorem is true for @xmath251 for @xmath252 the critical part is the one that raises the boost weight by one . however , by using the identity t_c_1 c_i c_p=_i=1^pr^e _ c_i abt_c_1ec_p , [ commutator]recursively , and also the bianchi identity , we see that it is also true for @xmath252 . hence , by induction , the theorem follows . these results consequently imply that the fibered vsi spacetime given by eq.([eq : csi ] ) is a @xmath0 spacetime . a crucial part of the proof is that @xmath253 is of constant curvature over the @xmath1 fibers , which implies that @xmath250 . consider the metric ( [ eq : csi ] ) . if for all @xmath9 @xmath241 has maximal boost order 0 , where all boost weight 0 components are constants , then there exists a homoegenous space @xmath254 with the same curvature invariants as the metric ( [ eq : csi ] ) . the proof essentially follows along the same lines as above by noticing that the functions @xmath255 and @xmath256 do not contribute to the curvature invariants . the homogeneous space can be obtained by setting all of these to zero . we have obtained a number of general results . we have studied warped product @xmath0 spacetimes ( theorem 3.1 ) . we have presented the canonical form for the kundt @xmath0 metric ( theorem 4.1 ) , and studied higher dimensional kundt spacetimes in some detail . in sections 3 & 5 we have constructed higher dimensional examples of @xmath0 spacetimes that arise as warped products and that belong to the kundt class . in the appendices we shall present all of the 3-dimensional @xmath1 metrics , explicitly construct the metrics in the 4-dimensional @xmath0 spacetimes , and we shall establish the canonical higher dimensional @xmath1 metric . motivated by our results and examples we propose the following conjectures : a spacetime is @xmath0 iff there exists a null frame in which the riemann tensor and its derivatives can be brought into one of the following forms : either 1 . the riemann tensor and its derivatives are constant , in which case we have a locally homogeneous space , or 2 . the riemann tensor and its derivatives are of boost order zero with constant boost weight zero components at each order . this implies that riemann tensor is of type ii or less . whereas the proof of the ` if ' direction is trivial , the ` only if ' part is significantly more difficult and would require considering second order curvature invariants . we also point out that if we relax @xmath0 to @xmath0@xmath257 then in ( 1 ) we must now include all 1-curvature homogeneous spacetimes as well , since these naturally imply @xmath0@xmath257 and are not necessarily homogeneous spaces . assuming that the above conjecture is correct and that there exists such a preferred null frame , then if a spacetime is @xmath0 , then the spacetime is either locally homogeneous or belongs to the higher dimensional kundt @xmath0 class . finally , we have that if a spacetime is @xmath0 , then it can be constructed from locally homogeneous spaces and @xmath1 spacetimes . this construction can be done by means of fibering , warping and tensor sums . from the results above and these conjectures , it is plausible that for @xmath0 spacetimes that are not locally homogeneous , the weyl type is @xmath2 , @xmath3 , @xmath4 or @xmath5 , and that all boost weight zero terms are constant . we intend to study the general validity of these conjectures in future work . the relationship to curvature homogeneous spacetimes is addressed in @xcite . support for these conjectures in the 4d case is discussed in the next section . let us now consider the 4d @xmath0 spacetimes in more detail . we assume that a @xmath0 spacetimes is _ either _ of petrov ( p)-type i or plebanski - petrov ( pp)-type i , _ or _ of p - type ii and pp - type ii ( branch a or b of table 1 , respectively ) . in branch a , we explicitly assume that the spacetime is not of p - type ii and pp - type ii , otherwise we would be in branch b. it is plausible that in branch a the spacetimes must be of p - type i _ and _ pp - type i ( but we have not established this here ) . since the spacetime is not of p - type ii and pp - type ii , it is necessarily completely backsolvable ( cb ) @xcite . since all of the zeroth order scalar curvature invariants are constant , there exists a frame in which the components of the riemann tensor are all constant , and the spacetime is curvature homogeneous @xmath258 ( i.e. , @xmath259 ) . if the spacetime is also @xmath260 , then it is necessarily locally homogeneous @xmath15 @xcite . therefore , if it is not locally homogeneous , it can not be @xmath260 . finally , by considering differential scalar invariants , we can obtain information on the spin coefficents in the @xmath0 spacetime . although a comprehensive analysis of the differential invariants is necessary , a preliminary investigation indicates that @xmath261 and @xmath67 must all be constant and are likely zero . therefore , in branch a , there are severe constraints on those spacetimes that are not locally homogeneous , and it is plausible that there are no such spacetimes . in branch b of table 1 , we have that the spacetime is necessarily of p - type ii and pp - type ii . it then follows from theorem 7.1 below , that all boost weight zero terms are necessarily constant ( i.e. , the spacetime is @xmath262 ) . the spacetime is then either cb , in which case all of the results in branch a apply , or they are ncb and a number of further conditions apply ( these conditions are very severe @xcite ) . in either case , there are a number of different classes characterized by their p - type and pp - type ( in each case at least of type ii ) , and in each class there are a number of further restrictions . by investigating the differential scalar invariants we then find conditions on the spin coefficients . by considering each class separately , we can then establish that @xmath263 , and that the spacetime is necessarily @xmath264 . all of these spacetimes are constructed in appendix b. the results are summarized in table 1 . in branch a of table 1 , we have established that if the @xmath0 spacetime is not locally homogeneous , then necessarily it is of p - type i or pp - type i ( and does _ not _ belong to @xmath23 , @xmath18 , or @xmath7 ) , it belongs to @xmath265 , but not @xmath260 and there are a number of further constraints arising from the non - cb conditions and the differential constraints ( i.e. , constraints on the spin coefficients ) . this exceptional set is very sparse , and mostly likely empty ; but in order to demonstrate this the conjectures in 4d must be proven , utilising a number of 4d invariants . we have explicitly constructed the 4d @xmath18 , @xmath23 , and @xmath7 , and established the relationship between these @xmath0 subsets themselves and with @xmath266 , thereby lending support to the conjectures in the previous section . the metrics of these @xmath0 spacetimes are described in appendix b. from this calculation of all of the members of the set @xmath186 , we see that they all belong to @xmath7 . that is , we have established the uniqueness of @xmath7 within this particular context . in particular , we have found all @xmath0 in branch b of table 1 . finally , let us consider petrov ( p)-type ii and plebanski - petrov ( pp)-type ii lorentzian @xmath0 spacetimes . we shall show that if all zeroth order curvature invariants are constant then the boost weight zero curvature scalars , @xmath267 , @xmath268 and @xmath269 are constant . for these algebraically special spacetimes the converse follows straightforwardly . if all zeroth order curvature invariants are constant in petrov - type ii and plebanski - petrov - type ii lorentzian @xmath0 spacetimes , then the boost weight zero curvature scalars , @xmath267 , @xmath268 and @xmath269 are constant . we make use of the following curvature invariants @xmath270 it is worth noting that in general there is an additional independent degree 4 ricci invariant @xmath271 ; however , in pp - types ii or less there always exists a syzygy for @xmath271 in terms of @xmath272 and @xmath273 ( or else it vanishes ) . this is analogous to the well known syzygy for the weyl invariants , @xmath274 , signifying an algebraically special spacetime . assuming first that the weyl and ricci canonical frames are aligned , then the non - vanishing curvature scalars are @xmath267 , @xmath275 , @xmath268 , @xmath276 , @xmath277 and we have @xmath278 , @xmath279 , @xmath280 . clearly @xmath281 constant implies @xmath267 constant . if @xmath282 then @xmath269 and @xmath268 can be expressed in terms of @xmath272 and @xmath273 ; therefore , constant @xmath272 and @xmath273 gives constant @xmath268 and @xmath269 . if @xmath283 or @xmath284 then @xmath272 can be used to give the same result . in this aligned case we find that constant @xmath59 , @xmath281 , @xmath272 and @xmath273 implies that all curvature scalars are constant ( including the boost weight 0 scalars ) ; therefore , we have a curvature homogeneous spacetime . in the non - aligned case we refer to carminati and zakhary @xcite where it was shown that for this pp - type , complete backsolving ( cb ) of the carminati - zakhary ( cz ) invariants can be achieved and all curvature scalars can be expressed in terms of zeroth order invariants ; thus constant cz invariants implies constant curvature scalars . in the exceptional case in which @xmath285 in the ricci canonical frame , complete backsolvability is not possible since @xmath286 and @xmath287 can not be determined from any zeroth order curvature invariant @xcite . however , this exceptional case occurs when the weyl and ricci canonical frames are aligned , for p - type ii @xmath288 and @xmath275 and the remaining curvature scalars were shown above to be constant ( this is a particular instance of a not completely backsolvable ( ncb ) case becoming completely backsolvable ; see comment in ref . 3 of @xcite ) . we have consequently also shown that for p - type ii and pp - type ii spacetimes , if the zeroth order invariants are constant then all curvature scalars are constant . these results support the higher dimensional conjectures discussed in the previous section . we would like to thank r. milson , v. pravda and a. pravdova for helpful comments . this work was supported by nserc ( ac and np ) and the killam foundation ( sh ) . by analogy with the four dimensional @xmath1 spacetimes , we begin by considering a real null frame @xmath289 such that the only non - vanishing inner products are @xmath290 . in three dimensions , the riemann tensor is equivalent to the ricci tensor , therefore all zeroth order invariants vanish if @xmath291 we identify @xmath292 and @xmath293 and we use the n - dimensional version of the @xmath1 theorem to conclude that the analogues of @xmath294 , @xmath67 and @xmath295 must vanish . in three dimensions we find that that @xmath296 , @xmath297 , and @xmath298 , where @xmath299 are the ricci - rotation coefficients . therefore , we have that a three dimensional spacetime is @xmath1 if and only if @xmath300 and has segre type \{3 } , \{(21 ) } or \{(111 ) } with the possibility that the ( non-)vanishing of @xmath301 may lead to distinct subclasses . the method we use to explicitly obtain the three dimensional @xmath1 spacetimes involves writing the four dimensional @xmath1 spacetimes in terms of real coordinates @xmath302 then considering the resulting frame when either @xmath303 or @xmath304 we set @xmath305 and @xmath306 throughout , and in every case the restriction to @xmath304 yields nothing new , hence the relevant null frame , in coordinates @xmath307 , is @xmath308\partial_{v}+w_{1}\partial_{x } , & m=\partial_{x}\end{aligned}\ ] ] upon restriction to @xmath303 most of the four dimensional @xmath1 spacetimes become either flat or a special case of a null frame given in the following tables . in addition , whenever @xmath301 is nonzero then it is always equal to @xmath309 . cl @xmath301 & constraints on @xmath310 , @xmath311 and @xmath312 + + @xmath313 & a1 ) ( p - iii ; @xmath314 ; pp - n ) + & @xmath315 , @xmath316 , @xmath317 + + @xmath318 & b1 ) ( p - iii ; @xmath319 ; pp - n ) + & @xmath320 , @xmath316 , @xmath321 cl @xmath301 & constraints on @xmath310 , @xmath311 and @xmath312 + + @xmath313 & d1 ) ( p - n ; @xmath314 ; pp - o ) special case of a1 ) + & @xmath322 , @xmath323 , @xmath324 + + @xmath318 & f1 ) ( p - iii ; @xmath319 ; pp - o ) special case of b1 ) + & @xmath325 , @xmath326 , @xmath327 cl @xmath301 & constraints on @xmath310 , @xmath311 and @xmath312 + + @xmath313 & i1 ) ( p - iii ; @xmath314 ; vacuum ) + & @xmath328 , @xmath326 , @xmath329 + + @xmath318 & l1 ) ( p - n ; @xmath319 ; vacuum ) + & @xmath330 , @xmath331 , @xmath332 $ ] tables 24 provide a list of the three - dimensional @xmath1 spacetimes . here , we have chosen to characterise the 3d cases according to segre type . in this regard , the classification in @xcite of three dimensional spacetimes is worth noting . in 4d the segre types are related to the pp types @xcite . we have also indicated the corresponding class of four dimensional @xmath1 spacetimes ( in brackets ) from which the 3d solutions were obtained . it is also worth noting that some of the spacetimes given here may be equivalent and so these tables may reduce further . for example , in table 4 the vanishing of the ricci tensor implies that all spacetimes listed there are flat ; therefore , using the full 3 parameter lorentz group , @xmath333 , it should be possible to set @xmath334 ( i.e. , the flat metric should appear in this table ) . in tables 2 and 3 it is necessary to consider the frame freedom left after the ricci tensor has been brought to its canonical form . this can then be used to determine if @xmath301 can be made to vanish ; alternatively , it may be an invariant of this left over frame freedom and the metrics would be inequivalent . if @xmath301 can be set to zero , then it still remains to be shown if a coordinate transformation can be found relating the two metrics . we know that ( lorentzian ) homogeneous spaces and @xmath1 spacetimes are necessarily @xmath0 spacetimes . let us display all 4d spacetimes that are @xmath18 and @xmath23 . we shall find that all such spacetimes are necessarily in @xmath7 . the 4d spacetimes in @xmath173 are as follows . [ [ appb1 ] ] @xmath335 s^2&=&e^-2py+y^2 , [ eq:4dcsi]where _ x & = & v_x^(1)(u , x)+ _ x^(0)(u , x , y ) , + _ y & = & _ y^(0)(u , x , y ) , + & = & ( ^(1)_x)^2+v^(1)(u , x , y)+^(0)(u , x , y).and @xmath336 satisfy the 3d @xmath1 equations ( see appendix a ) . since @xmath337 depends on @xmath338 , in addition to @xmath164 and @xmath339 , this metric is fibered . due to the terms @xmath340 and @xmath341 in the metric , the spacetime is a warped product . by redefining @xmath342 in the metric above by omitting the term @xmath343 ( the @xmath344 term ) and in @xmath256 omitting the term @xmath345 ( the @xmath346 term ) , we can rewrite the metric as the tensor sum of the redefined metric ( [ eq:4dcsi ] ) and the metric s^2_+ & = & e^-2py+y^2 , [ eq:4dcsisum ] clearly the metric can be written in terms of the combined operations of fibering , warping and tensor summing , and hence belongs to @xmath7 . [ [ section ] ] @xmath347s^2&=&e^-2py+e^-2qyx^2 + y^2 , [ eq:4dcsi3]where _ x & = & _ x^(0)(u , x , y),_y = _ y^(0)(u , x , y ) , + & = & v^(1)(u , x , y)+^(0)(u , x , y).this metric is fibered and warped . [ [ section-1 ] ] @xmath348 : s^2&=&2u(v+hu+w_i*m*^i)+_ij*m*^i*m*^j , [ eq:4dcsi4]where _ i & = & vw^(1)_i+_i^(0)(u , x , y ) , + h & = & v^2h^(2)+v^(1)(u , x , y)+^(0)(u , x , y).and @xmath349 and @xmath350 are given in the table below . this metric is fibered and warped , and due to the @xmath344 and @xmath346 terms , can be written as a tensor sum . [ cols="^,^,^,^",options="header " , ] [ [ section-2 ] ] @xmath351 : metrics of the form ( [ eq : csi2 ] ) , ( [ eq : csi3 ] ) and ( [ eq : csi4 ] ) . these metrics are fibered , warped and tensor summed . [ [ appb5 ] ] @xmath352 : metrics of the form ( [ eq : csi2 ] ) , ( [ eq : csi3 ] ) and ( [ eq : csi4 ] ) . in addition , for @xmath353 s^2&=&^2(y)^2(x ) + & & + ^2(y)x^2+y , where _ i = _ i^(0)(u , x , y ) , = + v^(1)(u , x , y)+^(0)(u , x , y ) . and in the case of @xmath335 , similar versions of the metrics ( [ eq : csi3 ] ) and ( [ eq : csi4 ] ) are obtained . these metrics are fibered , warped and tensor summed . all of the metrics in sections [ appb1]-[appb5 ] can be written in terms of the combined operations of fibering , warping and tensor summing . therefore , all of the spacetimes constructed belong to @xmath7 . we consequently have the result that in 4d @xmath18 , @xmath23 and @xmath7 are equivalent . it is plausible that a similar result applies in higher dimensions . finally , let us consider the tensor sum in a little more detail . let us consider the @xmath7 metric ( [ eq : csi2 ] ) . redefining the metric by omitting the @xmath346 term , it can be rewritten as the tensor sum of the redefined metric and s^2_+&= & + x^2 . [ eq : csi2sum]the term @xmath354 $ ] can be rewritten as @xmath355 $ ] , which is a metric of constant ricci scalar curvature . it is plausible that we have constructed all of the 4d @xmath0 spacetimes here . to prove this , we must prove the conjectures outlined above . this is perhaps best done by considering higher order differential scalar invariants in 4d @xcite , which we hope to do in a future paper . we can now write down the metric for higher dimensional vsi spacetimes in a canonical form . an immediate corollary of theorem [ thm : kundt ] is : in @xcite it was proven that all @xmath1 metrics have @xmath70 and @xmath356 . the first condition implies that the metric is kundt , hence of the form ( [ kundt ] ) , while the second implies that the spatial metric , @xmath157 , is flat . using theorem [ thm : kundt ] the corollary now follows . j. bicak and j. podolsky , 1999 , j. math . phys . * 40 * , 4495 ; j. podolsky and m. ortaggio , 2003 , class . * 20 * , 1685 & j. b. griffiths , p. docherty and j. podolsky , 2004 , class . q. grav . * 21 * , 207 .

the masses of 40 neutron - rich nuclides from z = 51 to 64 were measured at an average precision of using the canadian penning trap mass spectrometer at argonne national laboratory . the measurements , of fission fragments from acf spontaneous fission source in a helium gas catcher , approach the predicted path of the astrophysical process . where overlap exists , this data set is largely consistent with previous measurements from penning traps , storage rings , and reaction energetics , but large systematic deviations are apparent in-endpoint measurements . differences in mass excess from the 2003 atomic mass evaluation of up to kev are seen , as well as systematic disagreement with various mass models .

1203.4470

neutron - rich isotopes have become increasingly more accessible to experiments requiring accelerated beams or trapped ions in recent years , and regions long unexplored are now being probed . penning trap mass spectrometers are taking advantage of stopped fission and reaction products from gas catchers @xcite and isol facilities @xcite to give direct measurements of these nuclidic masses to high precision . the masses of nuclides far from stability are of interest for a variety of fields , including astrophysics and nuclear structure @xcite . in particular , the rapid neutron - capture process ( @xmath2 process ) @xcite path is predicted to lie mostly in the region of unmeasured neutron - rich nuclides which is beginning to come into reach of precision study . determination of neutron separation energies ( @xmath5 ) from nuclear masses is critical for establishing the path of the @xmath2 process , which is thought to lie near the line @xmath6 mev , and is important for numerical simulations of @xmath2-process dynamics in different environments @xcite . much of the path is still inaccessible to experiment , but extending mass measurements closer to it provides information to better constrain mass models and extrapolations to the proposed @xmath2-process path and further still to the neutron dripline . additionally , the appearance of neutron sub - shells , regions of deformation @xcite , and shell quenching @xcite could be discovered via examination of @xmath7 trends . continuing a previous program of fission fragment mass measurements @xcite , the canadian penning trap ( cpt ) at argonne national laboratory has measured the masses of 40 neutron - rich nuclides from z=51 to 64 near the target precision of @xmath8 , or . this paper reports on the results and techniques of those measurements , and compares the results to previous measurements and the mass models on which @xmath2-process simulations depend . the cpt and associated equipment have been described elsewhere @xcite , but details emphasizing recent additions relevant to these measurements will be presented here . the system , illustrated in fig . [ fig : system ] , consists of a gas catcher for stopping reaction products , a radio frequency quadrupole ( rfq ) cooler and buncher , an isotope separator penning trap , and a linear paul trap for accumulation and staging of ions before injection in the precision penning trap . upgrades since the last cpt measurements in the neutron - rich region @xcite include a larger - volume gas catcher with a stronger fission source and a higher - resolution isotope separator in a superconducting magnet . the gas catcher @xcite was designed to stop and cool ions produced either from an internal spontaneous fission source or from reactions using beams from the argonne tandem - linac accelerator system ( atlas ) . this work was performed with fission fragments from a @xmath1cf source placed behind a gold degrader foil with thickness optimized for collection of fragments in the heavy peak . the @xmath9 m - length cylindrical gas catcher is filled with @xmath10@xmath11 torr of purified helium gas which stops the fission products through collision and ionization . a large fraction , roughly @xmath12@xmath10% , of these fission products stop as singly- or doubly - charged ions in the helium gas . an electrostatic gradient placed along the catcher axis pushes the ions toward the exit nozzle , while an rf ion guide @xcite keeps the ions from touching the walls and neutralizing . at the downstream end the ion guide forms an rf funnel leading to the exit nozzle , where gas flow pushes the ions into the next stage . ions are carried away from the gas catcher nozzle by an rfq ion guide @xcite , and are separated from the helium gas by the combined effects of electric fields and differential pumping . the ions are accumulated and bunched in a linear paul trap at the end of the rfq , where they are cooled further in @xmath13 torr room - temperature helium gas . for these measurements , after @xmath14 ms of accumulation the ion bunch is ejected from the trap into the beamline to the rest of the system . next , the ions are transferred to the isotope separator , where the ion bunches are purified . the isotope separator is a gas - filled cylindrical penning trap @xcite with a superconducting magnet . penning traps allow two orbital motions of the trapped ions : the modified cyclotron motion with frequency @xmath15 and the much slower magnetron motion with frequency @xmath16 , which are coupled by a cyclotron excitation @xmath17 . in the isotope separator , all ions are pushed outward by a mass - insensitive rf dipole field driven at @xmath16 while the ions of interest are re - centered in the trap by the application of a mass - sensitive rf quadrupole field at the @xmath18 frequency of those ions . the result is the accumulation of a purified sample of the desired ions with a mass resolving power of approximately @xmath19 over the mass range studied here . this is sufficient to remove ions of different mass number as well as any hydrocarbon contaminants that may be at the same mass number . in cases where remaining molecular contamination is suspected , a strong dipole excitation at the reduced cyclotron frequency is applied to break up the molecules via gas collisions during the cleaning process . after cleaning , ions are transferred to a second linear paul trap where multiple bunches from the isotope separator are accumulated and cooled before transfer to the precision penning trap . the cpt mass spectrometer consists of a hyperbolic penning trap @xcite in a highly stable magnetic field and ultra - high vacuum , where masses are measured using the time - of - flight technique @xcite . the main electrodes approximate hyperboloids of revolution , with apertures in the endcaps for ion entry and ejection and a splitting of the ring electrode into quadrants so that quadrupole excitations may be applied . additional electrodes are placed between the endcaps and the ring to correct for the finite extent of the trap , and correction electrodes placed outside the endcaps prevent field penetration through the endcap apertures @xcite . after an ion bunch is captured in this trap , the highest energy ions are evaporated away by briefly lowering the voltage of a correction tube and allowing them to escape . this is done both to keep only those ions which are in the region of the trap with the most homogeneous magnetic field and to decrease the ion time - of - flight spread . any contaminants suspected to survive the isotope separator are then removed by a @xmath14- to @xmath20-ms dipole excitation at the @xmath15 of those ions . next , a dipole excitation at @xmath16 is applied for @xmath21 ms to position the ions in an orbit with the desired radius . the last step inside the penning trap is the application of a quadrupole excitation at a candidate cyclotron frequency for @xmath22 to @xmath23 ms . if the frequency applied matches the actual cyclotron frequency , the slow magnetron motion previously induced is converted to the much faster modified cyclotron motion at the same orbital radius @xcite . a typical @xmath24 ratio is @xmath25 , thus the orbital kinetic energy increases by a factor of @xmath26 . after this final excitation the ion bunch is ejected from the trap and drifts down a long beam line to either an mcp or channeltron time - of - flight detector outside the magnet . as the ions travel through the gradient of the main magnetic field , the orbital motion is converted adiabatically to linear motion , accelerating the ions down the line . thus an application of a quadrupole field closer to the ion s true @xmath18 will result in a lower time of flight than a frequency farther away . by scanning frequencies over successive bunches of ions , the minimum in time of flight can be found , and a measurement of the the ion s @xmath18 made . because the frequency is applied with a square amplitude envelope , the time - of - flight spectrum reflects that envelope s fourier transform : a sinc function . the cyclotron frequency of a calibrant ion of well - known mass is measured in the same manner , and the ratio of the frequencies is taken to cancel out the magnetic field term . the mass of the neutral atom is then given by the calibrant mass and frequency ratio , with additional terms to compensate for the charge states and the masses of electrons not present given the charge states : @xmath27 where the subscript ` cal ' refers to the calibrant , @xmath28 is the integer charge state , and @xmath29 is the mass of the electron . alternatively , if multiple calibrant species are used , those frequency measurements can be combined as a single magnetic field strength measurement via @xmath30 , which can then be used to determine the unknown masses from their cyclotron frequency measurements . in either case , the calibrations need to be taken only as frequently as is required to monitor magnetic field drift , rather than after every measurement of an unknown ion . atomic binding energies are small enough to be neglected for these measurements . example time - of - flight scans are shown in fig . [ fig : resonances ] . following fission - fragment measurements made with a smaller gas catcher and source and a previous isotope separator @xcite , a new series of measurements in the @xmath1cf fission heavy peak began in april 2008 . the nuclides studied , identified in fig . [ fig : chart ] , can be divided into two components : the heaviest nuclides , from pr to gd , were measured in the @xmath31 charge state over @xmath32weeks from april to july 2008 ; the lighter nuclides in the heavy peak , from sb to cs , were measured in the @xmath33 charge state over 4 weeks in february and march 2009 . high charge states are advantageous for mass measurements because their higher cyclotron frequencies lead to lower uncertainties in mass for the same uncertainty in frequency . however , due to interactions with the helium gas used to stop and cool the ions in preparation for the precision penning trap , they are limited to only @xmath33 or @xmath31 in practice . the division of charge states in these measurements is due to the high ionization potential at the xe electron shell closure , which allows the fission fragments with @xmath34 to survive as @xmath31 , but not those with @xmath35 . excitation times used for candidate cyclotron frequencies range from 200 to 2000 ms , depending on the lifetime of the nuclide being studied and the conditions inside the trap . the total duration of each measurement typically 1 to 30 h was dictated by the target precision of one part in @xmath36 , the yield of ions out of the gas catcher , and the ultimate purity of the sample in the penning trap . to calibrate the @xmath31 measurements , which range in mass - to - charge ratio from @xmath37 to @xmath38 , the cyclotron frequencies of @xmath39c@xmath40h@xmath41 , @xmath42kr@xmath43 , and @xmath44kr@xmath43 were measured approximately once per measurement week , and all of the calibration measurements were combined to give a single value for the magnetic field . for the @xmath33 measurements , @xmath45xe@xmath43 was measured at least once per measurement week . in all weeks but one , at least one calibration resonance was taken with identical excitation time as used in that period to determine the fitting - function parameters to be used . for @xmath45i , no calibration was taken with the same excitation time as the measurement because of an unplanned interruption in the experiment . to calibrate this measurement , the results of a calibration were used , and the width of the resonance was scaled from that calibration fit . table [ tbl : calibrants ] lists the mass values used for each calibrant , and tables [ tbl:1+_cal ] and [ tbl:2+_cal ] list the cyclotron frequency ratios of each measured ion to each calibration ion . .mass values used for magnetic field calibrations . for the molecule , the mass listed is that of the ion plus the mass of an electron . the noble gas masses are those of neutral atoms . relative weights are given for the influence of each calibrant s mass on the corresponding set of measurements : @xmath45xe@xmath43 for the @xmath33 and the others for @xmath31 . [ cols="^,>,<,^,^ " , ] after the first week of @xmath33 measurements , an electrical discharge occurred in or near the precision penning trap . a lasting result of this discharge was an additional trap imperfection , possibly due to some surface charge deposited on a trap electrode . resonances taken under these conditions show an asymmetry , the effect of which had to be mitigated in the analysis as described below . a study undertaken with various stable xe isotopes @xcite found a systematic effect of only @xmath46 under these conditions , which is well below the statistical uncertainties in these measurements . most time - of - flight spectra were fit with a modified sinc function , which includes a parameter accounting for the possible over- or under - conversion from the @xmath16 to @xmath15 motions @xcite . for the data taken in the week following the discharge the time - of - flight spectrum showed an asymmetry in the side bands , pulling the modified sinc fit to the low - frequency side . a gaussian function was used to fit that week s data to minimize this problem , and the uncertainties were appropriately inflated due to the increased @xmath47 of these fits . in all cases cuts were placed on the number of detected ions from a single ion bunch to minimize systematic effects discussed below . there are several possible sources of systematic error , the largest of which are listed here . as the magnetic field drifts , @xmath18 will change , but the superconducting magnet is extremely stable ( fig . [ fig : calibrants_time ] ) , with the field drift measured to be @xmath48 ppb / day over a two - month period . effects of trap misalignment and electric field imperfections are suppressed to high order @xcite . in the weeks following the aforementioned discharge , the measured @xmath18 value for @xmath45xe@xmath43 drifted up @xmath49 ppb / day as the trap recovered . this small potential source of error during this time was mitigated by using a calibration taken within one week of each measurement to calculate the mass . the ion cloud s charge alters the electric field of the trap , but the effect on the @xmath18 value has been measured to be @xmath50 ppb per detected ion under normal conditions @xcite , and @xmath51 ppb per detected ion with the post - discharge effect during these measurements . this effect will be suppressed if the trap population during measurement and calibration are kept low , as was the case here with the average number of detected ions per ejection ranging from 0.03 to 3.1 for the ions of interest , and below 7 for the calibration ions . contaminant ions of similar mass can have a separate effect as the ion motions interfere with each other . based on a 100 ppb limit at 15 ions @xcite , we estimate the effect to have an upper limit of @xmath52 ppb for the 3-ion maximum rate here . the reference frequency used by the @xmath18 signal generator was compared to a rb frequency standard and measured to be stable to @xmath53 ppb over a two - week period . mass - dependent effects in the precision penning trap were investigated as part of a very high precision study of xe isotopes @xcite . a frequency - dependent attenuation of the applied @xmath18 amplitude was found at @xmath54/khz over the frequency range corresponding to @xmath55 from 113 to 165 . this , combined with a measured @xmath18 centroid dependence on amplitude due to the dodecapole moment of the trap , gives a variation of @xmath56 , which is again mitigated by using calibrants of similar mass . because these potential sources of error are much smaller than our statistical uncertanties of at least 32 ppb , or more typically 100 ppb , no systematic uncertainty was added to the quoted results . l l c & & excitation + & & time ( ms ) + @xmath57sb@xmath43&1.022510200(79)&1000 + @xmath58sb@xmath43&1.01483272(27)&500 + @xmath58te@xmath43&1.014904071(49)&1000 + @xmath59te@xmath43&1.007342824(75)&1000 + @xmath45te@xmath43&0.999905180(37)&1000 + @xmath60te@xmath43&0.99256230(13)&1000 + @xmath59i@xmath43&1.007391356(62)&1000 + @xmath45i@xmath43&0.999944000(36)&2000 + @xmath60i@xmath43&0.992617351(65)&1000 + @xmath61i@xmath43&0.985386600(46)&1000 + @xmath62i@xmath43&0.97826690(22)&1000 + @xmath60xe@xmath43&0.992664183(83)&1000 + @xmath61xe@xmath43&0.985447843(42)&1000 + @xmath62xe@xmath43&0.978321228(81)&1000 + @xmath63xe@xmath43&0.971309239(75)&1000 + @xmath64xe@xmath43&0.964381577(69)&1000 + @xmath64cs@xmath43&0.96442761(15)&1000 + @xmath65cs@xmath43&0.957603370(75)&1000 + l l l l c & & excitation + & & & & time ( ms ) + @xmath66pr@xmath67&0.99430250(26)&1.04511012(27)&1.12350053(29)&200 + @xmath68pr@xmath67&0.98142559(20)&1.03157521(22)&1.10895041(23)&500 + @xmath66nd@xmath67&0.994342931(34)&1.045152616(32)&1.123546210(39)&500 + @xmath68nd@xmath67&0.98147230(11)&1.03162431(11)&1.10900319(12)&500 + @xmath69nd@xmath67&0.96892546(29)&1.01843635(30)&1.09482604(32)&200 + @xmath66pm@xmath67&0.99436601(15)&1.04517687(16)&1.12357228(17)&500 + @xmath68pm@xmath67&0.981503965(56)&1.031657596(57)&1.109038971(63)&500 + @xmath70pm@xmath67&0.975190689(43)&1.025021719(42)&1.101905358(48)&1000 + @xmath69pm@xmath67&0.968964141(81)&1.018477002(84)&1.094869742(92)&500 + @xmath71pm@xmath67&0.96280782(15)&1.01200610(16)&1.08791348(17)&500 + @xmath72pm@xmath67&0.95673359(11)&1.00562148(12)&1.08104997(13)&500 + @xmath68sm@xmath67&0.98152610(15)&1.03168086(16)&1.10906398(17)&500 + @xmath69sm@xmath67&0.968993178(51)&1.018507523(52)&1.094902552(58)&500 + @xmath71sm@xmath67&0.962848141(58)&1.012048482(58)&1.087959039(65)&1000 + @xmath72sm@xmath67&0.956770122(65)&1.005659884(67)&1.081091252(74)&200 & 500 + @xmath73sm@xmath67&0.950775181(67)&0.999358608(69)&1.074317338(76)&1000 + @xmath74sm@xmath67&0.944844874(74)&0.993125271(77)&1.067616458(84)&500 + @xmath71eu@xmath67&0.96286130(15)&1.01206231(16)&1.08797391(17)&500 + @xmath72eu@xmath67&0.956794898(63)&1.005685926(64)&1.081119248(71)&500 + @xmath73eu@xmath67&0.95079589(10)&0.99938038(11)&1.07434074(12)&500 + @xmath74eu@xmath67&0.94487714(11)&0.99315919(12)&1.06765292(13)&500 + @xmath75gd@xmath67&0.933275822(91)&0.980965055(94)&1.05454414(10)&500 + l@l@l@l@r@lll & & + & & & & & & + @xmath57sb&132.915277(10)&-78918.7(9.5)&-78943(25)&24&(27)&&-78986(120 ) + @xmath58sb@xmath76&133.920812(35)&-73763(33)&-74170(40)&407&(52 ) & & + @xmath58te&133.9113976(65)&-82532.6(6.0)&-82559(11)&26&(13)&&-82758(121 ) + @xmath59te&134.916550(10)&-77733.2(9.3)&-77830(90)&97&(90)&&-77725(123 ) + @xmath45te&135.9201024(50)&-74424.2(4.6)&-74430(50)&6&(50 ) & & + @xmath60te&136.925622(18)&-69282(17)&-69560(120)&280&(120)&&-69290(120 ) + @xmath59i&134.9100503(82)&-83787.6(7.7)&-83790(7)&2&(10 ) & & + @xmath45i@xmath76&135.9148257(49)&-79339.3(4.5)&-79500(50)&161&(50 ) & & + @xmath60i&136.9180282(90)&-76356.2(8.3)&-76503(28)&147&(29)&&-76518(121 ) + @xmath61i&137.9227265(64)&-71979.8(6.0)&-72330(80)&350&(80 ) & & + @xmath62i&138.926506(31)&-68460(29)&-68840(30)&380&(42)&&-68527(121 ) + @xmath60xe&136.911569(11)&-82373(11)&-82379(7)&6&(13)&-82382.2(1.8 ) & + @xmath61xe&137.9141550(59)&-79964.1(5.5)&-80150(40)&186&(40)&-79975.1(3.3 ) & + @xmath62xe&138.918791(11)&-75645(11)&-75644(21)&-1&(24)&-75644.6(2.1 ) & + @xmath63xe&139.921658(11)&-72976(10)&-72990(60)&14&(61)&-72986.5(2.3)&-72870(121 ) + @xmath64xe&140.926785(10)&-68199.5(9.4)&-68330(90)&130&(90)&-68197.3(2.9)&-68521(127 ) + @xmath64cs&140.920058(21)&-74466(19)&-74477(11)&11&(22)&-74475(15)@xmath77 & + @xmath65cs&141.924303(11)&-70511(10)&-70515(11)&4&(15)&-70521(15)@xmath77 & + @xmath66pr@xmath78&152.933895(15)&-61576(14)&-61630(100)&54&(100 ) & & + @xmath68pr&154.940508(32)&-55416(30)&-55780(300)@xmath79&360&(300)@xmath79 & & + @xmath66nd&152.9277156(47)&-67332.5(4.4)&-67349(27)&16&(27 ) & & + @xmath68nd&154.933134(17)&-62285(16)&-62470(150)@xmath79&190&(150)@xmath79 & & + @xmath69nd&156.939383(46)&-56464(43)&-56790(200)@xmath79&330&(200)@xmath79 & & + @xmath66pm&152.924167(24)&-70638(22)&-70685(11)&47&(25 ) & & + @xmath68pm&154.9281350(85)&-66941.8(7.9)&-66970(30)&28&(31 ) & & + @xmath70pm&155.9311155(64)&-64165.5(5.9)&-64220(30)&54&(31 ) & & + @xmath69pm&156.933119(13)&-62299(12)&-62370(110)&70&(110 ) & & + @xmath71pm&157.936563(24)&-59091(23)&-59090(130)&0&(130 ) & & + @xmath72pm&158.939284(19)&-56557(17)&-56850(200)@xmath79&290&(200)@xmath79 & & + @xmath68sm&154.924642(24)&-70196(22)&-70197.2(2.6)&1&(22 ) & & + @xmath69sm&156.9284166(80)&-66679.5(7.4)&-66730(50)&50&(51 ) & & + @xmath71sm&157.9299497(91)&-65251.5(8.5)&-65210(80)&-42&(80 ) & & + @xmath72sm&158.933215(10)&-62209.6(9.7)&-62210(100)&0&(100 ) & & + @xmath73sm&159.935333(11)&-60237(10)&-60420(200)@xmath79&180&(200)@xmath79 & & + @xmath74sm&160.939158(12)&-56674(12)&-56980(300)@xmath79&310&(300)@xmath79 & & + @xmath71eu&157.927791(25)&-67262(23)&-67210(80)&-52&(83 ) & & + @xmath72eu&158.929100(10)&-66043.2(9.5)&-66053(7)&10&(12 ) & & + @xmath73eu&159.931849(17)&-63482(16)&-63370(200)@xmath79&-110&(200)@xmath79 & & + @xmath74eu&160.933662(19)&-61793(18)&-61780(300)@xmath79&0&(300)@xmath79 & & + @xmath75gd&162.934175(16)&-61316(15)&-61490(300)@xmath79&170&(300)@xmath79 & & + + + + & + + of the 40 nuclides measured , 8 had been measured previously by penning traps , 20 by @xmath3-endpoint only , and 3 had no previous mass measurement of any kind . measurement uncertainties in the cpt results range from 5 to 46 @xmath80u , with most below or near the target @xmath81 fractional uncertainty . isomers with lifetimes long enough to be captured in the trap are not expected for any of the measured nuclides except @xmath58sb @xcite and @xmath45i @xcite , for which the identities of the measured states are unclear . table [ tbl : masses ] shows the new cpt measurements and compares them to the 2003 atomic mass evaluation ( ame03 ) @xcite as well as some more recent measurements @xcite . the overall trend of these measurements versus the ame03 is of increasing mass as distance from stability increases , as seen in fig . [ fig : ame - me ] . this effect was also seen in the cpt s previous measurements of ba , la , ce , and pr fission fragments @xcite , proton - rich nuclides of nb to rh @xcite , as well as measurements by other penning traps @xcite . because the source data for the ame and its extrapolations are mostly from @xmath3-endpoint measurements , the observed deviations suggest an unaddressed systematic problem with that technique . the frequency of such disagreement suggests that @xmath3-endpoint results as a whole should not be trusted as a reliable source of data for extrapolations . the trends evident in fig . [ fig : ame - me ] demand separate comparisons between the cpt and literature mass values for different measurement techniques . four categories have been selected for individual discussion below , grouped so that systematic trends in differences may be identified and each method evaluated for accuracy . following these is a brief comparison of each new cpt measurement with the literature values for that nuclide . some of the masses measured here have been previously measured in the isoltrap or cpt penning traps . these are ideal checks on the accuracy of the experiment as a whole . the cpt had previously measured and published the mass of @xmath66pr @xcite , and the present measurement is in agreement . isoltrap has measured isotopic chains of cs @xcite and xe @xcite from proton - induced fission at isolde to neutron numbers beyond the current reach of the cpt for these elements . [ fig : isol - cpt ] shows a comparison between the seven nuclides measured by both traps , and there is no evidence of any systematic or isolated differences , with @xmath82 . on average , the reported isoltrap masses are @xmath83 lower than the cpt masses . the frs - esr @xcite cooler - storage ring at gsi uses isochronous mass spectrometry to measure masses of fission products in a radioactive beam via time of flight . eight nuclides of sb , te , i , and xe have been measured at both the frs - esr and the cpt , and are compared in fig . [ fig : esr - cpt ] . the two methods are in rough agreement but there is some scatter with an rms difference of @xmath84 kev , despite the uncertainty typical of the frs - esr results . comparing the two data sets yields @xmath85 , giving a statistical @xmath86-value of @xmath87 . if this large scatter is a real effect , then given the strong consistency of the penning traps discussed above it may be due to a flaw in either the frs - esr experiment or calibrant mass values . this last possibility is a significant one , given that some of those calibrant masses are largely determined via @xmath3-endpoint . unfortunately , not enough information is given in the frs - esr publication @xcite to accurately recalculate new mass values with updated calibrant masses and compare to the cpt results . five of the masses presented here had previously been established from nucleon - transfer or neutron - capture experiments . the @xmath88 reaction offers an opportunity for exquisitely precise neutron - separation energy ( @xmath5 ) measurements due to the monoenergetic gamma rays and the state of gamma - ray absorption detector technology . @xmath5(@xmath60xe ) has been measured to a precision of @xmath89 ev by this method @xcite , in agreement with the cpt result . @xmath5(@xmath68sm ) has twice been measured to sub - kev precision @xcite and these are consistent , both with each other and with the cpt result . transfer reactions offer similar benefits , and four such measurements have been made on nuclides presented here . the mass of @xmath59i has been measured via @xmath45xe@xmath90he@xmath91 @xcite , @xmath66pm via @xmath92sm@xmath90he@xmath91 @xcite and @xmath92sm@xmath93 @xcite , and @xmath72eu via @xmath73gd@xmath93 @xcite , to precisions of @xmath21 , @xmath94 , @xmath95 , and @xmath96 kev/@xmath97 , respectively . the differences between these and the cpt measurements , using ame03 mass values for the sm and gd parent nuclides , are @xmath98 , @xmath99 , @xmath100 , and @xmath101 , respectively , confirming the accuracy and reliability of transfer reaction measurements . @xmath3-endpoint measurements dominate the ame03 on the neutron - rich side of stability because of the applicability of that kind of measurement with a small number of nuclei . for 15 of the nuclides presented here where isomeric states are not suspected , the @xmath3-decay @xmath102-values can be calculated entirely from these cpt measurements , and a total of 22 @xmath3-endpoint measurements exist of these nuclides . a comparison of these data sets shows a systematic problem with the @xmath3-endpoint method , as illustrated in fig . [ fig : q - betas ] . of the 22 measurements , 10 are within @xmath103 and another 6 within @xmath104 , but a long tail exists up to @xmath105 . comparison of the data sets gives @xmath106 , which has a statistical @xmath86-value of @xmath107 . of the 22 @xmath3-endpoint measurements , 17 had values of @xmath108 smaller than the cpt results , which gives support to the notion that feeding to higher - lying states is erroneously pushing these measurements to lower masses . two recent sets of measurements demonstrate the importance of considering systematic uncertainties in @xmath3-endpoint experiments . et al . _ @xcite give new @xmath108 measurements for some of the eu and gd nuclides presented here and a new analysis of that group s earlier pm and sm @xcite measurements , and fogelberg _ et al . _ @xcite measured @xmath108 for neutron - rich te and i isotopes . the cpt results have good statistical agreement with the hayashi _ results , but less so with fogelberg _ _ , where deviations up to @xmath105 are seen . however , fogelberg _ et al . _ report only a statistical uncertainty , and if the systematic uncertainty suggested by shibata _ et al_. @xcite and used by hayashi _ et al . _ for the root - plot analysis method is applied , the disagreements decrease to less than @xmath104 . what follows is a comparison to previous measurements of each nuclide presented here , ordered by element : * @xmath57sb : * the ame03 cites a single experiment for the mass of @xmath57sb , a @xmath3-endpoint measurement @xcite . the value of @xmath108 in that paper is @xmath109 mev , leading to an evaluated mass excess of @xmath110 mev . this is consistent with the cpt value of @xmath111 mev . a more recent experiment by the frs - esr facility measured @xmath57sb via time of flight , and reported a mass excess of @xmath112 mev @xcite , which is also consistent with the cpt value . * @xmath58sb : * there is a metastable state of @xmath58sb with a lifetime and excitation energy estimated at @xmath113 kev @xcite , and with a measurement from @xmath114 coincidence at 279 kev @xcite . only one resonance was observed with the cpt in the expected mass region , so the state of the measured ions is unclear . if the two states are within 400 kev of each other they would not be separately resolved in the cpt given the excitation time of 500 ms , so the observed resonance may be some weighted average of the two states . the ame03 cites two @xmath3-endpoint measurements for the mass determination of @xmath58sb : @xmath115 mev @xcite and @xmath116 mev @xcite , both of which are significantly lower ( 6.9 and 2.8 @xmath117 ) than the cpt value of @xmath118 mev . if the measured state is assumed to be the metastable state , then a combination of the cpt measurement and the @xmath3-endpoint measurements above gives an excitation energy of @xmath119 kev . we do not offer an assignment of the measured state . * @xmath58te : * a single @xmath3-endpoint measurement is cited by the ame03 with @xmath120 mev @xcite , leading to a mass excess of @xmath121 mev . this is marginally consistent with our result of @xmath122 mev . the frs - esr measurement gives @xmath123 mev @xcite , a @xmath124 difference from the cpt . * @xmath59te : * two @xmath3-endpoint measurements were used in the ame03 to calculate the mass excess of @xmath59te : @xmath125 mev @xcite , and @xmath126 mev @xcite . these are combined in the ame03 to give @xmath127 mev , consistent with the value from cpt measurements of @xmath128 mev . the recent frs - esr result gives a mass excess of @xmath129 mev @xcite , which is consistent with the cpt value of @xmath130 mev . a @xmath3-endpoint result by fogelberg _ et al . _ @xcite more recent than the ame03 gives @xmath131 mev , a 9.4 @xmath117 difference from the cpt . it is important to note that there was no systematic uncertainty assigned to that @xmath108 measurement , only a statistical one . other authors @xcite suggest a systematic uncertainty for the root - plot method employed in that study . increasing the fogelberg _ _ uncertainty to 100 kev reduces the disagreement to @xmath132 . * @xmath45te : * three measurements are used by the ame03 to determine this mass . the determination is dominated by a beta - delayed neutron - emission @xmath102-value measurement of @xmath133 mev @xcite , which agrees with the cpt value of @xmath134 mev . the other citations are of beta - decay @xmath102-value measurements of @xmath135 mev @xcite and @xmath136 mev @xcite . assuming the cpt @xmath45i measurement is of the ground state ( see discussion below ) both of these are somewhat higher than the cpt result of @xmath137 mev . since the ame03 publication , a new @xmath3-endpoint measurement has been made with @xmath138 mev @xcite , a disagreement of 171 kev and @xmath139 with the cpt result . adding a systematic uncertainty to the @xmath3-endpoint method as discussed above decreases the inconsistency to @xmath140 . * @xmath60te : * two @xmath3-endpoint experiments are cited in the ame03 for @xmath60te : @xmath141 mev @xcite and @xmath142 mev @xcite . these are combined in the ame03 to @xmath143 mev , marginally consistent with the cpt result of @xmath144 . the ame03 mass excess value of @xmath145 mev is 278 kev lower than our value of @xmath146 mev . over half of this difference , 147 kev , comes from the difference between the mass of @xmath60i used by the ame03 and that determined in our experiment , discussed below . the recent frs - esr measurement agrees with the cpt , with mass excess @xmath147 mev @xcite . * @xmath59i : * in the ame03 , two different methods primarily constrain the @xmath59i mass : a @xmath3-endpoint measurement of @xmath148 mev @xcite and a @xmath102(@xmath45xe(@xmath149,@xmath150he)@xmath59i ) measurement of @xmath151 mev @xcite . the cpt @xmath59i measurement gives @xmath152xe@xmath153 mev , in excellent agreement with the ame03 input value . there is also a slight contribution to the @xmath59i mass in the ame03 from a @xmath45te @xmath154 measurement @xcite of @xmath155 mev , consistent with cpt s @xmath134 mev . the cpt results can not be directly compared to the @xmath108 measurement . * @xmath45i : * there is a metastable state of @xmath45i @xcite which , with the ground state , is expected to be produced in fission . only one resonance was clearly measured during this experiment , and its identity is uncertain . if the states lie within 100 kev of each other they would not have been resolved given the excitation time of 2000 ms . the determination of the number of states trapped by the cpt was made difficult by the possible presence of @xmath59xe@xmath156h@xmath43 less than 1 hz ( 200 kev ) away from the observed @xmath45i or @xmath45i@xmath157 resonance . because the mass of @xmath59xe@xmath156h@xmath43 is well known we can be certain that the resonance measured is not that molecule . the mass excess of the observed @xmath45i state is found to be @xmath158 mev , and @xmath108 is @xmath159 mev . the ame03 cites three papers for a total of four measurements of @xmath45i and its metastable state , each a @xmath3-endpoint measurement . the most recent of those @xcite gives @xmath160 mev and @xmath161 mev for the two states . an older paper @xcite was disregarded for its measurement of the ground - state @xmath108 value , but its measurement of @xmath162 mev assigned to the isomer state was included in the evaluation . a third paper @xcite gives only the ground state @xmath108 value at @xmath163 mev . the adopted ame03 @xmath108 values are @xmath164 mev for the ground state and @xmath165 mev for the metastable state . these suggest that it is the ground state which was measured here , however , a paper published since the ame03 claims @xmath3-endpoint measurements of both the ground and metastable states , with @xmath166 and @xmath167 mev @xcite , respectively . this is the same paper discussed above which did not include systematic uncertainties . given the conflicting measurements it is difficult to assign a state to the measured ion . * @xmath60i : * the ame03 cites a @xmath3-delayed neutron study for this mass determination , with @xmath168 mev @xcite . the cpt measurement gives @xmath169 mev , which disagrees by @xmath170 . disregarded by the ame03 is a @xmath3-endpoint measurement of @xmath171 mev @xcite , which is inconsistent with the cpt result of @xmath172 mev . the recent frs - esr measurement gave a mass excess of @xmath173 mev @xcite , which differs from the cpt result of @xmath174 mev by @xmath175 . * @xmath61i : * only one experiment is cited in the literature for @xmath61i , a @xmath3-endpoint measurement of @xmath176 mev @xcite , a @xmath177 difference from our value of @xmath178 mev . the ame03 mass excess value of @xmath179 mev disagrees with the cpt value of @xmath180 mev , a larger , difference . of this difference , 186 kev is accounted for by our disagreement on the mass of @xmath61xe discussed below . * @xmath62i : * a single previous measurement was available for the ame03 , a @xmath3-endpoint measurement that gives @xmath181 mev @xcite , a , difference from the cpt result of @xmath182 mev . curiously , both the @xmath62xe and @xmath62i @xmath108 measurements that the ame03 relies upon are in this same paper but have wildly different agreement with our results . the recent frs - esr mass excess result of @xmath183 mev @xcite agrees with the cpt result of @xmath184 mev . * @xmath60xe : * by virtue of the adjacency of @xmath60xe to the stable @xmath45xe , the mass of @xmath60xe can be derived quite precisely from ( @xmath185 ) measurements . the ame03 cites such a measurement in a draft iaea technical document . that document has since been published , with @xmath186 kev @xcite , consistent with the cpt result of @xmath187 kev . another penning trap , isoltrap , recently measured all xenon isotopes from @xmath188 to @xmath189 with similar precision as these measurements @xcite . the reported isoltrap mass excess for @xmath60xe of @xmath190 mev is in agreement with the cpt value of @xmath191 mev . * @xmath61xe : * two @xmath3-endpoint measurements are used for the mass determination by the ame03 . those results are @xmath192 mev @xcite and @xmath193 mev @xcite . these are combined in the ame03 to give the adopted value of @xmath194 mev . as the mass of @xmath61cs was not measured in our experiment , a calculation of @xmath108 entirely from cpt data is not possible . it was previously measured by isoltrap , however , with an adjusted mass excess of @xmath195 mev @xcite ( see @xmath64cs , below , for adjustment ) . combining these two penning - trap measurements gives @xmath196 mev : a , disagreement . isoltrap s recent measurement of @xmath61xe with mass excess @xmath197 mev @xcite differs from the cpt value of @xmath198 mev by @xmath140 . * @xmath62xe : * two experiments have measured @xmath108 for this nuclide and are used by the ame03 : @xmath199 @xcite and @xmath200 mev @xcite , averaged in the ame03 to @xmath201 mev . using the adjusted mass excess value of @xmath62cs from isoltrap of @xmath202 mev @xcite ( see @xmath64cs , below , for adjustment ) , our @xmath203 mev is consistent . the recent isoltrap measurement of the @xmath62xe mass excess is @xmath204 mev @xcite , which is also consistent with the cpt value of @xmath205 mev . * @xmath63xe : * there is only one existing @xmath108 measurement for @xmath63xe , of @xmath206 mev @xcite . using the adjusted isoltrap @xmath63cs mass excess result of @xmath207 mev @xcite ( see @xmath64cs , below , for adjustment ) , the cpt @xmath208 mev agrees . the cpt mass excess of @xmath209 mev is consistent with both the isoltrap measurement of @xmath210 mev @xcite and frs - esr measurement of @xmath211 mev @xcite . * @xmath64xe : * this xenon isotope also has a single @xmath108 measurement , with @xmath212 mev @xcite . the cpt result of @xmath213 mev differs by @xmath175 . the cpt mass excess result of @xmath214 mev is consistent with isoltrap s result of @xmath215 mev @xcite . the recent frs - esr measurement disagrees by @xmath216 , with mass excess @xmath217 mev @xcite . * @xmath64cs : * neutron - rich cesium isotopes up to @xmath65cs @xcite as well as @xmath218cs @xcite have already been measured quite precisely in isoltrap , and to @xmath219cs with the orsay double - focusing mass spectrometer @xcite , also at isolde . therefore cesium was not investigated extensively in this experiment , with only @xmath220cs measured as a consistency check . for @xmath64cs , the ame03 quotes three measurements . most heavily weighted is the isoltrap measurement of the mass @xcite , which used @xmath57cs as the calibrant . the accepted mass of that calibrant had increased between the original paper and the ame03 by @xmath221 kev@xmath222 . taking this into account the isoltrap mass excess is @xmath223 mev , consistent with our value of @xmath224 mev . the other two cited measurements are a @xmath3-endpoint measurement of @xmath225 mev @xcite , and a beta - delayed neutron measurement @xcite . we can compare with the @xmath108 measurement by using an earlier cpt measurement of @xmath64ba @xcite to calculate @xmath226 mev , a difference of @xmath175 . we can compare to the ame03 input @xmath154 value of @xmath227 kev @xcite using the similarly adjusted isoltrap mass for @xmath63ba @xcite which yields @xmath228 kev , in agreement . * @xmath65cs : * for this nuclide the ame03 considers the isoltrap mass measurement @xcite and a @xmath108 measurement @xcite with an additional contribution from the orsay mass spectrometer @xcite , with which we can not make a direct comparison . the adjusted ( see @xmath64cs , above ) isoltrap mass excess of @xmath229 mev is in agreement with our result of @xmath230 mev . again utilizing previous cpt mass measurements @xcite , we can calculate @xmath108 to be @xmath231 mev , which is 1.2 @xmath117 from the value from @xcite of @xmath232 mev . * @xmath66pr : * this mass has previously been measured by the cpt @xcite at @xmath233 u. this is in agreement with the value found in the present experiment of @xmath234 u. the combined cpt result for this nuclide is therefore @xmath235 u. the sole input for the ame03 is a @xmath3-endpoint measurement of @xmath236 mev @xcite , which agrees with our result of @xmath237 mev . * @xmath68pr : * there are no previous mass measurements of this nuclide , direct or indirect . the ame03 extrapolates a mass excess of @xmath238 mev @xcite , which is @xmath101 from our result of @xmath239 mev . * @xmath66nd : * there is only one previous mass measurement of this neodymium isotope , a @xmath3-endpoint measurement of @xmath240 mev @xcite , which agrees with our result of @xmath241 mev . * @xmath68nd : * there is also only a single previous measurement of this isotope , a @xmath3-endpoint measurement @xcite that was discarded by the ame03 due to severe disagreement with systematic trends . that measurement was part of the same experiment as the accurate @xmath66nd measurement , above . that result of @xmath242 mev is inconsistent with our result of @xmath243 mev , which is closer to the ame03 interpolated value of @xmath244 mev @xcite . * @xmath69nd : * the mass of this neodymium isotope has never before been measured by any means , directly or indirectly . the ame03 extrapolates a mass excess of @xmath245 mev @xcite , @xmath132 lighter than this work s @xmath246 mev . * @xmath66pm : * three measurements are listed in the ame03 . one is a @xmath3-endpoint @xmath102-value experiment with a result of @xmath247 mev @xcite . no measurement of @xmath66sm was made by the cpt , but using the well - established ame03 value for that mass the @xmath108-value is found to be @xmath248 mev , a difference . a @xmath92sm@xmath90he@xmath249pm @xmath102-value measurement of @xmath250 mev @xcite is also used . using the well - established mass of @xmath92sm @xcite and our measurement , we find @xmath251 mev , which is consistent . a different proton - transfer reaction @xmath102-value , @xmath92sm@xmath252pm @xcite , was also measured at @xmath253 mev , which differs from our value of @xmath254 mev by @xmath255 . * @xmath68pm : * the only previous measurement is a @xmath3-endpoint measurement of @xmath256 mev @xcite , and the cpt result of @xmath257 mev agrees . * @xmath70pm : * two measurements are listed in the ame03 . both are @xmath3-endpoint experiments @xcite , but the cpt has not measured @xmath70sm so a direct comparison of @xmath108 is not possible . however , the mass excess determination based on these measurements given by the ame03 of @xmath258 mev differs from the cpt value of @xmath259 mev by @xmath260 . * @xmath69pm : * the ame03 cites a single @xmath3-endpoint measurement for this nuclide , with @xmath261 mev @xcite . the cpt measured both parent and daughter , finding @xmath262 mev , which agrees . * @xmath71pm : * this has also been measured only by @xmath3-endpoint , with a sole experiment in the ame03 giving @xmath263 mev @xcite . that experiment has since been reanalyzed by the same group in hayashi _ et al . _ @xcite resulting in @xmath264 mev . both values agree with the cpt value of @xmath265 mev . * @xmath72pm : * this nuclide had not been measured by any means as of ame03 publication , but has since been subject of @xmath3-endpoint measurements @xcite subsequently reanalyzed by hayashi _ @xcite resulting in @xmath266 mev . this differs from the cpt value of @xmath267 mev by @xmath268 . the ame03 systematic mass extrapolation gives a mass excess of @xmath269 mev , which differs from the cpt value of @xmath270 mev by @xmath255 . * @xmath68sm : * high - precision measurements of the neutron - separation energy of @xmath68sm have been made in two experiments via the @xmath88 reaction , and these are used exclusively by the ame03 to determine its mass . these measurements were adjusted in the ame03 to @xmath271 mev @xcite and @xmath272 mev @xcite . using the well - established ame03 value for the @xmath92sm mass , the cpt result is @xmath273 mev , which is consistent . * @xmath69sm : * a single @xmath3-endpoint measurement is used in the ame03 determination of this nuclide , with @xmath274 mev @xcite . the cpt has not measured the daughter of this decay , so a direct comparison is impossible ; however the cpt mass excess value of @xmath275 mev is consistent with the ame03 value of @xmath276 mev . * @xmath71sm : * a single @xmath3-endpoint experiment forms the basis for the ame03 determination of this nuclide s mass , with @xmath277 mev @xcite , consistent with the cpt value of @xmath278 mev . * @xmath72sm : * the ame03 cites a single @xmath3-endpoint measurement for this nuclide of @xmath279 mev @xcite , which has since been reanalyzed by the same group in hayashi _ @xcite resulting in @xmath280 mev . both values agree with the cpt value of @xmath281 mev . * @xmath73sm : * no previous mass measurement of any kind exists for @xmath73sm . the ame03 extrapolated a mass excess of @xmath282 mev , which is consistent with the cpt value of @xmath283 mev . * @xmath74sm : * no measurement of this nuclide existed as of ame03 publication . it has since been subject of the @xmath3-endpoint measurements @xcite subsequently reanalyzed by the same group in hayashi _ @xcite which found @xmath284 mev , which agrees with the cpt result of @xmath285 mev . the cpt mass measurement differs by 310 kev@xmath222 with the ame03 s extrapolation , which has an uncertainty of @xmath20 kev@xmath222 . * @xmath71eu : * this nuclide has two separate @xmath3-endpoint measurements , listed in the ame03 as @xmath286 @xcite and @xmath287 mev @xcite , combined in the ame03 to @xmath288 mev . all of these are consistent with the cpt value of @xmath289 mev , using the well - established ame03 mass value for @xmath71gd which was not measured by the cpt . * @xmath72eu : * this nuclide was the subject of a transfer - reaction experiment , @xmath73gd@xmath93 , which gave a @xmath102 value for that reaction of @xmath290 mev @xcite . using the well - known mass of the nearly stable @xmath73gd from the ame03 , the cpt measurement gives @xmath291 mev , a difference of @xmath101 . * @xmath73eu : * no previous mass measurements for this nuclide were used in the ame03 , it having rejected two beta - endpoint measurements of @xmath292 @xcite and @xmath293 mev @xcite . it has since been the subject of the @xmath3-endpoint measurements by hayashi _ et al . _ which found @xmath294 mev @xcite . using the well - known mass of the nearly stable @xmath73gd from the ame03 , the cpt measurement disagrees significantly , with @xmath295 mev , which is @xmath296 kev or @xmath297 from the hayashi _ the cpt mass value is consistent with the ame03 systematic extrapolation . * @xmath74eu : * no previous mass measurement existed for this nuclide as of ame03 publication , but it has since been the subject of the @xmath3-endpoint measurements by hayashi _ et al . _ which found @xmath298 mev @xcite . the cpt mass value is consistent with the ame03 systematic extrapolation . * @xmath75gd : * no previous mass measurement existed for this nuclide as of ame03 publication , but it has since been the subject of the @xmath3-endpoint measurements by hayashi _ et al . _ which found @xmath299 mev @xcite . the cpt mass value is consistent with the ame03 systematic extrapolation . l@ c@ [email protected] model & @xmath117 ( mev ) & + ame03 @xcite & 0.171 & -0&105 + frdm @xcite & 0.538 & -0&379 + hfb2 @xcite & 0.555 & 0&281 + hfb9 @xcite & 0.467 & -0&175 + hfbcs1 @xcite & 0.546 & -0&025 + duzu @xcite & 0.234 & 0&062 + ktuy05 @xcite & 0.611 & 0&438 + etfsi2 @xcite & 0.396 & -0&120 + because most of the @xmath2-process path is outside the region of known masses , @xmath2-process simulations are forced to use mass models for neutron - separation energy values . available mass models often disagree significantly for unmeasured masses , so it can be enlightening to compare first - time measurements to mass models which were created before these masses were available . table [ tbl : models ] summarizes the overall precision and accuracy of , and figs . [ fig : frdm - me][fig : etfsi12-me ] compare the new cpt measurements to , various popular mass models . the frdm is a commonly used model for @xmath2-process simulations , and a comparison with the new cpt results is shown in fig . [ fig : frdm - me ] . the @xmath300 shell closure is followed by an oscillation about the true masses at higher @xmath301 , though these oscillations appear to be damped away by @xmath302 . this region is important for the @xmath2 process , because its path is expected to exit the @xmath300 shell after the possible @xmath303cd waiting point @xcite . if the overshooting in binding energy past @xmath300 continues for @xmath304 lower than measured here , then the location of the @xmath2-process path may be closer to stability for @xmath305 and 50 than the frdm would indicate . clearly more mass measurements around @xmath305 and 50 are vital for an accurate path determination past @xmath300 . for astrophysics modeling it is not the absolute errors in mass models that are relevant but systematically increasing errors with @xmath301 , because this is what affects @xmath5 and thus the process paths . for example , table [ tbl : models ] would suggest that the duzu model is one of the most accurate in this region , but fig . [ fig : duzu - me ] shows that for most elements the deviations from our results have negative slope at @xmath10 to @xmath14 kev per neutron . given the @xmath5 slope of duzu here , the difference in @xmath5 corresponds to a shift in the @xmath2-process path of about one neutron closer to stability were the @xmath5 difference to continue changing with the same slope . ktuy05 ( fig . [ fig : ktuy05-me ] ) shows the same trend but with even larger magnitude . etfsi2 ( fig . [ fig : etfsi12-me ] ) on the other hand has a slope of similar magnitude in the opposite direction . any such trend in hfb2 or hfb9 ( fig . [ fig : hfb2-me ] and [ fig : hfb9-me ] ) is harder to identify given the apparent difficulty with pairing effects at large @xmath301 , with many changes of over @xmath306 kev in mass deviation from isotope to isotope . the hfbcs1 model ( fig . [ fig : hfbcs1-me ] ) had the lowest mean mass difference from the cpt among these models , but had among the largest scatters in difference , spanning over 2 mev . [ fig : frdm_s2n ] compares neutron separation energies for the cpt and frdm and suggests a more complicated story for this model . of course these new measurements are still very far from the predicted @xmath2-process path , so reliable predictions of the path location are not yet possible . given the varied problems with mass models , forming any expectation that one would be more accurate than the others at yet higher neutron excess would be a gamble . it is clear that mass measurements on or close to the @xmath2-process path are sorely needed , especially at the @xmath300 waiting points . mass measurements of 40 nuclides were made , most of which improved the precision and accuracy over literature values . results are largely consistent with previous results from penning traps , reaction energetics , and storage ring measurements , but are frequently much lower in mass than results from @xmath3-endpoint measurements . comparison with mass models shows only sporadic agreement , and suggests unsuitability of these models for precise astrophysical @xmath2-process simulations . reliable mass measurements yet closer to and on the @xmath2-process path are still needed . the cpt has been moved to the new californium rare isotope breeder upgrade ( caribu ) facility @xcite now operating at the atlas accelerator , and has begun a successor campaign to these measurements . the extent of the cpt s previous measurements was limited by the purity and rate of the available beam , and caribu offers several improvements over the system used here to overcome these limitations . the @xmath25-@xmath307 times more intense fission source and larger gas catcher will allow access to nuclides 3 - 5 neutrons farther from stability . a new compact higher - resolution in - flight isobar separator @xcite promises purification with a resolving power of four times that of the previous system while avoiding the saturation of ion traps that would have occurred otherwise with such an intense beam . the in - flight separation will also defer the limitation in nuclide lifetime that would be imposed by in - trap purification . many of the newly available nuclides lie directly on potential @xmath2-process paths near @xmath308 . the cpt will conduct a survey of all accessible @xmath2-process nuclides in this vicinity , as well as any other neutron - rich species made available with the goal of finding isomers , regions of deformation , and shell quenching . over 200 additional masses are expected to be measured by the cpt . this work performed under the auspices of nserc , canada , application number 216974 , and the u.s . doe , office of nuclear physics , under contract nos . de - ac02 - 06ch11357 , de - fg02 - 91er-40609 , de - fg02 - 98er41086 , and de - ac52 - 07na27344 . u. keyser , h. berg , f. mnich , b. pahlman , k. hawerkamp , b. pfeiffer , h. schrader , and e. monnand , in _ atomic masses and fundamental constants , 6 _ , edited by j. a. nolen , jr . and w. benenson ( plenum press , new york , 1980 ) , p. 485 . m. shibata , t. shindou , y. kojima , m. asai , k. tsukada , s. ichikawa , h. haba , y. nagame , and k. kawade , japan atomic energy research institute , tandem vdg annual report , 2001 ( 2002 ) p. 26 ; jaeri - review 2002 - 029 ( 2002 ) . m. shibata , o. suematsu , k. kawade , m. asai , s. ichikawa , y. nagame , a. osa , k. tsukada , y. kojima , and a. taniguchi , japan atomic energy research institute , tandem vdg annual report , 2002 ( 2003 ) p. 32 ; jaeri - review 2003 - 028 ( 2003 ) . kratz , a. schreder , h. ohm , h. gabelmann , w. ziegert , b. steinmller , and b. pfeiffer , in _ proceedings of the 7th international conference on atomic masses and fundamental constants amco-7 _ , edited by o. klepper ( technishe hochschule darmstadt lehrdruckerei , damstadt , 1984 ) , p. 127 . m. shibata , t. shindou , k. kawade , y. kojima , a. taniguchi , y. kawase , and s. ichikawa , in _ exotic nuclei and atomic masses , enam2001 _ , edited by j. yst , p. dendooven , a. jokinen , and m. leino ( springer - verlag , berlin , heidelberg , 2001 ) , p. 479 .

using the o(4 ) linear model , we address the topic of non - equilibrium relaxation of an inhomogeneous initial configuration due to quantum and thermal fluctuations . the space - time evolution of an inhomogeneous fluctuation of the condensate in the isoscalar channel decaying via the emission of pions in the medium is studied within the context of disoriented chiral condensates . we use out of equilibrium closed time path methods in field theory combined with the amplitude expansion . we give explicit expressions for the asymptotic space - time evolution of an initial inhomogeneous configuration including the contribution of thresholds at zero and non - zero temperature . at non - zero temperature we find new relaxational processes due to thermal cuts that have no counterpart in the homogeneous case . within the one - loop approximation , we find that the space time evolution of such inhomogeneous configuration out of equilibrium is effectively described in terms of a rapidity dependent temperature $ ] as well as a rapidity dependent decay rate . this rate is to be interpreted as the production minus absorption rate of pions in the medium and approaches the zero temperature value at large rapidities . an initial configuration localized on a bounded region spreads and decays in spherical waves with slower relaxational dynamics at large rapidity .

hep-ph9602232

the dynamics of relaxation of inhomogeneous field configurations out of equilibrium is an important problem and a common theme in cosmology and high energy physics . in high energy physics the experimental possibility of studying the chiral and quark - gluon phase transition with high luminosity hadron colliders and upcoming heavy - ion colliders makes imperative the understanding of relaxation and transport processes in extreme environments@xcite . a very exciting possibility@xcite that can be studied in high energy - high luminosity hadron collisions or relativistic heavy ion collisions within the energy range of the upcoming rhic and lhc colliders , is that after the chiral phase transition , regions in which the chiral condensate is misaligned with respect to the vacuum state are formed . in these ultra - high energy heavy ion collisions ( @xmath3 ) a large energy density ( a few @xmath4 ) is deposited in the collision region corresponding to temperatures above the critical value for chiral symmetry restoration @xmath5 . in this situation , it is possible that within a volume of a few @xmath6 an inhomogeneous coherent field configuration is formed corresponding to a region in space in which the chiral symmetry is restored . the idea@xcite is that as these regions cool down they relax towards the equilibrium situation emitting a large number of soft pions . such a configuration has been dubbed a ` disoriented chiral condensate ' ( dcc ) . bjorken , kowalski and taylor proposed a ` baked alaska'@xcite scenario in which this configuration relaxes via the copious emision of pions strongly correlated in isospin . such a scenario could also explain the centauro events observed in cosmic rays in which the ratio of charged to neutral pions is different from @xmath7@xcite . a microscopic , field theoretical description of the dynamics of relaxation of these field configurations with typically large amplitudes faces at least two major obstacles . the first is a non - perturbative treatment of inhomogeneous , large amplitude field configurations . the second is a consistent description of non - equilibrium processes that can allow a _ real time _ calculation of the evolution . recently there has been a surge of interest in the description of non - equilibrium processes both in cosmology and high energy physics . we refer to non - equilibrium processes to those corresponding to the the time evolution of quantum states in field theory which * are not * hamiltonian eigenstates as well as the time evolution density matrices that do not commute with the hamiltonian . in particular the relaxation of strong electric fields@xcite and the evolution of homogeneous scalar order parameters during phase transitions in minkowski and cosmological spacetimes @xcite . the non - equilibrium aspects of disoriented chiral condensates have been studied within different approximation schemes by several authors@xcite , but mainly focusing on homogeneous expectation values or correlation functions . however , to the best of our knowledge , these studies focused on the evolution of a homogeneous but time ( or proper time ) dependent mean field and fluctuations around it but did not address the relaxational dynamics of _ inhomogeneous _ field configurations including _ quantum and thermal _ effects . moreover , inhomogeneous field configurations appear in cosmology when topological objects , such as textures or cosmic strings involving inhomogeneous field configurations are considered . their relaxational dynamics is thought to have a bearing on the spectrum of fluctuations in the cosmic microwave background radiation @xcite . the focus of this article is to study the _ linear _ relaxational dynamics of inhomogeneous , coherent field configurations both at zero and finite temperature in the @xmath8 linear sigma model . as is known such model is the relevant field theory for the description of chiral condensates . we use out of equilibrium closed time path methods @xcite combined with the amplitude expansion @xcite to study the time evolution of @xmath9 . we consider small initial field amplitudes such that we can keep just the linear term in the amplitude expansion . in such approximation the field evolution equations linearize . whereas the focus is on the description of the dynamics within the setting of the chiral phase transition , the analysis presented below applies to a wide variety of physical situations in which a inhomogeneous scalar order parameter relaxes to the equilibrium situation via the production of lighter scalar fields in a medium . the main idea is that after a phase transition ( within the framework of dcc s the chiral phase transition ) regions are formed within which the scalar order parameter is inhomogeneous . as the zero momentum component rolls towards the equilibrium value , these inhomogeneous configurations will relax in such a way that the spatial gradients of the field configuration become smaller so as to decrease the energy . this relaxation will be accompanied by the production of quanta , such as pions in the case of dcc s . here we study the o(4 ) linear sigma model in the broken symmetry state , at temperatures below the chiral phase transition as an effective low energy theory of @xmath10 ( up and down quarks ) which presumably incorporates the effects of strongly interacting qcd on chiral dynamics@xcite . wilczek and rajagopal@xcite have argued that the o(4 ) linear sigma model effectively describes the same _ equilibrium _ universality class of qcd with two flavors of light quarks and proposed to study it within the context of dcc . clearly this simple model misses a great number of degrees of freedom , hadrons , vector mesons , etc . and can only be justified as an effective description at temperatures well below the chiral phase transition when the degrees of freedom with much higher masses are not relevant . the linear sigma model may also be obtained as a landau - ginsburg effective theory from a nambu - jona - lasinio model@xcite which is a popular description of the phenomenology of chiral symmetry at the quark level and which has also been used as an alternative description of dccs@xcite . the o(4 ) linear sigma model has also been studied within the context of texture type configurations deemed relevant in certain cosmological models of structure formation@xcite , hence our study may be relevant in these cosmological scenarios as well . we will restrict our study to the case in which the condensate occurs in the isoscalar channel , without isospin violation . in section ii we introduce the model , discuss its range of validity and summarize aspects of non - equilibrium field theory relevant to the calculation . section iii presents the bulk of the results , with explicit analysis of the space - time evolution , including the contribution from thresholds and thermal cuts . we summarize our results in the conclusions section . we study is the o(4 ) linear sigma model as an effective low energy description of chiral symmetry aspects of qcd with two ( light ) flavors of quarks . the lagrangian density is given by @xmath11 where @xmath12 is an @xmath8 vector , @xmath13 . the field @xmath0 describes the @xmath14 condensate while the ( isospin ) triplet @xmath15 describes the pions . the small magnetic field @xmath16 accounts for the explicit symmetry breaking arising from the up and down quark masses and gives the pions a small mass @xmath17 . the pion decay constant @xmath18 and @xmath19 is fixed by the value of the ` mass ' of the @xmath0 field @xmath20 . in what follows we will neglect the magnetic field , but explicitly give a mass to the pions and use @xmath21 as the pion decay constant . at this point we recognize that the linear @xmath0-model is a strongly coupled effective theory and any perturbative approximation will be subject to criticism . this important objection notwithstanding , we pursue the study of the non - equilibrium evolution of the condensate with the hope of learning new features of the decay of the condensate that will perhaps be rather generic and persist in higher order calculations or eventually in a non - perturbative treatment . since we want to understand the relaxation of inhomogeneous perturbations of the condensate around the equilibrium value @xmath21 , we write @xmath22 where the average above stands for the expectation value in the non - equilibrium state discussed below , and the fluctuation field @xmath23 describes the @xmath0 mesons in the broken symmetry state . the quantum state that leads to ( [ expvalue ] ) is an inhomogeneous coherent state of @xmath0 mesons . the field theoretical methods to describe processes out of equilibrium are known and described at length in the literature @xcite . the basic ingredient is the time evolution of an initially prepared density matrix , which leads to the generating functional of non - equilibrium green s functions in terms of a path integral representation along a complex contour in time . the contour involves a forward time branch , a backward time branch and a third branch down the imaginary time axis to time @xmath24 if the initial density matrix describes an ensemble at initial temperature @xmath25 . it can be proven ( see references above ) that the third , imaginary branch only determines the boundary conditions on the green s functions , but does not enter in the calculation of _ real - time _ correlation functions . the fields living on the forward and backward branches will be labelled with @xmath26 and @xmath27 , respectively , and the effective lagrangian that enters in the path integral representation of the non - equilibrium generating functional is given by @xmath28-{\cal l}[\vec{\phi}^- ] \label{noneqlag}\;.\ ] ] from this path integral representation it is possible to construct a perturbative expansion of the non - equilibrium green s functions in terms of modified feynman rules : 1 . the number of vertices is doubled . those in which all the fields are on the @xmath26 branch are the usual interaction vertices , while the vertices in which the fields are on the @xmath27 branch have the opposite sign . the combinatoric factors are the same as in usual field theory . the spatial fourier transform of the ( bosonic ) propagators are @xmath29 where @xmath30 is a generic bose field . that is @xmath31 or @xmath32 . now we have to specify the properties of the initial state . a particularly convenient choice is that of a thermal initial state at temperature @xmath33 . then the density matrix of this initial state is @xmath34 , where @xmath35 is the hamiltonian for times @xmath36 . this choice of the initial state determines the boundary conditions on the green s functions ; these are the usual periodicity conditions in imaginary time ( kms conditions ) : @xmath37 finally , the free - field green s functions for generic bose fields are constructed from the following ingredients : @xmath38 e^{-i\omega_k(t - t')}+n_{b}(\omega_k ) e^{i\omega_k(t- t')}\right\}\ ; , \label{ggreat } \nonumber \\ & & g_k^{<}(t , t')=\frac{i}{2\omega_k}\left\{[1+n_{b}(\omega_k ) ] e^{i\omega_k(t - t')}+n_{b}(\omega_k)e^{-i\omega_k(t- t')}\right\}\ ; , \label{gsmall } \nonumber \\ & & \quad\quad \omega_k=\sqrt{\vec{k}^2+m^2}\;,\quad\quad\quad n_{b}(\omega_k)=\frac{1}{e^{\beta \omega_k}-1 } \ ; , \label{bosefactor}\end{aligned}\ ] ] where @xmath39 is the mass of the boson . an important property that will be used in the calculations that follow is the relation @xmath40 our goal is to study the dynamics of the expectation value of the chiral order parameter @xmath41 . we obtain the equation of motion for @xmath42 using the tadpole method@xcite . to implement this method in the non - equilibrium formulation , set @xmath43 in the lagrangian [ noneqlag ] , and consider @xmath44 as a c - number background field , keeping the _ linear _ , cubic and quartic terms as interactions . the expectation value of the fluctuation @xmath45 is computed in presence of the background of fields @xmath42 and the condition @xmath46 is imposed to all orders in perturbation theory . we restrict our study to the case of _ linear relaxation _ , so that we obtain the equation of motion linearized in the amplitude of @xmath44 , an approximation that is valid for small amplitude fluctuations from the minimum of the tree level potential . this equation can be obtained in a systematic loop expansion using the usual feynman rules with the doubled interaction vertices and the non - equilibrium propagators . to one loop order there are two tadpole contributions to the equation of motion : one from the quartic vertex that is absorbed in a mass renormalization and the other from the cubic vertex that is absorbed in a renormalization of @xmath21 . the remaining one loop feynman diagrams that contribute to the equation of motion are shown in figure 1 . the details have already been presented elsewhere@xcite . the equation @xmath47 leads to @xmath48 \right\}=0 \;. \nonumber \end{aligned}\ ] ] we have absorbed a momentum independent but temperature dependent tadpole in a renormalization of @xmath49 , and used the property ( [ relation ] ) . @xmath50 is the retarded self - energy . we are interested in the decay of an inhomogeneous chiral condensate into pions . in the linear @xmath0 model the chiral condensate is represented by the @xmath0 field . the dynamics of the decay process of such configuration will be determined by the imaginary part of the self - energy _ on mass shell_. in a medium , different kind of processes contribute to the imaginary part of the on - shell self - energy of the external particle@xcite . in particular , collisional processes are always present and are responsible for a collisional lifetime of the particle in the medium . decay ( and recombination ) processes are only present if the kinematics of decay is allowed ; these contribute to an imaginary part of the self - energy _ on shell _ only when the lowest multiparticle threshold is _ below _ the single particle particle pole in the spectral density of the one - particle propagator . since the ` mass ' of the @xmath0 field ( @xmath51 600 mev ) is larger than twice the pion mass ( @xmath51 130 mev ) , the one - loop diagram with pions in the internal loop will contribute to an imaginary part on mass shell of the sigma particle through processes in which the @xmath0 decays into two pions in the medium , as well as the inverse process of pion recombination into @xmath0 . decay and recombination processes will also be present in higher order contributions to the self - energy , but we will only consider the contribution from the one - loop diagrams shown in figure 1 . the bilinear and trilinear interaction vertices needed for the one loop calculation are obtained from the following part of the interaction lagrangian density @xmath52 the retarded self - energy to this order is found to be : @xmath53 ^ 2 - [ g^<_{\sigma}(\vec{x}-\vec{x'},t - t')]^2 \ } \theta(t - t ' ) \ ; , \label{sigmachi } \nonumber\\ \sigma_{ret,\pi } ( \vec{x}-\vec{x'},t - t ' ) & = & 6\ i g^2 \{[g^>_{\pi}(\vec{x}-\vec{x'},t - t')]^2 - [ g^<_{\pi}(\vec{x}-\vec{x'},t - t')]^2\ } \theta(t - t ' ) \label{sigmapi } \;.\end{aligned}\ ] ] the green s functions in ( [ sigmachi ] , [ sigmapi ] ) are given by ( [ ggreat ] ) appropriate for the quanta of the fields @xmath23 and @xmath15 respectively . we have also used the fact that the pion green s functions are diagonal in isospin . since we are dealing with a real field , the retarded self - energy @xmath50 is given by @xcite @xmath54 translational invariance of the self - energy makes it convenient to write the equation of motion for the spatial fourier transform of @xmath42 . introduce @xmath55 in order to solve the integro - differential equation we will impose the initial condition @xmath56 and that this configuration is ` released ' at time @xmath57@xcite . under these conditions the evolution equation becomes @xmath58 using the free - field green s functions given above for the @xmath59 fields we find @xmath60 \nonumber \\ & & \quad-2 \ ; n_{k,\sigma } \sin[(\omega_{k+p,\sigma}-\omega_{k,\sigma})(t - t')]\ } \label{sigmachik } \\ \sigma_{\pi}(\vec p , t - t ' ) & = & -6g^2 \int\frac{d^3k}{(2\pi)^3 } \frac{1}{2\omega_{k,\pi } \omega_{k+p,\pi}}\ { ( 1 + 2 n_{k,\pi})\sin[(\omega_{k+p,\pi}+\omega_{k,\pi})(t - t ' ) ] \nonumber \\ & & \quad-2 \ ; n_{k,\pi } \sin[(\omega_{k+p,\pi}-\omega_{k,\pi})(t - t')]\ } \label{sigmapik } \end{aligned}\ ] ] where @xmath61 the equation of motion ( [ finaleqofmot ] ) can now be solved via a laplace transform . with the boundary conditions @xmath62 and denoting the laplace transforms of @xmath63 by @xmath64 respectively ( here @xmath65 is the laplace transform variable ) we find @xmath66 with @xmath67 given by ( [ sigmafreq ] ) . the time evolution is found by integration in the complex s - plane along the bromwich contour @xmath68 with @xmath69 chosen so that the contour lies to the right of the real part of all the singularities of @xmath70 . the self energies ( [ sigmachik],[sigmapik ] ) are proportional to the kernel @xmath71 \nonumber \\ & & -2 n_{k } \sin[(\omega_{\vec k + \vec p}- \omega_{\vec k})(t - t')]\ } \ ; , \label{kernel}\end{aligned}\ ] ] whose laplace transform can be written as a dispersion integral @xmath72 in the expressions above , the frequencies and bose factors correspond to the particle in the loop . we are considering the situation near the end of the chiral phase transition , in the hadronization stage with temperature different from zero but well below the transition temperature @xmath73 . however , the zero temperature limit can be taken at any stage in the calculation . since this kernel will determine the singularities of the laplace transform in the complex s - plane it is important to understand its analytic structure . from ( [ specdens ] ) we find the imaginary part of the kernel to be @xmath74 \ ; , \label{imagpart}\ ] ] where the @xmath75 reflects the retarded nature of the kernel@xcite . there are two different processes that contribute to the imaginary part of the kernel that deserve to be studied in detail , for @xmath76 these give the following contribution : ( for @xmath77 changes sign ) @xmath78 \;. \label{imsigma2}\end{aligned}\ ] ] the processes that contribute to @xmath79@xcite are the decay @xmath80 with boltzmann weight @xmath81 minus the recombination process @xmath82 with boltzmann weight @xmath83 when the particles in the loop are pions , or the process @xmath84 and its inverse @xmath85 with similar weights when the particles in the loop are @xmath0 . clearly these processes will only contribute above the two - particle threshold . the processes that contribute to @xmath86@xciteare the @xmath87 with weight @xmath88 minus the inverse @xmath89 with weight @xmath90 when the particles in the loop are pions , or the process @xmath91 minus its inverse @xmath92 with the corresponding weights when the particles in the loop are the @xmath0 mesons . clearly these latter process can only occur in the heat bath for non - zero momentum transfer and will lead to new thresholds and discontinuities . these are identified with the landau damping processes in the medium@xcite as they only occur for space - like momenta . an important point to consider is that these processes do not contribute to the relaxation of a _ homogeneous condensate _ ( zero momentum transfer ) and are , therefore , a new feature of the inhomogeneous situation . after analyzing the kinematical regions , we find : @xmath93 \right\ } \nonumber \\ & & \qquad\times\theta(\omega^2-p^2 - 4m^2 ) \ ; , \label{impartk1 } \\ & & imk^{(2)}(\vec p , i\omega + 0^+,t)= \frac{t}{4\pi p } \ln\left[\frac{1-e^{-\beta \omega_p^+ } } { 1-e^{-\beta \omega_p^-}}\right ] \theta(p^2-\omega^2 ) \ ; , \label{impark2}\\ & & \omega_p^{\pm } = \left|\frac{\omega}{2}\pm \frac{p}{2 } \sqrt{1-\frac{4m^2}{\omega^2-p^2}}\right| \ ; , \label{omegaplusmin } \end{aligned}\ ] ] where @xmath39 is the mass of the particles in the loop , @xmath94 for @xmath95 or @xmath96 for @xmath97 . in the expression for @xmath79 we have explicitly separated out the @xmath98 and @xmath99 contributions . note that whereas @xmath79 is non - zero above the two particle threshold and has a contribution that survives in the zero temperature limit . @xmath86 is non - zero only for @xmath99 , and for space - like momenta and does not contribute _ directly _ to the imaginary part _ on shell _ but * does contribute * to the relaxational dynamics as discussed below . explicit _ form of the imaginary part at finite temperature is one of the novel results of this study . for small momentum @xmath100 we find @xmath101 \ ; , \label{impar1p0}\\ & & i m k^{(2)}(\vec p , i\omega + 0^+,t ) \approx \frac{1}{8\pi}\sqrt{1-\frac{4m^2}{\omega^2-p^2 } } 2n(\omega^- ) \theta(p^2-\omega^2 ) \ ; , \label{impar2p0}\end{aligned}\ ] ] and we recognize ( [ impar1p0 ] ) as the imaginary part obtained in the homogeneous case@xcite . the @xmath102 self - energies are now given in terms of the above kernel as : @xmath103 the real part of the kernel can be obtained from the imaginary part from the dispersion integral ( [ laplacekernel ] ) , but is very difficult to compute in the general case . its zero temperature limit can be found analytically . the integral in ( [ laplacekernel ] ) has a logarithmic divergence which is subtracted at @xmath104 . this subtraction can be absorbed in a further renormalization of @xmath49 and we find that for @xmath98 : @xmath105\;,\ ] ] where again , @xmath39 refers to the mass of the particle in the loop . along the imaginary axis and below the two particle cut , the kernel is real and given by @xmath106\;,\ ] ] whereas above the two - particle cut @xmath107 the kernel has an imaginary part , @xmath108 with @xmath109 \ ; , \label{realkt0 } \\ & & k_i(\vec p , \omega ) = \frac{1}{8\pi}\sqrt{1-\frac{4m^2}{\omega^2-p^2 } } \;. \label{imkt0 } \\\end{aligned}\ ] ] using the parameters of the linear @xmath0 model , we find that @xmath110 thus we see that although the linear @xmath0 model is strongly coupled , the one - loop correction to @xmath49 is very small . however the imaginary part on shell ( related to the width of the particle , see below ) is very large . to perform the inverse laplace transform we need to understand the analytic structure of the laplace transform in the complex s - plane . from the analysis presented above , we find that for arbitrary temperature , @xmath70 is analytic except for 1 . isolated single particle poles at @xmath111 with @xmath112 being the solutions of @xmath113 and 2 . discontinuities across the imaginary axis determined by the imaginary part of the self - energy obtained above : @xmath114 where we have introduced the spectral density @xmath115 given by @xmath116 ^ 2 + \sigma_i(\omega,\vec p , t)^2 } \;. \label{specdens2}\ ] ] the real - time evolution of the spatial fourier transform of the condensate is obtained via the inverse laplace transform @xmath117 the integral is performed by deformation of the contour , wrapping around the single particle poles which are solutions of the equation ( [ polecondition ] ) and slightly to the right and left of the multiparticle cuts along the imaginary axis . thus the inverse laplace transform will have two contributions : from the poles @xmath118 and from the cuts @xmath119 and we can write in general @xmath120 there are several different cases that we can study . although this is _ not _ a realistic case for the chiral phase transition described by the linear @xmath0 model , this case is still worth studying in order to compare it to the unstable case @xmath122 to be described later . furthermore , we will see that many results from this particular case do apply in the unstable case which _ is _ of interest for the chiral phase transition . at zero temperature the self - energy is manifestly lorentz covariant as can be seen from the expressions ( [ realkt0 ] , [ imkt0 ] ) . we first analyze the one - particle pole contribution , which is a straightforward generalization of the homogeneous @xmath123 case@xcite . the condition ( [ polecondition ] ) is satisfied for @xmath124 then the pole contribution to the inverse laplace transform is @xmath125}^{-1 } \;. \label{wavefuncren}\end{aligned}\ ] ] where @xmath126 is the ( finite ) wave - function renormalization constant , defined _ on shell_. we will now consider the case in which @xmath127 ( independent of @xmath128 ) and later study the general case by convolution . at this point it is convenient to introduce the proper time and ( radial ) spatial rapidity variables as these exhibit the lorentz properties more clearly : @xmath129 \ ; , \nonumber \\ \tau & = & \sqrt{t^2-r^2}\ ; , \quad r = \tau \sinh[\vartheta_r ] \ ; , \quad t = \tau \cosh[\vartheta_r ] \;. \label{rvars } \end{aligned}\ ] ] the spatial fourier transform leads to the pole contribution @xmath130}\over{\tau}}\ ; j_2(m_0 \tau)\,\theta(\tau^2 ) \ ; , \label{stablepole}\ ] ] where @xmath131 is a bessel function . recalling that @xmath132 it is clear that this solution is lorentz invariant , since the product @xmath133 $ ] is lorentz invariant . the pole contribution then gives fluctuations which propagate inside the light - cone as a massive relativistic wave as expected . for large times and distances eq.([stablepole ] ) gives @xmath134}{\tau^{3/2}}\ ; \cos(m_0 \tau-\frac{\pi}{4})\,\theta(\tau^2 ) \label{asympole}\;.\ ] ] we now evaluate the cut contribution @xmath119 by performing the integral over the cut and using the change of variables @xmath135 @xmath136 where @xmath137 is the smallest of @xmath138 and @xmath139 is the spectral density given by ( [ specdens2 ] ) . the amplitude is then @xmath140\over \tau}\;\int_{2{\cal{m}}}^\infty { \omega^2 \ ; s(\omega , \vec p=0 , t=0 ) \ ; j_2(\omega\tau)\;d\omega } \label{corte } \;.\ ] ] for large proper time @xmath141 this integral is dominated by the two - particle thresholds ( @xmath142 ) and we find @xmath143\over \tau^3 } \left[{\cal{c}}_1 m^2_{\sigma } \sin(2m_{\sigma}\tau)+{\cal{c}}_2 m^2_{\pi}\sin(2m_{\pi}\tau)\right ] \ ; , \label{inhd}\ ] ] where @xmath144 are constants determined by the spectral densities at the respective two - particle thresholds . when @xmath145 , the amplitude of @xmath146 is smaller than the one of @xmath147 by a factor @xmath148 and is clearly subleading at long distances . this is analogous to the factor @xmath149 for the homogeneous case@xcite . unlike the @xmath98 case , we can not make use of manifest lorentz covariance , since this property is not available in the rest frame of the heat bath . however the situation is qualitatively similar to the zero temperature case . the positions of the one - particle pole are now given by @xmath150 the pole contribution is @xmath151 where the wave - function renormalization ( defined on shell ) is now @xmath152^{-1 } \;. \label{wavefunt}\ ] ] the lack of lorentz invariance introduces a complicated dependence on the momentum @xmath153 , and the inverse fourier transform can not be performed analytically as it was the case for @xmath98 . however , for @xmath154 ( @xmath155 ) one can compute @xmath156 by the stationary phase approximation . we have @xmath157 \;. \label{deltapole}\ ] ] this integral has stationary points at @xmath158 for large @xmath159 and @xmath160 and time - like intervals , where by @xmath161 we refer to terms that are small in the formal weak coupling expansion or in the strongly coupled @xmath0 model numerically small as given by equation [ realpartonshell ] . these saddle point values have an illuminating physical interpretation which is better displayed in terms of ( radial ) rapidity ( [ rvars ] ) and momentum rapidity variables defined by : @xmath162 \ ; , \label{momrap}\\ p & = & m_{\sigma } \sinh[\vartheta_p ] \ ; , \quad \omega = m_{\sigma } \cosh[\vartheta_p ] \;. \label{pvars}\end{aligned}\ ] ] using these variables , the saddle point condition becomes @xmath163 and we will just refer to @xmath164 as the rapidity variable irrespective of coordinate or momentum . the contribution from the saddle points yields @xmath165,t ) \left[\frac{m^{3/2}\cosh[\vartheta]}{\tau^{3/2 } } + { \cal o}(g^2)\right ] \nonumber \\ & & \quad\quad \times\cos\left\{m_{\sigma}\tau \left[1+\frac{1}{2m^2_{\sigma } } \sigma(im , p = m_{\sigma}\sinh[\vartheta],t)+{\cal o}(g^4)\right]- \frac{\pi}{4}\right\ } \nonumber \\ & & \quad\quad\times\left\{1+{\cal o}(\frac{1}{t^2},\,\frac{1}{r^2 } , \,\frac{1}{tr})\right\}\label{asympolet}\;.\end{aligned}\ ] ] the fluctuations for @xmath99 propagate as spherical waves , similarly to the zero temperature case displayed in eq.([asympole ] ) . however the phase and amplitude depend on @xmath166 and @xmath164 through the temperature dependence of @xmath167 as well as of z , which breaks manifest lorentz covariance . the cut contribution is @xmath168 where we have separated the ` normal ' ( zero temperature ) two - particle cut @xmath169 from the thermal contribution with support below the light - cone @xmath170 again , as in the zero temperature case @xmath171 is the smallest mass . let us first study @xmath172 . at large @xmath159 the most important contribution to the integral will be from the region near the thresholds and the largest contribution will be from the smallest threshold ( two pion ) . from the expression for the imaginary part given by eq . ( [ impartk1],[omegaplusmin ] ) we see that near the thresholds @xmath173 and @xmath174 \;,\label{impartnearthresh}\ ] ] with @xmath175 the bose occupation number . performing the change of variables @xmath176 the integral over @xmath177 is similar to ( [ corte ] ) but with the finite temperature factor @xmath178 $ ] in the integrand . for large distances , the spatial fourier transform can be done by the stationary phase approximation . the values of the stationary phase are @xmath179 \;. \label{newsaddle}\ ] ] finally , at large times the integral over @xmath177 obtains the largest contribution near the thresholds @xmath180 and we find the leading behavior for large times and distances ( for time - like intervals ) @xmath181\over\tau^3}\left\ { { \cal{c}}_1(t ) \left[1 + 2n\left(m_{\sigma}\cosh[\vartheta]\right)\right ] m^2_{\sigma } \sin(2m_{\sigma}\tau ) + \right . \nonumber \\ & & \left . { \cal{c}}_2(t ) \left[1 + 2n\left(m_{\pi}\cosh[\vartheta]\right)\right ] m^2_{\pi}\sin(2m_{\pi}\tau)\right\}\ ; , \label{ficut1}\end{aligned}\ ] ] with @xmath182 the finite temperature counterpart of the zero temperature constants and the bose enhancement factors are a result of the enhancement factors near thresholds as given by eq.([impartnearthresh ] ) . the temperature dependence of the constants @xmath183 is rather weak through the temperature dependence of the pole which can be neglected to the lowest order . then to this order we see that the most important effects of temperature are through the bose - enhancement factors which are the same as in the homogeneous case but with an effective , rapidity dependent temperature @xmath184 $ ] . the second cut gives the following contribution to the amplitude @xmath185 for large time and distances it is dominated by the point @xmath186 . defining @xmath187 , we obtain @xmath188 where the function @xmath189 can be found from the appropriate self - energy after some algebra but will not be relevant for our purposes , and the function @xmath190 is given by @xmath191 @xmath97 has an expression similar to @xmath192 , but with the factor 3 replaced by 9 and @xmath193 by @xmath49 . we will approximate @xmath194 by the ` pole mass ' which at low temperature ( compared to @xmath49 ) can be approximated by @xmath49 as discussed above . it is convenient to change integration variables in eq . ( [ thermalcut ] ) setting @xmath195 . for large @xmath160 and @xmath159 and @xmath196 the contribution from the loop with @xmath0 mesons is negligible , since only the lowest mass intermediate states contribute a long distances and the @xmath0 mesons are not thermally excited . in this case we find @xmath197 ^ 2 } \,\cos(\frac{ty\rho}{r } ) \ ; .\ ] ] using the symmetry of the integrand we can recast this result as @xmath198 ^ 2 } \sin[p(r - t\rho)]\ ; .\ ] ] the integral over @xmath153 yields the third derivative of the dirac delta and we finally obtain @xmath199\cosh^4[\vartheta]}\ ; { { \theta(\tau^2)}\over { \tau^5 } } \ ; \left . { { d^3}\over { d\rho^3}}\left [ \frac{\rho f_{\pi}(\rho , t)}{1+\left[\frac{3g^2}{m^2_{\sigma}}f_{\pi}(\rho , t)\right]^2 } \right ] \right|_{\rho=\tanh[\vartheta ] } \label{asinfi}\ ; .\ ] ] in the general case , there is a similar contribution from the @xmath0 meson loop , but with @xmath200 replaced by @xmath201 as mentioned above . finally the large distance , large time behavior is given by @xmath202 with the asymptotic results given by ( [ asympolet],[ficut1],[asinfi ] ) . we emphasize that the contribution from the thermal cut yields a non - oscillatory contribution decaying as @xmath203 or @xmath204 and is subleading with respect to the contributions from the pole and from the ` normal ' two - particle cut given by eq.([ficut1 ] ) at long distances . thus we obtain one of the important results of this work : the contribution from processes such as @xmath87 ( and the inverse process ) , that can occur only in a plasma of excitations , are asymptotically subleading compared to the one - particle pole and the contributions from processes such as @xmath80 ( and its inverse ) . in this subsection we will compute the time evolution in the case in which the initial state is a gaussian wave packet . that is , we will assume @xmath205 where we included a convenient normalization factor . we have to convolute this initial condition with the results obtained in the previous subsection . namely @xmath206 where @xmath207 is the amplitude for the evolution with the initial condition corresponding to a @xmath208 as studied in the previous sections . within the setting of the chiral phase transition and the possibility of the formation and evolution of dcc s , we are interested in initial packets of spatial sizes @xmath209 ( therefore @xmath210 ) , and the evolution for distances and times @xmath211 . we first analyze the pole contribution . by using ( [ deltapole ] ) in ( [ convolution ] ) and performing the simple integral over @xmath212 , we get the result @xmath213\;.\ ] ] we see that the only difference with ( [ deltapole ] ) is the gaussian factor in the integrand . for @xmath214 the stationary phase points are @xmath215 \left[1 + { \cal o}\left(\frac{m_{\sigma}\xi^2}{\tau}\right)\right ] + { \cal o}(g^2 ) \;. \label{finitesizesaddle}\ ] ] this analysis holds for all other contributions ( from both cuts ) , with the final conclusion that asymptotically for large distances and times compared with the size of the initial packet , the results obtained for the @xmath216-function initial condition can be used with the simple modification of the additional gaussian pre - factor @xmath217 } \;.\label{prefactor}\ ] ] thus at time and distances much larger than @xmath218 we obtain @xmath219 } \phi_{\xi = 0}(\vec x , t , t ) \;. \label{packetevol}\ ] ] only for distances comparable to the size of the initial packet or very close to the light cone do the corrections to the above results arising from the initial size of the packet become important . it is clear that the asymptotic space - time evolution is completely determined by the lowest energy thresholds and the fact that higher mass degrees of freedom are not described by the linear @xmath0 model should not modify the asymptotic behavior for distances larger than a few fm . a remarkable consequence of the saddle point conditions which determine the long time - long distance behavior of the contributions from the multiparticle cuts is that the temperature effects enter in terms of an effective temperature that _ depends on rapidity _ given by @xmath220 } \label{teffofrapid}\ ] ] in the sense that the boltzmann factors that enter into the expressions have this effective temperature dependence . thus the space - time evolution can be interpreted as that the inhomogeneous , non - equilibrium configuration is relaxing with a rapidity dependent temperature that becomes smaller at large rapidities . this is a one - loop result and may be modified by higher order corrections , a possibility worthy of further study . as is well known ( see @xcite ) , if the @xmath0 particle is unstable ( resonance ) , namely if @xmath222 , the pole is above the two - particle threshold and moves off into the second ( unphysical ) riemann sheet at a distance @xmath161 from the imaginary axis in the s - plane . thus in this case we have no pole contribution . however , if the theory is weakly coupled , the pole is very close to the cut , and will give the dominant contribution to the integral over the cut itself . in this case we have to consider only the cut contribution since , as explained in detail for the homogeneous case @xcite , the poles of @xmath223 get a real part and move off into the second riemann sheet in the @xmath65-plane . thus @xmath224 thus in this case we find @xmath225 ^ 2 + \sigma_i(\omega,\vec p , t)^2 } \;. \label{intform}\ ] ] this expression leads to the following sum rule obtained previously by pisarski@xcite in a different manner : @xmath226 ^ 2 + \sigma_i(\omega,\vec p , t)^2 } = 1 \;. \label{sumrule}\ ] ] in this unstable case , this sum rule illuminates the relationship between the landau damping contribution to the absorptive part of the self - energy and the on - shell contribution . notice that the landau damping contribution to the absorptive part of the self - energy @xmath227 given by eq.([impark2 ] ) has the same form as the finite temperature contribution to @xmath228 ( [ impartk1 ] ) . thus the sum rule above determines that the contribution to the spectral density below the light cone , that is the landau damping term given by eq.([impark2 ] ) is borrowed from the spectral density above the physical two - particle cut . therefore , although @xmath227 does not contribute _ directly _ to the width of the meson and its decay rate in the medium , it does so _ indirectly _ through the sum rule . in weakly coupled theories such that @xmath229 the spectral density features a narrow resonance and has the breit - wigner form . thus the integral over the discontinuity feels the pole and can be approximated by a breit - wigner resonance in the second riemann sheet . when the position of the resonance is far away from the two particle thresholds ( many widths ) the integral is dominated by the resonance for long times but eventually the large time behavior will be determined by the behavior of the spectral density at threshold@xcite . in what follows we will focus on the weakly coupled case and address the strongly coupled linear @xmath0 model afterwards . in the case @xmath229 , and after the change of variables @xmath230 the spectral density can be approximated by the breit - wigner form @xmath231 the integral over the discontinuity of the self - energy , given by [ stable ] , can be approximated by @xmath232 where the wave function renormalization @xmath126 and @xmath233 @xmath153-independent and given by @xmath234}^{-1}\ ; , \quad\quad\quad\quad \alpha =- z\frac{\partial\sigma_i(im_0,p=0)}{\partial m^2_0}\;,\ ] ] and the damping rate is given by @xmath235 at one loop order we find from ( [ imkt0 ] ) and ( [ spit0 ] ) @xmath236 the expression ( [ onelupdamp ] ) displays the lorentz contraction of the width ( dilation of the lifetime ) with respect to that in the rest frame . the fourier transform of @xmath119 can be evaluated for large @xmath159 and @xmath160 with the stationary phase method . we find @xmath237}\over \tau } \ ; e^{-\gamma(0 ) \tau } \nonumber \\ & & \quad\times\sin\left\{m_{\sigma}\tau \left[1 + 2\left({{2\gamma(0 ) } \over{m_{\sigma } } } \tanh[\vartheta]\right)^2+{\cal o } \left({{2\gamma(0)}\over{m } } \tanh[\vartheta]\right)^4\right]+\alpha \right\}\ ; .\label{corteunst } \end{aligned}\ ] ] there is now an exponential damping in the amplitude . however for proper times longer than @xmath238 $ ] , the most important contribution to the integral arises from the region near the two - pion threshold leading to a long time tail given by the pion contribution to @xmath239 in equation ( [ inhd ] ) . for an initial gaussian wave packet ( [ gaussian ] ) the field @xmath240 is given for large @xmath159 and @xmath160 by @xmath219 } \phi_{\xi = 0}(\vec x , t , t ) \label{packetevolunst}\ ] ] just as in the stable case because the stationary phase condition is the same as in that case . as shown in equations ( [ totalcut],[cut1],[cut2 ] ) , at finite temperature the are several different cuts which contribute to @xmath241 . the most important contribution arises from the two - particle cut starting at @xmath242 corresponding to the process of @xmath0 decay into two pions and pion recombination . this is clearly seen because the resonance will be at a position @xmath243 because the @xmath0 meson can now kinematically decay ` on - shell ' in two pions . for weak coupling the sharp resonance dominates the integral and one obtains a result for @xmath244 analogous to ( [ inhomunstable ] ) , with the finite temperature self - energy . the two - particle cut starting at @xmath245 and the thermal cut below the light - cone will give a contribution similar to the stable case studied in the previous section . thus we concentrate on studying the contribution from the two - pion cut . since manifest lorentz covariance is lost ( in the rest frame of the thermal bath ) , in the @xmath99 case all the parameters will depend on @xmath153 @xmath246^{-1 } \ ; , \label{zoft } \\ \gamma(p , t ) & = & z(p , t ) { { \sigma_i(i \omega(p , t),p , t ) } \over{2 \omega(p , t ) } } \ ; , \label{dampingratet } \\ \alpha(p , t ) & = & -z(p , t ) \left . \frac{\partial\sigma_i(im , p , t ) } { \partial m^2}\right|_{m= \omega(p , t ) } \ ; , \label{alfat}\end{aligned}\ ] ] where @xmath247 is given by equation ( [ tpolequa ] ) . the inverse fourier transform of @xmath244 can be evaluated for large times and distances with the saddle point method to obtain , in the breit - wigner approximation @xmath248}\over \tau } \ ; e^{-\gamma(\vartheta , t)\,\tau } \ ; \sin[m_{\sigma}\tau + \alpha(h , t ) ] \;\theta(\tau ) \label{fibw}\ ; , \ ] ] where @xmath249 $ ] . more explicitly for weak coupling , @xmath250 takes the form @xmath251 \ , \cosh[\vartheta]}}\ ; \log{{\sinh{{\cal a}_+(\vartheta)}}\over { \sinh{{\cal a}_-(\vartheta)}}}\label{bodrio}\;,\ ] ] with @xmath252 \over { 4 t}}\ ; \left ( 1 \pm \tanh[\vartheta ] \sqrt{1-{{4m^2_{\pi}}\over{m^2_{\sigma } } } } \right)\;.\ ] ] for high temperatures and @xmath253 one finds the remarkable result ( valid in the weak coupling case ) @xmath254 } } \;.\ ] ] notice again that the boltzmann factors depend on the effective temperature @xmath255 $ ] and that @xmath256 is a function of @xmath164 _ and _ @xmath257 . in particular , notice that ( [ bodrio ] ) approaches the zero temperature limit at very large rapidities . for long proper times , @xmath258 , the threshold will dominate the long time- large distance behavior with a contribution which is given by eq . ( [ ficut1 ] ) . the contribution from the thermal cut is _ the same _ as in the stable case and given by eq . ( [ asinfi ] ) . in the weak coupling limit the contribution from the breit - wigner resonance and that of the threshold can just be added to yield the total contribution from the cut , and since the contribution of the thermal cut is added to that of the ` normal ' cut , we finally find in the unstable case : @xmath259 with the asymptotic results given by ( [ fibw],[ficut1],[asinfi ] ) it is now straightforward to show as in the previous section for the stable case , that for an initial gaussian wave packet the final result is just multiplied by the factor @xmath260}$ ] because of the saddle point condition . thus at time and distances much larger than @xmath218 we obtain @xmath261 } \phi_{unst\ ; ; \xi = 0}(\vec x , t , t ) \;. \label{packetevolunstt}\ ] ] the remarkable result of this section is that the space - time description of the evolution of a wave packet that decays , is that the effective decay rate is a function of rapidity which itself depends on the rapidity dependent effective temperature @xmath257 . we recall that in a medium this decay rate _ does not _ give the rate of production of pions through the decay of the sigma meson , but the _ difference _ between the production and annihilation rate as discussed previously in section ii . let us now discuss the strongly coupled case . the approximations and conclusions obtained above are within the framework of a weakly coupled theory . as was discussed in section ii , the imaginary part on shell for the linear sigma model description of pion physics is very large ( see equation [ impartonshell ] ) , thus the validity of the breit - wigner approximation to describe the space - time evolution must be called into question . figure 2 shows @xmath262 as a function of @xmath263 for @xmath264 compared to the breit - wigner approximation . we see that in most of the range , but in a small region near the lowest threshold the breit - wigner approximation is excellent . we find the same result for temperatures all the way up to the pion mass and for a large range of momenta from very small to very large compared to @xmath49 . clearly what is less accurate within the breit - wigner approximation is the extension of the integration region in @xmath177 to @xmath265 because the resonance is rather broad . however one can extend the integration region and subtract the contribution from threshold , which is the dominant one at large rapidities and proper times . this is precisely what was done to arrive to the result given by eq . ( [ totalfitunst ] ) above . thus we conclude that even in the strong coupling case the asymptotic behavior is well approximated by eq.([packetevolunstt ] ) for the case of a gaussian initial packet . we have studied the relaxational dynamics of an inhomogeneous condensate fluctuation in the o(4 ) linear sigma model near the broken symmetry state , both at zero and non - zero temperature . explicit expressions are obtained for the self - energies at zero and finite temperature and we point out that at finite temperature there are new relaxational processes with origin in thermal cuts and that are * only * present in the inhomogeneous case with no counterpart in the relaxation of an homogeneous condensate . for initial gaussian fluctuations we have given explicit expressions for the asymptotic space - time evolution of the inhomogeneous fluctuation including the effect of thresholds both at zero and finite temperature . at finite temperature we obtained the decay rate of this inhomogeneous non - equilibrium configuration , in the medium this quantity is the rate of production _ minus _ the rate of absorption of pions . the space - time evolution is described in terms of an effective temperature and `` decay rate '' that depend on rapidity . we find to one loop that relaxational processes are described in terms of @xmath266 $ ] and @xmath2 , and for large rapidities the finite temperature decay rate approaches the zero temperature limit . we systematically compute the field behaviour for large times and distances compared with the inverse of the typical mass scale ( @xmath267 ) in the model . this means @xmath268 * and * @xmath269 . for @xmath270 the behaviour will be quantitatively different . * * stable case : @xmath273 * + @xmath274 . : : the pole contribution @xmath275 dominates , giving an amplitude that decays as @xmath276 and oscillates as a function of @xmath271 with frequency @xmath277 ( physical @xmath0 mass ) . the cut gives contributions smaller than @xmath278 by a factor @xmath276 and oscillating with frequencies equal to the threshold positions [ eqs.([asympole]-[inhd ] ) ] . @xmath99 . : : the pole contribution dominates for large @xmath166 . @xmath278 decays with the same power @xmath276 as for @xmath274 , but the amplitude and frequency change quantitatively [ see eq.([asympolet ] ) ] . + a new cut running from @xmath279 to @xmath280 appears at non - zero temperature . it gives a * non - oscillating * contribution for large @xmath166 that decays as @xmath281 [ see eq.([asinfi ] ) ] . * * unstable case : @xmath282 * + @xmath274 . : : @xmath42 is now a resonance ( a pole in the second riemann sheet ) yielding an * exponentially damped * amplitude that oscillates with frequency @xmath283 . the damping being the width of the resonance [ see eq.([corteunst ] ) ] . + eventually , for very large @xmath271 , the power - like tail coming from the cut contribution will dominate over the exponentially damped contribution of the resonance . @xmath99 . : : it is similar as for @xmath274 , except that the damping rate becomes a non - trivial function of @xmath284 and @xmath277 given by eq.([bodrio ] ) . + the thermal cut with its non - oscillating contribution to @xmath272 is also present here and dominates for @xmath285 . all the asymptotic results hold to first order in the field amplitude . however , since the field vanishes for @xmath286 , our results are true for * any initial amplitude * provided @xmath271 is large enough . the relaxation of an initially gaussian inhomogeneous fluctuation is described in terms of the spreading of the packet and decay in spherical waves , and we found that the time scale for relaxation , production and absorption of pions is a function of rapidity such that for larger rapidities the relaxational and decay processes are slower . we are currently extending these studies to the case of non - equilibrium fluctuations produced during the stage of parametric amplification as the @xmath0 rolls to the ground state@xcite . we believe that the techniques developed in this work and the non - equilibrium aspects found here will be of use in a variety of physical contexts . one that comes to mind concerns the emission of goldstone bosons from an axion string@xcite . this emission is important in determining the exact upper bound on the axion decay constant , and thus on whether axion models for solving the strong cp problem are still viable . this problem will entail some modifications of the tools described here , the most notable one having to do with the existence of zero modes for the string configuration . m. da . has been supported in part by a della riccia fellowship and would like to thank lpthe ( paris vi - vii ) for kind hospitality . d. b. gratefully acknowledges partial support from n.s.f . through grants : phy-9302534 and int-9216755 ( binational collaboration with france ) . he also thanks the warm hospitality at lpthe where part of this work was done and r. pisarski and h. a. weldon for illuminating conversations during the early stages of this work . m. da . and h. j. de v. acknowledge support from the european commission through the human and mobility program erbchrx - ct94 - 0488 . r.h . was supported in part by doe grant @xmath287 de - fg02 - 91er40682 . acta physica polonica b23 , 561 ( 1992 ) . j. p. blaizot and a. krzywicki , phys . d46 ( 1992 ) , 246 . k. l. kowalski and c. c. taylor , ` disoriented chiral condensate : a white paper for the full acceptance detector ' cwru report 92- he - ph/9211282 ( unpublished ) . j. d. bjorken , k.l . kowalski and c. c. taylor , ` baked alaska ' , proceedings of les rencontres de physique de la valle daoste , la thuile ( 1993 ) ; ( slac pub 6109 ) . y. kluger , j. m. eisenberg , b. svetitsky , f. cooper and e. mottola , phys . rev . d 51 , 4419 ( 1995 ) . k. rajagopal and f. wilczek , nucl . b399 , 395 ( 1993 ) . k. rajagopal and f. wilczek , nucl . b404 , 577 ( 1993 ) . s. gavin , a. gocksch and r. pisarski , phys . 72 , 2143 ( 1994 ) . these authors have studied the non - equilibrium evolution of expectation values that only depend on proper time but not on rapidity as suggested by the landau model of the collision . s. gavin and b. muller , phys . b 329 , 486 ( 1994 ) . s. mrowczynski and b. muller , phys . lett . b 363 , 1 ( 1995 ) . d. boyanovsky , h. j. de vega and r. holman , phys . rev . d 51 , 734 ( 1995 ) . see for example : a. vilenkin and e.p.s . shellard , ` cosmic strings and other topological defects ' , cambridge mon . on math . , cambridge univ . press ( 1994 ) . m. b. hindmarsh and t. w. b. kibble , s. p. klebansky , rev . of mod . 64 , 649 ( 1992 ) . p. f. bedaque and a. das , mod phys . a8 , 3151 ( 1993 ) . s. weinberg , phys . d 9 , 3357 ( 1974 ) . weldon , phys . d 28 , 2007 ( 1983 ) ; ann . of phys . 228 , 43 ( 1993 ) . blaizot in ` proceedings of fourth summer school and symposium on nuclear physics ' , eds . min and m. rho ( 1991 ) . w. keil , phys . d40 , 1176 ( 1989 ) . this mathematical condition is imposed by a lagrange multiplier that cancels the ( non - local ) integral up to @xmath57 and that vanishes for @xmath288 . since the field configuration will decay , this must be treated as an initial value problem and this particular condition defines such a problem . for a discussion on this point see : one - loop feynman diagrams contributing to the equation of motion . the dashed line corresponds to the insertion of the background field @xmath42 . thin lines correspond to pion propagators , thick lines to @xmath0 propagators . the tadpoles had been absorbed in mass and @xmath21 renormalization .

surface properties of mixtures of charged platelike colloids and salt in contact with a charged planar wall are studied within density functional theory . the particles are modeled by hard cuboids with their edges constrained to be parallel to the cartesian axes corresponding to the zwanzig model and the charges of the particles are concentrated in their centers . the density functional applied is an extension of a recently introduced functional for charged platelike colloids . analytically and numerically calculated bulk and surface phase diagrams exhibit first - order wetting for sufficiently small macroion charges and isotropic bulk order as well as first - order drying for sufficiently large macroion charges and nematic bulk order . the asymptotic wetting and drying behavior is investigated by means of effective interface potentials which turn out to be asymptotically the same as for a suitable neutral system governed by isotropic nonretarded dispersion forces . wetting and drying points as well as predrying lines and the corresponding critical points have been located numerically . a crossover from monotonic to non - monotonic electrostatic potential profiles upon varying the surface charge density has been observed . due to the presence of both the coulomb interactions and the hard - core repulsions , the surface potential and the surface charge do not vanish simultaneously , i.e. , the point of zero charge and the isoelectric point of the surface do not coincide .

cond-mat0607321

clay particles are examples of charged colloidal plates with diameters in the range from nanometer to micrometer . besides their technological importance in applications such as agriculture ( as soil ) , construction ( in concrete ) , filling ( in cosmetics , rubber , plastic , etc . ) , coating ( of paper ) , and oil - drilling ( as rheological fluids for oil recovery ) , clay suspensions play an important role in basic research because they exhibit a complex phase behavior in terms of packing and orientation of the platelike particles @xcite . examples include sol - gel transitions , flocculation , aging , and even liquid crystalline phases depending on numerous parameters such as particle size , shape , charge , polydispersity , ionic strength , solvent , temperature , and external fields . the challenge is to understand the relation between these parameters and the phase behavior in order to control the properties of such fluids . most studies in this direction have been devoted to _ bulk _ properties . some experimental progress has been made due to advances in the synthesis of well defined and experimentally convenient model systems @xcite . for a recent review of these investigations see , e.g. , ref . computer simulations for platelike particles in bulk systems have been performed in recent years @xcite , facing particular challenges as compared with systems consisting of rodlike particles . most theoretical work has been concentrated on the investigation of models of hard platelets @xcite and anisotropically charged discs @xcite in spatially homogeneous configurations . in contrast to the meanwhile well studied case of bulk suspensions of platelike particles , only a few theoretical investigations of such systems with _ spatial inhomogeneities _ such as interfaces , surfaces , or due to external fields have been undertaken @xcite because of the technical difficulties implied by reduced translational symmetry . on the other hand , understanding the influence of surfaces is of particular importance for applications because , e.g. , omnipresent walls have a strong influence on this sort of material and electrodes are a common means to manipulate fluids of charged particles . a first important theoretical step for understanding such surface properties requires to determine the corresponding surface phase diagrams . whereas surface phase transitions for simple fluid models have been investigated for decades @xcite , no such studies are available for complex fluids involving charges _ and _ anisotropically shaped particles . here we present the investigation of a density functional theory for a fluid of charged platelike colloids and salt in the presence of a charged hard wall . due to the expected spatial inhomogeneities of the fluid close to the surface , density functional theory @xcite is the method of choice for addressing these issues @xcite . recently , we have proposed a density functional for fluids of charged platelike colloids which also takes long - ranged coulomb interactions into account @xcite . here a straightforward extension of that functional is presented in order to capture the influence of a charged hard wall . as introduced in ref . @xcite , the fluid is modeled by a mixture of colloidal rectangular plates and cubelike salt ions with the platelet orientation restricted to three mutually perpendicular directions ( zwanzig model @xcite ) . the charges are concentrated in the centers of the particles . the bulk phase diagram of this model exhibits one isotropic and one nematic phase separated by a first - order phase transition @xcite . the salt concentration at coexistence of the isotropic and the nematic phase is larger for the isotropic than for the nematic state @xcite . this phenomenon is well - known for membrane equilibria @xcite where it is commonly referred to as the donnan effect @xcite . by bringing an isotropic and a coexisting nematic bulk system into spatial contact , an electric double layer forms at the free interface . for negatively charged platelets , a negatively charged layer forms on the nematic side of the free interface whereas a positively charged layer is found on the isotropic side @xcite . the corresponding electrostatic potential profile changes from one side of the interface to the other by the so - called donnan potential @xcite . the present analysis is devoted to the investigation of the model fluid studied in ref . @xcite in contact with a hard wall which acquired charges upon releasing counterions into the fluid . section [ sec : generalformalism ] presents the model and extends the functional used in ref . @xcite in order to incorporate a wall potential . in sec . [ sec : phasediagram ] , the resulting bulk and surface phase diagrams are discussed . further details concerning the wetting ( drying ) behavior of a charged wall in contact with an isotropic ( nematic ) bulk system are given in sec . [ sec : wetting ] ( [ sec : drying ] ) . electrostatic potential profiles and the electrostatic surface potential are investigated in sec . [ sec : electrostaticpotential ] . section [ sec : summary ] summarizes the results . we consider a fluid of charged platelike colloids and salt in three - dimensional space confined by a single planar structureless charged hard wall . the fluid model used here is the same as in ref . @xcite . it is a ternary mixture of charged hard square cuboids ( see macroions @xmath0 are square cuboids of size @xmath1 , @xmath2 with charge @xmath3 whereas salt ions @xmath4 and counterions @xmath5 are cubes of side length @xmath6 with charges @xmath7 and @xmath8 , respectively . the pointlike charges ( @xmath9 ) are localized at the centers of the cuboids . the macroions can adopt three possible orientations @xmath10 , @xmath11 , and @xmath12 corresponding to the @xmath13-edges being parallel to the @xmath14- , @xmath15- , and @xmath16-axis , respectively . ] [ fig : geo ] ) with their edges parallel to the cartesian axes ( zwanzig model @xcite ) and dissolved in a dielectric solvent ( e.g. , water ) with dielectric constant @xmath17 . the solvent is treated as a continuum . for simplicity , the charges are fixed , monodisperse , and concentrated at the centers of the particles . the particles of the first component represent the macroions @xmath0 of size @xmath18 , @xmath2 , and charge @xmath3 . within the zwanzig approximation , macroions can take three different orientations , denoted as @xmath10 , @xmath11 , or @xmath12 corresponding to whether the @xmath13-edges are parallel to the @xmath14- , @xmath15- , or @xmath16-axis , respectively ( see fig . [ fig : geo ] ) . the second component consists of salt ions @xmath4 modeled as cubes of side length @xmath6 and charge @xmath7 ( see fig . [ fig : geo ] ) . the salt ion charge @xmath7 is chosen to have the same sign as the macroion charge @xmath3 if the latter does not vanish . finally , the third component consists of counterions @xmath5 guaranteeing overall charge neutrality . they are also described by cubes with the same side length @xmath6 but opposite charge @xmath19 ( see fig . [ fig : geo ] ) . due to the choice of the same sign for @xmath3 and @xmath7 , ions of component @xmath5 are counterions in the usual sense for both the macroions @xmath0 and the salt ions @xmath4 . the wall , which confines the fluid from one side , is modeled as impenetrable for the fluid particles . moreover , it carries a fixed homogeneously distributed surface charge of areal density @xmath20 which , depending on the sign of @xmath20 , can be assumed to have been generated by releasing an appropriate amount of ions of either component @xmath4 or component @xmath5 into the fluid . @xmath21 denotes the number density at point @xmath22 of the centers of macroions with orientation @xmath23 , salt ions , and counterions , respectively . note that the position @xmath24 with @xmath25 denoting the system volume is a _ continuous _ variable in contrast to the orientation of macroions , which varies within a _ discrete _ set . we introduce the abbreviation @xmath26 . the system under consideration is coupled to two particle reservoirs : one supplies neutralized macroions ( chemical formula @xmath27 ) and the other neutral salt ( chemical formula @xmath28 ) ; @xmath29 and @xmath30 denote the corresponding chemical potentials . these molecules dissociate upon entering the system : @xmath31 these equilibrium chemical reactions lead to the following relations between the reservoir chemical potentials ( @xmath29 and @xmath30 ) and the particle chemical potentials ( @xmath32 ) : @xmath33 configurations of the system are characterized by the set of number density profiles @xmath34 . the equilibrium states minimize the grand canonical density functional @xcite @xmath35 & = & \sum_i\int{v}{3}{r}\rho_i(\vec{r})\big(\ln(\rho_i(\vec{r } ) ) - 1 - \mu^*_i + v_i(\vec{r})\big ) \nonumber\\ & & + f^\m{ex}[\set{\rho } ] + u^\m{self}(\sigma ) , \label{eq : df1}\end{aligned}\ ] ] where @xmath36 is the wall ( or substrate ) potential exerted on particles of type @xmath37 , @xmath38 is the free energy in excess over the ideal gas contribution , and @xmath39 is the self energy of the charged wall . the latter is _ independent _ of @xmath34 and hence merely shifts @xmath40 by a constant value . it has been introduced for later convenience . finally , the reduced particle chemical potentials @xmath41 with the thermal de broglie wavelength @xmath42 ( @xmath43 ) for particles of type @xmath37 have been introduced . with the reduced reservoir chemical potentials @xmath44 eq . takes the form @xmath45 for given reservoir chemical potentials @xmath46 and @xmath47 , the particle chemical potentials @xmath48 are fixed by eq . and the constraint of global charge neutrality : @xmath49 where @xmath50 is the area of the charged surface . the euler - lagrange equations of the minimization problem read @xmath51 with the one - particle direct correlation function @xmath52 equation has to be solved under suitable boundary conditions in accordance with the bulk states for given reservoir chemical potentials @xmath53 and @xmath54 . these bulk phases have already been determined in ref . @xcite . as introduced in ref . @xcite , the interactions between charges are approximated by @xmath55 where the usual euclidean norm @xmath56 is replaced by the supremum norm @xmath57 which is the adapted norm in the context of a zwanzig model for cuboids . employing a debye charging process , @xmath38 can be expressed as @xcite @xmath58 the first term in eq . is the fundamental measure functional introduced by cuesta and martnez - ratn @xcite : @xmath59 : = \int{v}{3}{r}\phi(\set{n}(\vec{r } ) ) \label{eq : fexh}\ ] ] with the weighted densities @xmath60 for @xmath61 and the excess free energy density @xmath62 it describes the free energy due to the hard - core interactions @xmath63 between two particles of type @xmath37 and @xmath64 at positions @xmath22 and @xmath65 , respectively . @xmath63 is infinite if the hard cores of these particles overlap and zero otherwise . the electrostatic part @xmath66 : = \frac{1}{2}\sum_{ij}\int{v}{3}{r}\int{v}{3}{r'}\rho_i(\vec{r})\rho_j(\vec{r ' } ) u^\m{c}_{ij}(\vec{r}-\vec{r ' } ) \label{eq : fexcel}\ ] ] accounts for the particle - particle coulomb interaction within the random phase approximation ( rpa ) whereas the correlation part @xmath67 & : = & \frac{1}{2}\sum_{ij}\int{v}{3}{r}\int{v}{3}{r'}\rho_i(\vec{r})\rho_j(\vec{r ' } ) u^\m{c}_{ij}(\vec{r}-\vec{r ' } ) \big ( \exp(-u^\m{h}_{ij}(\vec{r}-\vec{r'}))-1 \nonumber\\ & & + \exp(-u^\m{h}_{ij}(\vec{r}-\vec{r ' } ) ) g_{ij}(\kappa(\vec{r},\vec{r'};[\set{\rho}]),\snorm{\vec{r}-\vec{r ' } } ) \big ) \label{eq : fexccorr}\end{aligned}\ ] ] with the function @xmath68 and the screening factor @xmath69 ) : = \frac{1}{2}(\tilde\kappa(\vec{r};[\set{\rho } ] ) + \tilde\kappa(\vec{r'};[\set{\rho } ] ) ) \label{eq : kappa}\ ] ] using @xmath70 ) : = \sqrt{4\pi q_s^2(\rho_s(\vec{r } ) + \rho_c(\vec{r } ) ) } \label{eq : kappatilde}\ ] ] represents corrections beyond rpa . since @xmath71 ( see ref . @xcite ) , the contribution @xmath72 in eq . _ beyond _ rpa due to the coulomb interactions has the _ same _ @xmath73 asymptotics and sign as the contribution _ within _ rpa of nonretarded dispersion forces . the amplitudes of nonretarded van der waals forces for typical hamaker constants of clay ( e.g. , @xmath74 for @xmath75 @xcite ) are two orders of magnitude smaller than those of @xmath76 for @xmath77 . moreover , for long distances ( @xmath78 ) dispersion forces cross over to an @xmath79 asymptotics due to retardation @xcite . dispersion forces are negligible in this retardation limit compared to @xmath76 . for these reasons we have not included dispersion forces into our model . since the thermodynamic limit of globally charge neutral systems of hard particles with coulomb interactions exists @xcite , i.e. , the bulk free energy density exists and it depends neither on the shape nor on the nature of the boundaries of the system volume @xmath25 , the system volumes @xmath80 $ ] of size @xmath81 are considered in the limit @xmath82 . here @xmath83 is a square in the @xmath14-@xmath15 plane of side length @xmath84 , i.e. , @xmath85 , with ( lateral ) periodic boundary conditions . furthermore , the charged hard surface is chosen as the intersection of @xmath86 with the plane @xmath87 . the notation @xmath88 in the above definition of @xmath86 is used in order to emphasize the fact that the charged wall at @xmath87 belongs _ entirely _ to the system volume . the wall potential @xmath36 for particles of type @xmath37 can be decomposed as @xmath89 with the hard wall contribution @xmath90 0 & , z > \frac{s^z_i}{2 } \end{array}\right . , \label{eq : vh}\ ] ] where @xmath91 is the @xmath16-extension of particles of type @xmath37 , and the contribution due to the surface charge @xmath92 similarly , the wall self energy is given by @xmath93 introducing the local charge density @xmath94 and the electrostatic potential @xmath95 one readily finds @xmath96 according to ref . @xcite , the expression for the electrostatic potential reduces to @xmath97 due to the lateral translational symmetry . by differentiating twice one can show that @xmath98 fulfills the poisson equation @xmath99 combined with eqs . and , the density functional in eq . takes the final form @xmath35 & = & |a|\sum_i\int\limits_{0 ^ -_{\vphantom{|}}}^l\!\d z\ ; \rho_i(z ) \big ( \ln(\rho_i(z ) ) - 1 - \mu^*_i \nonumber\\ & & + v_i^\m{h}(z ) + \frac{1}{2}q_i\psi(z)\big ) + \frac{1}{2}|a|\sigma\psi(0 ) \nonumber\\ & & + f^\m{ex , h}[\set{\rho } ] + f^\m{ex , c}_\m{corr}[\set{\rho } ] \label{eq : df3}\end{aligned}\ ] ] which has to be minimized under the constraint of global charge neutrality @xmath100 the corresponding euler - lagrange equations read @xmath101 expanding the euler - lagrange equations ( eq . ) for @xmath102 around the bulk values provides the number density deviations @xmath103 with @xmath104 the electrostatic potential difference @xmath105 depends on @xmath106 via the local charge density @xmath107 ( see eqs . and ) . however , this dependence will not be used for the derivation of , c.f . , eq . . inserting eqs . and into eq . leads to @xmath108 and @xmath109 where @xmath110 . the spatially constant @xmath111-matrices @xmath112 and @xmath113 depend only on model parameters ( particle sizes and charges ) and bulk quantities ( see appendix a ) . by defining @xmath114 and since @xmath115 for @xmath116 ( eq . ) , eq . is equivalent to @xmath117 firstly , eq . implies an asymptotic decay of @xmath118 _ not faster _ than proportional to @xmath119 ; otherwise @xmath120 with some spatially constant amplitude @xmath121 would fulfill the relation @xmath122 _ simultaneously _ for _ all _ @xmath37 which is impossible because the second term depends on @xmath37 whereas the first does not . moreover , it can be shown that the deviations @xmath118 exhibit a purely algebraic decay in leading order . finally , one is led to the conclusion that the asymptotic euler - lagrange equations in conjunction with the poisson equation leads to the properties @xmath123 , @xmath124 , and @xmath125 . note that on the right - hand side of eq . for @xmath102 the leading and the next to leading order contributions to the density profiles @xmath34 cancel . interestingly , the asymptotic decay proportional to @xmath119 of @xmath118 within the current model for charged particles equals the corresponding one for systems governed by isotropic nonretarded van der waals forces in ref . this behavior is in sharp contrast to the results obtained within multicomponent poisson - boltzmann theories which give rise to exponentially decaying density profiles @xcite . upon solving the euler - lagrange equations ( eq . ) for mixtures of platelike macroions and monovalent salt in contact with a charged hard wall , one finds the equilibrium state of this system . here the equilibrium states are calculated as function of the chemical potential @xmath46 , the macroion charge @xmath3 , and the surface charge density @xmath20 . the remaining parameters of the model have been fixed as @xcite ( see fig . [ fig : geo ] ) @xmath126 , @xmath127 , @xmath128 , @xmath129 , and salt density @xmath130 in the bulk . as discussed in ref . @xcite , the current model yields reasonable results only in the range @xmath131 , i.e. , close to the point of zero charge of the colloidal platelets . it turns out that the equilibrium states exhibit rotational symmetry with respect to the wall normal ( @xmath16-axis ) , i.e. , @xmath132 . therefore , the local equilibrium structure of the macroions is captured completely by the two densities @xmath133 and @xmath134 . equivalently , the ( total ) macroion density @xmath135 and the nematic order parameter along the @xmath16-axis @xmath136\ ] ] may serve to identify various structures . the definition of @xmath137 coincides with the well - known scalar liquid - crystal order parameter @xmath138 for the special case of a zwanzig model , within which the only possible macroion orientations are parallel ( @xmath12 , @xmath139 ) and perpendicular ( @xmath140 , @xmath141 ) to the @xmath16-axis . therefore , structures with @xmath142 and @xmath143 are called _ isotropic _ and _ nematic _ , respectively . bulk and surface phase diagram of mixtures @xcite of platelike macroions ( @xmath144 ) and monovalent salt ( @xmath145 ) for salt density @xmath146 in contact with a charged hard wall of surface charge density @xmath147 in terms of macroion charge @xmath3 and chemical potential difference @xmath148 . the bulk equilibrium states for @xmath149 and @xmath150 are isotropic ( @xmath151 ) and nematic ( @xmath152 ) , respectively . isotropic - nematic bulk coexistence corresponds to @xmath153 ( solid line ) . for isotropic boundary conditions in the bulk ( @xmath102 ) at isotropic - nematic coexistence ( @xmath154 ) , two first - order wetting transition points @xmath155 ( at @xmath156 $ ] ) and @xmath157 ( at @xmath158 $ ] ) have been found . the tolerance intervals are indicated by frames of corresponding widths . complete wetting by the nematic phase occurs for @xmath159 upon approaching coexistence from the isotropic side . for nematic boundary conditions in the bulk at isotropic - nematic coexistence ( @xmath160 ) , a first - order drying transition point @xmath161 ( at @xmath162 ) has been found . the accompanying predrying line ( dashed line ) terminates at a critical point @xmath163 ( at @xmath164 ) . complete drying by the isotropic phase occurs upon approaching coexistence from the nematic side for @xmath165 . this implies that for @xmath166 or @xmath167 there is neither complete wetting ( by the nematic phase ) nor complete drying ( by the isotropic phase ) . a second drying transition point @xmath168 appears ( not shown ) for @xmath3 sufficiently large and @xmath20 sufficiently small so that there is reentrance of complete drying for large positive values of @xmath3 . ] figure [ fig : a ] displays the bulk and surface phase diagram for the surface charge density @xmath169 in terms of the macroion charge @xmath3 and the chemical potential difference @xmath170 . the solid line ( @xmath171 ) denotes the states of bulk coexistence between the isotropic ( @xmath151 ) phase and the nematic ( @xmath152 ) phase , corresponding to the chemical potential @xmath172 at coexistence . the bulk equilibrium states for @xmath149 and @xmath150 are isotropic and nematic , respectively . at isotropic - nematic bulk coexistence with isotropic boundary conditions in the bulk ( @xmath154 ) , two first - order wetting transition points @xmath155 and @xmath157 have been found which could be located only within the intervals of the coexistence line indicated by frames of corresponding widths : whereas @xmath155 ( @xmath156 $ ] ) is known rather precisely , there remains some uncertainty with respect to the location of @xmath157 ( @xmath158 $ ] ) . the corresponding prewetting lines are so close to the coexistence line such that they could not be resolved numerically . complete wetting by the nematic phase occurs for approaching isotropic - nematic bulk coexistence from the isotropic side for @xmath173 . the phenomenon that only _ partial _ wetting is found for sufficiently large macroion charges @xmath174 may be qualitatively understood as follows : the macroion number density profiles @xmath133 , @xmath175 , and @xmath134 close to the surface are influenced by the hard - core interactions which give rise to a preference of nematic order close to the wall , the macroion - substrate coulomb interactions proportional to @xmath174 , and the macroion - macroion coulomb repulsion proportional to @xmath176 . the latter dominates for large macroion charges @xmath174 leading to a depression of the values of the macroion number densities near the wall which in turn prevents the growth of a nematic film . for small macroion charges @xmath174 , @xmath133 , @xmath175 , and @xmath134 near the surface are determined by the balance between the hard - core interactions and the macroion - substrate interactions which lead to complete wetting for attractive walls . further details of the wetting behavior will be given in sec . [ sec : wetting ] . in fig . [ fig : a ] , a first - order drying transition point @xmath161 at @xmath162 is found for isotropic - nematic bulk coexistence with nematic boundary conditions in the bulk ( @xmath160 ) . the first - order character of the drying transition at state point @xmath177 implies the existence of a predrying line ( dashed line in fig . [ fig : a ] ) , along which the excess adsorption of macroions @xmath178 which is proportional to the thickness of the emerging film , exhibits a finite discontinuity . the predrying line is expected to meet the isotropic - nematic bulk coexistence line tangentially @xcite . our numerical data are consistent with this behavior but not definitely conclusive due to the numerical difficulties arising from the large thickness of drying films very close to isotropic - nematic bulk coexistence . the predrying line terminates at a critical point @xmath163 located at state point @xmath179 . complete drying by the isotropic phase occurs for @xmath165 upon approaching isotropic - nematic bulk coexistence from the nematic side . as for the wetting scenario discussed above , the macroion - macroion repulsion will prevail over the macroion - surface interactions for a sufficiently large macroion charges @xmath174 . therefore , for large @xmath174 , the formation of an isotropic film is initiated by the strongly depressed macroion number density close to the surface . hence one expects complete drying for sufficiently large macroion charges @xmath174 . for a surface charge density @xmath169 ( see fig . [ fig : a ] ) , _ partial _ drying is found for @xmath180 $ ] , i.e. , the expected second drying transition point @xmath168 is located in the range @xmath181 . for smaller surface charge densities @xmath20 , indeed _ two _ drying transition points have been found within the range @xmath182 $ ] . more details of the first - order drying transition and the corresponding predrying line can be found in sec . [ sec : drying ] . within the intervals @xmath183 $ ] and @xmath184 $ ] only partial wetting and drying occurs . in order to validate the topology of the bulk and surface phase diagram in fig . [ fig : a ] , we investigated a modified version of the density functional ( eq . ) in which the correlation term @xmath185 $ ] is omitted . an asymptotic analysis analogous to sec . [ sec : asym ] leads to exponentially decaying profiles of @xmath106 , @xmath105 , and @xmath107 . the corresponding phase diagram is qualitatively the same as in fig . [ fig : a ] . in particular , there are also first - order wetting and drying transition points which are separated by intervals of only partial wetting _ and _ drying . due to the exponentially decaying electrostatic potential , this modified model is similar to a recently investigated model of hard rods interacting with an exponentially decaying wall potential @xcite . whereas the authors of ref . @xcite relied entirely on a numerical approach , which provided them only with _ evidences _ of first - order wetting transitions , we shall show analytically that the wetting and drying transitions shown in fig . [ fig : a ] are of first order ( secs . [ sec : wetting ] and [ sec : drying ] ) . in this section , we discuss the wetting behavior of the model fluid of platelike macroions and salt in contact with a charged hard wall . for a wetting scenario , the boundary conditions imposed on the solutions of the euler - lagrange equations ( eq . ) require the isotropic bulk structure far from the wall . isotropic - nematic coexistence with such boundary conditions is denoted as @xmath154 . in principle , the asymptotic wetting behavior can be completely inferred from the _ effective interface potential _ @xmath186-\omega_b)/|a|$ ] where @xmath187 denotes the _ bulk _ contribution to the grand potential and the density profiles @xmath188 are the solutions of the euler - lagrange equations ( eq . ) under the _ constraint _ of a prescribed film thickness @xmath189 characterized by the position of the isotropic - nematic interface @xcite . since exact solutions are out of reach for the present model , we have determined @xmath190 approximately by considering the following subspace of density profiles @xcite : @xmath191 where @xmath192 and @xmath193 correspond to the wall - nematic and the free nematic - isotropic density profiles with the surface and the interface located at @xmath87 , respectively . therefore , in the interval @xmath194 $ ] the trial density profile @xmath195 is described by the wall - nematic profile whereas in the interval @xmath196 it is given by the free nematic - isotropic profile shifted to position @xmath189 . due to @xmath197 , the discontinuity of @xmath195 at @xmath198 vanishes in the limit @xmath199 . the transition regime around @xmath198 does not contribute to the leading asymptotic terms of @xmath200 @xcite . substituting the trial density profiles @xmath195 into the density functional in eq . and using the asymptotic behavior of @xmath201 and @xmath202 ( sec . [ sec : asym ] ) leads to the effective interface potential @xmath203 with the amplitudes @xmath204 and @xmath205 being independent of @xmath189 . since the density profiles decay proportional to @xmath119 towards the bulk values ( sec . [ sec : asym ] ) , the results of ref . @xcite can be used directly : @xmath204 depends only on the particle charges and bulk densities whereas @xmath205 , in addition , contains contributions due to the wall - nematic and the free nematic - isotropic excess adsorption ( see appendix a ) . the only difference between the density functional in eq . and the one investigated in ref . @xcite is the presence of the electrostatic term eq . . it can be shown that it merely contributes a term @xmath206 to @xmath207 due to the more rapid asymptotic decay of the charge density @xmath125 . by inspection one recognizes the _ same _ leading asymptotic decay of @xmath207 in eq . as for systems governed by isotropic _ nonretarded dispersion forces _ we regard this as an surprising result because our model ( sec . [ sec : fex ] ) does _ not _ include dispersion forces . the asymptotic behavior @xmath208 in eq . is ultimately generated by the assumption of a pair distribution function @xmath209 which decays asymptotically with a debye - hckel ( screened coulomb ) form @xcite . this assumption concerning the pair distribution function @xmath209 is valid because the linearization approximation underlying the debye - hckel theory is justified at large distances . minimizing the effective interface potential @xmath190 in eq . with respect to the interface position @xmath189 leads to the ( equilibrium ) excess adsorption @xmath210 for @xmath211 as long as @xmath212 . evaluating the analytic expression for @xmath204 along the isotropic - nematic coexistence line , which depends only on bulk quantities ( see ref . @xcite and appendix a ) , one finds two wetting transition points corresponding to @xmath155 and @xmath157 in fig . [ fig : a ] with complete wetting , i.e. , @xmath213 , in between . excess adsorption @xmath214 of platelike macroions along vertical thermodynamic paths in fig . [ fig : a ] characterized by fixed macroion charges @xmath3 and parameterized by the chemical potential difference @xmath215 , which measures the thermodynamic distance from isotropic - nematic bulk coexistence . ( a ) the numerically determined excess adsorption @xmath214 remains finite upon @xmath211 for @xmath216 ( see also fig . [ fig : b ] ) , whereas a divergence is suggested for the macroion charges @xmath217 and @xmath218 , i.e. , there is complete wetting of the surface by a nematic film . ( b ) the comparison of the numerical solution for @xmath217 ( dashed line ) with the analytical asymptotic power law behavior @xmath219 ( dash - dotted line ) ( see main text ) indicates that the asymptotic regime is not yet reached within the numerically accessible range of undersaturations . ] figure [ fig : j ] displays the excess adsorption @xmath214 ( eq . ) as a function of the chemical potential difference @xmath149 for fixed macroion charges @xmath3 . for @xmath211 , isotropic - nematic bulk coexistence is approached along vertical thermodynamic paths in fig . [ fig : a ] . in fig . [ fig : j](a ) the curves for @xmath217 and @xmath218 suggest a divergence of @xmath214 in this limit , i.e. , _ complete wetting _ of the surface by a nematic phase occurs for these macroion charges . on the other hand , for @xmath216 the excess adsorption remains finite , i.e. , _ partial wetting _ occurs . the complete wetting curves for @xmath217 and @xmath218 in fig . [ fig : j](a ) exhibit _ no discontinuity _ in the shown range . thus , the prewetting lines attached to the first - order wetting transition points @xmath155 and @xmath157 must be closer to the isotropic - nematic coexistence line than the numerically accessible values of @xmath148 . figure [ fig : j](b ) compares the analytically obtained asymptotic behavior for the case @xmath217 ( dash - dotted line ) with the corresponding numerical solution ( dashed line ) . the differences between these curves indicate that the ultimate asymptotic regime is not yet reached within the numerically accessible range of undersaturations . the order of the wetting transitions at state points @xmath155 and @xmath157 in fig . [ fig : a ] is determined by @xmath207 for @xmath171 . critical wetting occurs for @xmath220 , provided @xmath221 ; if @xmath222 the wetting transition is of first order and does not necessarily occur at the point given by @xmath220 @xcite . according to ref . @xcite and appendix a the analytical expression for @xmath205 contains a contribution due to the wall - nematic excess adsorption , which is influenced by the surface charge density @xmath20 . therefore , there is the possibility that the order of the wetting transition depends on the surface charge density @xmath20 . for the values of @xmath20 used in the present study , however , we always find @xmath223 , i.e. , the wetting transitions at @xmath155 and @xmath157 are of first order . whereas the asymptotical analysis above is reliable with respect to the _ order _ of the wetting transitions at @xmath155 and @xmath157 , this is not the case concerning the _ location _ of @xmath155 and @xmath157 , because the wetting transitions are of first order @xcite . therefore , we have to use numerical methods . excess adsorption @xmath214 of platelike macroions at isotropic - nematic coexistence with isotropic boundary conditions in the bulk ( @xmath154 ) ( see fig . [ fig : a ] ) in terms of the surface charge density @xmath20 . curves for @xmath224 and @xmath225 can be mapped upon each other by reflecting them at the axis @xmath226 . for fixed macroion charge , the excess adsorption increases with increasing surface charge density . for fixed surface charge density , the excess adsorption _ decreases _ upon increasing @xmath174 , even for attractive walls . ] in fig . [ fig : b ] the numerically determined excess adsorption at coexistence is shown as a function of the surface charge density @xmath20 . without loss of generality , only curves for negative macroion charges @xmath3 are displayed : since the density functional described in sec . [ sec : generalformalism ] is invariant under the simultaneous inversion of the signs of all charges ( @xmath3 , @xmath7 , and @xmath20 ) , the curve for @xmath227 is mapped onto the curve for @xmath225 by reflecting it at the axis @xmath226 . as expected , the excess adsorption @xmath214 of the macroions increases with increasing surface charge density @xmath20 for fixed macroion charge @xmath228 , because the surface becomes increasingly attractive ( or decreasingly repulsive ) for the macroions . however , for a fixed surface charge density @xmath20 and sufficiently large macroion charges @xmath174 , @xmath214 decreases upon increasing @xmath174 , irrespective of the sign of @xmath3 , i.e. , even for @xmath229 , for which the wall attracts macroions . this depression of the macroion number density near the surface occurs because the macroion - macroion repulsion dominates the macroion - surface interactions ( see sec . [ sec : phasediagram ] ) . from fig . [ fig : b ] one can indeed infer that there is _ partial _ wetting for sufficiently large macroion charges @xmath174 . excess adsorption @xmath214 of platelike macroions at isotropic - nematic coexistence with isotropic boundary conditions in the bulk ( @xmath154 ) for a surface charge density @xmath169 ( see fig . [ fig : a ] ) as function of the macroion charge @xmath3 . it is finite within the range @xmath230 which implies the location of the lower wetting transition point @xmath155 ( fig . [ fig : a ] ) to be within the range @xmath231 $ ] . ] the variation of the macroion excess adsorption @xmath214 upon @xmath232 along isotropic - nematic coexistence is shown in fig . [ fig : c ] . @xmath214 is finite for @xmath230 . on the other hand , @xmath217 corresponds to hard colloidal platelets for which the occurrence of complete wetting is well known @xcite . hence the lower wetting transition point @xmath155 in fig . [ fig : a ] is located within the range @xmath233 $ ] . in order to locate the upper wetting transition point @xmath157 one may use the fact that the true wetting transitions points @xmath234 can only be located _ within _ the interval @xmath235 $ ] , @xmath236 , where @xmath237 are the locations of the wetting transition points inferred from the asymptotic analysis . this statement follows from the observation that @xmath238 leads to @xmath239 which renders @xmath240 as a local _ maximum _ of @xmath190 @xcite . together with the numerically found complete wetting for @xmath218 one concludes that the location of the upper wetting transition point @xmath157 lies within the range @xmath241 $ ] . macroion density profiles @xmath242 of a mixture of platelike macroions and salt at isotropic - nematic coexistence with isotropic boundary conditions in the bulk ( @xmath154 ) for a surface charge density @xmath169 ( see fig . [ fig : a ] ) . upon decreasing @xmath174 , i.e. , upon approaching the wetting transition point @xmath155 in fig . [ fig : a ] , a nematiclike film forms at the surface . ] finally , fig . [ fig : f ] displays the increase of nematic film thicknesses upon increasing @xmath228 in terms of the macroion density profiles @xmath242 . in this section , we study the fluid composed of platelike macroions and salt in contact with a charged hard surface for nematic boundary conditions at large distances from the wall . isotropic - nematic coexistence with nematic boundary conditions in the bulk will be denoted as @xmath160 . we have performed an asymptotic analysis of the effective interface potential for drying similar to the one described in sec . [ sec : wetting ] . by means of this analysis two first - order drying transition points @xmath161 and @xmath168 have been found . due to the first - order character of the drying transitions , the loci of @xmath161 and @xmath168 have been determined numerically ( see sec . [ sec : wetting ] ) . note that the upper drying transition point @xmath168 is not visible in fig . [ fig : a ] because it is located at @xmath243 . however , for sufficiently small surface charge densities @xmath244 both drying transition points have been found within the interval @xmath245 $ ] . in contrast to the wetting scenario in sec . [ sec : wetting ] , we have been able to numerically detect the discontinuity of the excess adsorption at the drying transitions . moreover , the locations of the predrying lines in the phase diagrams ( e.g. , fig . [ fig : a ] ) can be numerically determined . excess adsorption @xmath214 of platelike macroions at isotropic - nematic coexistence with nematic boundary conditions ( @xmath160 ) for surface charge density @xmath169 ( see fig . [ fig : a ] ) in terms of the macroion charge @xmath3 . the excess adsorption is finite and bounded from below for @xmath246 , whereas at @xmath247 it jumps to @xmath248 . it has been verified by a comparison of surface tensions that the numerical solutions with finite excess adsorption correspond to equilibrium structures and not only to metastable states . the _ discontinuity _ of @xmath214 at @xmath249 identifies state point @xmath161 in fig . [ fig : a ] as a _ first - order _ drying transition point . the latter conclusion can also be drawn from an asymptotic analysis of the effective interface potential ( see main text ) . ] figure [ fig : d ] displays the macroion excess adsorption @xmath214 close to a charged hard wall with surface charge density @xmath147 as a function of the macroion charge @xmath250 . for @xmath251 , the excess adsorption is finite and bounded from below . it jumps to @xmath248 at @xmath247 . the wall - nematic surface tensions of the numerical solutions with finite excess adsorption are smaller than the sums of the corresponding wall - isotropic surface tensions and the isotropic - nematic interfacial tensions . therefore , these numerical solutions indeed describe equilibrium structures and not only metastable states . the _ discontinuity _ of @xmath214 corresponds to the occurrence of a first - order drying transition at @xmath249 , which is displayed as state point @xmath161 in fig . [ fig : a ] . a first - order drying transition is accompanied by a _ predrying line _ in the surface phase diagram ( see the dashed line in fig . [ fig : a ] ) , which connects the drying transition point @xmath161 with a critical point @xmath252 , located at @xmath253 . the predrying line may be parameterized in terms of , e.g. , the macroion charge : @xmath254 for @xmath255 denotes the chemical potential difference @xmath148 for which the excess adsorption @xmath214 as a function of @xmath3 and @xmath148 exhibits a _ finite _ discontinuity @xmath256 . excess adsorption discontinuity @xmath256 ( solid line ) of a mixture of platelike macroions and salt at the predrying line parameterized by the macroion charge @xmath3 for a surface charge density @xmath169 ( see fig . [ fig : a ] ) . the excess adsorption difference vanishes for @xmath257 according to a power law @xmath258 with the mean field critical exponent @xmath259 ( dashed line ) . ] figure [ fig : e ] displays this discontinuity as a function of the macroion charge @xmath3 ( solid line ) . it vanishes according to a power law @xmath260 with the mean field critical exponent @xmath259 ( dashed line ) . _ beyond _ mean field theory , one expects an exponent @xmath261 , corresponding to the two - dimensional ising universality class . macroion density profiles @xmath242 of a mixture of platelike macroions with @xmath262 and salt in contact with a surface of charge density @xmath169 upon crossing the predrying line at @xmath263 ( fig . [ fig : a ] ) . at the predrying line , a quasi - isotropic film with _ thickness appears at the surface . ] figure [ fig : g ] depicts the formation of an isotropic film upon approaching isotropic - nematic coexistence for @xmath150 with macroion charge @xmath262 and surface charge density @xmath169 . a _ finite _ discontinuity of the film thickness upon crossing the predrying line at @xmath264 can be inferred . whereas the three preceeding sections have been focused on the fluid structure , this section addresses the electrostatic properties of the surface due to the contact with the fluid of charged particles . the electrostatic potential profile difference @xmath265 ( see eq . ) relative to the electrostatic bulk potential @xmath266 for @xmath267 at isotropic - nematic bulk coexistence with isotropic boundary conditions in the bulk ( @xmath268 ) is shown in electrostatic potential difference profiles @xmath105 relative to the electrostatic bulk potential in a mixture of platelike macroions with @xmath267 and salt at isotropic - nematic bulk coexistence with isotropic boundary conditions in the bulk ( @xmath154 ) ( see fig . [ fig : a ] ) . the slope of @xmath105 at the surface ( @xmath87 ) is governed by the surface charge density @xmath20 whereas the bulk value @xmath269 is approached proportional to @xmath119 from below . hence upon increasing the surface charge , a crossover from monotonic to non - monotonic electrostatic potential profiles occurs . ] [ fig : h ] . for large distances @xmath16 from the surface , @xmath270 decays proportional to @xmath119 ( see sec . [ sec : asym ] ) . upon increasing the surface charge density @xmath20 a crossover from monotonic to non - monotonic electrostatic potential profiles occurs at @xmath226 . according to eq . , the slope of the electrostatic potential difference at the wall is given by the surface charge density : @xmath271 . on the other hand , an electric double - layer is formed with a negatively charged layer on the nematic side and a positively charged layer on the isotropic side if a quasi - free interface between the isotropic bulk and a nematic film of finite thickness is present @xcite . thus @xmath105 approaches its bulk value @xmath272 from below , i.e. , @xmath273 for large distances @xmath16 from the surface . therefore , for @xmath274 , i.e. , @xmath275 , @xmath270 is minimal at some finite distance @xmath276 , whereas for @xmath277 , i.e. , @xmath278 , @xmath105 attains its minimal value at @xmath87 . in the aforementioned case , a quasi - free isotropic - nematic interface is formed at isotropic - nematic coexistence for a state of partial wetting . additionally , nematic films of finite thickness are present for states slightly below isotropic - nematic coexistence . quasi - free isotropic - nematic interfaces also occur for nematic boundary conditions in the bulk close to complete drying . in the latter case , the electrostatic potential _ decreases _ upon approaching the bulk because @xmath279 . moreover , we have found that the electrostatic potential is a _ function of the distance from the surface if _ no _ isotropic - nematic interfaces form . electrostatic surface potential @xmath280 as a function of the surface charge density @xmath20 in a mixture of platelike macroions and salt at isotropic - nematic coexistence with isotropic boundary conditions in the bulk ( @xmath154 ) ( see fig . [ fig : a ] ) . all curves are monotonically increasing but they do _ not _ pass through the origin ( @xmath281 ) at ( @xmath282 ) if @xmath283 . thus , the point of zero ( surface ) charge ( @xmath226 ) does _ not _ coincide with the isoelectric point ( @xmath284 ) . note that curves for @xmath224 and @xmath225 can be mapped upon each other by reflecting them at the origin . ] figure [ fig : i ] displays the electrostatic surface potential @xmath285 as a function of the surface charge density @xmath20 at isotropic - nematic coexistence with isotropic boundary conditions in the bulk ( @xmath154 ) . curves for @xmath224 and @xmath225 can be mapped upon each other by reflecting them at the origin ( @xmath281 ) . as expected , the surface potential increases monotonically with the surface charge . however , the points of zero charge ( @xmath226 ) do _ not _ coincide with the isoelectric points ( @xmath284 ) , i.e. , the curves in fig . [ fig : i ] do _ not _ pass through the origin ( @xmath281 ) , in contrast to poisson - boltzmann theories @xcite . this behavior arises from the hard - core interaction of the particles , which leads to a depletion attraction of the larger macroions towards the wall on purely entropic grounds . therefore , for a hard and uncharged wall ( @xmath226 ) , negatively charged macroions ( @xmath279 ) are accumulated close to the surface leading to a negative electrostatic surface potential ( @xmath286 ) . our findings show that the sign of the electrostatic surface potential @xmath285 for vanishing surface charge density @xmath226 depends only on the sign of the macroion charge and not on the boundary conditions in the bulk or the chemical potential difference @xmath148 . this work has been devoted to the investigation of a recently introduced model fluid composed of charged platelike colloids and salt ( fig . [ fig : geo ] ) in contact with a homogeneously charged planar wall . the density functional introduced in ref . @xcite has been extended by a wall potential incorporating a short - ranged hard - core repulsion as well as a long - ranged coulomb interaction with charged fluid particles . by solving the corresponding euler - lagrange equations for appropriate boundary conditions , spatially inhomogeneous equilibrium states are obtained . bulk and surface phase diagrams ( fig . [ fig : a ] ) in terms of macroion charge , chemical potential supersaturation , and surface charge density have been calculated by means of asymptotical analysis of effective interface potentials and numerical solutions of the euler - lagrange equations . the effective interface potentials exhibit the same asymptotic behavior as nonretarded van der waals forces although our model does not include dispersion forces . the origin of this phenomenon can be traced back to the application within the density functional of a particle - particle pair distribution function which decays asymptotically with a debye - hckel form . complete wetting by the nematic phase occurs in between two first - order wetting transition points on the isotropic - nematic bulk coexistence curve for isotropic boundary conditions in the bulk ( fig . [ fig : j ] ) whereas only partial wetting occurs at the remaining state points of the isotropic - nematic coexistence line ( fig . [ fig : b ] ) . the continuous but limited increase of the excess adsorption ( fig . [ fig : c ] ) and of the wetting film thickness ( fig . [ fig : f ] ) upon approaching the wetting transition points is used to locate the wetting transition points within the phase diagrams . whereas the first - order character of the wetting transitions and the complete wetting behavior @xmath287 have been inferred from analytical considerations , the location of the wetting transition points must be obtained numerically . the corresponding prewetting lines could not be resolved . first - order drying by the isotropic phase , characterized by a discontinuous divergence of the excess adsorption upon crossing the drying transition point , is found at isotropic - nematic coexistence for nematic boundary conditions in the bulk ( fig . [ fig : d ] ) . the predrying line , where the excess adsorption shows a finite discontinuity ( fig . [ fig : g ] ) , terminates at a critical point ( fig . [ fig : e ] ) . if quasi - free isotropic - nematic interfaces between the bulk and surface films of finite thickness form , a crossover is found from monotonic to non - monotonic electrostatic potential profiles upon varying the surface charge density ( fig . [ fig : h ] ) . surface potential and surface charge do _ not _ vanish simultaneously ( fig . [ fig : i ] ) , i.e. , the point of zero charge and the isoelectric point of the surface do not coincide due to the presence of both coulomb interactions and hard - core repulsion . the @xmath288-matrix @xmath112 in eq . is given by @xmath289\big),\ ] ] where @xmath290 denotes the two - particle direct correlation function of the hard core term @xmath291 ( eq . ) and @xmath292 is the set of _ constant _ bulk density profiles . by defining @xmath293 which renders eq . in the form @xmath294 = \frac{1}{2}|a|\sum_{ij}\int\limits_{0 ^ -_{\vphantom{|}}}^\infty\!\mathrm{d}z \int\limits_{0 ^ -_{\vphantom{|}}}^\infty\!\mathrm{d}z ' \rho_i(z)\rho_j(z')w_{ij}(z - z'),\ ] ] the @xmath111-matrix @xmath113 in eq . can be expressed as @xmath295 according to ref . @xcite , the amplitudes @xmath204 and @xmath205 in eq . are given by @xmath296 and @xmath297 with the abbreviation @xmath298 the quantities @xmath299 and @xmath300 are related to excess adsorptions of the wall - nematic surface and the free nematic - isotropic interface , respectively . 00 a. mourchid , a. delville , j. lembard , e. lcolier , and p. levitz , langmuir * 11 * , 1942 ( 1995 ) . gabriel , c. sanchez , and p. davidson , j. phys . chem . * 100 * , 11139 ( 1996 ) . a. mourchid , e. lcolier , h. van damme , and p. levitz , langmuir * 14 * , 4718 ( 1998 ) . brown , c. ferrero , t. narayanan , and a.r . rennie , eur . j. b * 11 * , 481 ( 1999 ) . p. levitz , e. lcolier , a. mourchid , a. delville , and s. lyonnard , europhys . lett . * 49 * , 672 ( 2000 ) . f.m . van der kooij and h.n.w . lekkerkerker , langmuir * 16 * , 10144 ( 2000 ) . d. van der beek and h.n.w . lekkerkerker , europhys . lett . * 61 * , 702 ( 2003 ) . d. van der beek and h.n.w . lekkerkerker , langmuir * 20 * , 8582 ( 2004 ) . wijnhoven , d.d . vant zand , d. van der beek , and h.n.w . lekkerkerker , langmuir * 21 * , 10422 ( 2005 ) . b. ruzicka , l. zulian , and g. ruocco , langmuir * 22 * , 1106 ( 2006 ) . l. harnau and s. dietrich , in _ soft matter _ , vol . 3 , edited by g. gompper and m. schick ( wiley , berlin , 2006 ) , to appear . brown , s.m . clarke , and a.r . rennie , langmuir * 14 * , 3129 ( 1998 ) . van der kooij and h.n.w . lekkerkerker , j. phys . b * 102 * , 7829 ( 1998 ) . p. davidson and j .- ch.p . gabriel , curr . colloid interface sci . * 9 * , 377 ( 2005 ) . veerman and d. frenkel , phys . a * 45 * , 5632 ( 1992 ) . m. dijkstra , j .- hansen , and p.a . madden , phys . e * 55 * , 3044 ( 1997 ) . s. meyer , p. levitz , and a. delville , j. phys . chem . b * 105 * , 9595 ( 2001 ) . s. meyer , p. levitz , and a. delville , j. phys . b * 105 * , 10684 ( 2001 ) . g. odriozola , m. romero - bastida , and f. de j. guevara - rodrguez , phys . e * 70 * , 021405 ( 2004 ) . d. van der beek , t. schilling , and h.n.w . lekkerkerker , j. chem . phys . * 121 * , 5423 ( 2004 ) . b.s . john and f.a . escobedo , j. phys . b * 109 * , 23008 ( 2005 ) . forsyth , s. marelja , d.j . mitchell , and b.w . ninham , adv . colloid interface sci . * 9 * , 37 ( 1978 ) . cuesta and r.p . sear , eur . j. b * 8 * , 233 ( 1999 ) . wensink , g.j . vroege , and h.n.w . lekkerkerker , j. phys . b * 105 * , 10610 ( 2001 ) . rowan and j .- p . hansen , langmuir * 18 * , 2063 ( 2002 ) . l. harnau , d. rowan , and j .- hansen , j. chem . phys . * 117 * , 11359 ( 2002 ) . h.h . wensink and g.j . vroege , j. phys . : condens . matter * 16 * , s2015 ( 2004 ) . y. rosenfeld and w.m . gelbart , j. chem . phys . * 81 * , 4574 ( 1984 ) . y. rosenfeld and l. blum , j. phys . chem . * 89 * , 5149 ( 1985 ) . rowan , j .- hansen , and e. trizac , mol . phys . * 98 * , 1369 ( 2000 ) . l. harnau , d. costa , and j .- hansen , europhys . lett . * 53 * , 729 ( 2001 ) . l. harnau and j .- p . hansen , j. chem . phys . * 116 * , 9051 ( 2002 ) . e. trizac , l. bocquet , r. agra , j .- j . weis , and m. aubouy , j. phys . : matter * 14 * , 9339 ( 2002 ) . r. agra , e. trizac , and l. bocquet , eur . j. e * 15 * , 345 ( 2004 ) . d. costa , j .- hansen , and l. harnau , mol . * 103 * , 1917 ( 2005 ) . l. li , l. harnau , s. rosenfeldt , and m. ballauff , phys . e * 72 * , 051504 ( 2005 ) . moore and w.e . mcmullen , j. phys . chem . * 96 * , 3374 ( 1992 ) . l. harnau and s. dietrich , phys . e * 65 * , 021505 ( 2002 ) . l. harnau and s. dietrich , phys . e * 66 * , 051702 ( 2002 ) . m. bier , l. harnau , and s. dietrich , phys . e * 69 * , 021506 ( 2004 ) . wensink and h.n.w . lekkerkerker , europhys . lett . * 66 * , 125 ( 2004 ) . l. harnau and s. dietrich , phys . e * 71 * , 011504 ( 2005 ) . m. bier , l. harnau , and s. dietrich , j. chem . phys . * 123 * , 114906 ( 2005 ) . . de gennes , rev . phys . * 57 * , 827 ( 1985 ) . d.e . sullivan and m.m . telo da gama , in _ fluid interfacial phenomena _ , edited by c.a . croxton ( wiley , chichester , 1986 ) , p. 45 . s. dietrich , in _ phase transitions and critical phenomena _ , 12 , edited by c. domb and j.l . lebowitz ( academic , london , 1988 ) , p. 1 . m. schick , in _ les houches , session xlviii , 1988 liquides aux interfaces / liquids at interfaces _ , edited by j. charvolin , j.f . joanny , and j. zinn - justin ( north - holland , amsterdam , 1990 ) , p. 415 . r. evans , adv . * 28 * , 143 ( 1979 ) . r. evans , in _ fundamentals of inhomogeneous fluids _ , edited by d. henderson ( dekker , new york , 1992 ) , p. 85 . r. zwanzig , j. chem . phys . * 39 * , 1714 ( 1963 ) . donnan , z. elektrochem . * 17 * , 572 ( 1911 ) . donnan , chem . rev . * 1 * , 73 ( 1924 ) . adamson , _ a textbook of physical chemistry _ ( academic , new york , 1973 ) . dimensionless quantities are used in units of @xmath301 ( thermal energy ) , @xmath302 ( proton charge ) , and @xmath303 ( bjerrum length ) with the solvent dielectric constant @xmath17 . for @xmath304 and @xmath305 ( water ) the latter is @xmath306 . in order to avoid a clumsy notation , the corresponding dimensional quantities are denoted by the same symbol and can be recognized by the presence of units . dimensional number densities are sometimes given in units of @xmath307 . cuesta and y. martnez - ratn , phys . lett . * 78 * , 3681 ( 1997 ) . cuesta and y. martnez - ratn , j. chem . phys . * 107 * , 6379 ( 1997 ) . ackler , r.h . french , and y .- m . chiang , j. colloid interface sci . * 179 * , 460 ( 1996 ) . j.l . lebowitz and e.h . lieb , phys . * 22 * , 631 ( 1969 ) . lieb and j.l . lebowitz , adv . math . * 9 * , 316 ( 1972 ) . s. dietrich and m. napirkowski , phys . a * 43 * , 1861 ( 1991 ) . grahame , chem . rev . * 41 * , 441 ( 1947 ) . hauge and m. schick , phys . b * 27 * , 4288 ( 1983 ) . k. shundyak and r. van roij , europhys . lett . * 74 * , 1039 ( 2006 ) .

we briefly review the status of motivated beyond - the - mssm phenomenology in the light of the lhc searches to date . in particular , we discuss the conceptual consequences of the exclusion bounds , of the hint for a higgs boson at about 125 gev , and of interpreting the excess of direct cp violation in the charm sector as a signal of new physics . we try to go into the various topics in a compact way while providing a relatively rich list of references , with particular attention to the most recent developments . given at http://indico.cern.ch/conferenceotherviews.py?view=standard&confid=153252[``xx international workshop on deep inelastic scattering and related subjects '' ] , 26 - 30 march 2012 , bonn , germany . ] 1.0 cm * supersymmetry phenomenology beyond + the mssm after 5/fb of lhc data * 1.0 cm paolo lodone + institut de thorie des phnomnes physiques , epfl , lausanne , switzerland + 1.0 cm

1203.6227

the theoretical difficulty in understanding the smallness of the fermi scale in the standard model ( sm ) with respect to any new physics at very high energies is the main motivation for expecting new phenomena to show up already in the energy range that is probed by the cern lhc . one of the most appealing solutions to this ` hierarchy problem ' is supersymmetry ( susy ) , to the point that it is sometimes called ` the standard way beyond the sm ' . since superpartners have not been found at colliders , supersymmetry is apparently pushed more and more to higher energies and this in principle weakens its power in solving the hierarchy problem , or equivalently some amount of finetuning is apparently reintroduced in the determination of the electroweak scale . while this is more and more unavoidably true at least for its minimal implementation , the minimal supersimmetric standard model ( mssm ) , it is also clear that non - minimal implementations or non - standard configurations can still have the chance to be natural , i.e. not finely - tuned . it is then crucial , from a conceptual point of view , to keep an eye on these natural configurations until they are excluded since , if it were not for the hierarchy problem , susy could manifest itself at much higher energies . ] . the main recent experimental inputs that are relevent for our considerations are the following : * a sm - like higgs boson is now excluded from about 127 gev up to 600 gev @xcite@xcite . at the same time , there is a @xmath1 hint for a higgs - like scalar close to 125 gev . * direct searches of s - particles @xcite-@xcite have already set very strong lower bounds on their masses . these bounds are quite model dependent , however it can be said that typically the squarks of the first two generations have to be heavier than about 1 tev . the case in which only the third generation is light , which is very motivated by naturalness as we shall see , is instead much less constrained , see @xcite@xcite for theoretical estimates . in such a situation the gluino could still be as light as @xmath2 gev , with the third generation even down to @xmath3 gev , and the lhc phenomenology depends crucially on the mass difference between the gluino and the charginos / neutralinos ( see figure [ fig : gluinodecays ] ) . * lhcb @xcite , and later cdf @xcite , have measured with a high precision the direct cp violation in the decays of a @xmath4 meson into @xmath5 and @xmath6 . the world average now yields : @xmath7 which deviates by approximately 3.8@xmath8 from the no - cp violation point , and is quite larger than the value @xmath9 that one would expect in the sm . we focus on naturalness considerations in section [ sect : nat ] , we consider other possible issues and points of view in section [ sect : other ] , and then conclude in section [ sect : concl ] . the main purpose of this discussion is to conceptually clarify the various possibilities in a compact way , while providing a relatively rich list of references with particular attention to the most recent studies . we hope that this summary may be useful at least as an orientation in this subject that is now a very hot topic @xcite@xcite@xcite . as already said naturalness , or the problem of finetuning , is the main theoretical motivation for physics beyond the sm at the lhc energies . it makes thus sense to discuss this point in some detail . it is usually said that the problem stems from the ` quardatic divergences ' that are present in the radiative corrections to the mass of an elementary scalar , or from the fact that these corrections are quadratically sensitive to the ultraviolet cutoff of the theory . this statement may sound strange , because it seems to depend on the way the theory is regularized : for example , where is the problem if one uses dimensional regularization instead of a sharp momentum cutoff to make the loop integrals finite ? given the importance of the issue , let us briefly see how it can be expressed in terms of renormalized quantities : this will give us more insight in the meaning of finetuning in the supersymmetric case . consider a heavy fermion field @xmath10 coupled to the higgs boson @xmath11 through a yukawa interaction : @xmath12 and let us regularize the higgs self energy @xmath13 with a cutoff @xmath14 . the relevant diagram is shown in figure [ figura : loops ] left , and we have : @xmath15\right|_{p^2=0}\ ] ] where @xmath16 , @xmath17 is the momentum of the @xmath11 line and ` @xmath18 ' are finite terms . let us now consider a process involving energy scales @xmath19 much higher than the value @xmath20 of the mass of the higgs particle when it is produced on shell ( the ` pole mass ' , that corresponds to the pole of the propagator ) . if we do renormalized perturbation theory using @xmath20 , then quantum effects will give in general significant corrections to the physical observables , and it will be necessary to compute the relevant amplitudes at many orders in perturbation theory . it is instead convenient to treat @xmath21 as a parameter of perturbation theory , defined by a renormalization prescription at the scale @xmath19 , for example : @xmath22 where @xmath23 is the renormalized propagator of the scalar @xmath11 . as a result @xmath21 becomes an ` effective mass ' and starts running with the energy scale @xmath19 according to the renormalization group ( rg ) flow , like the other parameters of perturbation theory , with initial condition : @xmath24 in this way the most important quantum corrections are resummed provided that one uses the value of this ` running mass ' @xmath25 to compute the mass effects in a process involving the typical momentum scale @xmath19 . in words , @xmath25 is ` the mass that minimizes the quantum corrections if we probe the theory at the energy scale @xmath19 ' . [ cols="^,^,^ " , ] to summarize , natural supersymmetry with stops and left - handed sbottom well below the tev and heavier first two generations stands now as the only ` conventional ' left out possibility for supersymmetry to solve the hierarchy problem with @xmath26 , or equivalently 10% finetuning ( other ` less conventional ' possibilities are discussed in section [ sect : escaping ] ) . our considerations are briefly summarized in table [ table : naturalsusy ] . the main point , as already said , is that the spectrum shown in figure [ fig : natspectrum ] has now the right to be called ` natural supersymmetry ' . moreover , if the hint for a higgs boson at 125 gev is confirmed , @xmath27susy with singlet - doublet mixing stands as the most natural possibility . notice also that , due to the large increase in the higgs quartic , in @xmath27susy all the squarks are naturally allowed to be heavier by a factor of 2 or 3 ( see also figure [ naturalness12gen ] ) . thus , for example , if in the future no stops are found below 1 tev and/or the squarks of the first two generations are excluded up to about 5 tev , then only @xmath27susy - like configurations will have the right to be called ` natural supersymmetry ' ... in this last case , the easiest way to exclude natural supersymmetry might be to exclude the extended higgs sector . we will come back on these issues in the conclusions . as a final remark we quote that , maybe also motivated by similar considerations , there has been recent interesting model building activity focussed on susy breaking with split families , or natural supersymmetry as we now say : see e.g. the recent idea of ` flavor mediation ' @xcite @xcite or other possibilities @xcite @xcite-@xcite , including even natural supersymmetry from string theory @xcite . moreover , since a low @xmath28 can help in minimizing the amount of finetuning in ( [ eq : natbounds ] ) given the sparticle masses , we can also say that low - scale susy breaking is somehow favoured by data(+naturalness ) ; see e.g. @xcite@xcite for very recent studies . besides taking heavier superpartners , another way to keep supersymmetry alive is escaping the experimental bounds through less - standard or peculiar configurations . a first well - known possibility is r - parity violation . recall that , from the superfield point of view , the down - type higgs doublet @xmath29 has exactly the same quantum numbers of the left - handed lepton doublet . it is thus clear that the baryon number and the lepton number are not accidental symmetries of the theory at the renormalizable level , as instead is the case for the sm . introducing all the couplings allowed by the gauge symmetry without any suppression mechanism , one ends up with large contributions to excluded processes like proton decay . the standard way to avoid these terms is to impose by hand an additional discrete symmetry , namely the r - parity , whose quantum number is @xmath30 where @xmath31 is the baryon number , @xmath32 is the lepton number , and @xmath33 is the spin of the field @xmath34 and @xmath32 are separately violated by nonperturbative effects at high energies , while @xmath35 is conserved . ] . notice that all the sm particles have r - parity @xmath36 , while all the superpartners have r - parity @xmath37 . the most striking consequence of this fact is that the lightest supersymmetric particle ( lsp ) must be stable , and can thus be a very good dark matter candidate . from the point of view of collider signatures , r - parity leads to the crucial prediction of abundance of events with missing energy , since the lsp necessarily escapes the detector . most of the experimental susy - search analyses published so far assume in fact that the production of any supersymmetric particle is accompained by the missing energy carried away by the lsp at the end , eventually , of a decay chain . for the case of r - parity violating ( rpv ) supersymmetry , on the contrary , very few experimental analyses have been published so far , see e.g. @xcite @xcite , and it is clear that the present stringent bounds on the ` conventional ' r - parity conserving case can be escaped , at least in part . from the conceptual point of view , for rpv to be ` natural ' one needs a rationale to make the effect small enough in order not to be in trouble with proton decay and other very constrained processes . for example a mechanism that one could use , and that is well known to be very efficient in suppressing the flavor violation that additional flavor structures can introduce in beyond - the - sm contexts , is minimal flavor violation ( mfv ) . the possibility of rpv with mfv has been studied in @xcite@xcite and also recently in @xcite@xcite . see also @xcite@xcite for recent discussions . even with conserved r - parity , the signal of jets plus missing energy can be significantly reduced if the mass splittings between some superpartners are small . for example , a small gluino - lsp splitting reduces the phase space for the gluino dacay : as a consequence the typical missing transverse energy and momentum coming from processes involving gluinos are reduced , leading to a reduced acceptance for given selection cuts . for recent studies of this configuration , called ` compressed spectrum ' , see e.g. @xcite@xcite . a similar way to reduce the missing energy , that in some cases is even more efficient , is ` stealth susy ' @xcite@xcite ( see also @xcite ) . in this case , r - parity is conserved but the lsp of the ordinary sector decays into particles belonging to a ` stealth sector ' in which the relative mass splittings within a supermultiplet are small . moreover , r - parity - even stealth particles can decay back into sm particles ; the phase space is thus again reduced for the decay chain that terminates with the true ` stealth lsp ' . another example of possible non - standard signatures is given by dirac gauginos , which may distort the phenomenology in a significant way @xcite@xcite@xcite , see also @xcite@xcite@xcite for recent studies . in summary , given the negative results of the susy searches so far , trying to see whether the detection may have been escaped for a reason that is not simply large mass is probably being to become a very hot topic in the near future , as an alternative way in which a natural supersymmetry could be realized . known possibilities are suppressing the missing energy ( rpv ) , suppressing the ` visible energy ' that goes into jets ( compressed spectrum ) , suppressing both ( stealth susy ) , or suppressing the production cross section by giving the gauginos a large dirac mass ( dirac gauginos ) . new phenomenological studies and ideas about the above issues would probably be very timely . recall that the main motivations for supersymmetry are , broadly speaking : ( i ) it can solve the hierarchy problem , in a perturbative way ; ( ii ) it gives a good dark matter ( dm ) candidate ; ( iii ) gauge couplings unify at high energy ; ( iv ) it is very elegant and a necessary ingredient of string theory . moreover all this is done in a way that is almost automatically compatible with the ewpt . retaining all the features ( i)-(iv ) provides a very attractive picture . however , that nature must be without finetuning is not a necessity ( we will come back on this in the conclusions ) , and it is meaningful to study what happens retaining ( ii)-(iv ) while abandoning the requirement ( i ) . it is clear that for dm one needs at least some neutralino at low energy , although its mass may be relatively heavier than the usual susy dm . it can be seen that the only choice that is radiatively stable is to keep all the gauginos and higgsinos around the tev scale , while the squarks and one higgs scalar doublet can be at a very high mass scale @xmath38 . remarkably , it turns out that this is also what is required for obtaining a precise gauge coupling unification . this scanario is called ` split supersymmetry ' @xcite-@xcite , and the only finetuning that is needed is the ( large ) one that makes one of the two scalar doublets much lighter than its natural mass scale , which would be @xmath38 . split supersymmetry has many interesting peculiar features , whose discussion go beyond our present scopes ; we refer for example to @xcite for a review . to cite one of them , the gluino can be very non - standard and its dominant dacay width can be into a gluon and a gravitino , giving a single - jet signal with a distinctive energy distribution . alternatively , although it is not the lsp , the gluino can be very long lived if its dominant decay mode is mediated by the heavy squarks , with a variety of possible unusual signatures not only at the lhc @xcite-@xcite . the experimental discovery of a slowly decaying gluino would be a strong indication for split susy , and would also provide a way to measure the high scale @xmath38 . a more radical point of view is to retain only the motivation ( iv ) and to accept again a large finetuning in the value of the fermi scale . actually , having accepted the finetuning , string theory is a very plausible framework since it provides a huge number of vacua ( the ` landscape ' ) , and one can speculate about environmental / anthropic selection among them @xcite . in any case , what one ends up with is ` high - scale supersymmetry ' @xcite , in which all the superpartners are relegated at a very high energy and dm is plausibly made of axions . what can lhc data to - date say about these scenarios ? an interesting feature of both models is that , assuming that there is no additional field content besides the one of the mssm , the value of the higgs mass is very precisely determined in terms of the parameters of the theory , in particular the scale @xmath38 , with some variability depending on the boundary condition at the scale @xmath38 . by carefully performing the running and the various matchings @xcite-@xcite@xcite , it can be seen that a higgs mass of 125 gev implies @xmath39 tev for the case of split susy , while no bound can be derived on @xmath38 in the case of high - scale susy . finally , these models are practically designed so that the direct bounds on sparticle masses do not touch them , and in particular the second case would not be touched even by eventual strong exclusion bounds on higgsinos and gauginos . if , on the contrary , particles resembling charginos and neutralinos are found , then it will be crucial to measure their couplings with high precision and see whether they are compatible with supersymmetry or not . to do so , a linear collider would be probably necessary . so far we have assumed that the hint for a higgs boson at 125 gev stands indeed for a signal , and we discussed some of the consequences for supersymmetric models . given the importance of the issue , however , it is clear that we can not content ourselves with a @xmath1 evidence , and more data is definitely needed ( as the experimental collaborations also say ) . moreover the fact that the channel in which most of the excess is seen , namely the diphoton one , prefers a rate that is _ larger _ than in the sm is precisely what one would expect in case of a background fluctuation . let us than say something about the possibility ( that may be disproved quite soon ) that the present excess goes away and a sm - like higgs boson is excluded up to a mass of 500 - 600 gev . the sm alone would then be excluded by the combination of higgs searches and electroweak precision tests . in the context of supersymmetry , the most plausible explanation would be that the cp - even scalars of the higgs sector have a reduced production cross section times branching ratio with respect to the sm higgs boson . namely , what counts is the quantity : @xmath40 where @xmath41 is a given final state . in particular , above @xmath42 the higgs boson decay into vector bosons is the dominant channel in the sm . looking at the exclusion plots we see that a heavy ` not - so - much - sm - like higgs boson ' is still allowed by data provided that @xmath43 ( @xmath44 ) is smaller than @xmath45 . this suppression is not difficult to achieve , generally speaking , in supersymmetric models @xcite@xcite , thanks to mixing effecs ( see e.g. @xcite ) , or top - loop effects that modify the gluon - fusion efficiency @xcite , or a depletion of @xmath46 due to a significant decay width into two neutralino lsp . in the case of @xmath27susy , in which such non - sm - like higgs particles above @xmath42 can be achieved very easily , it can be seen that it is not implausible to have @xmath47 for both the cp - even scalars , and it can also happen that the next to lightest state is more easily detectable than the lightest one @xcite-@xcite . in conclusion , strictly speaking at the moment of writing it is not excluded that the hint for a higgs boson at 125 gev is just a fluctuation of the background . in this case there may be one or more not - so - much - sm - like higgs bosons with mass between 200 and 300 gev and @xmath48 . waiting for more data to disprove these possibilities , we can say that if no sm - like higgs boson is found at 125 gev , then in reasonable models we typically expect at least one almost - sm - like such scalar below 500 gev with . ] @xmath49 . if the lightest scalar , with reduced couplings , is found to be above 200 gev , then in the context of supersymmetry this would be a strong indication for models with largely increased quartic coupling , like @xmath27susy . on the contrary , if a sm - like higgs boson is indeed found at 125 gev , then eventual additional scalars can have very low @xmath50-values . as a last issue , let us give attention to the possibility that the recently measured direct cp - violation in the @xmath4-meson decays , reported in equation ( [ eq : deltaacp ] ) , is a signal of physics beyond the sm . whether the experimental value can be explained within the sm or not is still under discussion . what we can say is that it is _ possible _ that the sm accounts for the observed effect @xcite@xcite , but maybe not very easily or at least it is fair to say that _ new physics is not implausible _ ( see e.g. @xcite -@xcite ) . in the np interpretation , what one needs is a rationale to understand why the main effect comes out in a @xmath51 process without disturbing the very tightly - constrained @xmath52 ones . referring to @xcite for details , a way to understand it in the context of supersymmetric models may be that flavor violation in the squark sector comes mainly from left - right mixings ( i.e. from the a - terms ) , while the left - left and right - right mixings are subleading . the size of the experimental value can then be naturally explained in the configuration called ` disoriented a - terms ' , in which the size of the entries of the a - term matrices is the same of those of the corresponding yukawa matrices but without respecting exact proportionality . a concrete realization of this configuration that may actually be considered also independently of supersymmetry can be @xcite to invoke the paradigm of partial compositeness . in fact , if the flavor structure is originated by this mechanism , than one expects an effective lagrangian of the type : @xmath53 while the yukawa matrices are given by : @xmath54 where the @xmath55 are ` suppression factors ' generated by the strong dynamics . it is then clear that @xmath56 operators tend to be suppressed more than the @xmath57 ones , and thus we are going in the right direction . moreover as discussed e.g. in @xcite and @xcite , the best candidates for producing a sizable effect in the charm sector while being consistent with the other flavor constraints are the @xmath58 chromomagnetic operators , such as : @xmath59 together with the other one with opposite chiralities whose effect is typically relatively suppressed by a factor @xmath60 . in the notation ( [ eq : finalparametrization ] ) we have , following @xcite@xcite : @xmath61 where im@xmath62 is the hadronic matrix element of @xmath63 , and we used the fact that @xmath64 is expected to be @xmath65 times the cabibbo angle . since a reasonable estimate can be @xmath66 , in order to be compatible with the observed value ( [ eq : deltaacp ] ) one needs either an enhanced sm contribution @xmath67 , or a np contribution that corresponds in our case to : @xmath68 in a supersymmetric context these flavor - violating operators are generated by the flavor - mixings in the squark mass matrices , that are of the form ( looking at the order of magnitudes , no proportionality between matrices ! ) : @xmath69 moreover the effects come at the loop level , so that ( [ eq : fixingthescale ] ) can translate into : @xmath70 where @xmath71 is the typical scale of superpartner masses . notice that until last year this would have been more than welcome , as a possible signal of supersymmetric particles , while now it starts being in tension with the bounds from direct detection ! for a detailed study along this direction we refer to @xcite , in which this possibility is thoroughly discussed . it can be seen that the other bounds on flavor - violation in the quark sector are easily satisfied , while the electric dipole moments ( edm ) turn out to be the strongest constraint . this picture could also be extended to the lepton sector , and in this case the strongest bounds come from @xmath72 and the electon edm . we discussed the status of a few motivated beyond - the - mssm models in the light of the recent lhc data . our conclusions can be summarized as follows : 1 . * insisting on naturalness : natural supersymmetry . * the present hint of a higgs boson at 125 gev tells us that , in the mssm , the stop should be too heavy to be natural according to the criterion @xmath26 . this instead is not a problem for non - minimal extensions , and thus we can at least say that beyond - the - mssm models with increased quartic coupling are more motivated than before . apart from that , the most solid input so far is that the first two generations must be above the tev , thus either one tolerates @xmath73 or typically one needs the third generation to be lighter than the first two . for this reason , ` more minimal ' or ` hierarchical ' scenarios can now be renamed ` natural supersymmetry ' ( since it is the only left natural possibility ... unless point 4 , below ) . actually in @xmath27susy the stop at about 1 tev , eventually degenerate with the other squarks , can be compatible with @xmath26 . in this case however also the first two generations can naturally be much heavier . low scale susy breaking is also favoured in this view , since broadly speaking it reduces the amount of tuning . + how to exclude natural supersymmetry ? a relatively fair statement may be : _ not finding the higgs boson _ ( which would actually exclude also most of the finely - tuned susy scenarios ) , _ and/or especially not finding superpartners at the lhc_. 2 . * escaping the lhc bounds . * r - parity violation after all is not compulsory , and without it the present bounds are partly escaped . other possibilities can be susy with a compressed spectrum , stealth mechanisms , or dirac gauginos . it is important to be ready to these distortions of the ` usual ' lhc phenomenology . + this is clearly a very hot topic ! in the future , if the bounds get stronger , we might be forced to call it something like ` alternative natural supersymmetry ' ... 3 . * insisting only on dm and unification and/or strings : split and high - scale susy . * by construction , these possibilities are not touched by direct bounds on squarks . thus for the moment the only experimental input for these models is the value of the higgs boson mass , from which we can say that split susy has to be ` not - so - split ' ( @xmath39 tev ) , while nothing can be said so far in case of high - scale susy . * higgs boson not at 125 gev . * the importance of the issue is such that we can not content ourselves with a @xmath74 evidence : although there are a lot of good reasons for the higgs boson to be there , the significance of the present hint is such that it may well be a background fluctuation . more importantly , heavier scalars with reduced couplings are present in susy as well as in other motivated extensions of the sm , and they must be looked for . + what can be said is that , if the present hint turns out to be a fluctuation , then it is obviously crucial to look for the most - sm - higgs - like particle ( with reduced couplings ) . in this case , it is fair to say that such a particle can not be considered to be reasonably excluded until the exclusion bound at 95% c.l . is at least down to @xmath75 . * direct cpv in @xmath4 decays . * in the context of supersymmetry , assuming that the sm contribution is not enough to explain the data , a very natural possibility is to have ` disoriented a - terms ' . this may be explicitly realized in a context in which the flavor structure has something to do with the mechanism of partial compositeness . if this is the case , new effects are predicted to be around the corner in the neutron edm and , if the picture is extended to leptons in the most straightforward way , in the electron edm and in @xmath72 . to conclude , in the near future a very fundamental question has a chance to be answered : _ was naturalness a good guiding principle ? _ the lhc will tell us . the very minimal requirement is a colored sector mainly coupled to the top , at or below the tev . if supersymmetry is the way of nature , this will be the scalar partner of the top . thus a very important point is whether the stop is there or not . notice that if this sector is made of scalars the first option would be susy , while in case of fermions the most plausible possibility would be compositeness . in any case , also being open to the possibility that naturalness is implemented in an unexpected way , at least new particles / resonances within the lhc reach and/or a non - standard higgs boson are requred . thus , in case the higgs boson is discovered , it will be crucial to measure its couplings and check whether it is sm - like or not , also looking at ww - scattering and looking for a second higgs doublet . if the higgs particle is discovered to be exactly sm - like and no new particles are found , this will be a _ _ strong indication that there is fundamental finetuning in nature _ _ , we like it or not . on the contrary , a non - standard higgs boson and especially new particles would probably tell us that naturalness arguments were correct . confirming or discarding the naturalness of the fermi scale will be a very nontrivial conceptual input for the scientific though . thus for the moment it makes sense to keep insisting on natural supersymmetry , until it is found or excluded . a quite non - conventional alternative is that _ nothing _ is found at the lhc , not even the higgs particle , which is probably what most people expect least . finally , we must not forget about flavor physics : even in presence of a totally - sm - like higgs boson and no extra particles , non - ckm flavor effects would at least imply that there is some new physics relatively at hand , although maybe beyond the lhc reach for direct discovery . figure [ fig : outlook ] is a pictorial summary of these last considerations . i thank riccardo barbieri , gino isidori , and riccardo rattazzi for useful discussions . i also thank the organizers of the session `` electroweak and searches '' at the `` xx international workshop on deep inelastic scattering and related subjects '' for their kind invitation to talk about this subject . this work is supported by the swiss national science foundation under grant 200021 - 125237 . g. aad _ et al . _ [ atlas collaboration ] , phys . b * 710 * ( 2012 ) 49 http://arxiv.org/abs/1202.1408[_arxiv:1202.1408_ ] [ hep - ex ] . updated results : http://cdsweb.cern.ch/record/1430033[atlas-conf-2012-019 ] . g. aad _ et al . _ [ atlas collaboration ] , phys . b * 710 * ( 2012 ) 67 http://arxiv.org/abs/1109.6572[_arxiv:1109.6572_ ] [ hep - ex ] . updated results : http://cdsweb.cern.ch/record/1432199[atlas-conf-2012-033 ] . r. aaij _ et al . _ [ lhcb collaboration ] , http://arxiv.org/abs/1112.0938[_arxiv:1112.0938_ ] [ hep - ex ] . [ cdf collaboration ] , http://www-cdf.fnal.gov/physics/new/bottom/120216.blessed-cpvcharm10fb/cdf10784.pdf[cdf note 10784 ] r. barbieri , http://www.sns.it/didattica/scienze/menunews/personale/docenti/barbieri/materiale/supersymmetry/ download / supersymmetry.pdf[``where are we with the search for supersymmetry '' ] , talk given at the sns of pisa , 12 october 2011 ; http://download.sns.it/barbieri/barbieri_paris_nov_2011.pdf[``susy phenomenology , including susy higgs '' ] , talk given at the 2011hcp , paris , 14 november 2011 ; http://download.sns.it/barbieri/cms_school_2012.pdf[``physics at the electroweak scale after the first 5 inverse femtobarn at the lhc '' ] , talk given at the cms data analysis school , pisa , 23 january 2012 . s. dimopoulos and g. f. giudice , phys . b * 357 * , 573 ( 1995 ) , http://arxiv.org/pdf/hep-ph/9507282[_arxiv:hep-ph/9507282_ ] . m. a. luty , http://arxiv.org/abs/hep-th/0509029[_arxiv:hep-th/0509029_ ] . g. f. giudice , in * kane , gordon ( ed . ) , pierce , aaron ( ed . ) : perspectives on lhc physics * 155 - 178 http://arxiv.org/abs/0801.2562[_arxiv:0801.2562_ ] . r. barbieri , `` effective field theories and physics beyond the standard model , '' proceedings of the international school of subnuclear physics _ effective theories and fundamental interactions _ , the subnuclear series * 34 * , world scientific ( 1996 ) . s. weinberg , phys . * 59 * ( 1987 ) 2607 . n. arkani - hamed , s. dimopoulos and g. r. dvali , phys . b * 429 * ( 1998 ) 263 http://arxiv.org/abs/hep-ph/9803315[_arxiv:hep-ph/9803315_ ] . i. antoniadis , n. arkani - hamed , s. dimopoulos and g. r. dvali , phys . lett . b * 436 * ( 1998 ) 257 http://arxiv.org/abs/hep-ph/9804398[_arxiv:hep-ph/9804398_ ] . s. p. martin , `` a supersymmetry primer , '' http://arxiv.org/abs/hep-ph/9709356[_arxiv:hep-ph/9709356_ ] . j. a. casas , j. r. espinosa and i. hidalgo , jhep * 0401 * ( 2004 ) 008 , http://arxiv.org/pdf/hep-ph/0310137[_arxiv:hep-ph/0310137_ ] . m. dine , r. g. leigh and a. kagan , phys . d * 48 * ( 1993 ) 4269 , http://arxiv.org/pdf/hep-ph/9304299[_arxiv:hep-ph/9304299_ ] . p. pouliot and n. seiberg , phys . b * 318 * ( 1993 ) 169 , http://arxiv.org/pdf/hep-ph/9308363[_arxiv:hep-ph/9308363_ ] . a. pomarol and d. tommasini , nucl . b * 466 * , 3 ( 1996 ) , http://arxiv.org/pdf/hep-ph/9507462[_arxiv:hep-ph/9507462_ ] . r. barbieri , g. r. dvali and l. j. hall , phys . b * 377 * , 76 ( 1996 ) , http://arxiv.org/pdf/hep-ph/9512388[_arxiv:hep-ph/9512388_ ] . a. g. cohen , d. b. kaplan and a. e. nelson , phys . b * 388 * , 588 ( 1996 ) , http://arxiv.org/pdf/hep-ph/9607394[_arxiv:hep-ph/9607394_ ] . r. barbieri , l. j. hall and a. romanino , phys . b * 401 * ( 1997 ) 47 , http://arxiv.org/pdf/hep-ph/9702315[_arxiv:hep-ph/9702315_ ] . r. barbieri , e. bertuzzo , m. farina , p. lodone , d. zhuridov , jhep * 1012 * ( 2010 ) 070 . http://arxiv.org/abs/1011.0730[_arxiv:1011.0730_ ] . r. barbieri , p. lodone , d. m. straub , jhep * 1105 * ( 2011 ) 049 . http://arxiv.org/abs/1102.0726[_arxiv:1102.0726_ ] . r. barbieri , g. isidori , j. jones - perez , p. lodone , d. m. straub , eur . j. * c71 * ( 2011 ) 1725 . http://arxiv.org/abs/1105.2296[_arxiv:1105.2296_ ] . l. j. hall , d. pinner and j. t. ruderman , http://arxiv.org/abs/1112.2703[_arxiv:1112.2703_ ] . a. arbey , m. battaglia , a. djouadi , f. mahmoudi and j. quevillon , phys . b * 708 * ( 2012 ) 162 http://arxiv.org/abs/1112.3028[_arxiv:1112.3028_ ] . p. draper , p. meade , m. reece and d. shih , http://arxiv.org/abs/1112.3068[_arxiv:1112.3068_ ] . d. m. ghilencea , h. m. lee and m. park , http://arxiv.org/abs/1203.0569[_arxiv:1203.0569_ ] . m. carena , s. gori , n. r. shah and c. e. m. wagner , http://arxiv.org/abs/1112.3336[_arxiv:1112.3336_ ] . r. harnik , g. d. kribs , d. t. larson and h. murayama , phys . d * 70 * , 015002 ( 2004 ) http://arxiv.org/abs/hep-ph/0311349[_arxiv:hep-ph/0311349_ ] . s. chang , c. kilic and r. mahbubani , phys . d * 71 * , 015003 ( 2005 ) http://arxiv.org/abs/hep-ph/0405267[_arxiv:hep-ph/0405267_ ] . a. birkedal , z. chacko and y. nomura , phys . d * 71 * , 015006 ( 2005 ) http://arxiv.org/abs/hep-ph/0408329[_arxiv:hep-ph/0408329_ ] . a. delgado and t. m. p. tait , jhep * 0507 * ( 2005 ) 023 http://arxiv.org/abs/hep-ph/0504224[_arxiv:hep-ph/0504224_ ] . p. batra , a. delgado , d. e. kaplan and t. m. p. tait , jhep * 0402 * , 043 ( 2004 ) http://arxiv.org/abs/hep-ph/0309149[_arxiv:hep-ph/0309149_ ] . p. lodone , jhep * 1005 * ( 2010 ) 068 http://arxiv.org/abs/1004.1271[_arxiv:1004.1271_ ] . j. r. espinosa and m. quiros , phys . b * 279 * , 92 ( 1992 ) . p. batra , a. delgado , d. e. kaplan and t. m. p. tait , jhep * 0406 * , 032 ( 2004 ) http://arxiv.org/abs/hep-ph/0404251[_arxiv:hep-ph/0404251_ ] . a. delgado , http://arxiv.org/abs/hep-ph/0409073[_arxiv:hep-ph/0409073_ ] . a. maloney , a. pierce and j. g. wacker , jhep * 0606 * , 034 ( 2006 ) http://arxiv.org/abs/hep-ph/0409127[_arxiv:hep-ph/0409127_ ] . s. p. martin , phys . d * 81 * , 035004 ( 2010 ) http://arxiv.org/abs/0910.2732[_arxiv:0910.2732_ ] . a. brignole , j. a. casas , j. r. espinosa and i. navarro , nucl . b * 666 * ( 2003 ) 105 , http://arxiv.org/pdf/hep-ph/0301121[_arxiv:hep-ph/0301121_ ] . m. dine , n. seiberg and s. thomas , phys . d * 76 * ( 2007 ) 095004 http://arxiv.org/abs/0707.0005[_arxiv:0707.0005_ ] . i. antoniadis , e. dudas , d. m. ghilencea and p. tziveloglou , nucl . b * 808 * ( 2009 ) 155 http://arxiv.org/abs/0806.3778[_arxiv:0806.3778_ ] . m. carena , k. kong , e. ponton and j. zurita , phys . d * 81 * ( 2010 ) 015001 http://arxiv.org/abs/0909.5434[_arxiv:0909.5434_ ] . i. antoniadis , e. dudas , d. m. ghilencea and p. tziveloglou , nucl . b * 831 * ( 2010 ) 133 http://arxiv.org/abs/0910.1100[_arxiv:0910.1100_ ] . y. kats , p. meade , m. reece and d. shih , jhep * 1202 * ( 2012 ) 115 http://arxiv.org/abs/1110.6444[_arxiv:1110.6444_ ] . r. essig , e. izaguirre , j. kaplan and j. g. wacker , jhep * 1201 * ( 2012 ) 074 http://arxiv.org/abs/1110.6443[_arxiv:1110.6443_ ] . c. brust , a. katz , s. lawrence and r. sundrum , http://arxiv.org/abs/1110.6670[_arxiv:1110.6670_ ] . a. delgado and m. quiros , phys . d * 85 * ( 2012 ) 015001 http://arxiv.org/abs/1111.0528[_arxiv:1111.0528_ ] . z. kang , j. li and t. li , http://arxiv.org/abs/1201.5305[_arxiv:1201.5305_ ] . j. cao , z. heng , j. m. yang , y. zhang and j. zhu , http://arxiv.org/abs/1202.5821[_arxiv:1202.5821_ ] . u. ellwanger , http://arxiv.org/abs/1112.3548[_arxiv:1112.3548_ ] . d. a. vasquez , g. belanger , c. boehm , j. da silva , p. richardson and c. wymant , http://arxiv.org/abs/1203.3446[_arxiv:1203.3446_ ] . u. ellwanger and c. hugonie , http://arxiv.org/abs/1203.5048[_arxiv:1203.5048_ ] . n. craig , m. mccullough and j. thaler , http://arxiv.org/abs/1201.2179[_arxiv:1201.2179_ ] . n. craig , m. mccullough and j. thaler , http://arxiv.org/abs/1203.1622[_arxiv:1203.1622_ ] . n. craig , d. green and a. katz , http://arxiv.org/abs/1103.3708[_arxiv:1103.3708_ ] . g. larsen , y. nomura and h. l. l. roberts , http://arxiv.org/abs/1202.6339[_arxiv:1202.6339_ ] . t. gherghetta and a. pomarol , phys . d * 67 * ( 2003 ) 085018 http://arxiv.org/abs/hep-ph/0302001[_arxiv:hep-ph/0302001_ ] . m. gabella , t. gherghetta and j. giedt , phys . d * 76 * ( 2007 ) 055001 http://arxiv.org/abs/0704.3571[_arxiv:0704.3571_ ] . r. sundrum , jhep * 1101 * ( 2011 ) 062 http://arxiv.org/abs/0909.5430[_arxiv:0909.5430_ ] . l. j. hall and m. suzuki , nucl . b * 231 * ( 1984 ) 419 . j. r. ellis , g. gelmini , c. jarlskog , g. g. ross and j. w. f. valle , phys . b * 150 * ( 1985 ) 142 . g. g. ross and j. w. f. valle , phys . b * 151 * ( 1985 ) 375 . v. d. barger , g. f. giudice and t. han , phys . d * 40 * ( 1989 ) 2987 . s. dawson , nucl . b * 261 * ( 1985 ) 297 . r. barbieri and a. masiero , nucl . b * 267 * ( 1986 ) 679 . jackson [ on behalf of the atlas collaboration ] , http://arxiv.org/abs/1112.0369[_arxiv:1112.0369_ ] [ hep - ex ] . [ atlas collaboration ] , http://cdsweb.cern.ch/record/1432202/files/atlas-conf-2012-035.pdf[atlas-conf-2012-035 ] . g. dambrosio , g. f. giudice , g. isidori and a. strumia , nucl . b * 645 * ( 2002 ) 155 http://arxiv.org/abs/hep-ph/0207036[_arxiv:hep-ph/0207036_ ] . g. isidori and d. m. straub , http://arxiv.org/abs/1202.0464[_arxiv:1202.0464_ ] . e. nikolidakis and c. smith , phys . d * 77 * ( 2008 ) 015021 http://arxiv.org/abs/0710.3129[_arxiv:0710.3129_ ] . c. smith , http://arxiv.org/abs/0809.3152[_arxiv:0809.3152_ ] . c. csaki , y. grossman and b. heidenreich , http://arxiv.org/abs/1111.1239[_arxiv:1111.1239_ ] . g. arcadi , l. di luzio and m. nardecchia , http://arxiv.org/abs/1111.3941[_arxiv:1111.3941_ ] . t. j. lecompte and s. p. martin , phys . d * 84 * ( 2011 ) 015004 http://arxiv.org/abs/1105.4304[_arxiv:1105.4304_ ] . t. j. lecompte and s. p. martin , phys . d * 85 * ( 2012 ) 035023 http://arxiv.org/abs/1111.6897[_arxiv:1111.6897_ ] . l. j. hall and l. randall , nucl . b * 352 * ( 1991 ) 289 . l. randall and n. rius , phys . b * 286 * ( 1992 ) 299 . p. j. fox , a. e. nelson and n. weiner , jhep * 0208 * ( 2002 ) 035 http://arxiv.org/abs/hep-ph/0206096[_arxiv:hep-ph/0206096_ ] . n. arkani - hamed , s. dimopoulos , g. f. giudice and a. romanino , nucl . b * 709 * ( 2005 ) 3 http://arxiv.org/abs/hep-ph/0409232[_arxiv:hep-ph/0409232_ ] . n. bernal , a. djouadi , p. slavich , jhep * 0707 * ( 2007 ) 016 . http://arxiv.org/abs/0705.1496[_arxiv:0705.1496_ ] . w. kilian , t. plehn , p. richardson and e. schmidt , eur . j. c * 39 * ( 2005 ) 229 http://arxiv.org/abs/hep-ph/0408088[_arxiv:hep-ph/0408088_ ] . j. l. hewett , b. lillie , m. masip and t. g. rizzo , jhep * 0409 * ( 2004 ) 070 http://arxiv.org/abs/hep-ph/0408248[_arxiv:hep-ph/0408248_ ] . k. cheung and w. -y . keung , phys . d * 71 * ( 2005 ) 015015 http://arxiv.org/abs/hep-ph/0408335[_arxiv:hep-ph/0408335_ ] . l. anchordoqui , h. goldberg and c. nunez , phys . d * 71 * ( 2005 ) 065014 http://arxiv.org/abs/hep-ph/0408284[_arxiv:hep-ph/0408284_ ] . j. g. gonzalez , s. reucroft and j. swain , phys . d * 74 * ( 2006 ) 027701 http://arxiv.org/abs/hep-ph/0504260[_arxiv:hep-ph/0504260_ ] . d. s. m. alves , e. izaguirre and j. g. wacker , http://arxiv.org/abs/1108.3390[_arxiv:1108.3390_ ] . m. e. cabrera , j. a. casas and a. delgado , phys . * 108 * ( 2012 ) 021802 http://arxiv.org/abs/1108.3867[_arxiv:1108.3867_ ] . g. f. giudice and a. strumia , http://arxiv.org/abs/1108.6077[_arxiv:1108.6077_ ] . t. tsukamoto , k. fujii , h. murayama , m. yamaguchi and y. okada , phys . d * 51 * ( 1995 ) 3153 . j. l. feng , m. e. peskin , h. murayama and x. r. tata , phys . d * 52 * ( 1995 ) 1418 http://arxiv.org/abs/hep-ph/9502260[_arxiv:hep-ph/9502260_ ] . m. m. nojiri , k. fujii and t. tsukamoto , phys . d * 54 * ( 1996 ) 6756 http://arxiv.org/abs/hep-ph/9606370[_arxiv:hep-ph/9606370_ ] . k. -i . hikasa and y. nakamura , z. phys . c * 70 * ( 1996 ) 139 [ erratum - ibid . c * 71 * ( 1996 ) 356 ] http://arxiv.org/abs/hep-ph/9501382[_arxiv:hep-ph/9501382_ ] . d. m. pierce , j. a. bagger , k. t. matchev and r. -j . zhang , nucl . b * 491 * ( 1997 ) 3 http://arxiv.org/abs/hep-ph/9606211[_arxiv:hep-ph/9606211_ ] . h. -c . cheng , j. l. feng , n. polonsky , phys . d56 * ( 1997 ) 6875 - 6884 . http://arxiv.org/abs/hep-ph/9706438[_arxiv:hep-ph/9706438_ ] . h. -c . cheng , j. l. feng and n. polonsky , phys . d * 57 * ( 1998 ) 152 http://arxiv.org/abs/hep-ph/9706476[_arxiv:hep-ph/9706476_ ] . e. katz , l. randall , s. -f . su , nucl . phys . * b536 * ( 1998 ) 3 - 28 . http://arxiv.org/abs/hep-ph/9801416[_arxiv:hep-ph/9801416_ ] . m. spira , a. djouadi , d. graudenz and p. m. zerwas , nucl . b * 453 * ( 1995 ) 17 http://arxiv.org/abs/hep-ph/9504378[_arxiv:hep-ph/9504378_ ] . a. djouadi , phys . * 459 * ( 2008 ) 1 http://arxiv.org/abs/hep-ph/0503173[_arxiv:hep-ph/0503173_ ] . m. carena , p. draper , t. liu and c. wagner , phys . d * 84 * ( 2011 ) 095010 http://arxiv.org/abs/1107.4354[_arxiv:1107.4354_ ] . i. low and s. shalgar , jhep * 0904 * ( 2009 ) 091 http://arxiv.org/abs/0901.0266[_arxiv:0901.0266_ ] . p. lodone , int . j. mod . a * 26 * ( 2011 ) 4053 http://arxiv.org/abs/1105.5248[_arxiv:1105.5248_ ] . e. bertuzzo , m. farina , http://arxiv.org/abs/1105.5389[_arxiv:1105.5389_ ] . e. bertuzzo and m. farina , http://arxiv.org/abs/1112.2190[_arxiv:1112.2190_ ] . y. grossman , a. l. kagan and y. nir , phys . d * 75 * ( 2007 ) 036008 http://arxiv.org/abs/hep-ph/0609178[_arxiv:hep-ph/0609178_ ] . g. isidori , j. f. kamenik , z. ligeti and g. perez , http://arxiv.org/abs/1111.4987[_arxiv:1111.4987_ ] . h. -n . li , c. -d . lu and f. -s . yu , http://arxiv.org/abs/1203.3120[_arxiv:1203.3120_ ] . e. franco , s. mishima and l. silvestrini , http://arxiv.org/abs/1203.3131[_arxiv:1203.3131_ ] . d. b. kaplan , nucl . b * 365 * ( 1991 ) 259 . t. gherghetta and a. pomarol , nucl . b * 586 * ( 2000 ) 141 http://arxiv.org/abs/hep-ph/0003129[_arxiv:hep-ph/0003129_ ] . k. agashe , g. perez and a. soni , phys . d * 71 * ( 2005 ) 016002 http://arxiv.org/abs/hep-ph/0408134[_arxiv:hep-ph/0408134_ ] . r. contino , t. kramer , m. son and r. sundrum , jhep * 0705 * ( 2007 ) 074 http://arxiv.org/abs/hep-ph/0612180[_arxiv:hep-ph/0612180_ ] . r. contino , http://arxiv.org/abs/1005.4269[_arxiv:1005.4269_ ] . m. redi and a. weiler , jhep * 1111 * ( 2011 ) 108 http://arxiv.org/abs/1106.6357[_arxiv:1106.6357_ ] . r. barbieri , g. isidori and d. pappadopulo , jhep * 0902 * ( 2009 ) 029 http://arxiv.org/abs/0811.2888[_arxiv:0811.2888_ ] . s. davidson , g. isidori and s. uhlig , phys . b * 663 * , 73 ( 2008 ) , http://arxiv.org/abs/0711.3376[_arxiv:0711.3376_ ] . y. nomura , m. papucci and d. stolarski , phys . rev . d * 77 * , 075006 ( 2008 ) , http://arxiv.org/abs/0712.2074[_arxiv:0712.2074_ ] . s. davidson , g. isidori and s. uhlig , phys . b * 663 * , 73 ( 2008 ) , http://arxiv.org/abs/0711.3376[_arxiv:0711.3376_ ] . y. nomura , m. papucci and d. stolarski , jhep * 0807 * , 055 ( 2008 ) , http://arxiv.org/abs/0802.2582[_arxiv:0802.2582_ ] . y. nomura and d. stolarski , phys . d * 78 * , 095011 ( 2008 ) , http://arxiv.org/abs/0808.1380[_arxiv:0808.1380_ ] . e. dudas , g. von gersdorff , j. parmentier and s. pokorski , jhep * 1012 * , 015 ( 2010 ) , http://arxiv.org/abs/1007.5208[_arxiv:1007.5208_ ] . a. e. nelson and m. j. strassler , jhep * 0009 * ( 2000 ) 030 http://arxiv.org/abs/hep-ph/0006251[_arxiv:hep-ph/0006251_ ] . a. e. nelson and m. j. strassler , jhep * 0207 * ( 2002 ) 021 http://arxiv.org/abs/hep-ph/0104051[_arxiv:hep-ph/0104051_ ] . d. b. kaplan , h. georgi and s. dimopoulos , phys . b * 136 * ( 1984 ) 187 . d. b. kaplan and h. georgi , phys . b * 136 * ( 1984 ) 183 . h. georgi and d. b. kaplan , phys . b * 145 * ( 1984 ) 216 . m. j. dugan , h. georgi and d. b. kaplan , nucl . b * 254 * ( 1985 ) 299 . a. falkowski , s. rychkov and a. urbano , http://arxiv.org/abs/1202.1532[_arxiv:1202.1532_ ] . a. azatov , r. contino and j. galloway , http://arxiv.org/abs/1202.3415[_arxiv:1202.3415_ ] . j. r. espinosa , c. grojean , m. muhlleitner and m. trott , http://arxiv.org/abs/1202.3697[_arxiv:1202.3697_ ] . a. djouadi , http://arxiv.org/abs/1203.4199[_arxiv:1203.4199_ ] . giardino , k. kannike , m. raidal and a. strumia , http://arxiv.org/abs/1203.4254[_arxiv:1203.4254_ ] . c. englert , b. jager , m. worek and d. zeppenfeld , http://arxiv.org/abs/0904.2119[_arxiv:0904.2119_ ] . p. borel , r. franceschini , r. rattazzi and a. wulzer , http://arxiv.org/abs/1202.1904[_arxiv:1202.1904_ ] . a. ballestrero , d. b. franzosi and e. maina , http://arxiv.org/abs/1203.2771[_arxiv:1203.2771_ ] . k. blum and r. t. dagnolo , http://arxiv.org/abs/1202.2364[_arxiv:1202.2364_ ] . l. maiani , a. d. polosa and v. riquer , http://arxiv.org/abs/1202.5998[_arxiv:1202.5998_ ] . b. bellazzini , c. csaki , a. falkowski and a. weiler , phys . d * 80 * ( 2009 ) 075008 http://arxiv.org/abs/0906.3026[_arxiv:0906.3026_ ] . b. bellazzini , c. csaki , j. hubisz and j. shao , phys . d * 83 * ( 2011 ) 095018 http://arxiv.org/abs/1012.1316[_arxiv:1012.1316_ ] .

we give reconstruction formulas inverting the geodesic x - ray transform over functions ( call it ) and solenoidal vector fields on surfaces with negative curvature and strictly convex boundary . these formulas generalize the pestov - uhlmann formulas in ( established for simple surfaces ) to cases allowing geodesics with infinite length on surfaces with trapping . such formulas take the form of fredholm equations , where the analysis of error operators requires deriving new estimates for the normal operator . numerical examples are provided at the end .

1511.05516

for a riemannian metric @xmath2 on a manifold @xmath3 with strictly convex boundary @xmath4 , the x - ray transform @xmath5 of a function @xmath6 is the collection of integrals over all finite length geodesics with endpoints on @xmath4 . for a domain in @xmath7 equipped with the euclidean metric , this is nothing more than the radon transform . a natural inverse problem is to recover @xmath8 from @xmath5 , ie . to find an inversion procedure for @xmath0 . in the euclidean case , radon found an inversion formula @xcite , and this has been extended to hyperbolic spaces by helgason @xcite and berenstein - casadio tarabusi @xcite . this problem has been studied more generally for _ simple metrics _ , which are metrics with no conjugate points and such that each geodesic has finite length . let @xmath9 be the unit tangent bundle of @xmath3 and @xmath10 the canonical projection @xmath11 ; if @xmath12 denotes the inward unit pointing vector field to @xmath4 , we define @xmath13 the inward ( - ) and outward ( + ) boundaries . then if @xmath14 denotes the geodesic flow on @xmath15 at time @xmath16 and if @xmath2 is simple , we can view @xmath0 as the map @xmath17 where @xmath18 is the length of the geodesic @xmath19 relating boundary points of @xmath3 . the injectivity of @xmath0 for simple metrics has been proved by mukhometov @xcite and anikonov @xcite ( see also the book by sharafutdinov @xcite ) . more recently pestov - uhlmann @xcite gave an approximate inversion formula in dimension @xmath20 for simple metrics , which is exact when the curvature of @xmath2 is constant . krishnan @xcite then showed that the pestov - uhlmann formula can be made exact in a small enough @xmath21-neighborhood of a given simple metric @xmath22 with constant curvature , but the bound is not explicit and depends for instance on the diameter of @xmath23 . the pestov - uhlmann approach , exploiting the interplay between transport equations on @xmath15 and the fiberwise hilbert transform , was used by the second author to derive further inversion formulas for certain types of weighted x - ray transforms on simple surfaces @xcite . manifolds with geodesics having infinite length are called _ manifolds with trapping_. for instance any negatively curved riemannian manifold with strictly convex boundary which have non contractible loops have closed geodesics and thus geodesics with infinite lengths . in general we call @xmath24 the set of points @xmath25 such that @xmath26 belongs to the interior @xmath27 of @xmath15 for all @xmath28 , which means that the geodesic @xmath29 has infinite length . the trapped set @xmath30 is defined by @xmath31 and consists of geodesics never touching the boundary . when @xmath2 has negative curvature , @xmath32 is a hyperbolic set for the flow in the sense of and @xmath2 has no conjugate points . in the recent work @xcite , the first author proved that for a class of manifolds including those with negative curvature ( see for the precise assumptions ) , the @xmath33-ray transform @xmath5 can be defined by the expression outside @xmath34 and @xmath0 extends as a bounded and injective operator @xmath35 where @xmath36 is some smooth lebesgue type measure on @xmath37 multiplied by a weight vanishing on @xmath38 . in fact , the same result holds as a map @xmath39 by the boundedness proved in lemma [ i_0i_1 ] below . for @xmath40 , the odd extension of @xmath5 is defined as the @xmath41 element @xmath42 similarly , we can integrate @xmath43-forms along geodesics and we can define the x - ray transform on @xmath43-forms by @xmath44 where @xmath45 is defined by @xmath46 , this also extends as a bounded operator @xmath47 which is injective on the space of divergence free forms . we let @xmath48 be the adjoint operator when @xmath49 is viewed as a map @xmath50 . to state our result we need to define two more operators . let @xmath51 be the hilbert transform in the fibers of @xmath52 ( i.e. multiplication by @xmath53 on the @xmath54 fourier component in the fiber , with @xmath55 ) , and let @xmath56 be the scattering operator of the flow , defined by @xmath57 then we have the following approximate inversion formula extending pestov - uhlmann formula to the setting of manifolds with trapping . [ inversionintro ] let @xmath58 be a riemannian surface with strictly convex boundary and assume that the trapped set @xmath32 for the flow is hyperbolic , that @xmath2 has no conjugate points ( these conditions are satisfied in negative curvature ) . then there exists an operator @xmath59 with smooth integral kernel such that for each @xmath60 @xmath61 where @xmath62 is the hodge star operator and @xmath63 the exterior derivative . we also have an inversion formula for divergence free @xmath43-forms , see theorem [ inversion ] . we next prove a upper bound on the @xmath64 norm of @xmath65 in a neighborhood of constant negatively curved metrics , showing that @xmath66 can be inverted by the neumann series , and thus providing an exact inversion formula for @xmath0 . in constant curvature , the trapped set @xmath32 is a fractal set with hausdorff dimension @xmath67 when @xmath68 ( see @xcite ) . [ theo2 ] let @xmath23 be a manifold with strictly convex boundary and constant negative curvature @xmath69 and trapped set @xmath32 , and let @xmath70 . then for each @xmath71 so that @xmath72 , there is a an explicit constant @xmath73 depending only on @xmath74 such that for all metric @xmath2 on @xmath3 with strictly convex boundary and gauss curvature @xmath75 satisfying latexmath:[\[{\lambda}_1^{2}g_0\leq g\leq { \lambda}_1^{-2 } g_0 , \quad \kappa(x ) \leq -{\lambda}_2 ^ 2\kappa_0 , \quad the remainder operator @xmath65 obtained in theorem [ inversionintro ] has norm @xmath77 and thus the inversion formula allows to recover @xmath60 from @xmath5 by a convergent series . when @xmath78 , the constant @xmath73 does not depend on @xmath79 . the constant @xmath73 is given by @xmath80 where @xmath81 and @xmath82 is the constant given in [ bound1 ] . this shows that there is an explicit neighborhood of the space of constant curvature metrics where the inversion formula can be made exact . even when @xmath83 the result is new since our bounds on the size of the neighborhood of constant curvature metrics is independent of the volume and the diameter of @xmath58 , which was not the case in the previous works for simple metrics @xcite . to prove such a bound , we compute on convex co - compact hyperbolic surfaces an operator closely related to @xmath84 in terms of the laplacian on the surface , where @xmath85 is the x - ray transform with a constant attenuation @xmath86 . this is done in the appendix . finally , we remark that is is quite possible that helgason type inversion for @xmath0 on @xmath87 could be applied to deal with the case of constant curvature metrics ( passing to the universal cover ) , but that would involve checking that the inversion formula applies to functions which are periodic by a discrete group , thus not tending to @xmath88 at infinity of @xmath87 , and in any case this would not work in variable curvature , unlike our formula . * acknowledgement . is partially supported by anr grant anr-13-js01 - 0006 and anr-13-bs01 - 0007 - 01 . fm is partially supported by nsf grant dms-1514820 . part of this work was done during the research program inverse problems at ihp , paris , spring 2015 . let @xmath58 be an oriented surface with strictly convex boundary and let @xmath15 be its unit tangent bundle with canonical projection @xmath89 . we will assume that @xmath90 notice that these assumptions are satisfied if @xmath2 has negative curvature , see @xcite . we denote by @xmath33 the geodesic vector field on @xmath15 and @xmath14 its flow at time @xmath16 . recall that hyperbolicity of the flow means that there is a continuous , flow invariant , decomposition on @xmath32 @xmath91 where @xmath92 and @xmath93 are the stable and unstable subspaces satisfying for each @xmath94 @xmath95 some uniform @xmath96 . let @xmath97 be the vertical vector field on @xmath15 defined by @xmath98 where @xmath99 is the @xmath100 rotation in the fibers of @xmath15 . locally , in isothermal coordinates @xmath101 on @xmath3 , the metric can be written as @xmath102 for some smooth function @xmath103 , we have associated coordinates @xmath104 on @xmath15 if @xmath105 parametrizes the fibers elements by @xmath106 ; in these coordinates @xmath97 is simply given by @xmath107 . the tangent space @xmath108 of @xmath15 is spanned by the vector fields @xmath109 where @xmath110 $ ] ; the vectors @xmath111 span the horizontal space @xmath112 , while @xmath97 spans the vertical space @xmath113 in @xmath108 . the sasaki metric @xmath114 on @xmath15 is the metric such that @xmath115 is orthonormal . the liouville measure @xmath116 on @xmath15 is the riemannian measure of the metric @xmath114 and we denote by @xmath117 the measure induced by @xmath114 on the boundary @xmath52 of @xmath15 . we define @xmath118 and @xmath119 the @xmath120 spaces with respect to @xmath116 and @xmath117 . there is another natural measure @xmath121 on @xmath52 which is given by @xmath122 where @xmath12 is the inward pointing normal vector field to @xmath4 . let us also define the incoming ( - ) , outgoing ( + ) and glancing ( 0 ) boundaries of @xmath15 @xmath123 define @xmath124 to be a small extension of @xmath3 with no conjugate points and with strictly convex boundary , and so that each geodesic with initial point @xmath125 goes in finite time to @xmath126 ; the existence of @xmath127 is proved in section 2.1 and lemma 2.3 of @xcite . we keep the notation @xmath33 for the vector field of the geodesic flow on @xmath128 . the liouville measure @xmath116 extends to @xmath128 . for each point @xmath129 , define the time of escape of @xmath15 in positive ( + ) and negative ( - ) time : @xmath130,\\ \ell_- ( x , v)=\inf\ , \ { t\leq 0 ; \varphi_{t}(x , v)\in sm\}\subset [ -\infty,0 ] \end{gathered}\ ] ] and the incoming ( - ) and outgoing ( + ) tail are defined by @xmath131 we define @xmath132 as well on @xmath128 in the same way , so that they extend those on @xmath15 , and they are closed set in @xmath128 are we keep the same notation @xmath132 for both @xmath15 and @xmath128 . the trapped set is @xmath133 and is invariant by the flow . we can define the extended stable bundle @xmath134 over @xmath135 satisfying @xmath136 and @xmath137 for some uniform @xmath96 ; the extended unstable bundle @xmath138 over @xmath139 is defined similarly by taking negative times instead of @xmath28 . the bundles @xmath140 and @xmath141 are defined by @xmath142 it follows from @xcite that under the assumptions we have ( see ( * ? ? ? 2.4 ) for details ) @xmath143 following @xcite , there is an orthogonal decomposition corresponding to fourier series in the fibers @xmath144 where @xmath145 is the space of @xmath146 sections of a complex line bundle over @xmath147 . similarly , one has a decomposition on @xmath52 @xmath148 using fourier analysis in the fibers of the circle bundle . there is a canonical map @xmath149 and in particular @xmath150 is just the pull - back by the projection @xmath151 . we define the dual map @xmath152 on distributions @xmath153 by @xmath154 where the distribution pairing is @xmath155 when @xmath156 . we notice that @xmath157 and it is easily checked that @xmath158 are the orthogonal projection from @xmath159 to respectively @xmath160 and @xmath161 . the hilbert transform in the fibers is defined by using the decomposition : @xmath162 with @xmath55 by convention . we can extend continuously @xmath51 to @xmath163 by the expression @xmath164 similarly , we define the hilbert transform in the fibers on @xmath52 @xmath165 and its extension to distributions as for @xmath128 . we note that @xmath51 extends as a bounded operator on @xmath159 and @xmath166 as a bounded operator on @xmath167 . for smooth @xmath168 we have that @xmath169 and the identity extends by continuity to the space of distributions in @xmath170 with wave - front set disjoint from @xmath171 since , by ( * ? ? ? 8.2.4 ) , the restriction map @xmath172 obtained by pull - back through the inclusion map @xmath173 extends continuously to the space of distributions on @xmath170 with wavefront set not intersecting @xmath171 . the following relation between hilbert transform and flow was proved by pestov - uhlmann ( * ? ? ? 1.5 ) ( it holds on any surface ) : @xmath174w = x_\perp w_0+(x_\perp w)_0\ ] ] where @xmath175 for @xmath176 . we can use the odd / even decomposition of distributions with respect to the involution @xmath177 on @xmath128 , @xmath15 and @xmath52 : for instance @xmath33 maps odd distributions to even distributions and conversely , @xmath51 maps odd ( resp . even ) distributions to odd ( resp . even ) distributions . we set @xmath178 and @xmath179 . we write similarly @xmath180 and @xmath181 for the hilbert transform on ( open sets of ) @xmath52 and the relation also holds with @xmath182 replacing @xmath183 if @xmath184 is even . taking the odd part of , for any @xmath176 @xmath185 for convenience , we extend @xmath127 to a larger with boundary manifold @xmath186 and extend smoothly @xmath33 to @xmath187 in a way that the flow of @xmath33 in @xmath187 is complete and for each @xmath188 , if @xmath189 for some @xmath190 , then @xmath191 for all @xmath192 ; we refer to ( * ? ? ? * section 2.1 ) . let @xmath116 be the liouville measure on @xmath127 , which we extend smoothly to @xmath187 in any fashion . the invariance of liouville measure by the flow in @xmath128 is satisfied , ie . @xmath193 , so that @xmath33 is formally skew - adjoint operator when acting on @xmath194 . by ( * section 4.2 ) , under the assumptions , for each @xmath195 , the boundary value problem @xmath196 has a unique solution in @xmath197 given by @xmath198 with @xmath199 the operator @xmath200 defined by is bounded and the expression actually extends as a bounded map @xmath201 as well , see ( * ? ? ? * proposition 4.2 ) . it has the following mapping properties @xmath202 where @xmath203 is defined by . define the _ scattering map _ @xmath204 by @xmath205 by ( * ? ? ? * lemma 3.4 ) the following operator is unitary @xmath206 the _ x - ray transform _ @xmath207 is defined by @xmath208 from ( * ? ? ? * lemma 5.1 ) and ( * ? ? ? 2.4 ) , we have that , under the assumptions , it extends as a bounded operator @xmath209 we denote by @xmath210 its odd continuation defined by @xmath211 and similarly we define the even continuation by @xmath212 for @xmath213 . the operator @xmath214 for all @xmath215 is the adjoint of @xmath216 , it is obtained as follows : for @xmath217 , @xmath218 is the unique @xmath219 solution of @xmath220 by this we mean that @xmath184 has an extension in a small neighborhood @xmath221 of @xmath222 as an invariant distribution and the restriction @xmath223 makes sense as a distribution since , by elliptic regularity , @xmath184 has wave - front set @xmath224 which does not intersect the conormal @xmath225 , as @xmath226 over @xmath37 . a consequence of this is that the unique @xmath219 solution of @xmath227 , @xmath228 with @xmath229 and @xmath230 is given by @xmath231 since it will be convenient to apply the hilbert transform on @xmath167 , we first show [ l^2 ] under the assumptions , the x - ray transform is bounded as a map @xmath232 there exists @xmath233 small such that the following operator is bounded @xmath234 it suffices to consider @xmath235 as the part on @xmath236 is clearly the same . by proposition 2.4 in @xcite , we have @xmath237 for some @xmath238 if @xmath239 this directly implies that for all @xmath240 , we have @xmath241 . by strict convexity of @xmath4 , there is @xmath233 small such that if @xmath242 , one has @xmath243 for some uniform @xmath244 , see ( * ? ? ? * lemma 4.1.2 ) for example . we then get @xmath245 using hlder inequality and santalo formula , we get for @xmath246 and @xmath247 , if @xmath248 and @xmath249 @xmath250 where @xmath251 . this concludes the proof of the first statement . the boundedness of @xmath252 on @xmath146 is direct from the proof above by taking @xmath253 and using @xmath254 on @xmath37 if @xmath255 for @xmath233 small enough . finally the x - ray transform on functions @xmath0 and on 1-forms @xmath49 are defined as the bounded operators , for any @xmath247 , @xmath256 and @xmath257 for @xmath258 . in fact , we can show [ i_0l^2 ] under the assumptions , the x - ray transform on functions and on @xmath43-forms are bounded as maps @xmath259 the operators @xmath260 and @xmath261 extend to @xmath262 as smooth pseudo - differential operators of order @xmath263 by ( * ? ? ? * propositions 5.7 and 5.9 ) , thus one has by using the standard @xmath264 argument that @xmath265 and @xmath266 are bounded . this implies that @xmath267 is bounded as a map @xmath268 for all @xmath233 fixed , and the same for @xmath269 . but we also have @xmath270 bounded as a map @xmath268 if @xmath233 is small enough by lemma [ l^2 ] , and the same for @xmath271 . the proof is complete . notice however that when we will use the adjoints of @xmath0 or @xmath49 , this will be always adjoint with respect to the space @xmath272 . let us define the operator @xmath273 we start by proving the following [ regw ] the operator @xmath65 defined in has a smooth schwartz kernel on @xmath274 . if the gauss curvature @xmath75 is a negative constant , then @xmath275 . we decompose the operator @xmath276 as @xmath277 with @xmath278 where @xmath233 is smaller than the radius of injectivity of the metric . by @xcite , the schwartz kernel of @xmath279 is a distribution on @xmath280 with wavefront set @xmath281 where @xmath282 are defined in , @xmath283 is the conormal bundle to the diagonal @xmath284 of @xmath280 and @xmath285 the wave - front set of the schwartz kernel of @xmath286 is @xmath287 thus by the composition rule of wave - front sets given in ( * ? ? ? * theorem 8.2.14 ) , @xmath288 since a differential operator does not increase wavefront sets , we deduce that @xmath289 the schwartz kernel of @xmath290 is given by @xmath291 if @xmath32 is the kernel of @xmath292 . using , the same exact arguments as in proposition 5.7 of @xcite show that the push - forward @xmath291 is smooth outside the submanifold @xmath293 if @xmath2 has no conjugate points . we are reduced to analyzing @xmath294 and this is very similar to the case of a simple metric studied in ( * ? ? ? * prop 5.1 ) . we denote by @xmath295 the jacobi field along the geodesic @xmath296 satisfying @xmath297 and @xmath298 where @xmath299 is the rotation of @xmath300 in the fibers of @xmath15 , and let @xmath301 be the jacobi field satisfying @xmath302 and @xmath303 . they satisfy @xmath304 where we used the splitting @xmath305 . the vector field @xmath306 satisfy @xmath307 @xmath308 if @xmath309 and @xmath310 is the gaussian curvature . we can then write @xmath311 and using the change of variable @xmath312 satisfying @xmath313 , we deduce that the schwartz kernel of @xmath314 is given by @xmath315}(d_g(x , y ) ) \big[\frac{1}{b(x , v , t)}v\big(\frac{a(x , v , t)}{b(x , v , t)}\big)\big]\big|_{tv=\exp_{x}^{-1}(y ) } .\ ] ] clearly this kernel is smooth outside @xmath316 . we now study the singularity at @xmath317 , which is equivalent to @xmath318 in the @xmath319 coordinates . using @xmath320 and the same for @xmath321 , we make an expansion of @xmath322 at @xmath318 @xmath323 thus we get near @xmath318 @xmath324 which shows that the kernel extends continuously to the diagonal . let us show that @xmath314 extends smoothly to the diagonal @xmath317 . for this , note that for each @xmath325 the function @xmath326 is smooth in the variable @xmath327 in @xmath328 $ ] where @xmath329 is a neighborhood of @xmath330 in @xmath15 . then it suffices to show that for each @xmath331 the taylor expansion at @xmath318 of @xmath326 to any order @xmath332 satisfies @xmath333 where @xmath334 are smooth symmetric tensors of order @xmath54 . for this purpose , we have @xmath335 as @xmath336 where @xmath337 is the symmetrized covariant derivative . then , writing @xmath338 as @xmath336 , we get for each @xmath339 @xmath340 from this we deduce by a direct induction that @xmath341 is the restriction of an element @xmath342 to @xmath15 . the same argument applies to @xmath343 . therefore using that @xmath344 if @xmath345 , we see that has the desired expansion , showing that @xmath314 extends smoothly to the diagonal . since @xmath233 was arbitrary , we have that @xmath314 has smooth kernel . let us show that @xmath275 if @xmath346 is constant . for @xmath347 with @xmath244 large , we define the operator @xmath348 by @xmath349 due to the exponential damping in @xmath16 , it is easy to see that @xmath350 is bounded if @xmath244 is large enough . in @xcite it is shown that there exists @xmath351 such that @xmath352 admits an analytic extension to @xmath353 if @xmath354 is large enough depending on @xmath79 . thus @xmath348 also admits an analytic extension to the same half - plane as a map @xmath355 , and we will show it vanishes if @xmath310 is constant . since the flow is assumed to have no conjugate points and the trapped set is hyperbolic , there is a uniform lower bound @xmath356 for some @xmath233 , and by using gronwall lemma there exists @xmath244 such that @xmath357 for all @xmath358 . using these estimates and reasoning like above , we have the converging expression for each @xmath359 @xmath360 as a continuous map of @xmath331 if @xmath361 . now , by , the functions @xmath362 and @xmath363 are constant in @xmath25 if the curvature @xmath310 is constant , thus @xmath364 for all @xmath365 . by analyticity in @xmath366 , it implies that @xmath367 vanishes when @xmath310 is constant . using that @xmath368 on @xmath369 and @xmath370 if @xmath371 is odd with respect to @xmath177 , we obtain that the @xmath146-adjoint of @xmath65 is given by @xmath372 now , we can show that pestov - uhlmann inversion formula @xcite works also under our assumptions . [ inversion ] if @xmath58 has strictly convex boundary and assumptions hold , then the following identity holds for each @xmath60 and @xmath373 @xmath374 @xmath375 where @xmath65 is the smoothing operator defined in proposition [ regw ] , @xmath376 is its @xmath146-adjoint , @xmath62 denotes the hodge star operator on @xmath377forms and @xmath378 is defined by . we follow the proof of ( * ? ? ? * theorem 5.1 ) . first using we have as distribution on @xmath27 @xmath379 for any @xmath380 . applying @xmath51 to this identity , we get by @xmath381 but since @xmath382 as @xmath33 maps odd functions to even functions ( and conversely ) , we obtain @xmath383 the unique @xmath219-solution of @xmath227 , @xmath228 with @xmath229 and @xmath230 is given by @xmath231 . but @xmath384 by lemma [ i_0l^2 ] and , and @xmath385 by proposition [ regw ] , we apply this to and get @xmath386 notice that we have used that @xmath387 vanishes on @xmath236 since @xmath388 on @xmath37 . applying @xmath389 we get @xmath390 which gives by using , the identity @xmath391 and @xmath392 . the extension to @xmath393 for @xmath394 is obtained by density and boundedness of each of the operators in the formula on the correct spaces . notice that a priori @xmath395 if @xmath60 but actually shows it is in @xmath396 . next we prove the inversion formula for co - exact @xmath43-forms . let @xmath397 , then since @xmath398 is odd , we get @xmath399 and applying @xmath51 with , this gives @xmath400 by using that @xmath401 on @xmath161 and @xmath402 on @xmath160 . here @xmath403 and its restriction to @xmath52 is @xmath404 by using lemma [ i_0i_1 ] , thus @xmath405 we apply @xmath406 and obtain @xmath407 by density the same identity holds for @xmath373 . we notice that a divergence - free @xmath43-form @xmath408 with sobolev regularity @xmath409 can be decomposed under the form @xmath410 where @xmath411 and @xmath412 satisfies @xmath413 and @xmath414 . the formula allows to recover a divergence - free @xmath43 form @xmath415 from its x - ray transform , and the x - ray transform of @xmath416 of @xmath184 at @xmath417 is given purely in terms of the integral of @xmath184 on a curve in @xmath4 with endpoints @xmath331 and @xmath418 and the homology class of the geodesic @xmath419 in @xmath3 , which is a known data . we now show that in pinched negative curvature the operator @xmath65 has small norm on @xmath420 , allowing to invert @xmath421 by neumann series . first , we have by ( * ? ? ? * section 5 ) that @xmath422 and since @xmath423 by ( * ? ? ? * proposition 4.2 ) if @xmath393 with @xmath424 , we see that the integral @xmath425 is defined for almost every @xmath426 if @xmath393 with @xmath424 , by fubini theorem . in ( * proposition 5.7 ) , it is shown that @xmath260 is the restriction to @xmath3 of an elliptic pseudo - differential operator of order @xmath263 on the extended manifold @xmath262 , thus it maps @xmath420 to @xmath420 for each @xmath427 $ ] . [ normw ] let @xmath58 be a surface with strictly convex boundary and assume that the gauss curvature @xmath428 of @xmath2 is negative , with @xmath429 . for each @xmath427 $ ] , the @xmath430 norm of @xmath65 is bounded by @xmath431 let @xmath362 and @xmath363 the functions defined by . if we can prove that @xmath432 uniformly in @xmath358 , then the integral in is convergent for almost every @xmath331 when @xmath433 , for each @xmath434 . we get in this case that for each @xmath393 with @xmath435 $ ] @xmath436 clearly if @xmath437 , we have @xmath438 and @xmath439 , and both are increasing in @xmath16 . first , by wronskian constancy , we get for all @xmath28 and @xmath129 @xmath440 to simplify notations , we will often drop the @xmath25 dependence below . the function @xmath441 satisfies the riccati equation @xmath442 with @xmath443 and @xmath444 as @xmath445 . we claim that @xmath446 for all @xmath354 : indeed if it were not the case , there is @xmath233 small and @xmath354 so that @xmath447 . let @xmath448 be the first time so that this happens , then @xmath449 and thus @xmath450 for all @xmath451 close enough to @xmath452 , leading to a contradiction . we then get for @xmath453 @xmath454 notice that by sturm comparison theorem , we also have for @xmath455 @xmath456 then , when @xmath457 , @xmath458 exists and is continuous in @xmath25 , moreover for @xmath28 we deduce from @xmath459 in fact we have more generally the same estimate for all @xmath460 : for @xmath461 $ ] @xmath462 where we have denoted @xmath463 . let us set @xmath464 and @xmath465 . we have @xmath466 where @xmath467 and @xmath309 . we notice that @xmath468 . by duhamel formula , we obtain @xmath469 which gives @xmath470 we want to give a bound on this quantity . we see that for @xmath471 , @xmath472 combining with , we deduce the uniform pointwise estimates @xmath473 using and , we obtain @xmath474 and the double integral is a finite constant which , after some calculation turns out to be equal to @xmath475 . we will give an estimate in negative curvature of the norm of the operator @xmath476 by considering the geometry in the universal cover . we let @xmath477 be the universal cover of @xmath3 , equipped with the pull - back metric @xmath478 of @xmath2 . since @xmath2 has negative curvature , the manifold @xmath477 is a non - compact simply connected manifold with boundary and the fundamental group @xmath479 of @xmath3 acts properly discontinuously on @xmath477 by isometries with respect to @xmath480 . denote by @xmath481 the covering map , and by abuse of notation we also call @xmath482 the covering map where @xmath483 is the unit tangent bundle of @xmath484 . fix a fundamental domain @xmath485 for the action of @xmath486 , then @xmath487 is compact since @xmath3 is compact . if @xmath393 with @xmath488 , let @xmath489 be its lift to @xmath477 ( which is @xmath486-automorphic ) , then we have for almost every @xmath490 and @xmath491 @xmath492 where @xmath493 is the geodesic flow lifted to @xmath483 and @xmath494 . define the operator @xmath495 on @xmath477 as follows : for @xmath496 with compact support , let @xmath497 which is in @xmath498 by . with the same analysis as in proposition [ defw ] , we find the integral kernel of @xmath499 @xmath500\big|_{t\tilde{v}={\widetilde}{\exp}_{\tilde{x}}^{-1}(\tilde{x } ' ) } \in c^\infty({\widetilde}{m}{\times}{\widetilde}{m}\setminus { \rm diag } ) \end{gathered}\ ] ] if @xmath501 is the exponential map for @xmath477 and @xmath502 solves the same jacobi equation as but for @xmath503 . we clearly have @xmath504 if @xmath505 and the integral kernel @xmath506 of @xmath495 is positive , symmetric and satisfies @xmath507 for all @xmath508 . from , we can write for almost every @xmath509 @xmath510 let @xmath511 be supported in @xmath487 and such that @xmath512 if @xmath513 , then @xmath514 and we get for almost every @xmath513 and @xmath515 @xmath516 by using that @xmath517 is an isometry of @xmath478 and the invariance of @xmath506 by @xmath486 mentioned above . we can then view @xmath518 as an operator acting on @xmath519 ( for @xmath488 ) with integral kernel @xmath520 the sum makes sense as an @xmath521 function since @xmath522 and @xmath523 for all @xmath524 . in fact , we also know that @xmath518 maps @xmath525 to @xmath525 ( since it is the restriction of a pseudo - differential operator of order @xmath263 on @xmath147 ) thus , since @xmath526 , @xmath527 to estimate its @xmath430 norm for @xmath488 , we can use schur s lemma and since @xmath528 is symmetric , this gives directly the following [ normofpi ] with the notation above , the operator @xmath518 has norm bounded by @xmath529 from now on , we shall work in a neighborhood of a constant curvature metric @xmath22 and estimate the norm @xmath518 in that neighborhood . let us denote by @xmath87 the hyperbolic space with curvature @xmath263 , realized as the unit disk @xmath530 with metric @xmath531 . let @xmath23 have constant curvature @xmath532 for some @xmath533 , we can extend @xmath3 to a non - compact complete manifold @xmath534 with constant curvature . indeed , this can be done as follows . the connected components @xmath535 of the boundary @xmath4 are circles . let @xmath536 be the distance from @xmath537 to the unique geodesic @xmath538 homotopic to @xmath537 in @xmath3 , these geodesics form the boundary of the convex core @xmath539 of @xmath3 . if @xmath540 denotes the length of @xmath538 , then in normal geodesic coordinates @xmath541{\times}({\mathbb{r}}/c_i{\mathbb{z}})_{\alpha}$ ] from @xmath538 in the collar between @xmath537 and @xmath538 , the metric is given by @xmath542 then it suffices to replace the region between @xmath537 and @xmath538 by the full hyperbolic half - cylinder @xmath543 with the metric and we denote by @xmath534 the obtained complete surface . it is realized as a quotient @xmath544 by a convex co - compact group of isometries @xmath545 of the hyperbolic space @xmath546 with curvature @xmath69 ; this space @xmath547 is just the unit disk @xmath530 with metric @xmath548 . notice that @xmath486 is also convex a co - compact group of isometries of @xmath549 . the exponent of convergence of @xmath486 is defined to be @xmath550 where @xmath551 denotes the riemannian distance of @xmath549 and @xmath331 is any point fixed in @xmath552 . let @xmath553 be the limit set of the group @xmath486 ( ie . the set of accumulation points in @xmath554 of the orbits @xmath555 in @xmath556 ) . by a result of patterson @xcite and sullivan @xcite , @xmath557 where @xmath558 is the trapped set of the geodesic flow . the universal cover @xmath477 of @xmath3 is a subset of the open disk @xmath556 with infinitely many boundary components which are pieces of circles intersecting @xmath559 at their endpoints . the solution of is @xmath560 and the integral kernel of @xmath495 and @xmath518 are thus @xmath561 if @xmath562 , @xmath563 and @xmath564 with @xmath565 a fundamental domain . notice that the expression for @xmath528 extends to @xmath566 and this defines an operator @xmath567 so that @xmath568 if @xmath60 . we prove the following [ norml^2pi0 ] let @xmath23 be a manifold with strictly convex boundary and constant negative curvature @xmath69 , let @xmath545 be its fundamental group and @xmath569 the exponent of convergence of @xmath486 as defined in . let @xmath570 $ ] which satisfy the bound @xmath571 and set @xmath572 , then for all metric @xmath2 on @xmath3 with strictly convex boundary and gauss curvature @xmath75 satisfying @xmath573 the operator @xmath574 of @xmath2 has norm bounded by @xmath575 let @xmath576 so that @xmath577 . first , by the comparison result , we have the pointwise bound on the kernel @xmath506 of @xmath495 in the universal cover @xmath578 @xmath579 where we used @xmath580 and @xmath581 on @xmath477 . using the lower bound @xmath582 for @xmath455 , we get @xmath583 this implies that the kernel @xmath528 of @xmath518 satisfies @xmath584 provided that the sum converges , which is the case if @xmath585 . let @xmath586 be the metric with gauss curvature @xmath263 on @xmath3 and let @xmath587 be the operator on @xmath3 whose integral kernel with respect to the volume form @xmath588 is @xmath589 . let @xmath590 equipped with the hyperbolic metric @xmath591 ( with curvature @xmath263 ) and consider the operator @xmath592 on @xmath593 defined by . as explained above , the manifold @xmath593 can be viewed as a complete extension of the manifold with boundary @xmath594 , and we have for each @xmath595 extended by @xmath88 on @xmath593 @xmath596 the operator norm of @xmath592 on @xmath597 is computed in lemma [ pi0hyp ] and we denote this constant by @xmath598 . we then get @xmath599 , we use together with the fact that the integral kernels of @xmath518 and @xmath587 operator are positive and that @xmath600 this gives the bound @xmath601 the result then follows from and the value for @xmath602 given by lemma [ pi0hyp ] . combining proposition [ normw ] with theorem [ norml^2pi0 ] , we obtain theorem [ theo2 ] . we now illustrate with numerical examples the reconstruction formula proposed in theorem [ inversion ] , i.e. , we will reconstruct functions from knowledge of their ray transform , where the underlying surface with boundary is of one of the following two models : * a quotient of the poincar disk @xmath603 by a schottky group @xmath486 . * the cylinder @xmath604 with a metric of circular symmetry such that the trapped set is hyperbolic in example @xmath605 , the constancy of the curvature makes the operator @xmath65 vanish identically , so that the inversion is expected to be one - shot . in example @xmath606 , we will implement a partial neumann series to invert the operator @xmath607 . much of the ideas and implementation draw from the second author s previous implementation @xcite in the case of non - trapping surfaces . the numerical domain can be viewed as a domain in @xmath608 with an isotropic metric @xmath609 ( except for experiment 5 , where we use a metric of the form @xmath610 ) , and with certain identifications producing nontrivial topology and trapped geodesics . the underlying grid is cartesian ( subset of a @xmath611 uniform discretization of @xmath612 ^ 2 $ ] with @xmath613 here ) , which simplifies the computation of the operator @xmath614 . more specifically , in isothermal coordinates @xmath615 with an isotropic metric of the form @xmath609 , the operator @xmath616 appearing in takes the expression @xmath617 where , despite the fact that this formula might not make sense for those @xmath615 in the trapped set , the hyperbolicity of the flow is such that , when discretizing the integral over @xmath618 , numerical imperfections will make every geodesic exit the domain . the operators @xmath619 , @xmath620 can then conveniently computed by finite differences . on the other hand , this discretization implies that every pixel will carry a different volume , and that accuracy of reconstructions and resolution may be visually position - dependent . to the authors knowledge , litterature on numerical analysis of ray transforms on manifolds ( covering , e.g. , error estimates , appropriate sampling , regularization ) , is very scarce outside the well - understood euclidean case @xcite , an exception being on geodesic and horocyclic transforms on the hyperbolic plane in @xcite , whose homogeneity allows for harmonic analysis . the computation of geodesics is done in two ways depending on the purpose : * when computing the ray transform of a given function , we solve an ode and integrate the function along the way . the method used in that case is heun s scheme . * in the case of quotients of @xmath621 , where geodesics can be given by explicit formulas , we use these formulas to compute geodesic endpoints in a fast way during the backprojection step . computations are performed with matlab , using gpu in order to speed up calculations . as in @xcite , the main bottleneck is the backprojection step , which requires computing @xmath622 geodesic endpoints , with @xmath623 the gridsize and @xmath624 the number of directions used to compute integral over @xmath625 . this process requires computing whole geodesics , more of whom take longer to exit the domain when trapping increases ( this increase here is quantified by the fractal dimension of the limit set of the corresponding schottky group ) . accordingly , computations run within minutes on a personal computer with gpu card for the case of cylinders , hours for the case of schottky groups with two generators . here and below , we denote @xmath626 the unique geodesic passing through @xmath627 with direction @xmath628 . fix @xmath629 , let @xmath630 and define @xmath631 by @xmath632 the unique hyperbolic translation mapping @xmath331 to @xmath633 and define the surface @xmath634 . a fundamental domain for this surface is delimited by the geodesics @xmath635 and @xmath636 . this surface has no boundary , so we cut one up using an arc circle which intersects the geodesic @xmath635 at a right angle , which ensures that the boundary is smooth . this boundary has two connected components . the forward trapped set consists of , at every point , precisely two directions : the ones whose flow - out lands into either `` point at infinity '' @xmath637 or @xmath638 . the trapped set is simply the projection of the two oriented geodesics in @xmath639 relating @xmath637 to @xmath638 , and these are two oriented closed geodesics in the quotient . * experiment 1 : * we pick the function in fig . [ fig : model1 ] ( right ) , compute its ray transform on fig . [ fig : model1_recons ] ( left ) using @xmath640 boundary points on each top and bottom boundary , and @xmath641 directions for each boundary point uniformly within @xmath642 . we then apply the inversion formula , discretizing the @xmath618 integrals using @xmath640 directions . the error is visualized on fig . [ fig : model1_recons ] ( right ) . . the @xmath331-axis represents a point on the boundary parameterized with a parameter in @xmath643 $ ] , and the @xmath644-axis is the shooting direction with respect to the inner normal , an angle in @xmath645 . right : pointwise error on reconstruction after applying the formula from theorem [ inversion ] to the data.,title="fig : " ] . the @xmath331-axis represents a point on the boundary parameterized with a parameter in @xmath643 $ ] , and the @xmath644-axis is the shooting direction with respect to the inner normal , an angle in @xmath645 . right : pointwise error on reconstruction after applying the formula from theorem [ inversion ] to the data.,title="fig : " ] * observations : * based on the results displayed on fig . [ fig : model1_recons ] , one can notice artifacts of 10% relative size in certain areas . these artifacts lie in the region between the support of the function and the points at infinity ( this is especially noticeable once we periodize the reconstruction ) , and we propose two possible reasons for this : * as the computation of forward data does not use an exact computation of the geodesics , the accuracy degrades on the longer geodesics , i.e. the ones that wander near the closed one . * the data may have logarithmic singularities at trapped directions , which upon applying the hilbert transform , may create undesirable oscillations in the vicinity of these angles . these oscillations then create artifacts upon applying the backprojection @xmath48 , especially at the points which are such that the data is nonzero near their trapped directions ( see fig . [ fig : exp2_2 ] below ) . this is precisely the case when these points lie between the support of the function and the points at infinity . we now consider examples of quotients of @xmath603 by schottky groups with two generators . keeping @xmath629 , @xmath646 and @xmath647 defined as above , we define the surfaces @xmath648 and @xmath649 . both @xmath650 and @xmath651 can have the same fundamental domain , delimited by the geodesics @xmath652 and @xmath653 , though with different identifications , as represented fig . [ fig : schottky2 ] . in both examples , the trapping is now more complicated : at every point @xmath654 , the trapped directions form a fractal set at the tangent circle @xmath655 . in light of experiment 1 , this is expected to have quite a detrimental impact on the inversion . we study this influence by comparing noise on experiments 2 and 4 , where the same model @xmath650 is chosen for two different values of @xmath331 , yielding different trapping intensities . we quantify this by the dimension of the limit set @xmath656 $ ] of the group @xmath657 . we compute a numerical approximation of @xmath658 by running geodesic dynamics on a large number of points inside the domain and measuring their escape rate , i.e. the proportion that remains inside as a function of time . according to ( * ? ? ? * appendix b ) , this escape rate is of the form @xmath659 , and we use this to approximate @xmath658 by linearly interpolating @xmath660 , possibly after throwing away transient regimes . in experiments 2 - 3 - 4 , the influx boundary is discretized uniformly into @xmath661 boundary points and @xmath662 influx directions in the range @xmath663 per boundary point . * experiment 2 . * on @xmath664 , we pick the function in fig . [ fig : exp2 ] ( left ) , compute its ray transform on fig . [ fig : exp2 ] ( right ) and apply the inversion formula . the reconstruction is visualized on fig . [ fig : exp234 ] ( left ) . for later comparison , we have computed @xmath665 following the method described above . * experiment 3 . * on @xmath666 , we pick the function in fig . [ fig : exp3 ] ( left ) , compute its ray transform on fig . [ fig : exp3 ] ( right ) and apply the inversion formula . the reconstruction is visualized on fig . [ fig : exp234 ] ( middle ) . * experiment 4 . * we now repeat experiment 2 , this time on @xmath667 , whose topology is the same , though the trapping is now more prominent ( we have computed @xmath668 ) . we pick the function in fig . [ fig : exp4 ] ( left ) , compute its ray transform on fig . [ fig : exp4 ] ( right ) and apply the inversion formula . the reconstruction is visualized on fig . [ fig : exp234 ] ( right ) . * observations . * as in experiment 1 , the same comments hold regarding the influence of trapping : the more trapping , the more singularities in the data , which create oscillations upon applying the hilbert transform ( see fig . [ fig : exp2_2 ] ) , in turn yielding artifacts when applying the backprojection @xmath48 . comparing experiments 2 and 4 , the difference in trapping can be both seen on the singularities of the forward transforms ( compare figs . [ fig : exp2 ] , right and [ fig : exp4 ] , right ) , and on higher noise in reconstruction ( compare figs . [ fig : exp234 ] , left and right ) . let the topological cylinder @xmath604 endowed with the metric @xmath669 , where we have defined @xmath670 the gauss curvature is independent of @xmath331 and given by @xmath671 so we have @xmath672 for every @xmath644 , and we also get @xmath673 the structure of the trapped set is similar to that of experiment 1 , with the set @xmath674 the projection on the base of the only closed geodesic and @xmath675 . we can bound @xmath2 in terms of the constant curvature metric @xmath22 with @xmath676 via the constants @xmath677 and @xmath678 as in theorem [ norml^2pi0 ] . the requirement @xmath679 there gives the constraint @xmath680 . since @xmath681 , we deduce the estimate , with @xmath682 , @xmath683 combining this estimate with the estimate in proposition [ normw ] , we deduce @xmath684 in particular , @xmath685 and thus for @xmath686 small enough , inversion via a neumann series is justified . as in @xcite , when implementing the neumann series , the error operator @xmath687 is not implemented directly as it would be hopeless to satisfy at the discrete level . instead , once @xmath0 and its approximate inverse ( call it @xmath688 , say ) are discretized , we set @xmath689 and iterate the neumann series using this operator . * experiment 5 . * we pick the function in fig . [ fig : exp5 ] ( right ) , compute its ray transform on fig . [ fig : exp5_2 ] ( top - left ) and implement the approximate inversion formula ( visualized fig . [ fig : exp5_2 ] , top - right ) , followed by two neumann series iterations . pointwise errors after both one - shot and corrected reconstructions are visualized fig . [ fig : exp5_2 ] ( bottom ) . in light of these experiments , we make the following comments : * despite the theoretical exactness and stability of the proposed reconstruction formulas , geometries with trapping pose some numerical challenges , related to ( i ) the singularities it creates in data space , which in turn create oscillations after applying the hilbert transform , yieling reconstruction artifacts after backprojection ; and ( ii ) , the loss of accuracy on the computation of long geodesics which wander near the trapped set before exitting the domain . * as mentioned before in @xcite , another important challenge in surfaces with nonconstant curvature is to understand how to design an appropriate discretization of the influx boundary whose flow - out samples the manifold in as uniform a way as possible . * in the case of non - constant curvature , iterative corrections , which require a differentiation step at every iteration , may increase noise at high frequencies ( which , as mentioned earlier , becomes more prominent in the presence of trapping , unlike the case of simple surfaces covered in @xcite ) . regularization methods should then be considered . let @xmath691 be a quotient of hyperbolic space by a convex co - compact group @xmath545 and denote by @xmath658 the exponent of the group @xmath486 as defined by . such a manifold is non - compact but can be compactified into @xmath692 where @xmath693 is the set of discontinuity of @xmath486 ; @xmath694 is the largest open set in the closed unit disk @xmath554 where @xmath486 acts properly discontinuously ( we view @xmath87 as the unit disk @xmath556 in @xmath695 ) . the laplacian @xmath696 has continuous spectrum @xmath697 on @xmath396 and possibly finitely many eigenvalues in @xmath698 , which appear if and only if the exponent @xmath699 of @xmath486 satisfies @xmath700 , in which case @xmath701 . for further details on these topics , we refer to the book @xcite . in this appendix , we compute for small @xmath702 the operator on @xmath3 @xmath703 where @xmath14 denotes the geodesic flow at time @xmath16 on the unit tangent bundle @xmath15 . just like in section [ normofpi0 ] , the operator @xmath690 makes sense and has integral kernel @xmath704 , which can written as a converging sum on a fundamental domain @xmath565 of @xmath486 when @xmath705 @xmath706 for @xmath564 and @xmath707 the covering map , where @xmath708 is the integral kernel of the operator @xmath709 defined by the expression when @xmath710 . as in section [ normofpi0 ] , this kernel is given by @xmath711 which justifies that the sum above converges if @xmath705 . the expression can also be rewritten as @xmath712 using the symmetry @xmath713 in each fiber @xmath714 . [ pi0hyp ] on a convex co - compact quotient @xmath691 with exponent of convergence given by @xmath658 , the operator @xmath690 can be written when @xmath715 as @xmath716 where @xmath717 is the euler gamma function , and with the convention that @xmath718 when @xmath719 is an @xmath146-eigenvalue of @xmath720 with @xmath721 . the operator norm is given by @xmath722 since its schwartz kernel is a function of @xmath551 decaying like @xmath723 , the operator @xmath709 is a function @xmath724 of the hyperbolic laplacian @xmath725 ( see for example ( * ? ? ? * section 1.8 ) ) . notice that the series defining @xmath704 is absolutely convergent on compact sets of @xmath726 since @xmath727 if @xmath486 is convex co - compact . the operator @xmath728 is bounded on @xmath729 if @xmath730 : indeed , this is an operator in the class @xmath731 of the @xmath88-calculus of pseudo - differential operators introduced by mazzeo - melrose @xcite and the @xmath146-boundedness on @xmath87 follows for instance from mazzeo ( * ? ? ? * theorem 3.25 ) . to compute the function @xmath732 , we use the representation theory of @xmath733 , following @xcite . the ( time reversal ) principal series representations are unitary representations into @xmath734 defined as follows @xmath735 let @xmath736 , then @xmath737 ( if we identify @xmath738 in the canonical way ) and to compute @xmath739 we will use the plancherel formula of harish - chandra @xcite @xmath740 there is a unique @xmath32-invariant vector of norm @xmath43 in @xmath741 where @xmath32 is the action of @xmath742 via the representation @xmath743 , it is given by @xmath744 since @xmath745 is invariant by @xmath746 , to consider the action of the right multiplication by an element @xmath747 on @xmath745 with @xmath748 , it suffices to consider the action of @xmath749 on @xmath750 . the pull - back by the geodesic flow at time @xmath16 corresponds , in each unitary representation @xmath743 , to the unitary operator on @xmath741 @xmath751 the operator @xmath728 can be written as @xmath752 where @xmath753 is any element which projects to @xmath754 . since the average in the compact group @xmath746 amounts to project onto @xmath750 , the operator @xmath709 is unitarily equivalent to multiplication on the representation space @xmath741 by @xmath755 through the representation @xmath743 . by changing variable and by parity in @xmath331 , this term is of the form @xmath756 with @xmath757 notice that @xmath758 and that @xmath756 extends holomorphically in each strip @xmath759 and it can be written under the form @xmath760 for some @xmath732 holomorphic in the region containing @xmath761 and bounded by the parabola @xmath762 . the operator @xmath709 can then be written @xmath763 with @xmath764 the expression of @xmath765 for @xmath433 matches with the formula found by @xcite . in the image of the strip @xmath766 by @xmath767 for @xmath233 small enough @xmath768 for some @xmath769 depending only on @xmath686 . we can then write , using cauchy formula @xmath770 where @xmath771 is the contour given by the image of @xmath772 by the map @xmath767 , where @xmath233 is chosen small . the contour is parametrized by @xmath773 with @xmath774 and the @xmath775 norm of @xmath776 is @xmath777 showing that the integral converges in @xmath778 norm . now the resolvent @xmath779 has for integral kernel , viewed on the fundamental domain @xmath487 , @xmath780 when @xmath781 , and the sum is uniformly convergent on compact sets of @xmath782 for @xmath783 in that half - space : the proof is done in @xcite or ( * ? ? ? * proposition 2.2 ) , and follows from the uniform convergence of poincar series @xmath784 for @xmath785 on compact sets and from the pointwise estimate for @xmath786 ( which follows from the explicit formula for @xmath787 given for instance in @xcite ) @xmath788 if @xmath789 is fixed and @xmath790 depends only on @xmath791 and @xmath686 . this implies that , for @xmath792 and @xmath793 two relatively compact open sets of @xmath87 so that @xmath794 , there is @xmath244 depending only on the volume of @xmath792 and @xmath793 such that for all @xmath795 @xmath796 now , using that for @xmath434 supported in a compact set @xmath792 we have @xmath797r_{{\mathbb{h}}^2}(r^2)f= ( 1+r^2)\chi r_{{\mathbb{h}}^2}(r^2)f\ ] ] for each @xmath798 supported in a relatively compact set @xmath793 with @xmath799 , we get @xmath800 for @xmath801 where @xmath802 depends only on @xmath803 . this estimate together with the poincar series convergence implies that for each relatively compact open set @xmath804 , @xmath805 and @xmath806 , @xmath807 converges in the sense of schwartz kernels of bounded operator @xmath808 . this shows that for @xmath809 @xmath810 where @xmath732 is the holomorphic function defined before . now the bottom of the spectrum of @xmath696 corresponds to @xmath811 . let us show that the function @xmath812 is decreasing : we have @xmath813 where @xmath814 is the digamma function , and using its series expansion @xmath815 with @xmath816 the euler constant , we get for @xmath455 @xmath817 is negative . similarly , the function @xmath818 is increasing in @xmath819 : indeed , one has @xmath820 thus using the series expansion of @xmath821 and factorizing we obtain for @xmath822 @xmath823 from the spectral theorem , this implies that the @xmath824 norm of @xmath690 is obtained by evaluating @xmath732 at either @xmath88 if @xmath825 or at @xmath826 if @xmath700 . this ends the proof . using a similar analysis , the decomposition of @xmath827 into irreducible representations when @xmath486 is co - compact ( see for instance @xcite ) proves that the operator @xmath518 defined in @xcite is given on a compact hyperbolic manifold @xmath828 by the same formula as in lemma [ pi0hyp ] , but setting @xmath433 and replacing @xmath696 by @xmath829 if @xmath830 is the orthogonal projection onto @xmath831 in @xmath396 . b. fridman , p. kuchment , k. lancaster , s. lissianoi , m. mogilevski , d. ma , i. ponomarev , v. papanicolaou , _ numerical harmonic analysis on the hyperbolic plane _ , applicable analysis , taylor & francis , * 76 * ( 2000 ) , 351362 s. helgason , _ the totally - geodesic radon transform on constant curvature spaces . _ integral geometry and tomography ( arcata , ca , 1989 ) , 141149 , contemp . math . , 113 , amer . soc . , providence , ri , 1990 . j. radon _ uber die bestimmung von funktionen durch ihre integralwerte lngs gewisser mannigfaltigkeiten _ leipzig , math . * 69 * ( 1917 ) , 262277 ; english translation in deans , s. , the radon transform and some of its applications , krieger 1993 , 204217 .

we use a microscopic slater - koster tight - binding model with short - range exchange and atomic spin - orbit interactions that realistically captures generic features of ferromagnetic metal nanoparticles to address the mesoscopic physics of magnetocrystalline anisotropy and hysteresis in nanoparticle quasiparticle excitation spectra . our analysis is based on qualitative arguments supported by self - consistent hartree - fock calculations for nanoparticles containing up to 260 atoms . calculations of the total energy as a function of magnetization direction demonstrate that the magnetic anisotropy per atom fluctuates by several percents when the number of electrons in the particle changes by one , even for the largest particles we consider . contributions of individual orbitals to the magnetic anisotropy are characterized by a broad distribution with a mean more than two orders of magnitude smaller than its variance and with no detectable correlations between anisotropy contribution and quasiparticle energy . we find that the discrete quasiparticle excitation spectrum of a nanoparticle displays a complex non - monotonic dependence on an external magnetic field , with abrupt jumps when the magnetization direction is reversed by the field , explaining recent spectroscopic studies of magnetic nanoparticles . our results suggests the existence of a broad cross - over from a weak spin - orbit coupling to a strong spin - orbit coupling regime , occurring over the range from approximately 200- to 1000-atom nanoparticles .

cond-mat0209608

interest in the properties of magnetic nanoparticles has grown recently , partly because of advances in synthesis and measurement techniques and partly because of potential applications for high storage - density magnetic media and spin electronics . ferromagnetic nanoparticles with diameters of a few nanometers containing of order of 1000 or fewer atoms can now be reliably fabricated and studied with a variety of different methods@xcite . small monodomain magnetic particles have traditionally@xcite been described using classical _ micromagnetic _ theory , in which the total energy is expressed as a function of the magnetization orientation . shape and magnetocrystalline magnetic anisotropy leads to a dependence of energy on orientation , to barriers that separate minima that occur at _ easy _ magnetization orientations@xcite , and to hysteretic discontinuous changes in orientation as a function of the strength of an external magnetic field . when the size of a magnetic particle is only a few nanometers , the discrete nature of its quantum energy spectrum can be directly observable at low temperatures and starts to affect the magnetic properties of the particle . a milestone in the experimental study of _ individual _ ferromagnetic metal nanoparticles was achieved recently by deshmukh , guron , ralph _ et.al._@xcite , using single - electron tunneling spectroscopy . by exploiting the coulomb blockade effect , these experimentalists were able to resolve _ individual quantum states _ in the discrete many - body excitation spectrum of single ferromagnetic metal nanograins with sizes in the range from @xmath0 to @xmath1 nm . as in bulk ferromagnetic metals , the low - energy excitations of the nanoparticles that were probed in these experiments involve both particle - hole excitations of the electronic quasiparticles and quantized collective excitations of the magnetization - orientation field that appears in the classical micromagnetic theory . in an initial attempt to achieve an understanding of the novel physics evident in the external field dependence of the experimental excitation spectra , two of us recently@xcite analyzed a simple quantum model with long range exchange interactions . we were able to demonstrate explicitly that the low - energy excitations of a ferromagnetic metal nanoparticle are specified by the occupation numbers of its quasiparticle orbitals , as in a fermi liquid , _ and _ by the global orientation of the aligned spins of all single - occupied ( and therefore spin - polarized ) orbitals . this model is not , however , able to account realistically for the influence of spin - orbit interactions , which play the essential role in controlling the complex hysteretic behavior seen in these experiments@xcite . kleff , von delft , deshmukh and ralph@xcite have proposed that the single - electron tunneling spectroscopy experiments can be explained by accounting for non - equilibrium spin accumulation , and by assuming that the magnetic anisotropy energy of a small magnetic particle has surprisingly large fluctuations as a function of the number of electrons on the particle . this assumption leads to a non - trivial magnetic field dependence of tunneling resonances that resembles experimental behavior . more critically , these authors point out that if non - equilibrium spin and quasi - particle excitations both occur@xcite , the low energy spectra are characterized by many closely spaced resonances , consistent with experiment@xcite . in ref . we presented a possible approach toward achieving a unified and consistent quantum description of both collective and quasiparticle physics in magnetic nanoparticles . the attempts put forward in refs . to develop a quantum description of ferromagnetic metal nanoparticles do not address the microscopic origin of the magnetic anisotropy and the fluctuations of this quantity that are a necessary consequence of spin - orbit interactions in mesoscopic systems . this article addresses these issues and , more in general , investigates the changes in magnetic properties of small metallic nanoparticles that occur because of the finite spacing between quasiparticle levels near the fermi energy . our conclusions follow from qualitative arguments based on perturbation theory expressions for the influence of spin - orbit interactions on quasiparticle energy levels , and on numerical studies of a simplified model that we believe is sufficiently realistic to describe generic aspects of the interrelated mesoscopic physics of quasiparticle energy - levels and magnetic anisotropy energy in ferromagnetic metal nanoparticles . we are particularly interested in the variation of quasiparticle energies with fields on the scale of the coercive field , which is trivial in the absence of spin - orbit interactions , but entirely non - trivial in their presence . the model we study is based on a slater - koster tight binding hamiltonian and a mean - field treatment of exchange interactions . our aim is to understand the dependence of magnetic anisotropy and quasiparticle energy - level - spacing statistics on particle size and shape , external magnetic field , and , with single - electron - transistor systems in mind , also on electron number . we find that , because of the absence of strong correlations between angular momentum operator matrix elements and orbital energy differences , the size of spin - orbit induced energy shifts in nanoparticles and bulk perfect crystals are similar . on the basis of estimates for energy shifts and for avoided crossing gaps , we predict that a crossover from weak to strong quasiparticle spin - orbit scattering will occur over the range from approximately 200-atom to approximately 1000-atom nanoparticles . we find that for small particles the contribution of individual quasiparticle orbitals to the magnetocrystalline anisotropy energy has a wide distribution , characterized by a variance comparable to the spin - orbit - scattering lifetime broadening energy , @xmath2 , and a mean that is smaller by more than two orders of magnitude . surprisingly , we find no measurable correlation between the contributions to magnetic anisotropy from quasiparticles that are close in energy in this limit . as a result of the statistical properties of the quasiparticle magnetic anisotropies , the total magnetic anisotropy per atom fluctuates by several percents when the number of electrons in the nanoparticle changes by one , even for particles containing 260 atoms . finally , in agreement with experiment@xcite , we find that the quasiparticle excitation spectrum exhibits a complex non - monotonic behavior as a function of an external magnetic field , with abrupt jumps when the magnetization orientation of a nanoparticle changes discontinuously in response to the field . our analysis provides insight into mesoscopic fluctuations of the micromagnetic energy functional which appears in the classical theory of a magnetic nanoparticle . our paper is organized in the following way . in section [ model ] we introduce the model and describe the formalism . in section [ so_anis ] we analyze the qualitative change in quasiparticle energy - level statistics induced by spin - orbit interactions , and discuss the connection between the quasiparticle properties and the magnetic anisotropy of a ferromagnetic nanoparticle . fluctuations of the magnetic anisotropy as a function of spin - splitting field , atom number , and electron number are investigated in section [ fluct ] . magnetic hysteresis and the external field dependence of quasi - particle energy levels are discussed in section [ hysteresis ] . finally , in section [ conclusion ] we summarize our findings and present our conclusions . we model the nano - particle as a cluster of @xmath3 atoms located on the sites of a truncated crystal . the numerical results we present here are for a cobalt cluster whose truncated f.c.c . crystal is circumscribed by a hemisphere whose equator lies in the @xmath4-plane of the f.c.c . the choice of a hemisphere is motivated by the tunneling experiments of ref . the crystal structure could not be identified in these experiments . high resolution transmission electron microscopy measurements on co clusters prepared in a different way@xcite have shown that the nanoparticles are well crystallized in the f.c.c . structure even though bulk co has an h.c.p . structure . ] the model we use is intended to qualitatively capture the physics of a transition - metal , itinerant - electron ferromagnet . we use a s - p - d tight - binding model for the quasiparticle orbitals , with 18 orbitals per atom including the spin - degree of freedom . nine orbitals per co atom are occupied for in neutral nanoparticles . the full hamiltonian is , @xmath5 here @xmath6 is a one - body term describing the orbital motion of the electrons . in second quantization , it has the form @xmath7 where @xmath8 and @xmath9 are fermion creation and annihilation operators for single - particle states labeled by @xmath10 . the indices @xmath11 are atomic site labels , and @xmath12 couples up to second nearest - neighbors . the indices @xmath13 label the nine distinct atomic orbitals ( one @xmath14 , three @xmath15 and five @xmath16 ) . the spin degrees of freedom , labeled by the index @xmath17 , double the number of orbitals at each site . it is useful for us to vary the spin - quantization axis , which is specified by a unit vector @xmath18 where @xmath19 and @xmath20 are the usual angular coordinates defined with respect to the f.c.c . crystal axes . the parameters @xmath21 are slater - koster parameters@xcite obtained after performing a lwdin symmetric orthogonalization procedure@xcite on the set of slater - koster parameters for non - orthogonal atomic orbitals of bulk spin - unpolarized co@xcite . the electron - electron interaction term in eq . ( [ totham ] ) is simplified by introducing explicitlynotice however , that the one body term , @xmath22 implicitly includes a mean - field approximation to those portions of the interaction not captured by the exchange term . ] only the ferromagnetic exchange interaction , @xmath23 , between the electrons spins of @xmath24-orbitals on the same atomic site . these interactions are largely responsible for magnetic order in transition metal ferromagnets : @xmath25 where @xmath26 the parameter @xmath27 in eq . ( [ exchh ] ) determines the strength of the exchange interaction and is set equal to @xmath0 ev . this value of @xmath27 gives rise in our finite clusters to an average magnetic moment per atom of the order of @xmath28 , which is larger than the bulk value in co@xcite , in agreement with other calculations@xcite and experimental results@xcite for co clusters . this value of @xmath27 is also approximately consistent with a mean - field relationship , derived below , between band spin - splitting and magnetization , using bulk values@xcite for these two quantities . in eq . ( [ pauli ] ) , @xmath29 is a vector whose components @xmath30 , @xmath31 are the three pauli matrices . the third term in eq . ( [ totham ] ) , @xmath32 , is a one - body operator , essentially atomic in character , representing the spin - orbit interaction . it can be written as@xcite @xmath33 the atomic matrix elements @xmath34 can be been calculated explicitly as a function of the direction of the magnetization @xmath35@xcite . the energy scale @xmath36 , which characterizes the coupling between spin and orbital degrees of freedom , varies in the range from 50 to 100 mev in bulk 3d transition metal ferromagnets@xcite . for our calculations we have used @xmath37 mev , taken from ref . @xcite . spin - orbit coupling gives rise to a dependence of the total energy of a ferromagnet on the direction of its spontaneous magnetization , an effect known as magneto - crystalline anisotropy more precisely , this is the magneto - electric contribution to the magneto - crystalline anisotropy . ] ; with spin - orbit coupling the magnetization is partially orbital in character@xcite and is sensitive to orbital anisotropy due to crystal - field interactions on an atomic site , due to the spatial arrangement of the atoms neighboring that site to which electrons can hop , and due to the overall shape of the full nanoparticle that becomes available to an electron after many hops . the additional anisotropy due to magnetic dipole interactions is also sensitive to the overall nanoparticle shape , but is not included here because it is generally less important for nanoparticles than for macroscopic samples . this contribution to the anisotropy is additive to a good approximation , does not have substantial mesoscopic fluctuations , and can be simply added to the effects discussed here when it is important . ] we will see below that the hemispherical shape of the nanoparticles we have studied plays the dominant role in determining their magnetic anisotropy energy . the last term in eq . ( [ totham ] ) is a local one - body operator , representing the zeeman coupling of the orbital and spin degrees of freedom to an external magnetic field @xmath38 : @xmath39 the extreme sensitivity of magnetization orientation to external magnetic field is a combined effect of the collective behavior of many electrons in a ferromagnetic nanoparticle and the smallness of the magnetic anisotropy energy relative to the overall energy gain associated with ferromagnetic order . the most delicate physics of a ferromagnetic nanoparticle , and in our view the most interesting , is that associated with magnetization direction reorientation by weak external magnetic fields . thus the interplay between the zeeman term and the magnetic anisotropy produced by spin - orbit interactions is at the heart of the physics we intend to address . we seek a ferromagnetic solution to the mean - field equations for this model , decoupling the quartic term in the exchange interaction using the ansatz @xmath40 ignoring terms that are second order in @xmath41 , and determining ground state expectation values , @xmath42 , self - consistently . this standard procedure leads to a hamiltonian that can be diagonalized numerically and to a self - consistency condition that can be solved iteratively to determine the mean - field order parameters . for the present calculation we have simplified this procedure further , by averaging the spin - splitting exchange mean field , @xmath43 over all sites . our motivation for doing so is to simplify the magnetic anisotropy energy landscape discussed below , forcing all spins to change their orientations coherently . we recognize that complicated non - collinear spin configurations@xcite commonly occur in magnetic nanoparticles , and that under the action of an external field small groups of atoms can change their orientation relative to other parts of the nanoparticle . complex magnetization reorientation processes are obvious in addition spectroscopy experiments@xcite . by forcing atoms to change their magnetic orientations coherently , we are restricting our attention to relatively large nanoparticles ( @xmath44 ) in which most atoms have parallel spins and to nanoparticles with simple bistable hysteretic behavior in which the physics we address will be easier to study . the beautiful series of detailed squid magnetometry experiments by wernsdorfer and colleagues@xcite demonstrate that nanoparticles can be prepared that do have simple coherent magnetization reversal properties . by using a simplified model hamiltonian we are able to deal with larger nanoparticle systems than would be possible with a first principles calculation@xcite ; since we are interested only in generic aspects of the ferromagnetic nanoparticle physics there is little to gain from the additional realism that could be achieved by performing self - consistent spin - density functional calculations . given the averaged spin - splitting field @xmath45 , the mean - field hamiltonian is now a single - body operator @xmath46 the self - consistent spin - spitting field is also the field , denoted by @xmath47 below , at which the total ground - state energy function @xmath48 is minimized . notice that @xmath49 has the dimension of an energy . in fact , in absence of spin - orbit interaction , @xmath50 can be identified with the band spin - splitting @xmath51 , where @xmath52 and @xmath53 are the majority and minority spin quasi - particle fermi energies . the use of two fermi energies for majority and minority spins can generate some confusion : in this terminology , the energies are pure kinetic ( or band ) energies , _ i.e _ the exchange contribution is not included . when the exchange energy is included , as we do below in the context of mean - field theory , majority and minority spins have the same fermi level . ] diagonalization of the quadratic hamiltonian yields a set of quasi - particle energies @xmath54 , which in the ground state are filled up to a fermi energy determined by the number of valence electrons in the nanoparticle . strictly speaking , because of the spin - orbit interaction , the spin character of the corresponding eigenstates @xmath55 is not well defined . for small nanoparticles , most eigenstates have _ predominantly _ spin - up or spin - down character and we sometimes use this property to assign spin labels @xmath56 for spins along the order direction and @xmath57 for reversed spins . the magnetic anisotropy energy of small ferromagnetic particles@xcite has two fundamentally distinct origins , long - range magnetic - dipole interactions which cause a dependence on overall sample shape and short - range exchange interactions that , because of atomic - like spin - orbit interactions , are sensitive to all aspects of the electron hopping network including bulk crystal symmetry , facet orientations , and also overall sample shape . we concentrate here on spin - orbit - induced magnetocrystalline anisotropy which gives rise to the most interesting physics in ferromagnetic nanoparticles . when magnetostatic shape anisotropy is important , it can be added as a separate contribution . we begin our discussion with some qualitative estimates of the effect of spin - orbit interactions that are based on perturbation theory@xcite . in bulk 3d transition metal ferromagnets , spin - orbit interactions are relatively weak . their coupling strength is less than @xmath58 of the d - band width @xmath59 in bulk materials@xcite , allowing the energy shifts they produce to be estimated perturbatively . because of angular momentum quenching in the absence of external fields , the expectation value of @xmath60 is zero , even in case of ferromagnets@xcite . the quasi - particle energy shift due to spin - orbit interactions is given by second - order perturbation theory as @xmath61 where @xmath62 and @xmath63 are respectively the single - particle eigenstates and energies in the absence of spin - orbit interaction . in small particles , the importance of the spin - orbit interactions can be assessed by comparing the spin - orbit energy shift @xmath64 with the single - particle mean - level spacing @xmath65 . in an infinite periodic solid only states at the same @xmath66 are coupled and these are separated energetically by an energy comparable to the bandwidth @xmath59 . in a nanoparticle a given state will be coupled to many other orbitals , but the coupling matrix elements will be reduced in accord with the following sum rule : @xmath67 the estimate for the right hand side of eq . [ eq : sumrule ] is based on the atomic character of the angular momentum in our model , and uses that @xmath68/9 $ ] , with the estimate of the typical @xmath69 expectation value representing an average over @xmath24 , @xmath17 and @xmath70 orbitals . it follows that , unless there are important correlations between angular momentum matrix elements and quasiparticle energy differences , the typical shift in energy caused by the spin - orbit interaction is @xmath71 , in both bulk crystals and in nanoparticles , which is in the range between 1 to 10 mev . for example in co , using @xmath72mev and @xmath73ev , this rough estimate gives for the magnitude of the spin - orbit energy shift @xmath74 @xmath75mev . the sign of the shift might be expected to be sensitive to the spacing of nearby quasiparticle orbitals . the anisotropy energy , that is , the dependence of the total band energy on the magnetization orientation , is given to a good approximation by a partial canceling sum of spin - orbit induced energy shift dependences on magnetic orientation . in the approximation that the exchange field is orbital independent , majority spin and minority spin orbitals are identical and differ only in their occupation numbers . in this approximation there is no contribution to the anisotropy energy from doubly occupied orbitals . because of the cancellations , the anisotropy energy per atom is much smaller than @xmath76 . for example , the zero - temperature anisotropy energy per atom in bulk is @xmath77ev for h.c.p . co , and @xmath78ev for b.c.c . fe and f.c.c . ni@xcite . in a finite or disordered system , there will always be perturbative coupling to quasiparticle states close in energy in eq . [ eq : perttheory ] , but the matrix elements , which satisfy the sum rule of eq . [ eq : sumrule ] , will be distributed among many states and typical energy shifts in nanoparticles should generally be comparable to those in bulk perfect crystals . typically net anisotropy energies per atom in small magnetic particles are larger than the bulk because of the loss of symmetry at the surface . an important quantity used to characterize the strength of spin - orbit interactions in bulk systems and large nanoparticles is the the spin - orbit scattering time , @xmath79@xcite . in the weak coupling regime , it is given by the fermi s golden rule @xmath80\ ; , \label{eq : so_sc_time}\ ] ] where the @xmath81function is understood to be broadened to a width much larger than the level spacing . assuming that there is no correlation between angular momentum matrix elements and orbital energy differences , it follows from the sum rule mentioned above that @xmath82 the absence of strong correlations between energy differences and angular momentum matrix elements , a property that we find somewhat surprising , has been verified numerically as we discuss below . the character of the nanoparticle quasiparticle energy spectrum changes when these intensive energy scales become comparable to the nanoparticle level spacing @xmath65 . ( a ) magnetization in the z - direction . ( b ) magnetization in the x - direction . ] the qualitative considerations of the previous section provide a framework for thinking about the effects of spin - orbit interactions . our microscopic model , on the other hand , allows us to explore realistic magnetic nanoparticle systems in great detail . we have studied numerically nanoparticles containing up to 260 atoms . most of the results presented below are for hemispherical 143-atom nanoparticles . the calculations have been performed with a r12000 300mhz processor on an sgi origin 2000 computer . diagonalizations rely on lapack drivers . a single diagonalization of the hamiltonian for a 143-atom cluster has a running time of approximately 1 hour , and requires around 750 mb of internal memory . in fig . [ [ soeig.eps ] ] we plot the energy shifts caused by spin - orbit interaction for a hemispherical cobalt nanoparticle of 143 atoms with a f.c.c . crystal structure . for this nanoparticle size , the minority and majority single - particle mean - level spacings at the fermi level are @xmath83 mev and @xmath84 mev respectively , when spin - orbit interactions are absent . the single - particle mean - level spacing averaged over all states ( i.e. without distinguishing between majority and minority levels ) is @xmath85 mev at the fermi level this is approximately the value of the single - particle mean - level at the fermi level also when spin - orbit interaction is included . ] . the spin - orbit induced shifts are both positive and negative and their absolute values go from 1 mev up to 10 mev . the average absolute value of the energy shifts is 2.7 mev when the magnetization is in the z - direction , and 2 mev when the magnetization is in the @xmath4-plane , consistent with the rough estimates above . we also note that there is not a strong correlation between the sign of the energy shift and the energy of the orbital . . the fermi level is the dotted line . ( a ) @xmath35 lies in the @xmath4-plane and @xmath20 is the angle with @xmath86-axis . ( b ) @xmath35 lies in the @xmath87-plane and @xmath19 is the angle with the @xmath88-axis . ] because of the spin - orbit interaction , each individual eigenlevel has an energy dependence on the spin - splitting field ( or magnetization ) direction @xmath35 . to illustrate typical properties of these dependences , we plot in fig . [ [ aniseigcomb ] ] the variation of a few energy levels around the fermi level for cases in which the magnetization rotates in @xmath87-plane and @xmath4-planes respectively . in the absence of spin - orbit interactions there would be no dependence of any of these orbital energies on magnetization orientation . - and @xmath87-planes for a 143-atom nanoparticle . ] notice that the angle dependence in the @xmath4-plane is considerably weaker than in the @xmath87-plane , for the hemispherical nanoparticle we consider . this trend indicates that for our nanoparticles , it is the overall shape which dominates the spin - orbit induced anisotropy physics . for this size cluster , there are many narrowly avoided level crossings , a point to which we return below . the difference in eigenvalue sensitivity to magnetization rotations in the two planes is clearly visible in fig . [ [ eiganis3 ] ] , where we plot the single - particle level anisotropies @xmath89 and @xmath90 versus the eigenvalue index @xmath91 . here @xmath92 and @xmath93 are the magnitudes of the self - consistent spin - splitting field when its direction is along @xmath94 and @xmath95 respectively . we note that typical change in orbital energy between @xmath94 and @xmath95 direction magnetizations is only @xmath96 smaller than the typical total shifts induced by spin - orbit interactions . on the other hand the typical difference in orbital energy between @xmath94 and @xmath97 direction magnetization is five times smaller than the corresponding spin - orbit induced energy shift . all orbital energies are relatively insensitive to the magnetization orientation within the @xmath4-plane . in both cases the correlation between position within the band and the sign and magnitude of the energy shift is weak . in addition , energy shifts at nearby energies are weakly correlated . that is , the correlation function @xmath98 where the average @xmath99 is over the occupied levels @xmath91 , drops to zero very rapidly with @xmath100 , as clearly shown in fig . [ [ correlation ] ] . ] . the correlation function drops immediately to zero for @xmath101 . note that the @xmath102 value of the correlation function is equal to the width of the anisotropy distribution , plotted in fig . [ [ histeiganis2 ] ] . ] -plane , @xmath103 $ ] , for a 143-atom nanoparticle . the mean value is @xmath104ev , the width of the distribution ( enclosed by the vertical dashed lines ) is @xmath105 mev . ] it is useful to consider the _ distribution _ of the quasiparticle anisotropies , @xmath106 . as an example , in fig . [ histeiganis2 ] we plot the distribution of anisotropies in the @xmath87-plane , @xmath103 $ ] , constructed with the @xmath107 occupied single - particle states of a 143-atom nanoparticle . the distribution has a width characterized by the root mean square @xmath108 mev and a much smaller mean value @xmath109ev . note that @xmath110 is exactly equal to the @xmath102 value of the correlation function displayed in fig . [ [ correlation ] ] . single particle anisotropies of groups of orbitals over within a specified energy range tend to be anti - correlated , leading to typical averages smaller than the variance of the distribution . the large difference between the distribution mean and variance will play an important role in sec . [ anis_fluctsn ] , where we discuss fluctuations of the anisotropy energy as a function of electron number . the width of the distribution , @xmath110 , gives a measure of the average magnitude of the single - particle level anisotropy , and the ratio @xmath111 characterizes the strength of mixing between quasiparticle orbitals that results from spin - orbit interactions . as mentioned previously , this identification of weak and strong spin - orbit interaction regimes is equivalent to the usual one@xcite based on the comparison of the spin - orbit scattering time @xmath79 and the mean - level spacing @xmath65 . when @xmath112 , a limit achieved for small enough particle size for any value of @xmath36 , spin - orbit coupling is a relatively weak effect and there is little mixing between spin - up and spin - down states . as a consequence , the level crossings between states of predominantly opposite spins that occur as a function of the magnitude and orientation of @xmath45 , will be only weakly avoided . with increasing particle size , @xmath65 decreases and we enter the regime of strong - spin orbit interaction and strong level repulsion . the single - particle spectrum becomes relatively rigid , level crossing will be strongly avoided , and @xmath65 will limit the variation of individual levels as a function of the magnetization direction . within our model we have found that the cross - over between these two regimes is very broad , it starts for nanoparticles containing of order 200 atoms and , as we argue below , it will be completed when the nanoparticles contain approximately 1000 atom . for a 143-atom nanoparticle , we are already approaching the cross - over regime ; we find @xmath113mev while the single - particle mean - level spacing at the fermi level is @xmath85 mev . in this regime level crossings will be moderately avoided . and @xmath114 are single - particle levels in absence of spin - orbit interaction . ( a ) @xmath115 and @xmath116 have like - spins ; ( b ) @xmath115 and @xmath117 have opposite - spins . solid and dotted lines are for the cases of nearby levels ( @xmath118 ) and _ any _ pair of levels respectively . ] the typical size of avoided crossing gaps between opposite spin orbitals in small nanoparticles can be understood by the following argument . we first note that the unperturbed orbitals satisfy the following sum rule @xmath119 the estimate for the right hand side of eq . [ sum_rule ] is based on the on the same considerations leading to eq . [ eq : sumrule ] , and uses that @xmath120 . if angular momentum matrix elements between orbitals are not correlated with the energy differences between these orbitals , and the matrix elements are reasonably narrowly distributed , this sum rule can be used to estimate the typical matrix element . expectation values of the angular momentum operators are zero because of angular momentum quenching , and a finite fraction of the matrix elements vanish because of symmetries present in our rather regularly shaped nanoparticles . aside from these features , we find numerically that correlations between matrix elements and energy differences are too small to be clearly observable . [ hist4pub ] shows the distribution function we have obtained for matrix elements between opposite and like - spin states . we have considered both the matrix element distribution for closely spaced levels and the distribution for any pair of levels , not necessarily nearby . the distributions are found to be very similar . approximately @xmath121 and @xmath122 of matrix elements are zero for opposite - spin and like - spin cases respectively . based on these numerical results and the sum rule eq . [ sum_rule ] , we estimate the typical value of @xmath123 as @xmath124 divided by half the total number of s , p and d orbitals , @xmath125 . this implies a typical non - zero matrix element equal to @xmath126 . in fig . [ dist4 ] we plot the average square matrix element vs. energy difference for a 143 atom cluster , obtaining remarkably precise agreement with this estimate provided that the energy differences are much smaller than the @xmath24-band width . and @xmath114 are quasi - particle states without spin - orbit interactions . the four curves correspond to the four possible spin combinations . ] the average square matrix element for like spin orbitals is approximately a factor of two smaller , consistent with the type of argument presented above which would imply proportionality to @xmath127 in that case . we expect the dependence of quasiparticle energies on the magnitude and orientation of the order parameter , and also on external fields discussed below , to change in character when the typical matrix element becomes comparable to the level spacing , _ i.e. _ when @xmath128 our numerical calculations of quasiparticle spectra are consistent with using the condition @xmath129 as a criterion for the _ start _ of the crossover to the strong coupling limit , and the condition that the typical avoided crossing gap estimate be equal to @xmath65 , as a criterion for _ completion _ of the crossover to strong - coupling relatively rigid spectra . for cobalt nanoparticles the later condition is reached for @xmath130 ; for smaller nanoparticles , the quasiparticles generally have somewhat well - defined spin character , some poisson character in their spectral statistics , and complicated evolution patterns with external field and order parameter variations . for larger nanoparticles , which we are not however able to study numerically , we expect that quasiparticles will have strongly mixed spins , and more rigid spectra with smoother evolution patterns . all the ferromagnetic nanoparticles that we are able to study here , and many nanoparticles studied experimentally , are in the crossover regime . note that since @xmath131 , @xmath132 ; the two conditions differ quantitatively not parametrically . $ ] vs. the angle @xmath19 between @xmath35 and @xmath95 , when @xmath35 is in the @xmath87-plane . the black line represents @xmath133 $ ] vs. the angle @xmath20 between @xmath35 and @xmath94 , when @xmath35 lies in the @xmath4-plane . ] the total ground - state energy of the particle @xmath134 , obtained by summing over the lowest @xmath135 occupied orbitals , depends on the direction of the self - consistent spin - splitting field @xmath35 . in fig . [ [ anis2 ] ] we plot @xmath136 vs. the angle @xmath19 between @xmath137 and @xmath95 ( when @xmath137 lies in the @xmath87-plane ) and @xmath138 vs. the angle @xmath20 between @xmath137 and @xmath94 ( when @xmath137 lies in the @xmath87-plane ) . from the figure we can see that the @xmath4-plane is almost an _ easy _ plane for the model nanoparticle , except for a weak energy dependence which generates four easy axis in the directions @xmath139 . again this property reflects the large significance of the overall sample shape . the four easy axis directions are remnants of the magnetic anisotropy symmetry in bulk f.c.c . ferromagnets . group theory considerations demand that a bulk f.c.c . ferromagnet have an easy axis in one of the directions perpendicular to the 8 @xmath140@xcite planes . in a hemispherical nanoparticle the f.c.c . symmetry is partially lifted and the magnetization is forced to lie in the @xmath4-plane . we can define the following two anisotropy - energy - per - atom constants @xmath141 as expected on the basis of the qualitative considerations of section [ perturb ] , the anisotropy per occupied orbital is much smaller than the average single - particle level shift @xmath76 ( approximately 200 times smaller in both cases ) due to cancellations between positive and negative shifts mentioned above . we notice that @xmath142 for our nanoparticle is _ larger _ than the bulk anisotropy per atom @xmath143ev , as one would expect because of the hemispherical shape of the sample . however @xmath144 is actually _ smaller _ than @xmath145 . this comparison should be regarded with caution , since it is known that accurate theoretical estimates of the magnetic anisotropy for bulk crystals are very delicate and agreement with experiment values even in the bulk is still not completely satisfactory @xcite ; we have not evaluated the anisotropy energy that results from the bulk limit of our nanoparticle model and it may well not agree with experiment . nonetheless , the small value that we find for the anisotropy in the @xmath4-plane might be connected to the puzzling finding , in both tunneling@xcite and switching - field@xcite experiments on single ferromagnetic nanoparticles , of anisotropy energies per atom a factor of five _ smaller _ than bulk values . so far we have considered only average values of the anisotropy constants . in ferromagnetic nanoparticles , however , these quantities are also characterized by large fluctuations as a function of experimentally relevant parameters . anisotropy fluctuations are the topic of this section . near @xmath146 for a 143-atom nanoparticle . 200 levels are shown . the fermi level is the thick black line . the vertical white dashed lines indicate the positions of the charge redistributions displayed in fig . [ [ anisfluc143 ] ] . the vertical black dashed line corresponds to the self - consistent spin - splitting field . ] ] . charge fluctuations of the order of one electron result in fluctuations in the magnetic anisotropy . ( b ) anisotropy energy in the @xmath87-plane vs. the magnitude of the spin - splitting field taken as a free parameter . ] when spin - orbit interactions are included , all quasi - particle eigenstates have mixed majority - spin and minority - spin character . our specific calculations , however , are for small enough particles such that most eigenstates still have predominantly one spin character . it is convenient to use this `` predominant spin '' to label the states when discussing their dependence on @xmath147 or on an external magnetic field . in a paramagnetic system ( @xmath148 ) in absence of external field , there is a kramer degeneracy that pairs up eigenstates with opposite predominant spin character . the degeneracy is lifted in the ferromagnetic state . majority - spin states will move down in energy while minority - spin states will move up as @xmath147 increases . because of spin - orbit induced level repulsion , all of the level crossings are avoided ; the levels evolve continuously with @xmath147 , gradually changing their spin and orbital character . in fig . [ [ eigspectrum2 ] ] we plot the variation of two hundred levels as a function of @xmath149 near @xmath150 ev , for a 143-atom nanoparticle . the fermi level is the thick black line lying in a region of predominantly minority - spin quasiparticles ( lines with positive slope ) . by increasing @xmath147 , individual majority - spin quasiparticle energies ( negative slopes ) come down from regions above the fermi level , creating avoided crossing gaps as they approach minority - spin quasiparticles moving in the opposite direction . since a 143-atom nanoparticle is still toward the _ weak _ spin - orbit coupling limit , the level crossings will be avoided only weakly . whenever one of the majority - spin quasiparticles crosses the fermi level , there is a change in the spin character of one of the quasiparticle levels and the total spin of the nanoparticle increases approximately by one . due to the exchange interaction , such spin - flips bring about charge redistribution inside the nanoparticle , which in turns gives rise to fluctuations in the anisotropy energy . in fig . [ [ anisfluc143]a ] we plot the total atomic site charge redistribution , when @xmath151 , for @xmath152 mev @xmath153 where @xmath154 is the total charge at atom @xmath155 . it is seen that @xmath156 changes by one in the small interval @xmath157 mev where majority - spin quasiparticles weakly avoid crossing the fermi level . ( note that this is a charge redistribution , not a change in total charge . ) this result is not unexpected , since the majority and minority spin orbitals should have uncorrelated spatial distributions . the charge redistribution is likely overstated by our simplified model , however , since it does not account for long - range coulomb interactions and related screening effects . it is , however still a relatively small @xmath158 effect . the anisotropy energy fluctuations associated with the level crossings that occur as a function of @xmath159 , shown in fig . [ [ anisfluc143]b ] are much larger in relative terms . variation in @xmath159 corresponds to variation in the amplitude of the magnetic order parameter . in micromagnetic modeling of nanoparticles , it is implicitly assumed that the magnitude of the order parameter is fixed and that this collective degree of freedom can be ignored in modeling nanoparticle properties . the substantial dependence of anisotropy energy on @xmath159 that we find demonstrates that the amplitude and orientation fluctuations of the order parameter can be strongly coupled in small magnetic nanoparticles . related anisotropy fluctuations as a function of _ electron number _ are important in understanding the addition - potential - spectroscopy single - electron - transistor experiments@xcite which are the topic of the next section . the analysis of the field - dependence of tunneling resonances in experiments on magnetic nanoparticles suggests that the anisotropy energy fluctuates significantly from eigenstate to eigenstate@xcite . in refs . @xcite the effects of these fluctuations were mimicked by using two different anisotropy energy constants @xmath160 and @xmath161 for @xmath135- and @xmath162-electron states . by assuming @xmath163 in the range of a few percent , it was possible to explain the non - monotonic behavior of the tunneling resonances seen experimentally . within our microscopic model , we have calculated @xmath160 and @xmath164 as defined in eqs . ( [ anisconst1],[anisconst2 ] ) . in light of results of the type summarized in fig . [ [ anis2 ] ] , it suffices to compute the total energy for two different directions of the spin - splitting field and take the difference . the total energy for a @xmath135-electron system , for spin splitting fields along directions @xmath165 , is given by : @xmath166 where @xmath167 is calculated self - consistently for a given fixed direction . the anisotropy - energy - per - atom constant is then @xmath168 + \frac{({h}_1^{\star})^{2}-({h}_2^{\star})^{2}}{2u_{dd } } ( g_s \mu_{\rm b})^2\ ] ] where the subscript @xmath135 emphasizes the fact that the constant refers to a @xmath135-electron system . it turns out that the value of @xmath169 depends very weakly on @xmath170 . for example , in a 143-atom nanoparticle @xmath171 . this property reflects the large ratio between the total magnetic condensation energy and the anisotropy energy , even for nanoparticles . the strong coupling between amplitude and orientation fluctuations mentioned above , occurs only when a small change in @xmath159 leads to a crossing between majority and minority spin orbitals . evaluating @xmath172 at @xmath173 , expanding in powers of @xmath174 and remembering that @xmath175 minimizes @xmath172 yields @xmath176 + { \cal o}\bigg[{n\over { \cal n}_a } \big({\delta { h}^{\star}\over { h}^{\star}_1}\big)^2 \bigg]\ ] ] for any @xmath135 . of the anisotropy energy constant @xmath177 , when the electron number increases by one , @xmath178 , as a function of spin - splitting field . ( a ) 260-atom nanoparticle . ( b ) 143-atom nanoparticle . the self - consistent spin - splitting field is @xmath179 ev for both a 143-atom and a 260-atom nanoparticle . ] adding an electron to the system changes the magnitude of the spin - splitting field @xmath180 , but again @xmath181 . thus we obtain @xmath182 + { \cal o}\bigg[{n+1\over { { \cal n}_a } } \big({\delta { h}^{\star}\over { h}^{\star}_1}\big)^2\bigg]\nonumber\\ = k_n(\hat \omega_1 , \hat \omega_2 ) + { \epsilon_{n+1}({h}^{\star}_1\ , \hat\omega^{\phantom { \star}}_1 ) - \epsilon_{n+1}({h}^{\star}_1\ , \hat\omega^{\phantom { \star}}_2 ) \over { \cal n}_a } + { \cal o}\bigg[{n+1\over { { \cal n}_a } } \big({\delta { h}^{\star}\over { h}^{\star}_1}\big)^2\bigg]\end{aligned}\ ] ] replacing @xmath183\over { \cal n}_a}\ ; , \ ] ] where @xmath184 $ ] is the width of the distribution of the single - particle anisotropies in plane containing the two directions @xmath185 and @xmath186 , we finally obtain @xmath187\over { { \cal n}_a}}\ ; , \ ] ] the fluctuations of @xmath188 due to an additional electron are therefore of the order of @xmath184/{\cal n}_a$ ] , and are regulated by the mean level spacing , @xmath65 , which suppresses the magnitude of @xmath184 $ ] at large nanoparticle sizes as we discuss below . for a 143-atom dot , @xmath184 \approx 2.9 $ ] mev in the zx - plane , where @xmath189 mev . therefore @xmath190 /k_n(\hat z , \hat x)\approx 15\%$ ] . these estimates are confirmed by a direct calculation of @xmath191 and @xmath192 for 143-atom and 260-atom nanoparticles , shown in fig . [ [ aniscomp ] ] . here @xmath191 and @xmath192 are plotted as a function of @xmath193 , taken as a free parameter ( the self - consistent value is close to @xmath194 ev for both a 143-atom and a 260-atom nanoparticle ) . for a 143-atom nanoparticle the calculated anisotropy energy fluctuations are of the order of @xmath195 . for a 260-atom nanoparticle the anisotropy energy fluctuations are smaller , but still close to @xmath196 . we can understand the surprisingly large fluctuations in anisotropy energy with particle and electron number by the following observations . our numerical results ( see fig . [ eiganis3 ] and fig . [ histeiganis2 ] ) indicate that the contribution of a given orbital to the anisotropy energy is chosen essentially at random from a distribution which , for the @xmath87-plane anisotropy , has a width @xmath197 mev and a much smaller mean value @xmath198ev , where @xmath199 mev . the relative change in the anisotropy energy expected when one additional orbital is occupied is @xmath200 , which can easily be larger than @xmath201 for @xmath202 . we expect strong level repulsion in the rigid spectrum of larger nanoparticles to be accompanied by more regular behavior of the anisotropy energy per atom , with smaller variations as a function @xmath47 , and electron number . we note that fluctuations in the contribution of a given orbital to the anisotropy energy should not exceed @xmath203 , which vanishes in the limit of very large particles . we expect that the distribution function of contributions to anisotropy from orbitals near the fermi energy to become narrow when @xmath204 is smaller than the average anisotropy energy contribution @xmath205 . this condition is satisfied for particles containing more than @xmath206 atoms . . a splitting field of @xmath207 ev is assumed for all sizes . ] we conclude this section with a few remarks on the dependence of the magnetic anisotropy energy on nanoparticle size and shape . in fig . [ [ anisatom2 ] ] we plot the anisotropy energy per atom for nanoparticles of different sizes . for a small number of atoms ( below 60 ) the anisotropy per atom is very large and decreases rapidly with size . in this regime , non - extensive surface contributions , which are present because of the abrupt truncation of the lattice , are clearly playing a dominating role . by increasing the size of the nanoparticle , we must eventually reach a regime where total magnetic anisotropy becomes proportional to the nanoparticle volume . [ [ anisatom2 ] ] shows that this regime is not yet fully established even for nanoparticles with hundreds of atoms , and fluctuations are still pronounced . in this regime the anisotropy per atom is still 2 - 3 times larger than the bulk value for cobalt , which is 0.06 mev . there is no regime in which _ bulk _ ( proportional to @xmath208 ) and surface ( proportional to @xmath209 ) contributions to the anisotropy can be cleanly separated . -plane , @xmath210 $ ] for spherical nanoparticles of two sizes . @xmath19 is the angle between @xmath95 and @xmath35 . by symmetry , the anisotropy energy in the @xmath4-plane is now the same as in the @xmath87-plane . the spin - splitting field is assumed to be @xmath207 ev for both sizes . ] in is clear that for small particles the shape of the nanoparticle plays an important role in the determination of the magnetic anisotropy . for example , as seen in fig . [ [ sphereanis ] ] , for spherical particles we find anisotropy energies which are one order of magnitude smaller than the @xmath87 anisotropy of hemispherical particles , comparable instead to the @xmath4 anisotropy of hemispherical particles with the same number of atoms . a 140-atom spherical particle has an anisotropy energy per atom @xmath211mev . it is interesting to notice that both tunneling@xcite and switching - field@xcite experiments on single ferromagnetic nanoparticles find anisotropy energies per atom which are of the order @xmath212mev , a factor of five _ smaller _ than the bulk value . the nanoparticles in tunneling experiments are roughly hemispherical@xcite , whereas the nanoparticle shape in ref . [ ] is close to spherical . further theoretical studies that focus on the relationship between nanoparticle shape and anisotropy could be helpful to efforts to engineer ferromagnetic nanoparticles whose shape , size , and crystal structure are tuned to produce desired magnetic properties . , for different magnitudes of the external magnetic field . the external field is in the @xmath87-plane at an angle @xmath213 with the @xmath95-axis . @xmath19 is angle between @xmath35 ( lying in the @xmath87-plane ) and @xmath95 . ] ] , as a function of @xmath214 . @xmath19 is the angle between @xmath35 and @xmath95 . @xmath215 is in the @xmath87-plane at @xmath216 from the @xmath95- axis . ] -plane for different magnitudes of an external magnetic field . the direction of the external field is @xmath217 . @xmath20 is the angle between @xmath35 ( lying in the @xmath4-plane ) and @xmath94 . ] in the last section we shall investigate the effect of an external magnetic field on the total energy of the nanoparticle . this will allow us to make a connection between our microscopic model and more familiar classical micromagnetic energy functional expressions which are normally used to interpret the results of switching - field experiments@xcite . we shall also study the dependence of the quasiparticle energy levels on external magnetic fields ; this complex behavior has been probed directly in recent tunneling experiments@xcite . ideally one would like to study the three dimensional energy landscape @xmath218 as a function of @xmath219 , for several values of @xmath215 . here we consider two simplified cases . in fig . [ [ anismag ] ] we plot the total energy difference @xmath220 as a function of an external magnetic field @xmath215 . here @xmath35 is allowed to rotate in the @xmath87-plane . in this picture we recognize the familiar features of a micromagnetic energy functional characterized by a uniaxial magnetic anisotropy : at @xmath221 there are two degenerate minima , separated by an energy barrier , representing two equivalent easy directions @xmath222 . when a magnetic field is applied in the @xmath87-plane , the degeneracy is lifted and the minimum in the @xmath223 direction becomes a metastable local minimum , until a switching field is reached at which the local minimum disappears . this happens at @xmath224 t for a 143-atom nanoparticle . this simple model captures the essence of classical hysteresis in ferromagnetic nanoparticles . quantum mechanically , for sufficiently weak fields the system can still sit in the quantum state characterized by a magnetic moment pointing along the classically metastable direction until the switching field is reached , the system relaxes to its true ground state , and the magnetization orientation collective coordinate changes discontinuously . this scenario neglects the possibility of quantum tunneling between the two local minima through the energy barrier separating them . at low fields and for nanoparticles of 1 - 4 nanometers in diameter , quantum tunneling is exponentially small . ] with a further increase of the external field , the magnetization is gradually twisted from the easy @xmath86-axis toward the direction of the magnetic field , as shown in fig . [ [ angle1 ] ] . at low fields ( @xmath225 t ) , the direction of the magnetization corresponding to this local minimum is very close to the @xmath4-plane . because our hemispherical nanoparticles have a nearly isotropic easy @xmath4-plane , the magnetization will rotate in the @xmath4-plane in response to _ weak _ applied external fields clearly , if the external field is not in the @xmath4-plane the magnetization will not stay exactly in this easy plane . however the displacement of the magnetization from easy to hard directions is negligible at the weak fields ( @xmath226 ) ( see fig . [ [ angle1 ] ] ) and it becomes relevant only at larger fields , when the component of the magnetization in the @xmath4-plane is essentially frozen along the direction of the component of @xmath214 in that plane . ] this is exemplified in fig . [ [ xyanisdb ] ] , where we plot the the total energy as a function of @xmath214 , when @xmath227 lies in the @xmath4-plane . @xmath214 is oriented in the direction @xmath228 . in the absence of the external field , the ground state of the nanoparticle is four - fold degenerate , with the degeneracy corresponding to one of the four magnetization directions @xmath229 , associated to the four local minima of fig . [ [ xyanisdb ] ] . as the external field increases , three of these minima will become classically metastable with small barriers separating them from each other and the true ground state . in general the different local minima will lose their metastability at different external field strengths . in the case we consider , the first switching field is reached at @xmath230 t , when the minimum originally at @xmath231 disappears . if the system starts out in this local minimum at zero external field , at the switching field it will jump to the ground state with a corresponding rotation of the magnetization in the @xmath232-direction . since everything takes place essentially within the @xmath4-plane , where the typical anisotropy energies are one order of magnitude smaller than in the @xmath87-plane , the scale of the coercivity is also much smaller than the one in the @xmath87-plane . more complicated hysteretic behaviors are also possible , in which , by starting from a different minimum at zero external field , the system jumps first from one metastable to another metastable state and only at a second switching field reaches its true ground state . the hysteretic behavior that we have seen in the ground - state properties of a ferromagnetic nanoparticle has profound implications for the magnetic field dependence of its low - energy elementary excitations . in ferromagnetic metals there exist two kinds of elementary excitations : collective spin excitations associated with magnetization orientation degrees of freedom , and particle - hole excitations . in a ferromagnetic nanoparticle the distinction between these two kinds of excitations is partly obscured by the effect of spin - orbit coupling@xcite . sorting out this problem is particularly important in order to interpret current tunneling experiments in single - electron transistors@xcite . here we examine only particle - hole excitations around the fermi level , which are immediately available within our hartree - fock treatment of the many - body hamiltonian . in fig . [ [ mod1 ] ] we plot the magnetic field dependence of a few single - particle energy levels around the fermi energy . as a simplified illustration , the levels are calculated assuming that the ground - state dependence on the external field is as described in fig . [ [ anismag ] ] : at low field the ground - state magnetization is oriented around the @xmath223-axis until the switching field is reached , whereupon the magnetization is reversed along a direction around the @xmath94-axis . around the fermi level in an external magnetic field . the levels are calculated assuming that the ground - state of the nanoparticle changes in the external field in the way described in fig . [ [ anismag ] ] . ] around the fermi level in an external magnetic field . ( a ) the levels are calculated assuming that the ground - state of the nanoparticle changes in the external field in the way described in fig . [ [ xyanisdb ] ] . ( b ) for magnetic fields larger than @xmath0 t , the magnetization starts to develop a non - negligible component perpendicular to the @xmath4 plane , while its projection in the @xmath4-plane is parallel to the component of @xmath215 in that plane . ] the corresponding field dependence of the quasiparticle is quite complex . in the small - field regime ( @xmath233 t ) there is a hysteretic switching at a certain @xmath234 t , due to an abrupt change of the ground - state magnetic moment . there is basically no correlation between single particle states before and after reversal . notice in particular that the levels can jump either up or down at @xmath235 . in this case the jumps at the switching fields will cause a complete reshuffling of the quasiparticle levels . ] . furthermore the quasiparticle energies have continuous non - monotonic variations , which seem to differ randomly from level to level . in the large field regime ( @xmath236 ) , the quasiparticle energies depend roughly linearly on @xmath214 and their slopes almost all have the same sign . in the small - field behavior , the variation of the quasiparticle energies as a function of the external field is due to a combination of the rotation of the particle s magnetic moment direction @xmath227 , and zeeman coupling . we have shown above ( see fig . [ [ eiganis3 ] ] ) that the dependence of the quasiparticle energies on @xmath35 varies randomly from level to level . thus the complex non - monotonic behavior at small fields can be understood within our model . at large fields ( @xmath236 ) @xmath35 is slowly twisted toward the field direction and the zeeman coupling plays the dominating role in the field dependence . moreover , because the fermi level lies in a region of predominantly minority - spin energy levels , it is expected that almost all particle and hole excitations around the fermi level have the same high - field slope@xcite . in the discussion above we have assumed , out of simplicity , that the magnetization is arbitrarily constrained to rotate in the @xmath87-plane . at weak external fields , however , the magnetization will stay _ close _ to the @xmath4-plane . for a weak external field in the @xmath237 direction , the actual dependence of the ground - state energy on the magnetization direction at different field strengths will be the one represented in fig . [ [ xyanisdb ] ] . consequently , the change of quasi - particle energies as a function of the external field will reflect this more complicated behavior . we illustrate this point in fig . [ [ mod2]-a ] where , as an example , we assume that at zero field the magnetization points in the @xmath238 direction . as the magnetic field increases , there will be a first jump in the in the quasiparticle - level dependence at @xmath239 t , when the magnetization re - orients itself in the @xmath240-direction . there will a second hysteresis jump in the quasi - particle energies at @xmath241 t , when the magnetic moment finally switches to the @xmath242-direction , corresponding to the only metastable configuration left at this field . for a different starting local minimum at @xmath243 , the variation of the quasiparticle energies as a function of the external field will be in general different . by increasing the magnetic field beyond @xmath0 t , the magnetization will start to develop a non negligible component perpendicular to the @xmath4-plane , while its component in the @xmath4-plane will become essentially frozen along the direction of the component of @xmath215 in that plane . the quasiparticle - energy dependence on an external magnetic field presented here , has striking similarities with the field dependence of the tunneling resonance energies in single - electron tunneling experiments@xcite . notice that for the hysteretic behavior occurring as a result of the rotation of the magnetization in the vicinity of the easy @xmath4-plane , the order of magnitude of the coercivity that we find is close to the experimental value . the only discrepancy between the results of our theoretical model and the experiment is the mean level spacing @xmath244 of the low - energy excitations . since in our model we are only taking into account hartree - fock quasiparticles , the expected level spacing of the low - lying excitation is approximately @xmath245 , which is a few mev in our 143-atom nanoparticle and should be of the order of 0.5 mev for a 1500-atom nanoparticle considered in the experiment . this value is still larger than the value @xmath246 mev observed experimentally . we believe that a unified and consistent inclusion of collective spin excitations , and possibly also the non - equilibrium transport effects proposed in refs . @xcite , could resolve this confusion . in this article we have investigated the effects of spin - orbit interactions on the properties of ferromagnetic metal nanoparticles . in particular , we have focused on their novel microscopic magnetocrystalline anisotropy physics , and on hysteresis in the quasiparticle excitations spectra of metallic nanomagnets . our analysis , based on qualitative considerations backed up by numerical studies of a generic model for ferromagnetic transition metal nanoparticles , provides an understanding of some emergent properties of their quasi - particle states . we find two regimes separated by a broad crossover and characterized by the comparison of several characteristic energy scales . for small nanoparticles with fewer than @xmath247 atoms , the single particle mean - level spacing @xmath65 is larger than spin - orbit induced energy - shifts @xmath76 in the quasi - particle spectra . these shifts have the same typical size as the spin - orbit scattering lifetime broadening energies of very large particles , @xmath248 . the quasi - particle levels of small nanoparticles in which @xmath249 evolve in a complicated way as a function of magnetization orientation and external magnetic field , with relatively small avoided crossing gaps and spin - orbit shifts of nearby orbitals that are nearly uncorrelated . the size of the avoided crossing gaps is determined by matrix element of the spin - orbit coupling operator between quasiparticle energy levels that are adjacent in energy . surprisingly , even though expectation values of these matrix elements vanish because of angular momentum quenching , typical values between energetically adjacent orbitals are comparable to those for orbitals at any position in the spectrum . eventually , for nanoparticles with more than @xmath250 atoms , the typical avoided crossing gap estimated in this way becomes comparable to the level spacing and we expect the crossover to the strong coupling limit is complete . nanoparticles with fewer than @xmath251 atoms can easily be prepared with current synthesis techniques and systems of experimental interest are often in the middle of the crossover between small particle and bulk ( weak and strong spin - orbit coupling ) limits . for example , the nanoparticles investigated by electron tunneling experiments contain between 50 and 1500 atoms@xcite . for nanoparticles in the size range we are able to study numerically , @xmath208 smaller than a few hundred , we find that the anisotropy energy per atom displays large changes of order several percent when the electron or atom number changes by one . our analysis allows us to make a connection between the microscopic model of a metal nanomagnet and more familiar classical micromagnetic energy functional expressions which are normally used to interpret the results of switching - field experiments@xcite . the ground - state energy as a function of the magnetization direction is characterized by minima separated by energy barriers . the quasi - particle levels exhibit a complex non - monotonic behavior and abrupt jumps when the magnetization direction is reversed by an external magnetic field . the nanoparticles investigated by electron tunneling experiments contain between 50 and 1500 atoms@xcite . the results that we have presented here are therefore particularly relevant for the interpretation of these experiments . in particular , we find that the anisotropy fluctuations inferred from interpretations of these experiments are indeed to be expected . similarly , the dependence of the tunneling resonances on the external magnetic field is qualitatively similar to the behavior of the quasi - particle excitations of our model , although it appears necessary to invoke non - equilibrium quasiparticle configurations@xcite in the nanoparticle in order to understand the size of the quasiparticle level spacing . it is a pleasure to thank mandar deshmukh and dan ralph for several discussions about their experimental results , and jan von delft for stimulating and informative interactions . c.m.c . would like to acknowledge helpful discussions with l. samuelson and with the participants of the 2001 workshop `` spins in nanostructures '' , at the aspen center for physics , where part of this work was completed . a.h.m . acknowledges informative discussions with w. wernsdorfer and p. brouwer . this work was supported in part by the swedish research council under grant no:621 - 2001 - 2357 and in part by the national science foundation under grant dmr 0115947

we present a comprehensive review of nuclear magnetic resonance ( nmr ) studies performed on three nanoscale molecular magnets with different configurations of geometrically frustrated antiferromagnetic ( afm ) triangles , new spin frustration systems with different novel structures : ( 1 ) the isolated single afm triangle k[vaso(ho)]ho ( in short v15 ) , ( 2 ) the spin ball [ mofeo(moo(ho)) ( mooh(ho))(chcoo)(ho)ho ( in short fe30 spin ball ) , and ( 3 ) the twisted triangular spin tube [ ( cucltachh)cl]cl ( in short cu3 spin tube ) . in the v15 nanomagnet , fromv nmr spectra observed at only below 0.1 k , we directly determined the local spin configuration in both the nonfrustrated total spin = 3/2 state at higher magnetic fields ( t ) and the two nearly degenerate = 1/2 ground states at lower magnetic fields ( t ) . the dynamical magnetic properties of v15 were investigated by proton spin - lattice relaxation rate ( 1/ ) measurements . in the = 3/2 state , 1/ shows thermally activated behavior as a function of temperature . on the other hand , a temperature independent behavior of 1/ at very low temperatures is observed in the frustrated = 1/2 ground state below 2.7 tesla . possible origins for the peculiar behavior of 1/ will be discussed in terms of magnetic fluctuations due to spin frustrations . in fe30 spin ball , static and dynamical properties of fe ( ) have been investigated by proton nmr spectra and 1/ measurements . from the temperature dependence of 1/ , the fluctuation frequency of the fe spins is found to decrease with decreasing temperature , indicating spin freezing at low temperatures . the spin freezing is also evidenced by the observation of a sudden broadening ofh nmr spectra below 0.6 k. finallyh nmr data in the cu spin tube will be described . an observation of magnetic broadening ofh nmr spectra at low temperatures below 1 k directly revealed a gapless ground state of the system . the measurements revealed an usual slow spin dynamics in the quasi one dimensional cu3 spin tube .

1507.00374

@xmath23spin frustration@xmath24 is one of the most exciting frontiers of contemporary research in condensed matter physics.@xcite historically , the interest in spin frustration was raised long ago by anderson s prediction for a resonating valence bond ( rvb ) state ( sometimes called @xmath23spin liquid state@xmath24 ) , as an alternative to the classical nel state @xcite for a magnetic ground state in a two - dimensional triangular - lattice @xmath25 = 1/2 antiferromagnet . from a classical point of view,@xcite this generates a very high degeneracy of the ground state , translating into a huge density of low energy excitations , no long - range order at @xmath26 = 0 k , and a very short magnetic correlation length characteristic of a @xmath23spin liquid state@xmath24 . experimentally , many of the spin frustrated systems , such as triangular - lattice , kagom - lattice and pyrochlores , have been investigated to elucidate the magnetic ground state . however , most of the systems show short - range or long - range magnetic order , due to small perturbations such as defects , anisotropy , and lattice distortion.@xcite this prevents the effects purely associated with spin frustrations from direct observation . therefore , it is difficult to clarify what kind of state is actually realized as the ground state of such a strongly frustrated system and more ideal compounds to investigate the effects of spin frustration in the ground state are strongly desired . recent progress in synthesizing nanoscale molecular magnets offers the opportunity to investigate the effects of spin frustration at nanoscale levels.@xcite molecular magnets are composed of a controllable number of magnetic ions with spins located on a triangle , tetrahedron , ring , ball , and other geometries.@xcite since the distance between the magnetic molecules themselves is usually more than @xmath14 1.5 nm , the magnetic interaction between the molecules is very small and usually three - dimensional magnetic ordering does not occur even at very low temperatures down to mk . thus , one can investigate the magnetic properties of the single molecular system at very low temperatures without any disturbance from magnetic orderings . the simplest spin frustrated system is three @xmath25 = 1/2 antiferromagnetically coupled spins with a triangular configuration . in a molecular nanomagnet , this simplest spin frustration system can really exist and one can investigate this simplest system experimentally.@xcite an innovative idea to utilize the nanoscale molecular magnet as a magnetic unit produced a great breakthrough for synthesizing a new class of spin frustration systems called spin ball @xcite and triangular spin tube.@xcite the spin ball system has a spherical shape constructed from twenty afm triangles with spin frustration . in the spin tube , af triangles form a one - dimensional chain as shown in fig . although these peculiar structures have attracted much interest @xcite the systems were considered to be only toy models for theoretical study , due to the lack of real model materials before the synthesis breakthrough . in this review , we discuss the magnetic properties of these peculiarly - structured spin frustration systems studied by using nuclear magnetic resonance ( nmr ) technique . nmr has proved to be a powerful tool for investigating both static and dynamic properties of nanoscale molecular magnets.@xcite the nmr spectrum gives us information on the hyperfine interaction of the nuclei with the local magnetic moments , in particular the @xmath23on - site@xmath24 nmr spectrum of nuclei on magnetic ions provides direct information on the local magnetic moments of the magnetic ions . on the other hand , the nuclear spin lattice relaxation rate ( 1/@xmath18 ) directly probes the low - frequency spectral weight of spin fluctuations of the local magnetic moments . the polycrystalline powder samples were synthesized as described in ref . for v15 , ref . for fe30 and ref . for the cu spin tube . the nmr measurements were performed utilizing a homemade phase - coherent spin - echo pulsed nmr spectrometer on @xmath27h nuclei ( nuclear spin @xmath28 = 1/2 and gyromagnetic ratio @xmath29@xmath30/2@xmath31 = 42.5774 mhz / t ) and @xmath13v nuclei ( @xmath28 = 7/2 and @xmath29@xmath30/2@xmath31 = 11.193 mhz / t ) in the @xmath26 range 0.1 k @xmath32 @xmath26 @xmath32 300 k using a @xmath33he-@xmath34he dilution refrigerator ( kelvinox mx100 , oxford instruments ) installed at the ames laboratory by the author . a part of nmr data was taken at hokkaido university where a @xmath33he-@xmath34he dilution refrigerator ( kelvinox100 , oxford instruments ) was also installed by the author and collaborators . the nmr spectrum was obtained either by fourier transform of the nmr echo signal at a constant magnetic field or by sweeping the magnetic field . the nmr echo signal was obtained by means of a hahn echo sequence with a typical @xmath31/2 pulse length of 1.5 @xmath35s . the nuclear spin - lattice relaxation rate 1/@xmath18 was measured by the saturation method at the peak position of the nmr spectrum . the nuclear magnetization recoveries for @xmath21h were found to be slightly non exponential due to the existence of protons with inequivalent spatial locations . such non - exponential behavior has been reported in many magnetic molecules.@xcite @xmath22 values were determined from the initial slope of the recovery behavior , which corresponds to a weighted average relaxation rate of the nonequivalent nucleus.@xcite k@xmath0[v@xmath1as@xmath0o@xmath2(h@xmath3o)]@xmath48h@xmath3o ( in short , v15 ) is one of the interesting molecular magnet for investigating spin frustration since v15 can be considered a model system for an @xmath36 heisenberg single triangle antiferromagnet , the simplest spin frustration system.@xcite v15 is comprised of fifteen v@xmath37 ions with @xmath36 , which are arranged in a quasi - spherical layered structure with a triangle sandwiched between two hexagons as shown in fig . all exchange interactions between v@xmath37 spins are antiferromagnetic ( af).@xcite each hexagon of v15 consists of three pairs of strongly coupled spins with @xmath38 @xmath14 800 k. each spin of v@xmath37 ions in the central triangle is coupled with the spins in both hexagons with @xmath39 = 150 k and @xmath40 = 300 k,@xcite resulting in a very weak exchange interaction between the spins within the central triangle with @xmath41 = 2.44 k.@xcite at low temperatures , the magnetic properties of v15 are determined entirely by the three v@xmath37 spins on the triangle ( a frustrated @xmath36 triangular system ) because the v@xmath37 spins on the hexagon are assumed to form a total @xmath15 = 0 spin singlet state due to the strong af interaction of @xmath38 @xmath14 800 k , as confirmed directly by @xmath13v - nmr measurements at very low temperatures.@xcite ( @xmath36 ) ions ( solid circles ) and the exchange coupling scheme in the v15 . ( b ) magnetic energy level scheme of the v15 as a function of the external magnetic field . solid and broken lines show the nearly degenerated two @xmath15 = 1/2 branches and the @xmath15 = 3/2 branches , respectively . , width=321 ] the magnetic levels of v15 at low temperature can thus be described by a simple model spin hamiltonian @xmath42 where the exchange parameters between spins can be assumed equal in first approximation , i.e. , @xmath43 = @xmath44 = @xmath45 = @xmath41 and @xmath46 are the individual @xmath36 v sites . the energy scheme at low temperature is given by a ground state of two degenerate total spin @xmath15 = 1/2 and an excited state of @xmath15 = 3/2 which lies @xmath14 3.8 k above the ground state.@xcite in reality , the two @xmath15 = 1/2 states are split with a small energy gap which was estimated to be of the order of 0.08 k from the magnetization measurement.@xcite among the possible origins for the gap have been discussed in terms of a small dzyaloshiskii - moriya ( dm ) interaction,@xcite hyperfine interactions@xcite and lattice distortion.@xcite v - nmr spectra for the v@xmath37 ions in the triangle observed in its @xmath15=3/2 ground states at very low temperature for three different resonance frequencies . the inset shows the external field dependence of resonance frequency for the peak measured below 100 mk . typical @xmath13v - nmr spectra observed in the @xmath15=1/2 ground states are shown in ( b ) and ( c).,width=302 ] @xmath13v nmr signal was observed successfully only below 0.1 k.@xcite above @xmath47 = 2.7 t , where the ground state of the v15 cluster is @xmath15 = 3/2 , a single peak of @xmath13v nmr spectrum with full width of half maximum ( fwhm ) of @xmath14 5 koe is observed . the peak position of the spectrum shifts to lower magnetic field with increasing resonance frequency as shown in fig . the resonance frequency @xmath48 is proportional to the vector sum of the internal field @xmath49 and external field @xmath50 : @xmath51 where @xmath52/2@xmath31 = 11.285 mhz / t is the gyromagnetic ratio of the @xmath13v nucleus . by fitting the data points in the field above 2.7 t , as shown in the inset of fig . 3(a ) , @xmath49 is estimated to be 7.6 t. the internal field at the nuclear site in a v@xmath37 ion with @xmath25 = 1/2 is dominated by the core polarization mechanism which induces a large negative internal field of the order of 100 koe/@xmath53 at the nuclear site.@xcite the value of 76 koe is close to the value of 85 koe reported in vo@xmath3.@xcite thus this @xmath13v signal can be assigned to v@xmath37 ions with an almost full spin moment of 1 @xmath53 . in addition , the good fitting result shown by solid line in the inset of fig . 3(a ) indicates the direction of the spin moments is parallel to the external field . thus we conclude that each v@xmath37 ion on the triangle has almost the full spin moment of 1 @xmath53 whose direction is parallel to the external field , which gives the total spin @xmath15 = 3/2 state for the ground state of the v15 cluster above 2.7 t. this is microscopic evidence of the spin structure of the v15 cluster for the @xmath15 = 3/2 ground state . v - nmr signals observed in @xmath15 = 1/2 ground states at @xmath54 @xmath55 2 t . the solid circles represent the stronger signal ( p1 ) shown in fig . the solid squares and triangles refer to the other two weaker signals p2 and p3 , respectively , shown in fig . 3(c ) . the high field data in fig . 3(a ) have been replotted here for comparison . the solid lines show the fitting results using eq . the inset shows schematic view of the expectation values for the spin moment for @xmath56 , @xmath57 and @xmath58 sites on the triangle according to the eigenfunctions : ( a ) @xmath59 , ( b ) @xmath60.,width=283 ] on the other hand , three different v sites ( p1 , p2 and p3 ) are observed in the case of @xmath61 ground state below 2.7 t as shown in fig . 3(b ) and ( c ) . the external field dependence of those peak positions are plotted as circles , squares and triangles in fig . 4 . by fitting these data using eq . 2 , internal fields for three v sites are estimated to be 7.5 t , 4.5 t and 2.5 t for p1 , p2 and p3 , respectively . assuming the hyperfine field for v@xmath37 ions in v15 due to the core polarization is 7.6 t for @xmath25 = 1/2 , spin moments for each each v@xmath37 site can be estimated to be @xmath14 1 @xmath53 , 0.6 @xmath53 and 0.33 @xmath53 for p1 , p2 and p3 , respectively , where the positive and negative signs of the spin moments correspond to the direction of parallel and antiparallel with respect to the external field , respectively . now we compare the experimental findings with the theoretical predictions . the two eigenfunctions for the @xmath36 heisenberg triangular system with spin hamiltonian eq . 1 corresponding to the eigenstate given by the two @xmath15=1/2 ground states can be expressed using basis functions of @xmath62@xmath63@xmath64 in which up and down arrows represent up and down spin , respectively , for @xmath65 ; @xmath66 where @xmath67=e@xmath68 . these two eigenfunctions correspond to two different spin chiral states in which v@xmath37 spin rotates 120 degrees with respect to neighboring spins respectively in clockwise and counterclockwise directions on the triangle . other eigenfunctions for the same hamiltonian can be obtained by making linear combinations of the basis functions . the two functions below are also eigenfunctions for the @xmath15=1/2 frustrated spin state : @xmath69 although the total spin moment for both sets of eigenfunctions ( eq . 3 and eq . 4 ) is @xmath15 = 1/2 ( 1 @xmath53 ) as required , the expectation values for local spin moments for @xmath56 , @xmath57 and @xmath58 site are estimated to be 1@xmath53 , 0@xmath53 and 0 @xmath53 for the @xmath59 wavefunction and -1/3 @xmath53 , 2/3@xmath53 and 2/3 @xmath53 for the @xmath60 , respectively , as schematically shown in the inset of fig . 4 . on the other hand , the expectation values for the local spin moments for the @xmath70 and @xmath71 eigenfunctions in eq . 3 are calculated to be 1/3 @xmath53 for each spin . since experimentally three v@xmath37 ions in the @xmath15 = 1/2 ground state with different spin moments ( 1 @xmath53 , 1/3 @xmath53 and 2/3 @xmath53 ) are observed , we conclude that the two eigenfunctions to be associated with the two @xmath15 = 1/2 ground states can be expressed by @xmath59 and @xmath60 . each of the two eigenfunctions should be assigned to one of the two quasidegenerate @xmath15=1/2 ground states . since the measurements were performed at a temperature smaller than the splitting of the two quasi - degenerate states the lower energy state is more populated and thus should yield a larger nmr signal . since we observe a larger signal intensity for the @xmath13v nmr signal arising from a v@xmath37 moment of 1 @xmath53 ( see the inset in fig . 4 ) we conclude that the lower energy of the two @xmath15 = 1/2 states correspond to the eigenfunction @xmath59 ( see the inset ( a ) in fig . 4 ) . the other two weak @xmath13v - nmr signals ( see fig . 3(c ) ) arise from v@xmath37 ions with moments ( 1/3 @xmath53 and 2/3 @xmath53 ) and correspond thus to the eigenfunction @xmath60 ( see the inset ( b ) in fig . 4 ) pertaining to the higher energy @xmath15=1/2 state . the above conclusions are in agreement with spin densities calculations which indicate @xmath59 as one of the eigenfunctions for the @xmath15=1/2 ground state.@xcite figure 5 shows the temperature dependence of the proton-1@xmath72 under various magnetic fields.@xcite in magnetic fields above 2.7 t , where the ground state of the cluster is @xmath73 , 1@xmath72 decreases with decreasing temperature at a rate which increases with increasing magnetic field . below 2.7 t where the ground state of the cluster is the nearly degenerate @xmath74 states , 1@xmath72 decreases on lowering the temperature . but , at very low temperatures , 1@xmath72 shows temperature independent behavior with a constant value that does not depend on the magnetic field , as shown in fig . 5 . under various magnetic fields for v15 . solid lines show fitted fitting results discussed in the text . , width=302 ] except for the temperature independent behavior of 1@xmath72 at very low temperatures in the @xmath74 ground states , the temperature dependence of 1@xmath72 can be explained as thermally activated behavior 1/@xmath75 ) , with a gap magnitude of @xmath76 . the lines in fig . 5 show fitting results using a relation 1/@xmath77 ) + @xmath78 where @xmath78 is a constant value of @xmath14 0.1 ( msec@xmath79 ) for the @xmath74 ground states and is zero for the @xmath80 ground state . the @xmath47 dependence of @xmath76 estimated from the temperature dependence of 1@xmath72 follows a relation of @xmath76 = 1.33@xmath81 ( k ) where @xmath81 is the external field in tesla . since @xmath76 = 1.33@xmath81 corresponds to the zeeman energy between the ground state and the first excited magnetic sublevels in the same total spin state , the @xmath26 dependent behavior of 1@xmath72 can be explained by thermal fluctuations of the magnetization between the ground and first excited sublevels for each total spin state . on the other hand , the temperature independent behavior of 1@xmath72 at very low temperatures in the nearly degenerate @xmath74 ground state can not be explained by thermal fluctuations of the magnetization . since most of the clusters occupy the nearly degenerate two @xmath74 ground state sublevels at low enough temperature , magnetic fluctuations can be considered only between the two nearly degenerate @xmath74 ground states . the temperature and magnetic field independent behavior of 1@xmath72 indicates that the magnetic fluctuations between @xmath74 ground states do not depend on temperature and magnetic field . as shown by eq . ( 3 ) , in the case of an @xmath36 antiferromagnetic heisenberg triangular magnet , the two @xmath74 states can be regarded as two states with different spin chirality ( different spin structures).@xcite since the fluctuations between two chiral states with different spin structures would give rise to a local field fluctuation at the proton sites , such chiral fluctuations could be a relaxation mechanism which gives rise to temperature independent behavior of 1/@xmath18 . however , as shown in the sec . iii - a , the two @xmath74 states in the v15 cluster are described not by the two purely chiral states but by different states given by mixtures of the two chiral states . although it is not clear whether the temperature independent behavior of 1@xmath72 originates from the chiral fluctuations between the mixed chiral states of the @xmath74 states in the v15 system or not , the temperature and magnetic field independent behavior of 1@xmath72 apparently relates to the existence of the two @xmath74 states , which originates from the spin frustration . the compound [ mo@xmath5fe@xmath6o@xmath7(mo@xmath3o@xmath8(h@xmath3o))@xmath3 ( mo@xmath3o@xmath9h@xmath3(h@xmath3o))(ch@xmath10coo)@xmath11(h@xmath3o)@xmath12@xmath4150h@xmath3o ( in short fe30 ) is one of the largest molecular paramagnets prepared to date and has 30 fe@xmath19 ( @xmath25 = 5/2 ) ions occupying the 30 vertices of an icosidodecahedron ( see fig . 1 ) , resulting in a closed spherical structure of 2.5 nm diameter.@xcite this polyhedron contains 20 corner - sharing fe3 triangles with antiferromagnetic exchange coupling ( @xmath82 k ) between fe spins.@xcite 0.1 t in a temperature range of 4.0 - 300 k using a superconducting quantum interface device ( squid ) magnetometer ( quantum design mpms-7 t ) . data reported by mller _ et al . _ are shown by red triangles.@xcite , width=302 ] the compound shows a simple paramagnetic behavior down to @xmath14 20 k where the magnetic susceptibility departs from a simple curie - weiss law . the magnetic susceptibility shows a broad peak around 2 k and is almost constant below 1 k down to 0.12 k.@xcite no three dimensional magnetic ordering down to 60 mk was reported from magnetization measurements.@xcite figure 6 shows the temperature dependence of @xmath83 together with data reported by mller _ et al_.@xcite the broad peak around 2 k associated with the short - range ordering due to the antiferromagnetic interaction @xmath84 1.57 k between fe@xmath19 spins in the molecules . figure 7(a ) shows @xmath21h - nmr spectra at @xmath48 = 7.03 mhz for several temperatures . @xcite figure 7(b ) shows the temperature dependence of the full width at half maximum ( fwhm ) of the @xmath21h nmr spectrum at @xmath85 mhz ( @xmath47 = 0.16 t ) , together with data at @xmath86 mhz ( @xmath87 t).@xcite a dramatic broadening of the spectrum is clearly observed at @xmath14 600 mk for both magnetic fields . the fwhm is almost independent of magnetic field below 600 mk , which indicates freezing state of fe@xmath19 spins . as can be seen in fig . 7(a ) , the @xmath21h - nmr signal at @xmath88 and 0.19 k can be observed at @xmath89 0 t , which is also direct evidence of very slow fluctuations of the fe spin moments . almost no signal was observed around 1 k due to a shortening of the nuclear spin - spin relaxation time @xmath90 which follows a corresponding shortening of the nuclear spin - lattice relaxation time @xmath18 . h - nmr spectrum at @xmath91 7.03 mhz . ( b ) temperature dependence of fwhm of @xmath21h nmr spectrum in fe30 at @xmath89 0.61 t ( circles ) and @xmath89 0.16 t ( triangles).,width=302 ] the spin freezing of fe@xmath19 spins can also be observed in proton 1/@xmath18 measurements . figure 8(a ) shows the temperature dependence of 1@xmath72 under two magnetic fields . with decreasing temperature , 1@xmath72 for @xmath89 0.75 t decreases gradually and starts to increase around 20 k , then shows a peak around 2 k. similar temperature dependence of 1@xmath72 was reported by lago _ et al._,@xcite and the peak of 1@xmath72 is found to increase in magnitude and move toward low temperature by decreasing magnetic field . the temperature dependence of 1@xmath72 at @xmath89 0.16 t is also plotted in the figure . 1@xmath72 could not be measured between 0.6 k and 1.5 k because of shortening of @xmath90 ( i.e. , @xmath18 ) . in general , 1/@xmath18 is expressed by the fourier transform of the time correlation function of the transverse component @xmath92@xmath93 of the fluctuating local field at nuclear sites with respect to the nuclear larmor frequency @xmath94 as,@xcite @xmath95 where @xmath52 is the gyromagnetic ratio of the nuclear spin . when the time correlation function is assumed to decay as exp(-@xmath96 ) , one can write@xcite @xmath97 where @xmath98 and is a parameter related to the hyperfine field at nuclear sites . @xmath99 corresponds to the inverse of the correlation time of the fluctuating hyperfine fields at the h sites , due to the fe@xmath19 spins , and thus can be regarded as the fluctuation frequency of the spins . to analyze the temperature dependence of 1@xmath72 by using eq . ( 6 ) , it is useful to re - plot the data by changing the vertical axis from 1@xmath72 to 1/@xmath100 as shown in fig . 8(b ) , where the @xmath83 values for @xmath101 k are obtained from @xmath83 data reported by mller in ref . and for @xmath102 k we used our data . if @xmath99 is independent of temperature , 1/@xmath100 should be constant which is in fact observed above 30 k in fig . this indicates that the nuclear spin relaxation above @xmath14 30 k is explained by paramagnetic fluctuations of the fe@xmath19 spins whereby the spins fluctuate almost independently from each other . on the other hand , below 30 k 1/@xmath100 increases with decreasing temperature . this indicates the simple paramagnetic fluctuations model can not explain the @xmath26 dependence of 1/@xmath100 . according to eq . ( 6 ) , 1/@xmath100 is proportional to 1/@xmath99 when @xmath99 @xmath103 @xmath94 ( fast - motion regime ) , while 1/@xmath100 is proportional to @xmath99/@xmath104 in the case of @xmath99 @xmath105 @xmath94 ( slow - motion regime ) . when @xmath99 = @xmath94 , 1/@xmath18 reaches a maximum value . thus , the slowing down of the fluctuation frequency @xmath99 of fe@xmath19 spins produces a peak of 1/@xmath18 . assuming @xmath106 ( rad@xmath107@xmath4k@xmath4emu / mol / s@xmath107 ) and @xmath99 = 1.34 @xmath108 10@xmath109 @xmath110 ( rad / s ) , the experimental results at low temperature regions below 4 k are qualitatively reproduced by eq . ( 6 ) , as shown in fig . 8(b ) by solid lines for different magnetic fields . these results indicate that the peak observed in the temperature dependence of 1/@xmath100 originates from a crossover between the fast - motion regime and the slow - motion regime , whereby the fluctuation frequency of fe@xmath19 spins below the peak temperature is less than the nmr frequency range which is of the order of mhz . to extract the temperature dependence of the fluctuation frequency of fe@xmath19 spins for a wide temperature region , we estimate the temperature dependence of @xmath99 from the temperature dependence of 1/@xmath100 assuming eq . ( 6 ) is valid for the entire temperature region . the estimated temperature dependences of @xmath99 for the two different magnetic fields are shown in fig . this log - log plot shows very clearly that @xmath99 has a power low behavior ( @xmath111 ) at low temperatures below @xmath14 1 k and is nearly constant at @xmath112 hz at high temperatures . at low enough temperature fe@xmath19 spins can fluctuate with low frequency which is less than the nmr frequency of the order of mhz . such a slow spin dynamics is consistent with the observation of broadening of nmr spectrum below 600 mk . /d@xmath47 in fe30 at @xmath113 k. the experimental data from ref . are shown by the black line . the green , blue and pink broken lines are calculated results with no single anisotropy ( @xmath114 : uniform j model ) , easy axis type single ion anisotropy of @xmath115 , and @xmath116 , respectively . , width=302 ] the origin of the spin freezing in fe30 is pointed out to be intramolecular exchange disorder by theoretical studies using monte carlo simulation.@xcite in the simulation , the distribution of the intramolecular exchange interaction @xmath117 is introduced to reproduce the d@xmath118/d@xmath47 curve at low temperatures which shows a broad dip around 5 t due to 1/3 anomaly.@xcite for a model with uniform @xmath117 , the broad dip is not reproduced at all as shown in fig . 9.@xcite the simulation found that the exchange disorder leads to a spin freezing state and slow spin dynamics.@xcite in the previous simulation , any anisotropy effects and/or dzyalosinskii - moriya interactions were not included . in order to investigate the effects of single ion anisotropy @xmath119 ( easy - axis type ) for fe@xmath19 spins on the d@xmath118/d@xmath47 curve , we have calculated the d@xmath118/d@xmath47 curve using the same method in ref . . it turns out that , although the dip becomes broader with increasing @xmath119 ( up to the half of @xmath117 ) as shown in fig . 9 , the d@xmath118/d@xmath47 can not be reproduced well as in the case of @xmath23exchange disorder@xmath24 shown by the red line . we also found the easy - plane type anisotropy produces similar effects as in the case of @xmath23easy axis@xmath24 . thus we consider that the origin of the spin freezing is not due to single ion anisotropy effects but more likely to the distribution of the exchange interactions . however , we can not rule out the possible effects due to dzyalosinskii - moriya interactions . this is outside the scope of the present work and remains to be confirmed by detailed calculations . in any case , since it is important to know the actual structure of fe30 at low enough temperature , it would be very helpful to perform low temperature x - ray diffraction measurements . the circles show cu@xmath120 ( @xmath36 ) ions . the black and gray lines are corresponding to the intra - triangle exchange ( @xmath121 k ) and inter - triangle exchange ( @xmath122 k ) coupling paths . bottom : temperature dependence of the triangular spin tube measured at @xmath123 t. the inset shows the @xmath47 dependence of the peak temperature @xmath124 . , width=226 ] [ ( cucl@xmath3tachh)@xmath10cl]cl@xmath3 ( tach = cis , trans-1,3,5-triamino - cyclohexane ) , is a newly synthesized triangle spin tube.@xcite with a trigonal crystal structure ( space group p63/m , lattice parameters @xmath125 , @xmath126),@xcite the triangles formed by three cu@xmath120 ( 3@xmath127 , @xmath25 = 1/2 ) ions are aligned to construct infinite stacks of antiprisms in a one dimensional way along the @xmath128 direction . a schematic structure of the triangular copper chain is shown in the top part of fig . 10 . each triangle rotates 180 degree with respect to its neighbor triangle in the chain direction ( a twisted triangular spin tube ) . each cu spin in the triangle is coupled to two cu spins of each neighboring triangle . from the temperature dependence of the magnetic susceptibility measured between 2 and 300k , antiferromagnetic interactions of intratriangle and intertriangle are reported to be @xmath129 @xmath14 0.9 k and @xmath130 @xmath14 1.95k , respectively.@xcite although initially the system was reported to have a spin singlet ground state with a spin gap @xmath131 @xmath14 0.4 k from a magnetization curve at @xmath26 = 0.5 k,@xcite theoretical investigations indicate a gapless ground state , i.e , no gapped ground state for the twisted triangular spin tube.@xcite experimentally , a gapless ground state of the cu spin tube was revealed by an observation of proton nmr line broadening at very low temperatures below 1 k,@xcite which is shown below . specific heat measurements down to 100 mk also evidenced a gapless ground state of the cu spin tube from the observed @xmath26 linear dependence of the specific heat below 0.6 k , which is attributed to a tomonaga - luttinger liquid state.@xcite figure 11(a ) shows a typical temperature dependence of @xmath21h nmr spectra measured at @xmath48 = 42.5759mhz.@xcite a single peak of @xmath21h nmr spectrum can be observed at high temperature above @xmath14 200 k. the spectrum broadens at low temperatures and shows three peaks , denoted p1 , p2 and p3 , as seen in fig . the peak positions for p1 and p3 shift to lower and higher magnetic field , respectively , on lowering temperature . on the other hand , the peak position of p2 observed around the larmor magnetic field is independent of temperature . typical temperature dependence of nmr shift ( @xmath119 ) for the p1 line is shown in fig . below @xmath14 1.5k , the shifted peaks p1 and p3 disappear due to the shortening of spin - spin relaxation time @xmath90 and only the non - shifted peak p2 can be observed . at very low temperatures below 0.2 k , shoulder structure can be seen on either side of the main peak . t he fwhm of the p2 line as a function of temperature is shown in fig . 11 ( c ) , which scales well with @xmath26 dependence of @xmath132 above 2 k. below 2 k , the fwhm shows a local minimum around 1 k and levels off on lowering temperature . since the line width relates to the magnetization @xmath118 , these experimental results indicate @xmath118 is almost constant and has finite value at very low temperatures , which is consistent with a gapless ground state of the system reported from specific heat measurements . h - nmr spectra measured at @xmath133 mhz . ( b ) temperature dependence of nmr shift ( @xmath119 ) for the p1 line . ( c ) temperature dependence of full width of half amplitude ( fwhm ) for the p2 line . , width=328 ] in general , there are two possibilities for the origin of the hyperfine field at proton sites , that is , dipolar fields and contact hyperfine fields from the cu@xmath120 spins . in the case of powder sample , the dipolar fields yield a broadening of the line but no net shift , while the net shift of the line can be produced by the contact hyperfine fields due to the overlap of the @xmath25 electron wave function at the proton sites with the @xmath134 electron ( cu@xmath120 ) wave function . in the cu spin tube system , protons can be mainly classified into two different sites . one is the protons contributed to hydogen bonds between triangles and another is the protons in cyclohexan trimanio located between the spin tubes . the almost unshifted line ( p2 ) originates from the protons which are coupled to cu@xmath120 spins with dipolar interaction only , corresponding to the protons in cyclohexan trimanio located between the spin tubes . on the other hand , the shifted lines can be ascribed to the protons inside in the exchange path between cu@xmath120 spins because these protons experience the isotropic hyperfine field due to the cu@xmath120 spins . the existence of isotropic hyperfine field for the protons is microscopic evidence of the importance of the hydrogen bonds for magnetic interactions between the triangles , responsible for the one dimensional magnetic nature of the system . in order to see the temperature evolution of the line width at the shoulder part of the p2 line , we define @xmath135 which is estimated from the line width at 10 @xmath136 of the peak intensity . the @xmath135 are displayed as a function of the temperature in fig . 12(a ) together with results measured at different magnetic fields from 0.223 t to 1.46 t. the @xmath26 dependence of @xmath135 above 0.3 k are in good agreement with that of fwhm , indicative of the existence of the local minimum in the @xmath26 dependence of @xmath83 around 1 k. it is noted that a different behavior between fwhm and @xmath135 can be found below 0.2k . although the origin of the different behavior is still not clear at present , it is important to point out that the increase of @xmath135 at low temperatures indicates the growth of a distribution of internal fields at the proton sites from the spin moments on the cu@xmath120 ions . since we observed the increase of @xmath135 even at a low magnetic field of @xmath137 t , we conclude that spin moments exist at low temperatures and the ground state of the system is not a spin singlet state but a gapless state . as shown in fig . 12 ( b ) , the @xmath135 at @xmath26 = 0.1 and 1.5 k increases linearly with @xmath47 , again consistent with a gapless ground state . our conclusion is consistent with theoretical results@xcite where the ground state of the twisted cu spin tube is predicted to be not gapped but gapless . 12(c ) shows the dc magnetization curve at @xmath26 = 0.09 , 0.3 , 0.5 and 1.8 k. the magnetization has a small fraction which shows brillouin function - like behavior in the low magnetic field region at low temperatures . this is attributed to spin contributions due to magnetic impurities . the fraction of the excess magnetization is estimated to be @xmath14 2 @xmath136 of the total magnetization . the most important finding in the dc magnetization measurement is an obvious change in slope of the magnetization curves around 5 t below and above 0.5 k , which can be also seen in change of d@xmath118/d@xmath47 as shown in the inset of fig . this signature can be attributed to the so - called 1/3 plateau in magnetization , since the magnetization around 5 t is @xmath14 1 @xmath53/mol , which is one - third of an expected total magnetization of 3 @xmath53/mol and 5 t is close to one - third of the saturation magnetic field of 14 t.@xcite a phase diagram for a twisted spin tube as a function of @xmath138 and of magnetic field , @xmath47 , has been proposed theoretically by fouet @xmath139,@xcite , using a density matrix renormalisation group calculation . when @xmath138 is smaller than 1.22 , the system has a gapped ground state with a two - fold - degenerate state , due to spin chirality . on the other hand , if @xmath138 is large enough , the system is considered an effective @xmath140 chain with a gapless ground state . the phase transition between the gapped and gapless states takes place for a critical point at @xmath138 @xmath14 1.22 . the 1/3 plateau in the magnetization curve is proposed below @xmath138 @xmath14 1.6 . although the ratio , @xmath138 , is reported previously to be @xmath14 2.16 from the temperature dependence of magnetic susceptibility , the observation of 1/3 plateau with the gapless ground state indicates that the ratio should fall in the interval 1.22 @xmath55 @xmath138 @xmath55 1.6 , suggesting that the cu3 spin tube is located close to the critical point of the phase transition . dependence of @xmath76@xmath47 for the p2 line under various magnetic fields . ( b ) external field dependence of @xmath135 at @xmath26 = 0.1 k and 1.5 k. ( c ) @xmath47 dependence of dc magnetization @xmath118 at @xmath26 = 0.09 , 0.3 , 0.5 and 1.8 k. the arrow shows the magnetic field , where the slope in magnetization changes below 0.5 k. the inset shows d@xmath118/d@xmath47 for @xmath26 = 1.8 k and 0.09 k around 5 t. , width=340 ] now let us discuss dynamical properties of the cu@xmath120 spins . we have carried out proton @xmath18 measurements at the peak position of the p2 line in a wide temperature range of @xmath141 k.@xcite fig . 13 shows the temperature dependence of 1@xmath72 under various magnetic fields . with decreasing temperature from room temperature , 1@xmath72 decreases smoothly and starts to increase around 2k , then shows a peak around 0.6 k. as the external field is increased , the peak temperature of 1@xmath72 shifts to higher temperature and at the same time the height of 1@xmath72 becomes smaller . at very low temperatures below @xmath14 0.1 k , 1@xmath72 seems to merge at different magnetic fields , which is probably due to experimental error due to heat up problems for the @xmath18 measurements . the inset of fig . 13 shows the @xmath47 dependence of 1@xmath72 measured at @xmath142 k where 1@xmath72 is found to be proportional to 1/@xmath143 whose behavior is well explained by spin diffusion effects in one dimensional magnetic systems.@xcite this confirms again the one dimensionality of the twisted triangular spin tube . under various magnetic fields . temperature dependences of @xmath144 estimated from @xmath119 , fwhm and @xmath76 shown by open symbols are plotted with arbitrary units in vertical axis to show the same temperature dependence of 1/@xmath18 at high temperatures . the inset shows the @xmath47 dependence of 1/@xmath18 as a function of 1/@xmath143 at @xmath142 k. the solid line in the inset shows a relation of 1/@xmath18 @xmath145 1/@xmath143 expected from spin diffusion effects in one dimensional spin systems . , width=226 ] we analyze the 1/@xmath18 data with the same equation used for the case of fe30 spin ball . as shown in eq . ( 6 ) , if the @xmath99 ( here @xmath99 corresponds to the inverse of the correlation time of the fluctuating hyperfine fields at the h sites , due to the cu@xmath120 spins ) is independent of temperature , temperature dependence of 1@xmath72 is simply expressed by @xmath144 . the open symbols ( circles , squares and triangles ) in fig . 13 show the @xmath26 dependences of @xmath144 estimated from the @xmath26 dependences of @xmath119 , fwhm and @xmath135 of @xmath21h nmr experiment . as can be seen in the figure , the @xmath26 dependence of 1@xmath72 seems to be in good agreement with that of @xmath144 , although 1@xmath72 depends on the external field due to the spin diffusion effects as described above . thus the nuclear spin relaxations in the twisted spin tube above @xmath14 3 k is mainly explained by the paramagnetic fluctuations of the cu@xmath120 spins . on the other hand , the simple paramagnetic fluctuations model can not reproduce the experimental results below 3k , especially the peak behavior in 1@xmath72 . it is noted that the @xmath47 dependence of 1@xmath72 above the peak is different from that of 1@xmath72 below the peaks . the @xmath47 dependence above the peak temperature can be explained by the spin diffusion nature as discussed above , but 1@xmath72 depends on @xmath47 stronger than the @xmath47 dependence of spin diffusion effects below the peak temperature . in order to analyze the @xmath26 and @xmath47 dependences of 1@xmath72 at low temperatures below 3 k using eq . ( 6 ) , we re - plotted the 1/@xmath100 vs. @xmath26 as in the case of fe30 , as shown in fig . 14(a ) where the @xmath144 values used are estimated from @xmath26 dependence of @xmath135 . although there is a slight difference between @xmath144 estimated from fwhm and @xmath135 at very low temperatures , it is only at very low temperatures below @xmath14 0.2 k , so that our analysis is not modified very much . using @xmath146 ( rad@xmath107@xmath4k@xmath4emu / mol / s@xmath107 ) and @xmath147 ( rad / s ) , the experimental results are well reproduced by eq . ( 6 ) as shown in fig . 14(a ) by solid lines for different magnetic fields . these results indicate that the peak behavior observed in the @xmath26 dependence of 1/@xmath100 originates from a crossover between the fast - motion regime and the slow - motion regime , and fluctuation frequency of cu@xmath120 spins below the peak is less than the nmr frequency at low temperatures . under various magnetic fields . the solid lines are calculated results using eq . ( b ) temperature dependence of fluctuation frequency ( @xmath99 ) of the cu@xmath120 spins estimated from the @xmath26 dependence of 1/@xmath100 . the solid line in the figure corresponds to a relation of @xmath148 . , width=302 ] we extract the temperature dependence of @xmath99 from the temperature dependence of 1/@xmath100 assuming eq . ( 6 ) is valid for the entire temperature region . the estimated @xmath26 dependences of @xmath99 for different external magnetic fields are shown in fig . this log - log plot shows very clearly that @xmath99 has power low behavior ( @xmath149 ) at low temperatures below @xmath14 3 k and shows an almost constant value of the order of 10@xmath150 hz at high temperatures . thus we conclude that the fluctuation frequency of the cu@xmath120 spins is almost constant at high temperatures and becomes slower and slower with a relation of @xmath151 on lowering the temperature below 3 k. at low enough temperature , cu@xmath120 spins can fluctuate very slowly , less than nmr frequency ( @xmath14 10@xmath152 hz ) . such a spin freezing phenomena could explain the different behavior in @xmath135 and fwhm at low temperatures . the protons close to one cu@xmath120 spin experience non - vanishing dipolar fields from the spins , which leads to broadening of the spectrum . on the other hand , the dipolar fields at proton sites located at the middle position between the antiferromagnetically coupled cu spins could be canceled out , resulting in sharper line width . thus the slight decrease of the fwhm below 0.2 k is though to be a signature of growth of antiferromagnetic correlations between the cu@xmath120 spins in the slow - motion regime . in usual @xmath36 one dimensional quantum spin systems with a gapless ground state , 1@xmath72 is considered to be in the fast - motion regime because such a spin freezing is not expected due to a continuous energy dispersion for the spin excitation spectrum . 1@xmath72 shows temperature independent behavior at high temperatures and a logarithmic increase at low temperatures without any crossover between fast - motion and slow - motion regimes.@xcite this is in sharp contrast to the present case . on the other hand , a similar crossover between slow - motion and fast motion regimes was observed in various nanoscale magnetic molecules.@xcite based on a similar analysis , the temperature dependence of @xmath99 in the magnetic molecules is found to be proportional to @xmath153 . this @xmath26 dependence is explained by the spin - phonon interactions.@xcite on the other hand , @xmath99 @xmath14 @xmath151 behavior in the present system can not be explained by the spin - phonon interactions . recently the specific heat of the system has been measured down to 0.1 k,@xcite confirming no three dimensional magnetic ordering under zero magnetic field down to 0.1 k. the temperature dependence of specific heat shows a broad peak around 2 k and rapid decrease below the peak with decreasing temperature . below 0.5 k , the specific heat is almost proportional to @xmath26 , which indicates a gapless ground state of the system . this is consistent with our experimental results . interestingly , @xmath26 linear behavior observed in the specific heat is one of the signatures of the tomonaga - luttinger liquid ( tll ) state . in the case of tll state , 1@xmath72 is expected to follow the relation of 1@xmath72 @xmath14 1/@xmath154 with @xmath155.@xcite goto et al . reported a observation of 1/@xmath18 @xmath14 1/@xmath156 behavior in one dimensional @xmath157 quantum spin system ( ch@xmath10)@xmath158nni(no@xmath3)@xmath10 in a gapless state under magnetic fields , suggesting the realization of tll state in that system.@xcite in the present system , 1@xmath72 does not show such a behavior expected for the tll states . thus it seems to be difficult to explain the anomalous behavior of 1@xmath72 observed in the twisted heisenberg triangular spin tube by the conventional explanations described above . at present , although it is not clear the reasons for the slowing down of spin fluctuations at low temperatures , it would be interesting if these peculiar behaviors observed in the twisted cu spin tube were related to spin chirality . a review of nmr studies performed on the three nanoscale molecular magnets with peculiarly - structured frustrated systems has been presented . \(i ) v15 : a model system of @xmath25=1/2 heisenberg triangular antiferromagnet . the @xmath13v nmr spectrum measurements at very low temperatures below 0.1 k and as a function of the external applied field have allowed the determination of the local microscopic spin configuration in the frustrated ground state . it is found that the pure chiral states described by eq . ( 3 ) are broken down in the v15 nanomagnet . the two quasidegenerate @xmath15 = 1/2 ground states correspond to two different spin configurations which can be described by the two eigenfunctions @xmath59 and @xmath60 . we also found that the lower energy of the two @xmath15 = 1/2 split states is the one pertaining to the eigenfunction @xmath59 . the local spin configuration is consistent with a small structural distortion from an equilateral triangle to a nearly isosceles one . the dynamical magnetic properties of v15 were investigated by proton spin - lattice relaxation rate ( 1/@xmath18 ) measurements . in the @xmath15 = 3/2 state , 1/@xmath18 shows thermally activated behavior as a function of temperature . on the other hand , a temperature independent behavior of 1/@xmath18 at very low temperatures is observed in frustrated @xmath15 = 1/2 ground state below 2.7 tesla . possible origins for the peculiar behavior of 1/@xmath18 have been discussed in terms of magnetic fluctuations due to spin frustrations . \(ii ) fe30 : a spin ball . static and dynamical properties of fe@xmath19 ( @xmath20 ) have been investigated by proton nmr spectra and 1/@xmath18 measurements . from the temperature dependence of 1/@xmath18 , the fluctuation frequency of the fe@xmath19 spins is found to decrease with decreasing temperature , indicating spin freezing at low temperatures . the spin freezing is also evidenced by the observation of a sudden broadening of @xmath21h nmr spectra below 0.6 k. the origin of the spin freezing in fe30 has been suggested to be the distribution of the exchange interactions . \(iii ) cu3 : a twisted spin tube . @xmath21h nmr and dc magnetization data in the cu spin tube were described . an observation of magnetic broadening of @xmath21h nmr spectra at low temperatures below 1 k directly revealed a gapless ground state of the system . the dc magnetization data below 1 k are also consistent with the gapless ground state . we also found the clear change in the slope of the magnetization curves around 5 t attributed to the so - called 1/3 plateau in magnetization . the temperature dependence of @xmath22 under various magnetic fields revealed an usual slow spin dynamics in the cu3 spin tube . the temperature dependence of 1/@xmath18 was not simply explained by the spin - phonon interactions model , nor is it the tomonaga - luttinger liquid model . further theoretical studies are required to understand the peculiar behavior of 1/@xmath18 in the quasi one dimensional cu3 spin tube . it is important to point out that nmr appears to be a unique tool for determination of the local spin configuration , ground state properties and low energy spin dynamics of frustrated systems . the author would like to acknowledge collaborations and fruitful discussions with ferdinando borsa , ken - ichi kumagai , alessandro lasscialfari , yusuke nishisaka , yuri fujiyoshi , yuzuru sumida , yuichi hatanaka , edoardo micotti , marshall luban , ruslan prozorov , christian schrder , jrgen schnack , and hiroyuki nojiri . special thanks go to paul kgerler , xikui fang , and leroy cronin who have been succeeded in synthesizing valuable crystals which makes it possible for us to carry out this research . monte carlo simulation for the fe30 magnetization has been performed by christian schrder . dc magnetization for the cu spin tube at low temperatures below 1 k was measured by hiroshi amitsuka , ken - ichi tenya , and yusei shimizu at hokkaido university , japan . a part of nmr work has been performed at hokkaido university in japan , which is supported by 21@xmath159 century coe programs `` topological science and technology '' at hokkaido university and grant - in - aid from the ministry of education , culture , sports , science and technology of japan . the work at ames laboratory was supported by the u.s . department of energy ( doe ) , office of basic energy sciences , division of materials sciences and engineering . ames laboratory is operated for the u.s . doe by iowa state university under contract no . de - ac02 - 07ch11358 . 99 , edited by h. t. deip , world scientific , singapore , 2005 . , edited by c. lacroix , p. mendels , and f. mila , springer - berlin heidelberg , 2011 . p. w. anderson , mater . * 8 * , 153 ( 1973 ) . e. f. schender and p. c. holdsworth , in _ fluctuations and order : the new synthesis _ , springer - verlag , berlin , 1994 , and references therein ; a. keren , phys . lett . * 72 * , 3254 ( 1994 ) . j. n. reimers and a. j. berlinsky , phys . b * 48 * , 9539 ( 1993 ) . t. inami , t. morimoto , m. nishiyama , s. maegawa , y. oka , and h. okumura , phys . b * 64 * 2001 ) 054421 . d. gatteschi , l. pardi , a. l. barra , a. muller , j. doring , nature * 354 * , 463 - 465 ( 1991 ) . e. j. l. mcinnes , s piligkos , g. a. timco , r. e.p . winpenny , coordination chemistry reviews * 249 * , 2577 ( 2005 ) . d. gatteschi , r. sessoli and j. villain , molecular nanomagnets ( oxford university press , 2006 ) . a. mller , s. sarkar , s. q. nazir shah , h. bgge , m. schmidtmann , s. sarkar , p. kgerler , b. hauptfleisch , a. x. trautwein , and v. schunemann , angew . chem . , int . ed . * 38 * , 3238 ( 1998 ) . c. schrder , h. nojiri , j. schnack , p. hage , m. luban , and p. kgerler , phys . lett . * 94 * , 017205 ( 2005 ) . j. lago , e. micotti , m. corti , a. lasicalfari , a. bianchi , s. carretta , p. santini , d. procissi , s. h. beak , p. kgerler , c. baines and a , amato , phys . b * 76 * 064432 ( 2007 ) . g. seeber , p. kgerler , b. m. kariuki , and l. cronin , chem . commun . * 2004 * , 1580 ( 2004 ) . j. schnack , h. nojiri , p. kgerler , g. j. t. cooper , and l. cronin , phys . b * 70 * , 174420 ( 2004 ) . k. kawano , and m. takahashi , j. phys . * 66 * , 4001 ( 1997 ) . a. luscher , r. m. noack , g. misguich , v. n. kotov , and f. mila , phys . rev . b * 70 * , 060405(r ) ( 2004 ) . fouet , a. luchli , s. pilgram , r. m. noack , and f. mila , phys . b * 73 * , 014409 ( 2005 ) . k. okunishi s. yoshikawa , t. sakai , and s. miyashita , prog . supp . * 159 * , 297 ( 2005 ) . m. sato , phys . b * 75 * , 174407 ( 2007 ) . m. sato , and t. sakai , phys . b * 75 * , 014411 ( 2007 ) . s. nishimoto , and m. arikawa , phys . b * 78 * , 054421 ( 2008 ) . t. sakai , m. sato , k. okunishi , y. otsuka , k. okamoto , and c. itoi , phys . b * 78 * , 184415 ( 2008 ) . t. sakai , m. sato , k. okunishi , k. okamoto , and c. itoi , j. phys . : condens . * 22 * , 403201 ( 2010 ) . m. lajko , p. sindzingre , and k. penc , phys . 108 , 017205 ( 2012 ) . f. borsa , a. lascialfari , and y. furukawa , _ nmr in magnetic molecular rings and clusters _ , in `` novel nmr and epr techniques '' , eds . j. dolinsek , m. vilfan , s. zumer , springer - berlin heidelberg 2006 , pp.297 - 349 . f. borsa , y. furukawa , and a. lascialfari , inorganica chumica acta , * 361 * 3777 ( 2008 ) . j. choi , l. a. w. sanderson , j. l. musfeldt , a. ellern , and p. kgerler , phys . rev . * b 68 * 064412 ( 2003 ) . d. gatteschi , l. pardi , a. l. barra , and a. mller , mol . * 3 * , 157 ( 1993 ) . barra , d. gatteschi , l. pardi , a. mller , and j. dring , j. am . * 114 * , 8509 ( 1992 ) . g. chaboussant , r. basler , a. sieber , s. t. ochsenbein , a. desmedt , r. e. lechner , m. t. f. telling , p. kgerler , a. mller , and h .- u . gdel , europhys . lett . * 59 * 291 ( 2002 ) . y. nishisaka , y. furukawa , k. kumagai , and p. p. kgerler , aip conf . * 870 * 1143 ( 2006 ) . v. v. dobrovitski , m. i. katsnelson , and b. n. harmon , phys . * 84 * 3458 ( 2000 ) . i. chiorescu , w. wernsdorfer , a. mller , h. bgge , and b. barbara , phys . * 84 * 3454 ( 2000 ) . i. chiorescu , w. wernsdorfer , b. barbara , a. mller , and h. bgge , j. magn . magn . mater . * 221 * , 103 ( 2000 ) . s. miyashita and n. nagaosa , prog . theor . phys . * 106 * , 533 ( 2001 ) . h. de raedt , s. miyashita , k. michielsen , and m. machida , phys . rev . * b 70 * 064401 ( 2004 ) . s. miyashita , h. de . raedt and k. michielsen , prog . * 110 * , 889 ( 2003 ) . g. chaboussant , s.t . ochsenbein , a. sieber , h .- u . h. mutka , a. mller and b. barbara , europhys . lett . , * 66 * , 423 ( 2004 ) . y. furukawa , y. nishisaka , k. kumagai , p. kgerler , and f. borsa , phys . b * 75 * , 220402(r ) ( 2007 ) . r. e. watson and a. j. freeman , _ hyperfine interactions _ edited by a. j. freeman and r. b. frankel , academic press , 1967 . k. takanashi , h. yasuoka , y. ueda , and k. kosuge , j. phys . jpn . * 52 * , 3953 ( 1983 ) . c. raghu , i. rudra , d. sen and s. ramasesha , phys . b 64 * , 064419 ( 2001 ) . y. furukawa , y. nishisaka , k. kumagai , f. borsa , and p. kgerler , hyperfine interaction , * 176 * , 65 - 68 ( 2007 ) . j. vallain , j. phys . c. * 10 * , 4793 ( 1997 ) . h. kawamura , j. phys . : condens . matter * 10 * , 4707 ( 1998 ) . a. mller , m. luban , c. schrder , r. molder , p. kgerler , m. axenovich , j. schnack , p. c. canfield , s. l. budko , chemphyschem * 2 * , 517 ( 2001 ) . c. schrder , r. prozorov , m.vannette , x. fang , m. luba , a. matsuo , k. kindo , a. mller , and a. m. todea , phys b * 77 * , 224409 ( 2008 ) . y. furukawa , e. micotti , a. lascialfari , f. borsa , k. kumagai , x. fang , p. kgerler , j. phys . 400 * , 032012 ( 2012 ) . c. schrder , x. fang , y. furukawa , m. luban , r. prozorov , f. borsa , k. kumagai , j. phys . : condens . matter * 22 * , 216007 ( 2010 ) . a. abragam , _ the principle of nuclear magnetism _ , clarendon press , oxford , 1961 . y. furukawa , y. sumida , y. nojiri , p. kgerler , and l. cronin , j. phys . conf . ser . * 150 * , 042036 ( 2009 ) . n. b. ivanov , j. schnack , r. schnalle , j. richter , p. kgerler , g. n. newton , l. cronin , y. oshima , and h. nojiri , phys . lett . * 105 * , 037206 ( 2010 ) . y. furukawa , y. sumida , k. kumagai , f. borsa , h. nojiri , y. shimizu , h. amitsuka , k tenya , p kgerler , and l. cronin , j. phys . ser . * 320 * , 012047 ( 2011 ) . f. borsa , and m. mali , phys . b * 9 * , 2215 ( 1974 ) . m. takigawa , n. motoyama , h. eisaki , and s. uchida , phys . rev . lett . * 76 * , 4612 ( 1996 ) . y. furukawa , k. watanabe , k. kumagai , z. h. jang , a. lascialfari , f. borsa and d. gatteschi , phys . b * 62 * , 14246 , ( 2000 ) . y. furukawa , k. watanabe , k. kumagai , f. borsa and d. gatteschi , phys . b * 64 * , 104401 ( 2001 ) . t. kubo , t. goto , t. koshiba , k. takeda , and k. awaga , phys . b * 65 * , 224425 ( 2002 ) . t. goto , t. koshiba , t. kubo , and k. awaga , phys . b * 67 * , 104408 ( 2003 ) . a. morello , o.n . bakharev , h.b . brom , l.j . de jongh , polyhedron * 22 * , 1745 ( 2003 ) . y. furukawa , k. kumagai , a. lascialfari , s. aldrovandi , f. borsa , r. sessoli and d. gatteschi , phys . b * 64 * 094439 ( 2001 ) . m. ueda , s. maegawa , and s. kitagawa , phys . b * 66 * , 073309 ( 2002 ) . y. furukawa , s. kawakami , k. kumagai , s - h . baek and f. borsa , phys . b * 68 * 180405(r ) , ( 2003 ) . s. h. baek , f. borsa , y. furukawa , y. hatanaka , s. kawakami , k. kumagai , b. j. suh , and a. cornia , phys . b * 71 * , 214436 ( 2005 ) . s. h. baek , m. luban , a. lascialfari , e. micotti , y. furukawa , f. borsa , j. van slageren , and a. cornia , phys . b * 70 * , 134434 ( 2004 ) . i. rousochatzakis , a. luchli , f. borsa and m. luban , phys . b 79 , 064421 ( 2009 ) . p. santini , s. carretta , e. liviotti , g. amoretti , p. carretta , m. filibian , a. lascialfari , and e. micotti , phys . 94 , 077203 ( 2005 ) . t. goto , t. ishikawa , y. shimaoka , and y. fujii , phys . b * 73 * , 214406 ( 2006 ) .

the quantum mechanical propagator of a massive particle in a linear gravitational potential derived already in 1927 by earle h. kennard contains a phase that scales with the third power of the time during which the particle experiences the corresponding force . since in conventional atom interferometers the internal atomic states are all exposed to the same acceleration , this-phase cancels out and the interferometer phase scales as . in contrast , by applying an external magnetic field we prepare two different accelerations and for two internal states of the atom , which translate themselves into two different cubic phases and the resulting interferometer phase scales as . we present the theoretical background for , and summarize our progress towards experimentally realizing such a novel atom interferometer . example.eps gsave newpath 20 20 moveto 20 220 lineto 220 220 lineto 220 20 lineto closepath 2 setlinewidth gsave .4 setgray fill grestore stroke grestore

1609.02337

phases play an extraordinary role in quantum theory . on one hand , they represent the central ingredient of wave mechanics @xmath6 la schrdinger , and on the other , they build a bridge to classical mechanics @xmath6 la hamilton - jacobi @xcite . for these reasons they constitute a crucial ingredient of matter wave interferometers @xcite which nowadays represent standard tools for precision measurements . in this article we consider a new type of atom interferometer whose phase scales with the cube of the time @xmath0 the atom spends in the interferometer . during an impromptu seminar at the nato advanced summer institute at bad windsheim in 1981 eugene paul wigner expressed his discomfort with general relativity @xcite in view of quantum mechanics @xcite . he emphasized that the notion of a space - time trajectory which is a crucial element of gravity is incompatible with the uncertainty relation of quantum mechanics . guided by the work of niels bohr and leon rosenfeld on the limitations of the electromagnetic field @xcite due to quantum fluctuations he argued that quantum theory puts severe restrictions on the measurement of the metric tensor representing the gravitational field . wigner s thoughts expressed in this seminar were a consequence of his work several decades earlier . indeed , already in 1958 together with helmut salecker he had constructed a clock @xcite based on the reflection of light signals from two mirrors and had analyzed the restrictions of the uncertainty relation on the weight of the mirrors . in this context it is also worth mentioning that h. salecker at the _ conference on the role of gravitation in physics _ in chapel hill in 1957 triggered @xcite a discussion on the equivalence principle by a gedanken experiment involving a stream of particles being scattered off a diffraction grating . daniel m. greenberger @xcite a decade later considered a similar arrangement and even argued that mass in quantum mechanics should be an operator . the celebrated colella - overhauser - werner ( cow ) experiment @xcite performed in 1975 propelled these and many other thoughts about the interference of quantum and gravity to the real world . indeed , based on the de broglie wave nature of neutrons @xcite the cow experiment could measure for the first time the phase shift between two arms of a neutron interferometer induced by the gravitational potential of the earth @xcite . the development of new sources of cold atoms @xcite as well as molecules @xcite and , in particular , the realization of bose - einstein condensates @xcite has ushered in a new era of experiments at the interface of quantum mechanics and gravity . now , novel tests of the equivalence principle based on matter wave interferometry of different isotopes of the same atom @xcite such as @xmath7 and @xmath8 or even different species @xcite such as @xmath9 and @xmath8 could be performed . even the detection of gravitational waves based on atom interferometry is pursued today @xcite . recently ref . @xcite suggested that gravitational decoherence gives rise to a universal decoherence . moreover , the atom laser @xcite utilizes gravity to form an airy mode . furthermore , it is mind - boggling that nowadays measurements of atomic transitions @xcite are sensitive to the redshift of the gravitational field . an extremely interesting development in this realm is again taking place in the field of neutron optics due to the experimental realization of the quantum bouncer @xcite . here neutrons exposed to the gravitational field of the earth are reflected from a surface and oscillate up and down . in particular , they experience a potential consisting of the linear ramp and an infinitely steep wall . it is amazing that a measurement of the transitions between the resulting discrete energy levels can put upper bounds @xcite on dark energy and dark matter scenarios . hopefully these few examples convey the excitement in this field of quantum optics and gravity . notwithstanding the fact that we still do not have a complete understanding of quantum gravity @xcite we have come a long way since the early salecker - wigner discussions but many questions remain . indeed , goals such as gravitational wave detection or a compact gravimeter @xcite based on atom optics drive the strive for higher sensitivity of these devices . the kasevich - chu atom interferometer @xcite which is the work horse of atom optics is analogous to the neutron interferometer employed by the cow experiment . the bragg diffraction of the neutron from three crystal planes of a silicon slab are replaced by raman diffraction of the atom from three standing light crystals . as a result , the phase shift introduced by the gravitational potential is _ quadratic _ in the time @xmath10 the particle spends in the interferometer . needless to say a different scaling of the phase shift , for example with @xmath2 , could lead to an enhanced sensitivity of the interferometer . in the present article we propose such an interferometer and describe our progress towards realizing it . our device rests on three principles : ( i ) we take advantage of the cubic dependence of the phase in the propagator in a linear potential . ( ii ) we employ two different internal states of the atom which experience different accelerations . ( iii ) we close the interferometer in position _ and _ velocity by a sequence of four laser pulses . our article is organized as follows . section [ sec:2 ] serves as a motivation . here we recall that the propagator of a particle in a linear potential displays a phase which is cubic in time . we show that this global phase depends on the initial wave function and outline an interferometric measurement scheme . in this arrangement the resulting interferometer phase which is cubic in time is independent of the initial state of the center - of - mass motion . in section [ sec:3 ] we introduce our interferometer capable of measuring the cubic phase accumulated by a particle during its motion in a linear potential provided by a constant gravitational field and magnetic field gradient . the three - level atom probing these fields has two ground states corresponding to two different magnetic quantum numbers and experiences a sequence of four raman pulses . in order to close the interferometer in position _ and _ velocity , we choose the separation of @xmath11 between the pulses . the resulting probability for the atom to exit the interferometer in one of the two ground states is the familiar oscillatory function , which is then independent of the initial wave function . however , in contrast to standard interferometers the argument now depends on the phase cubic in @xmath0 and the discrete third derivative of the laser phase . we dedicate section [ sec:4 ] to a comparison of the present scheme to the kasevich - chu interferometer @xcite and distinguish the cubic phase shift from the ones caused by a gravity gradient or the continuous - acceleration bloch ( cab ) technique @xcite . in section [ sec:5 ] we discuss a possible experimental implementation of our proposal . we conclude in section [ sec:6 ] by briefly summarizing our results and providing an outlook . in order to keep our article self - contained but focused on the central ideas , we include lengthy calculations in five appendices . in appendix [ sec : a ] we recall the technique of creating coherent superpositions of the two ground states and interchanging their populations using raman pulses . we then turn in appendix [ sec : b ] to a description of our interferometer as a sequence of unitary operators . in appendix [ sec : c ] we derive the conditions to close our interferometer and obtain in appendix [ sec : d ] an explicit expression for the interferometer phase induced by a linear potential . finally in appendix [ sec : e ] we make contact with the formalism of ref . @xcite developed for a rather general form of interferometer and rederive the phase of our @xmath2-interferometer . this approach brings out again that in contrast to the corresponding global phase this phase is independent of the initial state of the center - of - mass motion . the propagator of a quantum particle experiencing a linear potential is determined by a phase factor governed by the corresponding classical action . since the relevant classical motion involves time in the coordinate and velocity in a quadratic and a linear way , both the kinetic as well as potential energies bring in time quadratically . as a result , the action being the integral over time must contain a term proportional to the cube of time . this cubic phase which is independent of the coordinate is at the center of our interest in the present section . we first recall the essential features of the propagator for the wave function in a linear potential . here we focus especially on this cubic phase . moreover , we note that due to the huygens principle for matter waves the integration over the initial coordinate leads to a dependence of this phase on the initial wave function . although we find this property interesting we emphasize that it is of no importance for the present discussion . indeed , due to the born rule we can not measure the global phase factor of a _ single _ quantum system . however , an interferometric measurement of the difference of two different global phase factors of _ two _ systems is possible . in such an interferometer the cubic phase is independent @xcite of the initial wave function . we start our analysis by discussing the propagator of a particle in a linear potential . here we emphasize especially the emergence of the phase factor cubic in time . we consider a particle of mass @xmath12 moving in a linear potential @xmath13 corresponding to a constant force @xmath14 directed along the @xmath15-axis with the unit vector @xmath16 . this problem occurs for ( i ) a particle , which experiences a constant gravitational acceleration @xmath17 with @xmath18 as indicated in fig . [ fig : g - and - electric]a , and ( ii ) a charge @xmath19 in an ideal capacitor with the constant electric field @xmath20 , for which @xmath21 as shown in fig . [ fig : g - and - electric]b . throughout this article we focus on the example of a linear gravitational potential . the wave function @xmath22 representing the probability amplitude to find the particle at the final position @xmath23 at time @xmath24 is determined by the propagator @xcite @xmath25|z_\mathrm{i}\rangle,\ ] ] where @xmath26 is the value of the wave function at the initial position @xmath27 and time @xmath28 , and @xmath29 and @xmath30 denote the position and momentum operators , respectively . the propagator @xmath31 defined by eq . ( [ propagator ] ) can be cast @xcite in terms of the classical action @xmath32 along the classical trajectory given by @xmath33 and @xmath34 here we have used the lagrange function @xmath35 of a particle in a linear potential . indeed , the representation @xmath36\ ] ] with the normalization @xmath37 brings out most clearly that @xmath31 contains the phase @xmath38 which is independent of the initial and final positions @xmath27 and @xmath23 , and scales with the third power of the time difference @xmath39 , that is the time during which the particle experiences the constant force @xmath40 . the cubic phase @xmath41 in the propagator of a quantum particle moving in a linear potential manifests itself in every wave function exposed to this situation . indeed , since @xmath41 is independent of the initial coordinate @xmath27 it can be factored out of the huygens integral for matter waves , eq . ( [ wave - function ] ) . nevertheless , the integration of the remaining parts of the propagator in combination with the initial wave function can change the numerical factor in the cubic phase . in order to address this feature in more detail we consider the normalized initial wave function @xmath42 in the form of a gaussian of width @xmath43 . when we substitute this expression into the huygens integral of matter waves , eq . ( [ wave - function ] ) , and use the expressions , eqs . ( [ scl ] ) and ( [ propagator - cl ] ) , for the propagator we arrive at the final wave function @xmath44^{1/2 } } \exp\left\{-\frac{\left(z_\mathrm{f}-\frac{f}{2m}t^2\right)^2}{2\left[\delta z(t)\right]^2}+i\beta\right\}\ ] ] with the time - dependent width @xmath45 and phase @xmath46 ^ 2}- \frac{f^2t^3}{6\hbar m}-\frac{1}{2}\arctan\left(\frac{t}{t_{\rm s}}\right).\ ] ] here @xmath47 denotes the spreading time of the wave packet . we combine the term proportional to @xmath48 arising from the square in the second contribution to eq . ( [ wave function phase ] ) with the cubic phase and find the total phase @xmath49 here we have introduced the time - dependent numerical factor @xmath50 depending on the dimensionless ratio @xmath51 of coordinate time and spreading time @xmath52 , which according to eq . ( [ t spreading ] ) is proportional to square of the initial width @xmath43 . for a plane wave we find @xmath53 and thus @xmath54 leading us to @xmath55 however , for an infinitely narrow gaussian with @xmath56 and thus @xmath57 we find @xmath58 so far we have only considered the extreme cases of @xmath51 . only in the domains where @xmath59 is approximately constant do we find a pure cubic phase dependence on @xmath60 . indeed , between the extremes the time dependence is more complicated as expressed in fig . 2 . the propagator @xmath31 defined in eq . ( [ propagator - cl ] ) contains a global phase which is cubic in time . due to the integration over the initial position in the huygens integral , eq . ( [ wave - function ] ) , this phase depends on the initial wave function . these two sentences summarize in a pregnant way our results of the preceding section . we now briefly outline our interferometric measurement strategy for this phase . in the course of our analysis we shall find that the phase in this interferometer is independent of the initial state . obviously a setup providing us only with the probability density @xmath61 is insensitive to any global phase like the @xmath2-phase . therefore , we need to involve an interferometric measurement either with a path - dependent mass of the particle , or a path - dependent strength of the constant force . throughout this article we focus on the second alternative although we can imagine possibilities to utilize the dependence of the mass on the internal state . key elements of our technique are : ( i ) an atom with both a magnetically sensitive and a magnetically insensitive internal states and ( ii ) an external time - independent magnetic field with a constant gradient along one direction . due to the zeeman effect , such an atom experiences a constant force determined by its internal state . it is the same force that acts on a classical magnetic dipole in a non - uniform magnetic field . in this section we introduce the two crucial elements of our @xmath2-interferometer : ( i ) the zeeman shift of the atomic levels induced by the external time - independent magnetic field with a constant gradient along one direction , and ( ii ) the population dynamics of the two resonant states of the atom driven by the raman laser pulses . we consider a three - level atom consisting of the ground state @xmath62 , the state @xmath63 , and the excited state @xmath64 , as indicated in fig . [ fig : scheme ] . here @xmath62 and @xmath63 are chosen such that the mean value of their magnetic moment differs , which leads to a state - dependent acceleration . without loss of generality , we simplify the following calculations by selecting a magnetic insensitive state for @xmath62 . the center - of - mass motion of the atom is assumed to be along the @xmath15-axis , which is the direction of the constant gravitational acceleration . the interaction of the atom with a time - independent magnetic field having locally the form @xmath65 results in the linear zeeman shift @xmath66 of the energy of @xmath67 , and no zeeman shift @xmath68 of the energy of @xmath69 . here @xmath70 , @xmath71 , @xmath72 , and @xmath73 denote the bohr magneton , the land@xmath74 @xmath17-factor of @xmath63 , and the magnetic quantum numbers associated with the @xmath15-component of the angular momentum for @xmath62 and @xmath63 , which are given by @xmath75 and @xmath76 , respectively . when we take into account the linear gravitational potential @xmath77 we arrive at the hamiltonian @xmath78 \otimes\mathds{1}_z,\end{split}\ ] ] of the three - level atom , where @xmath79 and @xmath80 are the identity operators corresponding to the hilbert space of the internal atomic states , and the center - of - mass motion along the @xmath15-axis , respectively . here @xmath81 , @xmath82 , and @xmath83 are the energies of the atom in the internal states @xmath64 , @xmath62 , and @xmath63 , respectively , when the atom is not exposed to any external field . the homogeneous magnetic field @xmath84 leads to an energy shift @xmath85 and @xmath86 denote the accelerations of the atomic center - of - mass corresponding to @xmath62 and @xmath63 , respectively . thus , the zeeman effect for @xmath62 and @xmath63 in the magnetic field of constant gradient gives rise to the two different accelerations @xmath87 and @xmath88 . we are now in the position to present our @xmath2-interferometer for atoms capable of measuring the cubic phase . the general scheme of this particular atom interferometer depicted in fig . [ fig : scheme ] consists of two distinct `` building blocks '' : ( i ) four raman pulses , that is two @xmath89- and two @xmath90-pulses , which form a @xmath91 sequence , and ( ii ) three regions of the atomic center - of - mass motion with constant accelerations @xmath4 and @xmath5 . each block is described in terms of an appropriate unitary operator . as discussed in more detail in appendix [ sec : a ] , the atom - light interaction is accounted for by the interaction operator @xmath92 where @xmath93 is the laser phase and @xmath94 denotes the total pulse area . moreover , we have assumed that each raman pulse consists of two co - propagating laser beams with nearly equal wavelengths which makes @xmath95 given by eq . independent of @xmath15 . on the other hand , the operator @xmath96\ ] ] with @xmath97 or @xmath98 given by eq . ( [ a1-a2 ] ) provides us with the center - of - mass motion . in appendices [ sec : b ] and [ sec : c ] we analyse the interferometer of fig . [ fig : scheme ] as a sequence of these unitary operators and find the expression @xmath99\ ] ] for the probability of observing the atoms in the excited state @xmath63 after the action of the four raman pulses . according to appendix [ sec : d ] the interferometer phase @xmath100 reads @xmath101 and in appendix [ sec : b ] we derive the expression @xmath102 for the total laser phase . we emphasize that this result is independent of the initial state of the center - of - mass motion . this property of the _ interferometer phase _ is in sharp contrast to the dependence of global phase of sec . [ subsec:2 - 2 ] corresponding to a _ single _ degree of freedom . as shown in ref . @xcite it is a consequence of the fact that our interferometer is closed in both position _ and _ velocity . indeed , in appendix [ sec : e ] we use the expression for the interferometer phase of a rather general interferometer obtained in ref . @xcite to rederive eq . ( [ interferometer - probability - interferometer phase ] ) . we conclude by recalling eq . to obtain the explicit dependence of @xmath100 on the magnetic field gradient and the gravitational acceleration and cast eq . into the form @xmath103 for a known magnetic field gradient the gravitational acceleration can then be measured with a cubic scaling in @xmath0 . by using different zeeman sub - levels for the state @xmath63 it is even possible to measure both , the magnetic field gradient @xmath104 _ and _ the gravitational acceleration @xmath17 . a simultaneous measurement of these quantities has already been demonstrated for a bragg interferometer with an interferometer phase scaling as @xmath3 @xcite . in the preceding section we have derived an expression for the probability of finding the atom in the state @xmath63 at one exit of our interferometer . in the present section we compare and contrast our device with the kasevich - chu interferometer , and make contact with other cubic phases such as the ones caused by a gravity gradient , or arising in the cab technique . in the kasevich - chu interferometer the probability corresponding to eq . reads @xcite @xmath105\ ] ] with the interferometer phase @xmath106 and the total laser phase @xmath107 we note four major differences between our scheme and that of kasevich - chu : ( i ) the _ four _ rather than the _ three _ raman pulses create the _ sum _ of the two terms appearing in the square brackets of @xmath108 given by eq . ( [ interferometer - probability - result ] ) rather than the _ difference _ in @xmath109 defined by eq . ( [ interferometer - probability - result - kc ] ) . in the absence of any potentials the different pulse sequences can be visualized by a @xmath110-rotation on the bloch sphere in the case of the kasevich - chu interferometer , and a @xmath111-rotation for our @xmath2-interferometer giving rise to the opposite signs . ( ii ) the phase @xmath100 following from eq . ( [ interferometer - probability - interferometer phase ] ) and induced by the linear potentials depends on the separation @xmath0 of the pulses in a _ cubic _ rather than _ quadratic _ way as in @xmath112 expressed by eq . . ( iii ) _ co - propagating _ laser beams together with a constant magnetic field gradient lead to a proportionality of @xmath100 to @xmath113 , while in the case of kasevich - chu the use of _ counter - propagating _ laser beams results in a momentum transfer of @xmath114 which reflects itself in @xmath112 . ( iv ) the total laser phase @xmath115 defined by eq . ( [ interferometer - probability - laser - phase ] ) is a discrete _ third _ derivative rather than the _ second _ one for @xmath116 given by eq . ( [ interferometer - probability - laser - phase - kc ] ) . indeed , this feature becomes obvious when we consider the limit of @xmath117 , for which @xmath118 while @xcite @xmath119 , with @xmath120 and @xmath121 being the third and second continuous derivatives of the phase @xmath122 of the raman pulse , respectively . we now compare and contrast the @xmath2-phase in our interferometer induced by the propagator of a particle in a linear potential to other phases cubic in time . here we focus on two different situations : ( i ) the presence of a gravity gradient , or ( ii ) the application of the cab technique . in the presence of a gravity gradient @xmath123 a phase cubic in time appears in the kasevich - chu interferometer as a consequence of a _ single quadratic _ potential @xmath124 rather than _ two linear _ potentials . in particular , it results from an expansion of the atomic center - of - mass motion in the limit of a weak gravity gradient , that is @xmath125 . indeed , a gravity gradient leads to a _ position - dependent _ acceleration , while the two linear potentials in our interferometer lead to a _ state - dependent _ acceleration . furthermore , we emphasize , that the kasevich - chu interferometer is no longer closed in position _ and _ velocity in the presence of a gravity gradient . this deficiency leads to a loss of contrast @xcite and a dependence of the phase shift on the initial state . the interferometer can be closed ( at least approximately ) by additional laser pulses @xcite or by suitably adjusting the laser wavelength of the intermediate pulse @xcite . however , this procedure also eliminates the cubic contributions to the phase shift , in contrast to the situation considered here . our @xmath2-interferometer shares the underlying idea of cab , that is applying different constant accelerations along each interferometer arm . however , instead of achieving these accelerations via state - dependent linear potentials , a beam splitter based on bragg - diffraction is used to load one of the two exit ports into an optical lattice , which is accelerated subsequently by the use of bloch oscillations . while in our scheme we can close the interferometer easily in position _ and _ velocity by simply choosing the correct timing between the pulses , the cab scheme requires a sophisticated control of the acceleration of the optical lattice . moreover , we emphasize that in contrast to the @xmath2-interferometer , in the cab scheme not only a phase proportional to @xmath2 , but also one proportional to @xmath3 emerges . in this section we discuss a possible experimental implementation of our proposal for a @xmath2-interferometer based on our current laboratory apparatus depicted in fig . [ fig : experimental setup ] . we first summarize the key features of our setup , describe our method to deduce the magnetic fields from the observed raman spectra , and conclude by briefly analysing the present limitation of our device due to decoherence . we use the d2 transition of @xmath7 and choose @xmath69 and @xmath67 from the @xmath126 and @xmath127 hyperfine state manifolds with a frequency separation of approximately @xmath128 @xcite . the atoms are loaded into a three - dimensional magneto - optical trap ( 3d mot ) emerging from a two - dimensional trap ( 2d mot ) as shown in fig . [ fig : experimental setup ] . the 3d mot consists of standard cooling and repump beams as well as magnetic coils . we rely on an all - glass chamber as our vacuum system . after @xmath129 of loading to obtain a sufficient signal - to - noise ratio , the atoms are launched upwards along the @xmath15-axis in a moving optical molasses configuration with a velocity of approximately @xmath130 such that they strike the top of a @xmath131 tall glass tower . it takes the atoms between @xmath132 and @xmath133 to reach this point depending on the launch velocity which can be adjusted by the voltage of the launch signal . we emphasize that the top of the tower does not coincide with the apex of the trajectory . after launch the atoms are freely moving in the dark and we are able to apply a single or several raman pulses involving two co - propagating laser beams along the @xmath15-axis with the same circular polarization . during their motion the atoms interact with a magnetic field which varies linearly along the @xmath15-axis due to coils in an anti - helmholtz configuration which are not the ones used to trap the atoms . moreover , they feel the field of a solenoid of finite length not depicted in fig . [ fig : experimental setup ] surrounding the glass tower . the ability to change the current independently in each of the gradient coils and the solenoid provides us with the control of the location of the zero crossing of the magnetic field , or an effective way to adjust the bias field . on the top of the tower a photomultiplier tube ( pmt ) detector performs a projective measurement of the population in the state @xmath67 by collecting the fluorescence emitted by the atoms caused by the vertical trapping beams , that is the @xmath134 and @xmath135 beams , which are switched back on for the measurement . observing the atomic population always at the top of the tower provides us with a convenient way of varying the position where the raman pulse is applied to the atoms along their flight path . with this setup we can map out the magnetic field along the atom trajectory by measuring raman spectra of the type shown in fig . [ fig : raman spectrum ] . for this purpose we first use an optical pumping stage to transfer all atoms to the ground state @xmath69 , that is the state with @xmath126 . then we apply a raman light pulse whose intensity and duration are chosen to be close to a zero - detuning @xmath90-pulse . we have found that this condition is satisfied for a pulse duration of @xmath136 , using our typical total raman power of approximately @xmath137 in a beam of diameter @xmath138 and a single photon detuning of @xmath139 . finally , we observe the number of atoms transferred from @xmath69 to @xmath67 , that is the state with @xmath127 . in this manner we obtain raman spectra such as the one presented in fig . [ fig : raman spectrum ] , that is the population of @xmath63 versus the two - photon detuning @xmath140 of the raman pulse of the frequencies @xmath141 and @xmath142 with respect to the `` clock '' transition @xmath143 , for which the resonance occurs at @xmath144 . here @xmath145 denotes the energy of the hyperfine state . for a magnetic field pointing in an arbitrary direction , the observed raman spectrum displays up to 11 peaks ; that is 5 peaks for @xmath146 transitions and 6 peaks for @xmath147 transitions , where @xmath148 is the change of the magnetic quantum number . for a magnetic field directed along the @xmath15-axis as required for the @xmath2-experiment , transitions with @xmath147 are suppressed in our experimental setup @xcite . the position of the relative two - photon resonance in the raman spectrum corresponding to the magnetically sensitive transition @xmath149 is determined by the first - order zeeman shift @xmath150 in frequency , where @xmath151 is the value of the magnetic field at the center - of - mass coordinate @xmath15 of the atom cloud during the interaction with the raman pulse . using the @xmath152 transition , that is @xmath153 , as indicated in fig . [ fig : raman spectrum ] , we determine from the zeeman shift , eq . ( [ detuning shift ] ) , the magnetic field @xmath154 where we have used the fact @xcite that @xmath155 and @xmath156 . it is the opposite signs of @xmath157 and @xmath158 that result in the non - degenerate spectrum . by measuring the resonance frequencies corresponding to the clock and first ( or second ) peak in the raman spectrum , we automatically correct for any possible drift in the ac stark shift caused by a drift in the intensity or frequency of the raman fields . we map out the magnetic field experienced by the atoms as a function of their location within the tower , by repeated launching them and applying a raman pulse at different times after their launch . the corresponding location @xmath15 of the atom cloud is determined beforehand by time - of - flight photography . figure [ fig : magnetic field ] presents the measured magnetic field with a gradient of approximately @xmath159 for a current of @xmath160 in the gradient coils and @xmath161 in the solenoid . the theoretical proposal for a @xmath2-interferometer presented in the preceding sections assumes that the raman pulses applied to the atoms are perfect @xmath89- and @xmath90-pulses . by definition , such pulses can occur only when the raman fields are in a two - photon resonance . since this resonance shifts as the atoms travel up the tower , a rapid detuning of the relative frequency of the raman fields is necessary . we have been able to achieve this task using a combination of a high - frequency acousto - optic modulator and digital frequency synthesizer , thereby keeping the atoms in resonance during their flight @xcite . however , a more severe restriction is decoherence . in order to observe the @xmath2-phase for a magnetic field gradient of @xmath159 , a time @xmath0 between pulses of about @xmath162 is required . moreover , the decoherence time should be larger than @xmath163 , that is @xmath164 . if we use the clock transition there is little to no decoherence on this time scale . however , if we use a transition involving magnetically sensitive states , our signals decohere on the order of @xmath165 . we are presently investigating the source of this decoherence . in the present article we have proposed an atom interferometer that is sensitive to the quantum mechanical @xmath2-phase emerging in the propagator of a particle in a linear potential . for this purpose we have considered an atom with a magnetic sensitive , and an insensitive internal state being exposed to a constant gravitational field and a magnetic field of constant gradient . by applying a sequence of four co - propagating raman pulses , the atom interferometer can be closed in position _ and _ velocity . the resulting interferometer phase @xmath100 displays the cubic scaling in @xmath0 but also depends on the gravitational acceleration and the magnetic field gradient . we have compared and contrasted this cubic term to the one appearing in the phase of the kasevich - chu interferometer in the presence of a gravity gradient , and to the one obtained by using the cab technique . furthermore , we have outlined a possible experimental realization of our interferometer , which can be used as a gravimeter and magnetic gradiometer . cubic phases appear frequently in quantum physics and give rise to mind - boggling effects . for example , the energy wave function of a linear potential is given by the airy function @xcite whose standard integral representation involves a cubic phase . this term emerges from the eigenvalue equation in momentum space due to the integration of the kinetic energy which is quadratic in momentum . when we suddenly turn - off the potential the so - created airy wave packet accelerates and its probability density keeps its shape @xcite during the free propagation . deeper insight @xcite into this surprising phenomenon springs from wigner phase space @xcite and the fact that the wigner function of the airy wave packet is again an airy function . closely related to the cubic phase in the airy integral and the dispersionless free propagation of an airy wave packet is the oscillatory probability density created by a point source @xcite located in a linear potential and continuously emitting particles into all three space directions with an identical speed . these oscillations appearing in the plane orthogonal to the gravitational force are a consequence of the interference between two classical trajectories of different inclinations . the knowledge of the two distinct paths encoded in the different arrival times is erased by the continuous stream of particles . again the origin of this particular interference pattern can be traced back to the cubic phase in the green s function . due to the analogy between the constant gravitational field and the constant electric field between two plates of a capacitor discussed in the beginning of this article one might wonder if it is possible to construct a similar charged particle fountain . indeed , in the case of electrons in a uniform electric field such type of fountain has already been realized in photoionization and photodetachment microscopes @xcite . it would be fascinating to illuminate the similarities and differences between these three examples of cubic phases and our @xmath2-interferometer . unfortunately this task goes beyond the scope of the present article and has to be postponed to future publications . in this appendix we describe the population dynamics @xcite of the two resonant atomic states driven by the raman laser pulses . for this purpose we consider the interaction between a three - level atom and two laser pulses of the form @xmath166 and @xmath167 where @xmath168 , @xmath169 , @xmath170 , and @xmath171 with @xmath172 denote the time - dependent envelope , frequency , wave vector , and phase of the @xmath173-th field , respectively . the laser frequencies @xmath141 and @xmath142 are assumed to only drive the transitions @xmath174 and @xmath175 , respectively . moreover , we assume that the laser pulses are so short that the atom does not move significantly during the interaction . therefore , the position of the center - of - mass of the atom is considered to be fixed during the laser pulses . within the rotating - wave approximation @xcite and in the limit of far - detuned laser pulses with identical detunings , that is when the rabi frequencies @xmath176 of the transitions @xmath177 are much smaller than the detuning @xmath178 of the two laser pulses , @xmath179 , and @xmath180 , we can eliminate the excited state @xmath64 and neglect the stark shifts @xmath181 . the resulting effective hamiltonian @xcite @xmath182}|g_1\rangle\langle g_2|\right . \\ & \left.+ e^{-i[\delta k z+\phi_\mathrm{l}]}|g_2\rangle\langle g_1|\right)\end{split}\ ] ] describes the transitions between the states @xmath62 and @xmath63 due to the raman pulses . here @xmath183 and @xmath184 are the dipole - moment matrix element and the frequency of the transition @xmath177 , respectively , with @xmath185 and the slowly varying laser phase @xmath186 . to avoid a momentum transfer during the raman transitions , we assume that the laser pulses propagate in the same directions along the @xmath15-axis , eq . ( [ electric field ] ) , and that the difference @xmath187 of the two wave vectors is small compared to the size @xmath188 of the atomic wave packet , that is @xmath189 . in this case the dependence on @xmath15 in eq . ( [ atom - light - hamiltonian ] ) can be neglected and we arrive at @xmath190 the interaction of the atom with the two far - detuned raman pulses , corresponding to eq . ( [ electric field ] ) , during the time interval @xmath191 , and with @xmath192 , is given by the evolution operator @xcite @xmath193 which can be expressed as @xmath194 where @xmath195 denotes the total pulse area . the case @xmath196 , which is a @xmath89-pulse , gives rise @xcite to the coherent superpositions @xmath197 and @xmath198 in contrast , the case @xmath199 , known as a @xmath90-pulse describes an exchange @xmath200 and @xmath201 of the level populations . unitary operators describe both the interaction of the atom with the four raman pulses and the time evolution associated with the center - of - mass motion . in the present appendix we derive the complete quantum state of the atom consisting of the internal states as well as the center - of - mass in the two exit ports of our interferometer following the procedure outlined in refs . @xcite . the dynamics in our interferometer consists of the following steps : \1 . before the first @xmath89-pulse , at @xmath202 , the initial state @xmath203 consists of the center - of - mass motion @xmath204 and the internal state @xmath62 . here and throughout this appendix @xmath205 is an infinitesimally small and positive number . \2 . after the first @xmath89-pulse at @xmath206 , the state reads @xmath207 where we have used eq . ( [ beam - splitter ] ) . \3 . before the first @xmath90-pulse at @xmath208 , we find @xmath209 \4 . after the first @xmath90-pulse at @xmath210 , we obtain @xmath211}|g_1\rangle\hat{u}_{a_2}(t_1,t_0)|\psi_0\rangle . \end{split}\ ] ] \5 . before the second @xmath90-pulse at @xmath212 , the state takes the form @xmath213}|g_1\rangle\hat{u}_{a_1}(t_2,t_1)\hat{u}_{a_2}(t_1,t_0)|\psi_0\rangle.\end{split}\ ] ] \6 . after the second @xmath90-pulse at @xmath214 , we arrive at the state @xmath215}|g_1\rangle\hat{u}_{a_2}(t_2,t_1)\hat{u}_{a_1}(t_1,0)|\psi_0\rangle \\&+\frac{i}{\sqrt{2}}e^{-i[\phi_\mathrm{l}(t_0)-\phi_\mathrm{l}(t_1)+\phi_\mathrm{l}(t_2)]}\\ & \times|g_2\rangle\hat{u}_{a_1}(t_2,t_1)\hat{u}_{a_2}(t_1,t_0)|\psi_0\rangle . \end{split}\ ] ] \7 . before the second @xmath89-pulse at @xmath216 , the state reads @xmath217}\\\ & \times|g_1\rangle\hat{u}_{a_1}(t_3,t_2)\hat{u}_{a_2}(t_2,t_1)\hat{u}_{a_1}(t_1,t_0)|\psi_0\rangle\\ & + \frac{i}{\sqrt{2}}e^{-i[\phi_\mathrm{l}(t_0)-\phi_\mathrm{l}(t_1)+ \phi_\mathrm{l}(t_2)]}\\ & \times|g_2\rangle\hat{u}_{a_2}(t_3,t_2)\hat{u}_{a_1}(t_2,t_1)\hat{u}_{a_2}(t_1,t_0)|\psi_0\rangle.\end{split}\ ] ] \8 . finally , after the second @xmath89-pulse at @xmath218 , we conclude with the state @xmath219}|g_1\rangle\left(e^{-i\varphi_\mathrm{l}}\hat{u}_{\mathrm{u}}-\hat{u}_{\mathrm{l}}\right)|\psi_0\rangle\\ & + \frac{i}{2}e^{-i[\phi_\mathrm{l}(t_1)-\phi_\mathrm{l}(t_2)+\phi_\mathrm{l}(t_3)]}|g_2\rangle \left(e^{-i\varphi_\mathrm{l}}\hat{u}_{\mathrm{u}}+\hat{u}_{\mathrm{l}}\right)|\psi_0\rangle,\end{split}\ ] ] where @xmath220 and @xmath221 are the unitary evolution operators associated with the center - of - mass motion for the lower and the upper paths of the interferometer shown in fig . [ fig : scheme ] , and @xmath222 is the total phase resulting from the action of the four laser pulses . in the preceding appendix we have derived an expression for the complete quantum state of the atom in the exit ports of the interferometer . here we have allowed arbitrary times for the interactions with the laser pulses . in the present appendix we choose these times in such a way as to maximize the contrast . the probability @xmath223 to observe atoms in the ground state @xmath62 after the action of the four raman pulses at @xmath218 , follows from the quantum state @xmath224 given by eq . ( [ fourth pulse ] ) and contains the state @xmath225 of the center - of - mass motion of atom in @xmath62 . it takes the form @xmath226,\ ] ] where the contrast @xmath227 and the phase @xmath100 of the interferometer are the modulus and the argument of the matrix element @xmath228 we maximize @xmath227 , that is we have @xmath229 , when we close our interferometer . in this case @xmath223 given by eq . ( [ interferometer - probability ] ) is independent of the initial velocity and position of the atom . in order to close the interferometer we have to find the time intervals @xmath230 with @xmath231 between the raman pulses shown in fig . [ fig : scheme ] , such that the final velocities @xmath232 and @xmath233 , as well as the final positions @xmath234 and @xmath235 on the upper and lower paths of the interferometer are identical . indeed , for the velocity we derive the following formulae : \i ) for the upper path @xmath236 \ii ) for the lower path @xmath237 as a result , the interferometer is closed in _ velocity _ space , if @xmath238 , that is , @xmath239 or , equivalently , @xmath240 as for the position , we obtain the following rather lengthy expressions : \i ) for the upper path @xmath241 \ii ) for the lower path @xmath242 as a result , the interferometer is closed in _ position _ space if @xmath243 , that is , @xmath244 when we solve the system of the two algebraic equations , ( [ v - condition ] ) and ( [ z - condition ] ) , for @xmath245 and @xmath246 in terms of @xmath247 , we obtain @xmath248 hence , in order to close the interferometer , the four raman pulses must be separated in time by @xmath0 , @xmath10 , and @xmath0 as indicated in fig . [ fig : scheme ] . in the preceding appendix we have used classical trajectories to find the separation @xmath11 between the four raman pulses leading to a closed interferometer . we now show that in this case the product @xmath249 of the evolution operators @xmath250 and @xmath251 defined by eq . ( [ u_l - u_u ] ) is proportional @xcite to the identity operator , that is @xmath252 where @xmath253 is the interferometer phase . therefore , a normalized state @xmath204 leads by virtue of eq . ( [ c - phi_i ] ) to a perfect contrast , that is @xmath229 , indicating that the interferometer is independent of @xmath204 . moreover , this calculation provides us with an explicit expression for @xmath253 . in order to evaluate the evolution operator @xmath254 for the lower path of our interferometer , shown in fig . [ fig : scheme ] , we use the baker - campbell - hausdorff and zassenhaus formulas @xcite to represent the operator @xmath255 given by eq . ( [ operator - u - gravity ] ) in the form of a product @xmath256 consisting of a phase factor , the displacement operator @xmath257,\ ] ] and the unitary operator @xmath258 of a free particle . the decomposition , eq . ( [ bch ] ) , allows us to rewrite eq . ( [ u_l product ] ) as @xmath259 \hat{\mathcal{d}}\left(\frac{1}{2}a_1 t^2,ma_1t\right)\hat{u}_{0}(t)\ ] ] @xmath260 with the help of the commutation relation @xmath261 and the addition identity @xmath262 for the operators @xmath263 and @xmath264 given by eqs . ( [ bch - displacement operator ] ) and ( [ bch - free evolution ] ) , we can shift all free - evolution operators @xmath265 in eq . ( [ u_l product total ] ) to the right and we arrive at @xmath259 \hat{\mathcal{d}}\left(\frac{1}{2}a_1 t^2,ma_1t\right)\ ] ] @xmath266 or @xmath267\ ] ] @xmath268 in the last step we have made use the addition identity @xmath269 with @xmath270 to combine all three displacement operators into a single one . since the evolution operator @xmath271 defined by eq . ( [ u_l - u_u ] ) for the upper path of our interferometer follows directly from the operator @xmath272 given by eq . ( [ u_l product final ] ) for the lower path by an exchange of the accelerations @xmath4 and @xmath5 , we arrive at @xmath273\ ] ] @xmath274 when we substitute eqs . ( [ u_l product final ] ) and ( [ u_u product final ] ) into the left - hand side of eq . ( [ interferometer phase - operator product ] ) and use the property that the operators @xmath263 and @xmath275 are unitary , the interferometer phase reads @xmath276 hence , @xmath253 is independent of the initial position @xmath277 and velocity @xmath278 as well as of the initial state . moreover , it scales with the third power of the time interval @xmath279 between the first and the second raman pulses . in this appendix we provide an alternative derivation of the phase shift for our @xmath2-interferometer by making use of the general results obtained in ref . @xcite . here both open and closed atom interferometers with branch - dependent forces were investigated . we first summarize the phase accumulated along a single trajectory and then consider the interferometer , that is we analyze the phase difference between two different paths . the framework outlined in ref . @xcite describes light - pulse atom interferometers where the time evolution of the center - of - mass motion between laser pulses is governed by a general quadratic hamiltonian and the time dependence of the corresponding quantum state along each interferometer branch is given by @xmath280 here @xmath281 denotes a _ centered _ state with vanishing expectation values for the position and momentum operators and evolves according to the purely quadratic part of the hamiltonian . it is convenient to employ a vector notation for phase - space quantities and the argument of the displacement operator @xmath263 is a displacement vector @xmath282 that corresponds to the classical phase - space trajectories @xmath283 and @xmath284 associated with the hamiltonian including the momentum kicks from the laser pulses along that branch . its value @xmath285 at the initial time @xmath286 coincides with the expectation values of the position and momentum operators for the initial state of the interferometer . the time - dependent phase @xmath287 reads @xmath288 j \boldsymbol{\chi}(t')+2v_0 ( t')\right\},\ ] ] where @xmath289 is the sum of the various phases for the laser pulses relevant to this branch and @xmath290 accounts for their momentum kicks . the quantity @xmath291 is a consequence of the linear terms in the hamiltonian and reduces for the case of a constant force considered in the present article to @xmath292 in addition , we have introduced the symplectic form @xmath293 finally the term with @xmath294 contains the contributions from the internal state energies and can also include the effect of uniform magnetic fields which can even be time - dependent due to the zeeman effect . since a solution @xmath295 of the classical equations of motion in phase space consists of a homogeneous solution containing the information on the initial conditions plus an inhomogeneous solution that accounts for the linear terms and the kicks from the laser pulses , @xmath291 and @xmath290 , we find @xmath296 here the transition matrix @xmath297 satisfies the homogeneous part of the equations of motion with the initial condition @xmath298 . moreover , we have employed the retarded propagator @xmath299 and introduced the notation @xmath300 when we restrict ourselves to the familiar kinetic term associated with the free evolution due to the purely quadratic part of the hamiltonian , the transition matrix simplifies to @xmath301 the oscillations in the number of atoms detected at each exit port as a result of the interference between the upper and lower branches of the interferometer is determined by the phase shift @xmath302 the last term arises only in _ open _ interferometers , where the central position and momentum of the two interfering wave packets do not coincide , that is @xmath303 . for the case of no momentum transfer from the laser pulses , that is @xmath304 , the phase difference @xmath305 reduces to @xmath306 here a bar over a quantity @xmath307 denotes the average of its values in the two branches , that is @xmath308 , and similarly their difference is @xmath309 in all cases except for @xmath310 , which is defined otherwise above . the quantity @xmath311 corresponds to the relative displacement between the two interfering wave packets at the exit port . since our @xmath2-interferometer is closed , the third term on the right - hand side of eq . vanishes . moreover , the atoms spend the same times in each internal state on both branches . as a consequence the second term on the right - hand side of eq . vanishes as well . therefore , when we substitute eqs . and into eq . , we arrive at @xmath312 where we have taken into account that @xmath313 in our case . between the first and the forth pulse , the internal states and the accelerations experienced by the atoms on the two branches are exchanged every time a pulse is applied . as a result , we find @xmath314 and @xmath315 with @xmath316 since @xmath317 is time independent we can immediately integrate over @xmath318 and obtain @xmath319 or @xmath320 which agrees with the result of appendix [ sec : d ] . we are grateful to e. giese , m. a. kasevich , s. kleinert , h. mller , g. welch , and w. zeller for many fruitful discussions on this topic . moreover , we thank n. ashby for pointing out ref . @xcite to us . this work is supported by dip , the german - israeli project cooperation , as well as the german space agency ( dlr ) with funds provided by the federal ministry for economic affairs and energy ( bmwi ) due to an enactment of the german bundestag under grants no . dlr 50wm1152 - 1157 ( quantus - iv ) and the centre for quantum engineering and space - time research quest . we appreciate the funding by the german research foundation ( dfg ) in the framework of the sfb / trr-21 . w.p.s . is grateful to texas a@xmath321 m university for a texas a@xmath321 m university institute for advanced study ( tias ) faculty fellowship . s.a.d . , a.s . , and f.a.n . gratefully acknowledge funding from the office of naval research and a grant from the naval air systems command chief technology office . s. fray , c. alvarez diez , th.w . hnsch , and m. weitz , phys . lett . * 93 * , 240404 ( 2004 ) e.h . kennard , zeitschrift fr physik * 44 , * 326 ( 1927 ) e.h . kennard , jour . * 207 , * 47 ( 1929 ) w.p . schleich , d.m . greenberger , d.h . kobe , and m.o . scully , proc . sci . * 110 * , 5374 ( 2013 ) , edited by p.r . berman ( academic press , san diego , 1997 ) a.d . cronin , j. schmiedmayer , and d.e . pritchard , rev . * 81 , * 1051 ( 2009 ) proceedings of the international school of physics `` enrico fermi '' course 188 `` atom interferometry '' , edited by g.m . tino and m.a . kasevich ( ios , amsterdam , 2014 ) c. misner , k. thorne , and j.a . wheeler , _ gravitation _ ( w. h. freeman , san francisco , 1973 ) d. bohm , _ quantum theory _ ( prentice hall , englewood cliffs , 1951 ) n. bohr and l. rosenfeld , kgl . danske videnskab . selskab , mat .- fys . medd . * 12 * , no . 8 ( 1933 ) ; n. bohr and l. rosenfeld , phys . rev . * 78 * , 794 ( 1950 ) h. salecker and e.p . wigner , phys . rev . * 109 * , 571 ( 1958 ) j.a . wheeler , in proceedings of the international school of physics `` enrico fermi '' course 72 `` problems in the foundations of physics '' ( north - holland , amsterdam , 1979 ) , edited by c.m . dewitt and d. rickles ( wright air development center , 1957 ) . these proceedings have been reprinted under the title _ the role of gravitation in physics _ by the max - planck research library for the history and development of knowledge . greenberger , j. math . 11 * , 2329 ( 1970 ) ; _ ibid . _ * 11 * , 2341 ( 1970 ) r. colella , a.w . overhauser , and s.a . werner , phys . * 34 , * 1472 ( 1975 ) h. rauch and s.a . werner , _ neutron interferometry _ ( new york , oxford university press , 2 edition , 2015 ) j .- l . staudenmann , s.a . werner , r. colella , and a.w . overhauser , phys . rev . a * 21 , * 1419 ( 1980 ) ; k.c . littrell , b.e . allman , and s.a . werner , phys . rev a * 56 * , 1767 ( 1997 ) k. hornberger , s. gerlich , p. haslinger , s. nimmrichter , and m. arndt , rev . * 84 * , 157 ( 2012 ) e.a . cornell and c.e . wieman , rev . * 74 * , 875 ( 2002 ) ; w. ketterle , rev . mod . phys . * 74 * , 1131 ( 2002 ) d. schlippert , j. hartwig , h. albers , l.l . richardson , c. schubert , a. roura , w.p . schleich , w. ertmer , and e.m . rasel , phys . rev . lett . * 112 * , 203002 ( 2014 ) s. dimopoulos , p.w . graham , j.m . hogan , and m.a . kasevich , phys . d * 78 * , 042003 ( 2008 ) i. pikovski , m. zych , f. costa , and @xmath322 . brukner , nature physics * 11 * , 668 ( 2015 ) m. khl , th.w . hnsch , and t. esslinger , phys . * 87 * , 160404 ( 2001 ) c.w . chou , d.b . hume , t. rosenband , d.j . wineland , science * 329 * , 1630 ( 2010 ) a. matveev , ch.g . parthey , k. predehl , j. alnis , a. beyer , r. holzwarth , th . udem , t. wilken , n. kolachevsky , m. abgrall , d. rovera , ch . salomon , ph . laurent , g. grosche , o. terra , th . legero , h. schnatz , s. weyers , b. altschul , and th.w . hnsch , phys . 110 * , 230801 ( 2013 ) t. jenke , p. geltenbort , h. lemmel , and h. abele , nature physics * 7 * , 468 ( 2011 ) ; h. abele and h. leeb , new j. phys . * 14 * , 055010 ( 2012 ) t. jenke , g. cronenberg , j. burgdrfer , l.a . chizhova , p. geltenbort , a.n . ivanov , t. lauer , t. lins , s. rotter , h. saul , u. schmidt , and h. abele , phys . lett . * 112 * , 151105 ( 2014 ) c. kiefer , _ quantum gravity _ ( oxford university press , oxford , 2007 ) s. abend , m. gebbe , m. gersemann , h. ahlers , h. mntinga , e. giese , n. gaaloul , c. schubert , c. lmmerzahl , w. ertmer , w.p . schleich , and e.m . rasel , submitted ( 2016 ) m. a. kasevich and s. chu , phys . * 67 , * 181 ( 1991 ) m. a. kasevich and s. chu , appl . b * 54 , * 321 ( 1992 ) e. giese , w. zeller , s. kleinert , m. meister , v. tamma , a. roura , w.p . schleich , proceedings of the international school of physics `` enrico fermi '' course 188 `` atom interferometry '' , edited by g.m . tino and m.a . kasevich ( ios , amsterdam , 2014 ) w.p . schleich , d.m . greenberger , and e.m . rasel , new j. phys . * 15 , * 013007 ( 2013 ) e. kajari , n.l . harshman , e.m . rasel , s. stenholm , g. sssmann , and w. p. schleich , appl . b * 100 , * 43 ( 2010 ) g.d . mcdonald , c.c.n . kuhn , s. bennetts , j.e . debs , k.s . hardman , j.d . close and n.p . robins , epl * 105 , * 63001 2014 a. roura , w. zeller , and w.p . schleich , new . j. phys . * 16 , * 123012 ( 2014 ) r.p . feynman and a.r . hibbs , _ quantum mechanics and path integrals _ ( new york , mcgraw - hill , 1965 ) for a different proposal to measure this cubic phase using electron optics , see for example f. fratini and l. safari , physica scripta * 89 , * 085004 ( 2014 ) k.s . hardman , p.j . everitt , g.d . mcdonald , p. manju , p.b . wigley , m.a . sooriyabandara , c.c.n . kuhn , j.e . debs , j.d . close , and n.p . robins , arxiv:1603.01967 ( 2016 ) k .- p . marzlin and j. audretsch , phys . a * 53 , * 312 ( 1996 ) a. roura , arxiv:1509.08098 ( 2015 ) s.a . desavage , j.p . davis , and f.a . narducci , j. mod . opt . * 60 , * 95 ( 2013 ) s.a . desavage , k.h . gordon , e.m . clifton , j.p . davis , and f.a . narducci , j. mod . opt . * 58 , * 2028 ( 2011 ) d.a . steck , rubidium 85 d line data , available online at http://steck.us/alkalidata ( revision 2.1.6 , 20 september 2013 ) c.l . adler , r. johnson , a. srinivasan , and f.a . narducci , to be submitted ( 2016 ) g. breit , phys . * 32 , * 273 ( 1928 ) m.v . berry and n.l . balazs , am . * 47 , * 264 ( 1979 ) w.p . schleich , _ quantum optics in phase space _ ( weinheim , wiley - vch , 2001 ) m.v . berry , j. phys . a * 15 , * l385 ( 1982 ) y.n . demkov , v.d . kondratovich , and v.n . ostrovskii , pisma zh . * 34 , * 425 - 427 ( 1981 ) c. bracher , t. kramer , and m. kleber , phys . a * 67 , * 043601 ( 2003 ) c. blondel , c. delsart , and f. dulieu , phys . * 77 , * 3755 ( 1996 ) s. kleinert , e. kajari , a. roura , w.p . schleich , phys . rep . * 605 , * 1 ( 2015 ) r.m . wilcox , j. math . * 8 , * 962 ( 1967 )

we study microstructure formation in two nonconvex singularly - perturbed variational problems from materials science , one modeling austenite - martensite interfaces in shape - memory alloys , the other one slip structures in the plastic deformation of crystals . for both functionals we determine the scaling of the optimal energy in terms of the parameters of the problem , leading to a characterization of the mesoscopic phase diagram . our results identify the presence of a new phase , which is intermediate between the classical laminar microstructures and branching patterns . the new phase , characterized by partial branching , appears for both problems in the limit of small volume fraction , that is , if one of the variants ( or of the slip systems ) dominates the picture and the volume fraction of the other one is small .

1507.04521

the study of spontaneous pattern formation in materials constitutes an important application of the calculus of variations to materials science . from a variational viewpoint , the origin of microstructure is related to a nonconvexity of the energy density , and to boundary conditions which favour states which correspond to a mixture of different minima of the energy density . in continuum mechanics typically the independent variable is the gradient of a vector field , which obeys the zero - curl differential condition , leading to strong constraints on the admissible microstructures . the theory of relaxation studies the effective behavior of nonconvex variational problems which lack lower semicontinuity and possibly existence of minimizers , but does not give detailed information on the type of microstructure expected @xcite . a finer analysis can be done if a regularization is included , in the form of a convex higher - order term , which physically may represent interfacial energies . an exact determination of the minimizers and the minimal energy is , for these more complex problems , typically impossible . already a study of the optimal scaling of the energy with respect to the parameters of the problem may , however , give very valuable information . starting with the works of landau @xcite on micromagnetism , branching - type patterns have been predicted and observed . they are characterized by coarse oscillations in the interior , which refine close to the boundary , as illustrated in figure [ figlambr ] . at a heuristic level , the transition between coarse and fine oscillations can be understood as the result of the competition between the miminization of the total length of the interfaces , the energetic cost of bending the domains , and the boundary conditions . the mathematical study of the subject began with the work of kohn and mller in the 90s @xcite , who proposed a simple scalar model for martensitic microstructures close to an interface with austenite , see section [ sec : introkm ] below for details . their basic finding was the presence of a transition between a regime in which the energy scales proportional to @xmath0 , with @xmath1 being the surface energy density and @xmath2 the ratio between the austenite and the martensite elastic coefficients , and a regime in which the energy scales proportional to @xmath3 . the first regime corresponds to a one - dimensional laminar pattern , the second one to a branching - type pattern , as illustrated in figure [ figlambr ] . .8 cm their results have been refined in specific regimes @xcite , extended to related scalar - valued models in different parameter regimes @xcite and to vector - valued models @xcite . similar results have been obtained in other variational models , including the magnetic structures in ferromagnets originally studied by landau @xcite , flux - domain patterns in type - i superconductors @xcite , diblock copolymers @xcite , blistering of thin compressed films @xcite , wrinkling of stretched thin films @xcite , dislocation patterns in crystal plasticity @xcite , and compliance minimization @xcite . in this paper we specifically focus on two problems in this class where the expected microstructure is essentially two - dimensional . the first one , and the simpler one , is the kohn - mller functional . the second one is the scalar , but three - dimensional , model of crystal plasticity from @xcite . we present the first model , its physical interpretation and the main results on this functional in section [ sec : introkm ] , the other one in section [ sec : introplastic ] . after mentioning some notation in section [ secnotation ] , we give the proofs for the scaling laws for the two models in sections [ secpfkm ] and [ sec : disloc ] respectively . we now introduce our model for martensitic microstructures . following kohn and mller @xcite we work for simplicity in two dimensions , in antiplane shear geometry , with the scalar field @xmath4 representing a deformation in the out - of - plane direction . we consider one interface between austenite and martensite , located at @xmath5 , with austenite on the right and martensite on the left , as sketched in figure [ fig - geometry ] . in the austenite , which for simplicity is assumed to extend to infinity , the minimum of the elastic energy is attained at @xmath6 . the coefficient @xmath7 represents the ratio of the elastic coefficients of austenite and martensite , respectively ( this parameter is often denoted by @xmath8 in the literature ) . in the martensite , which we assume to cover the domain @xmath9 , there are two minima of the elastic energy density . after scaling , we can write the energy as @xmath10 where admissible functions @xmath11 satisfy @xmath12 almost everywhere in @xmath13 , for some @xmath14}}$ ] . we denote by @xmath15 the distributional derivative , in the last term @xmath16 is assumed to be a measure and the term has to be understood distributionally . the partial derivative @xmath17 represents the order parameter of the martensitic phase transition , and its preferred values are dictated by crystallography . the first two terms in model the elastic energy contributions of the austenite and the martensite part , respectively . the last term in is a regularization term which penalizes changes in the order parameter , and prevents arbitrarily fine microstructures . it can thus be interpreted as a surface energy term , @xmath1 being a typical surface energy constant per unit length . the energy functional is normalized to set the elastic modulus of martensite to one . + mathematically , the parameter @xmath18 represents a compatibility condition . for @xmath19 , there exist trivial configurations with vanishing energy , while for @xmath20 , the minimal energy is strictly positive , and we expect the formation of microstructures . by swapping the two variants of microstructure one can assume without loss of generality that @xmath21 . experimental findings suggest that the size of @xmath18 is closely linked to the width of the thermal hysteresis loop in the austenite - martensite phase transition @xcite , and such low - hysteresis alloys have been found to exhibit peculiar microstructures , see e.g. @xcite . for the symmetric case @xmath22 , the scaling law for the minimal energy has been studied in @xcite . we point out that the same scaling law for holds if one restricts to periodic functions @xmath4 , i.e. , @xmath23 , and if @xmath24 , then the scaling results for @xmath25 can be generalized to the case @xmath26 with the obvious modifications in the lower bound , and the same kinds of test functions to prove the upper bound ( see @xcite ) . in all cases , one observes only scaling regimes that correspond to uniform structures , to laminated structures or to geometrically refining branching patterns ( see @xcite ) . we shall show in this paper that for @xmath27 there is an intermediate regime between laminates and branching , with the geometry sketched in figure [ fig : fig - new ] . a related intermediate regime between the branching constructions and the laminates in case of highly unequal volume fractions has been observed in a three - dimensional model of type - i superconductors @xcite . the scaling of the energy was different , and also the geometry was not the same as here . indeed , in the superconducting case the conservation of flux forces the minority ( normal ) phase to be connected , as in the left panel of figure [ fig : fig - new ] . here a second construction is possible , in which the majority phase is connected , as in the right panel of figure [ fig : fig - new ] , which has a smaller surface energy , as will become clear in lemma [ lem : branch ] below . for the dislocation problem only the second construction gives the optimal energy scaling . we consider the case of general @xmath2 , in particular @xmath27 . it turns out that in this case the choice of boundary conditions at the top and bottom boundaries matter . precisely , we consider two natural classes of admissible functions , the first one with neumann boundary conditions on the horizontal sides , @xmath28 and the one with periodic boundary conditions on the horizontal sides , @xmath29 we prove the following scaling laws for @xmath30 ( see propositions [ prop : kmub ] and [ prop : kmlb ] ) . [ th:1 ] for all @xmath1 , @xmath2 , @xmath31 and all @xmath32 , we have with @xmath33 , @xmath34 and @xmath35 here and in the rest of the paper @xmath36 means that there is a universal constant @xmath37 such that @xmath38 . the resulting phase diagrams are , for @xmath39 , illustrated in figure [ fig : fig2 ] . [ rem : pvsn ] if @xmath40 , the choice of boundary conditions affects the scaling behavior in a non - trivial way . precisely , for periodic boundary conditions we always have , by the constraint on @xmath41 , the lower bound @xmath42 . then , for @xmath43 , by theorem [ th:1 ] , @xmath44 while for @xmath45 , we have the different behavior @xmath46 the origin of the logarithmic correction in case of periodic boundary conditions will be outlined in remark [ rem : singlelam ] . . the regions with energy proportional to @xmath47 corresponds to the branching patterns of figure [ figlambr](b ) , the region marked `` tsb '' is where the new two - scale - branching regime illustrated in figure [ fig : fig - new ] is optimal , the region marked `` lam '' ( very small in the left panel ) is where the laminate pattern illustrated in [ figlambr](a ) is optimal.,title="fig:",height=226 ] . the regions with energy proportional to @xmath47 corresponds to the branching patterns of figure [ figlambr](b ) , the region marked `` tsb '' is where the new two - scale - branching regime illustrated in figure [ fig : fig - new ] is optimal , the region marked `` lam '' ( very small in the left panel ) is where the laminate pattern illustrated in [ figlambr](a ) is optimal.,title="fig:",height=226 ] [ sec : introplastic ] the second functional we study is a model for microstructures in crystal plasticity with large latent hardening that was proposed in @xcite . let @xmath48 describe a grain of a plastic crystal , in which two slip systems are active . in a similar spirit as in the kohn - mller model of section [ sec : introkm ] we focus on a scalar @xmath4 , which represents a component of the deformation , and assume that it is affected by two slip systems , with slip plane normals @xmath49 and @xmath50 . in the limit of large latent hardening the two slip systems can not be both activated at the same point in space , therefore the plastic strain @xmath8 is pointwise parallel to either @xmath51 or @xmath52 . we further assume that each slip system is only active with one orientation . the class of admissible deformations and plastic strains therefore takes the form @xmath53 here and in the rest of the paper , given a vector @xmath54 we denote the components along @xmath55 and @xmath56 by @xmath57 and @xmath58 , and the cartesian ones by @xmath59 , @xmath60 . the energy is given by the sum of the elastic energy and the line - tension energy of the geometrically necessary dislocations . dislocations are topological defects in crystals which are responsible for plastic deformation processes , they are necessarily present whenever the plastic part of the deformation gradient is not a gradient field itself . geometrically necessary dislocations are the dislocations whose presence can be inferred from the fact that the plastic strain @xmath8 is not a gradient field . starting from a variety of microscopic models it has been proven that the energetic contribution of such dislocations is linear in the curl of @xmath8 , with a coefficient which depends on the local burgers vector and slip - plane normal , see for example @xcite . working in the coordinates @xmath61 it is easy to see that the relevant terms are @xmath62 and @xmath63 . a detailed analysis of the latent - hardening condition from a relaxation viewpoint leads to the interpretation of @xmath64 as @xmath65 , where both derivatives are interpreted distributionally , see @xcite for a motivation . in the present setting there is only one burgers vector , since @xmath4 is a scalar . in order to study the scaling of the energy we can neglect the depencence on orientation and consider an isotropic penalization of @xmath66 . it is however important that not all components of the gradient @xmath67 enter the energy . the elastic energy can be computed from the elastic strain , which is @xmath68 in @xmath69 and @xmath70 outside . denoting by @xmath71 the line - tension energy of a dislocation , the functional takes the form @xmath72 + \mu\int_{\r^3\setminus\omega } |d(u - u_0)(x)|^2{\ , \mathrm{d } x}\,.\nonumber\\ \qquad\end{aligned}\ ] ] as in ( [ eq : km ] ) , the terms @xmath73 are interpreted distributionally . the relative activity of the two slip systems is fixed by the far - field deformation @xmath74 , which will be taken affine . we refer to @xcite for a more detailed physical motivation of the model and to @xcite for a discussion of the relaxation of @xmath75 . the energy scaling for the case of the far - field deformation @xmath76 , corresponding to a half - half mixture of the two slip directions @xmath51 and @xmath52 , was studied in @xcite . the key result was that the optimal energy scales , for small @xmath1 , as the minimum of @xmath0 and @xmath3 , similarly to the kohn - mller model . the first scaling regime corresponds to the experimentally known hall - petch law , which states that the critical stress for plastic deformation in a polycrystal scales as the grain size to the power @xmath77 . here we consider a situation in which the boundary data favor states in which one of the two slip systems dominates , corresponding to a macroscopic forcing of the type @xmath78 where @xmath79 . we derive the following scaling law for @xmath75 as defined in ( see propositions [ prop : dislocub ] and [ prop : disloclb ] ) . [ th:2 ] for all @xmath1 , @xmath80 , @xmath7 and all @xmath81 , we have with @xmath82 , @xmath83 similarly to the martensite case , we recover all regimes from the case of equal volume fraction , and show that there is an additional intermediate scaling regime , which is achieved by a two - scale branching construction . the proof of this result is given in section [ sec : disloc ] below . throughout the text , we denote by @xmath84 positive constants that do not depend on any of the parameters @xmath1 , @xmath80 , @xmath18 or @xmath2 but that may change from expression to expression . for @xmath85 , @xmath86 , we use the notation @xmath87 if there is @xmath37 such that @xmath88 , and similarly for @xmath89 and @xmath90 . for a measurable set @xmath91 , we denote by @xmath92 its @xmath93-dimensional lebesgue measure . for @xmath94 , @xmath95 denotes the total variation of the distributional derivative along @xmath96 . for the kohn - mller model , the elastic energy of the austenite part can be expressed as trace norm at the interface , i.e. , with @xmath97 , @xmath98_{h^{1/2}_n((0,1))}^2:=\inf\left\{\int_0^\infty\int_0 ^ 1 |\nabla v(x)|^2{\ , \mathrm{d } x}:\ v(0,x_2)=u_0(x_2),\ v\in w^{1,2}_{\text{loc}}((0,\infty)\times(0,1))\right\ } \text{\ for \ } u\in\mathcal{a}_n,\ ] ] and for @xmath99 @xmath98_{h^{1/2}_p((0,1))}^2:=\inf\left\{\int_0^\infty\int_0 ^ 1 |\nabla v(x)|^2{\ , \mathrm{d } x}:\ v(0,x_2)=u_0(x_2),\ v\in w^{1,2}_{\text{loc}}((0,\infty)\times(0,1)),\ v(\cdot , 0)=v(\cdot,1)\right\}.\ ] ] we denote by @xmath100 and @xmath101 the spaces of functions in @xmath102 for which the respective seminorms are finite . note that the above are two different seminorms , and the choice of boundary conditions affects the value of the trace norm , hence we use two different symbols for @xmath103 and @xmath104 . indeed , functions with jump discontinuities are not contained in @xmath105 , hence no jump discontinuity between the two boundary points is permitted in @xmath104 , at variance with @xmath103 ( see remark [ rem : singlelam ] ) . for later reference , we recall equivalent representations of the norms . first ( see , for example , @xcite ) , @xmath106_{h^{1/2}_p{{\color{black}}{((0,1))}}}^2=\int_0 ^ 1\int_{-\infty}^{+\infty}\frac{|u_0(z_1)-u_0(z_2)|^2}{|z_1-z_2|^2}{\ , \mathrm{d } z}\text{\qquad for all\quad}u\in\mathcal{a}_p , \end{aligned}\ ] ] where we identify @xmath74 with its @xmath107-periodic extension ; on the other hand , by gagliardo s inequality ( see , e.g. , ( * ? ? ? * chapter 15.3 ) ) , @xmath108_{h^{1/2}_n{{\color{black}}{(}}(0,1){{\color{black}}{)}}}^2\sim\int_0 ^ 1\int_{0}^{1}\frac{|u_0(z_1)-u_0(z_2)|^2}{|z_1-z_2|^2}{\ , \mathrm{d } z}\text{\qquad for all \quad}u\in\mathcal{a}_n.\end{aligned}\ ] ] further , for @xmath109 , the @xmath104-seminorm can be equivalently characterized in terms of the fourier series . precisely , if @xmath110 , then @xmath111_{h^{1/2}_p((0,1))}^2\sim\sum_{k\in\z}|k||u_k|^2.\end{aligned}\ ] ] we note the following interpolation result . [ lemmah12 ] there is a constant @xmath37 such that for all @xmath112 and @xmath113 , @xmath114_{h^{1/2}_n((0,1))}[\psi]_{h^{1/2}_n((0,1))}.\end{aligned}\ ] ] we first assume additionally that @xmath115 . let @xmath110 and @xmath116 be the fourier series . then by cauchy - schwarz and @xmath117_{h^{1/2}_p((0,1))}[\psi]_{h^{1/2}_p((0,1))}.\end{aligned}\ ] ] we turn to the general case . without loss of generality , we may assume that @xmath118 . we extend @xmath119 and @xmath120 to @xmath121 by setting @xmath122 and @xmath123 for @xmath124 . then @xmath119 , @xmath125 and by ( i ) , @xmath126_{h^{1/2}_p((0,2))}[\psi]_{h^{1/2}_p((0,2))}\sim [ v]_{h^{1/2}_n((0,1))}[\psi]_{h^{1/2}_n((0,1))},\end{aligned}\ ] ] where the last step follows directly from the definition by extending test functions for @xmath119 and @xmath120 on the smaller strip @xmath127 ( anti-)symmetrically to @xmath128 , and on the other hand restricting test functions on @xmath128 to the smaller strip . similarly , we define the @xmath103-seminorm for functions on rectangles @xmath129 via @xmath130_{h^{1/2}_n(r)}^2:=\inf \left\{\int_r\int_0^\infty |\nabla v(x)|^2{\ , \mathrm{d } x}:\ v(x_1,x_2,0)=v(x_1,x_2),\ v\in w^{1,2}_{\text{loc } } ( r\times(0,\infty))\right\},\ ] ] and for the boundary of a cube @xmath131 , we set @xmath130_{h^{1/2}(\partial\omega)}^2:=[v]_{h^{1/2}_n(\partial\omega)}^2:=\inf \left\{\int_{\r^3\setminus\omega } |\nabla v(x)|^2{\ , \mathrm{d } x}:\ v = v\text{\ on\ } \partial\omega,\ v\in w^{1,2}_{\text{loc}}(\r^3\setminus\omega)\right\}.\ ] ] we will use the following lemma . [ lem : h1/2average ] set @xmath132 . let @xmath133 and set @xmath134 , @xmath135 . then @xmath136_{h^{1/2}_n((0,b))}\lesssim a^{-1/2}[h]_{h^{1/2}_n(r)}$ ] . let @xmath137 be arbitrary , and let @xmath138 be such that @xmath139 and @xmath140_{h^{1/2}_n(r)}^2+\delta$ ] . define an extension @xmath141 for @xmath142 via @xmath143 . then the assertion follows by jensen s inequality and arbitrariness of @xmath137 . .8 cm .8 cm we use the following construction for the proof of the upper bounds in and , with slight modifications also in . our main new contribution in this section is the two - scale branching construction sketched in figure [ figkmub](c ) . an important building block for all constructions are the following two functions which also show the main difference between the two kinds of boundary conditions . [ [ uniform - configurationsingle - laminate . ] ] uniform configuration / single laminate . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + * for neumann boundary conditions , we consider a uniform configuration in the martensite part and set @xmath144\times[0,1],\\ ( 1-x_1)\theta x_2&\text{\qquad if\quad}(x_1,x_2)\in[0,1]\times[0,1],\\ 0&\text{\qquad otherwise}. \end{cases}\ ] ] then @xmath145 and @xmath146 * for periodic boundary conditions , we consider a single laminate in the martensite part and set @xmath147\times[0,\theta],\\ \theta ( x_2 - 1)&\text{\qquad if\ } -l\leq x_1\leq 1,\ \max\{\theta,\ ( 1-\theta)x_1+\theta\}\leq x_2\leq 1,\\ \theta x_2(1-\frac{1}{(1-\theta)x_1+\theta})&\text{\qquad if\quad}0\leq x_1\leq 1,\ 0\leq x_2\leq ( 1-\theta)x_1+\theta,\\ 0&\text{\qquad if\quad}x_1\geq 1 . \end{cases } \end{aligned}\ ] ] then @xmath148 , and @xmath149 [ rem : singlelam ] we note that , given the single laminate construction in the martensite part , the extension to the austenite part given in yields the optimal scaling for the elastic energy . precisely , @xmath150_{h^{1/2}_p{{\color{black}}{(}}(0,1){{\color{black}}{)}}}^2\sim\theta^2\ln\frac{1}{\theta}. \end{aligned}\ ] ] the logarithmic correction term is due to the periodicity assumption . in contrast , we have @xmath151_{h^{1/2}_n{{\color{black}}{(}}(0,1){{\color{black}}{)}}}^2\sim [ u_{u}(0,\cdot)]_{h^{1/2}_n{{\color{black}}{(}}(0,1){{\color{black}}{)}}}^2\sim\theta^2 . \end{aligned}\ ] ] to prove it remains to show a lower bound on the seminorm . for that , we use the equivalent form , we identify @xmath152 with its @xmath107-periodic extension and estimate @xmath153_{h^{1/2}_p{{\color{black}}{(}}(0,1){{\color{black}}{)}}}^2&\geq&\int_{7/8}^1\int_{1+\theta}^{{{\color{black}}{1+}}\theta+1/8}\frac{|u_{sl}(0,z_1)-u_{sl}(0,z_2)|^2}{|z_1-z_2|^2}{\ , \mathrm{d } z}\\ & \geq & \left(\frac{\theta}{4}\right)^2\int_{7/8}^1\int_{1+\theta}^{{{\color{black}}{1+}}\theta+1/8}\frac{1}{|z_1-z_2|^2}{\ , \mathrm{d } z}\gtrsim\theta^2\ln\frac{1}{\theta}.\end{aligned}\ ] ] to show , we first extend @xmath152 to @xmath154\times[0,1]$ ] by @xmath155 then @xmath156 , and consequently , since @xmath157 , we have @xmath158_{h^{1/2}_p(0,1)}^2\lesssim [ u_{u}]_{h^{1/2}_n(0,1)}^2+\theta^2 $ ] and @xmath159_{h^{1/2}_n(0,1)}^2\lesssim [ u_{sl}]_{h^{1/2}_p(0,1)}^2+\theta^2 $ ] , which concludes the proof since @xmath159_{h^{1/2}_n(0,1)}^2\sim\theta^2 $ ] as shown in ( i ) above . all remaining constructions are special cases of a two - scale branching construction , which relies on the following lemma . the latter is a straight - forward generalization of a construction from ( * ? ? ? * lemma 2.3 ) . [ lem : branch ] suppose that @xmath32 , and let @xmath142 , @xmath56 , @xmath160 be such that @xmath161 . then there is a function @xmath162 with the following properties : * @xmath163 , @xmath164 , @xmath165 and @xmath166 if @xmath167 with @xmath168 and @xmath169 , * @xmath170 if @xmath171 and @xmath172 is @xmath173-periodic , * @xmath174 almost everywhere , * @xmath175 , where @xmath176 does not depend on @xmath142 , @xmath177 , @xmath18 or @xmath56 , * @xmath178 * @xmath179 . the function @xmath180 is sketched in figure [ figlembranch ] ( right ) . by ( ii ) , it suffices to construct @xmath180 for @xmath181 , namely @xmath182 and to evaluate all integrals . we remark that all conditions except the last one would also be satisfied by the construction with a connected minority phase . the last one is not important for the kohn - mller model , but will become relevant for the dislocation model . @xmath183 extended so that @xmath184 is @xmath173-periodic in @xmath185 . . left : the connected pattern described by @xmath186 , which has a longer vertical component of the interface and does not satisfy ( vi ) . right : the disconnected pattern of @xmath180 , which fulfills all stated properties . , title="fig:",height=151 ] . left : the connected pattern described by @xmath186 , which has a longer vertical component of the interface and does not satisfy ( vi ) . right : the disconnected pattern of @xmath180 , which fulfills all stated properties . , title="fig:",height=151 ] .8 cm .8 cm [ [ two - scale - branching . ] ] two - scale branching . + + + + + + + + + + + + + + + + + + + + we present a new construction which as special cases comprises test functions from the literature , in particular laminated structures , the single laminate and branched patterns ( see below ) . let @xmath168 , @xmath187 and @xmath188 be such that @xmath189 if @xmath190 . we construct a deformation @xmath191 in three steps : + _ step 1 : construction for @xmath192 : _ we construct an admissible function @xmath191 that is periodic in @xmath185-direction with period @xmath193 , and describe the construction on @xmath194}}\times{{\color{black}}{[}}0,1/n{{\color{black}}{]}}$ ] . we set @xmath195,\\ \theta ( x_2-\frac{1}{n})&\text{\qquad if\quad}x_2\in[\frac{1+h}{2n},\frac{1}{n } ] , \end{cases}\ ] ] and construct a branching function on the rectangle @xmath196\times{{\color{black}}{[}}\frac{1-h}{2n},\frac{1+h}{2n}]$ ] as follows : subdivide @xmath197 into rectangles @xmath198}}\times{{\color{black}}{[}}\frac{1-h}{2n}+\frac{jh}{2^in},\frac{1-h}{2n}+\frac{(j+1)h}{2^in}{{\color{black}}{]}},\qquad i=0,1,\dots,\text{\quad and\quad}j=0,\dots,2^i-1,\ ] ] and consider the bottom rectangles @xmath199 first . on level @xmath200 choose @xmath201 , @xmath202 and @xmath203 and note that the assumptions of lemma [ lem : branch ] are satisfied . we use the function @xmath204 from lemma [ lem : branch ] and set @xmath205 the function @xmath191 is then extended to a continuous function on @xmath206 such that @xmath207 is @xmath208-periodic in @xmath185-direction . by lemma [ lem : branch ] this yields a continuous function @xmath191 on @xmath209 with @xmath210 almost everywhere , and @xmath211 _ step 2 : construction for @xmath212 : _ for @xmath213 and @xmath214 , we set @xmath215 with @xmath216 set @xmath217 if @xmath218 , and extend it periodically in @xmath185-direction with period @xmath219 . note that for @xmath220 , we have @xmath221 if @xmath212 . we obtain@xmath222 summarizing , if @xmath189 , we have constructed a function @xmath223 such that ( see and ) @xmath224 we remark that the term proportional to @xmath2 disappears for @xmath220 . _ step 3 : construction for @xmath225 : _ it remains to consider the case @xmath226 . recall that we allow @xmath227 only if @xmath228 ( this will be needed in case of neumann boundary conditions below ) . we distinguish between dirichlet and neumann boundary conditions . to construct a periodic function @xmath229 , we extend @xmath191 constantly in @xmath230 , i.e. , @xmath231 we obtain an additional energy contribution @xmath232 in case of neumann boundary conditions , we introduce an interpolation layer analogously to the truncated branching construction from ( * ? ? ? * theorem 1 ) . precisely , we set ( if @xmath233 , we use the restriction of the function ) @xmath234 and extend @xmath235 constantly in @xmath230 for @xmath236 , i.e. , @xmath237 therefore , the additional energy contribution is estimated above by @xmath238 summarizing , if @xmath226 and @xmath228 , we have constructed functions @xmath239 with ( see , and respectively ) @xmath240 and @xmath241 as above , the term proportional to @xmath2 disappears for @xmath220 . we note for later reference that optimizing the upper bounds for the total energies ( see , and ) in @xmath142 subject to the constraint @xmath188 yields @xmath242 [ rem : trunc ] in order to reuse this construction in the upper bound for the dislocations , we will need a slight modification of step 1 , similarly to constructions from @xcite , i.e. , we stop at some finite level @xmath243 such that the slopes of the interfaces remain bounded by a constant , i.e. , @xmath244 , which is equivalent to @xmath245 we note that if such an @xmath243 exists , then @xmath246 we are now in the position to prove the upper bound of theorem [ th:1 ] . we point out that @xmath247 , and thus , we may replace @xmath248 by @xmath249 in the scaling law . [ prop : kmub ] there are constants @xmath250 , @xmath251 such that or all @xmath1 , @xmath2 , @xmath31 and all @xmath252}}$ ] , we have with @xmath33 , @xmath253 and @xmath254 we prove first . we use mostly variants of the two - scale - branching construction ( tsb ) and provide appropriate choices for the parameters @xmath177 , @xmath142 and @xmath255 in the various regimes . * the uniform function @xmath256 is always admissible , and thus , @xmath257 . * for any given parameters we obtain an admissible function as restriction to @xmath258 of the tsb with @xmath228 , @xmath220 and @xmath259 . this yields the truncated branching construction from ( * ? ? ? * theorem 1 ) , which is sketched in figure [ figtruncb ] ( left ) . thus ( see also ) , we have @xmath260 . * in the regime in which the minimum in is attained by @xmath261 , we in particular have @xmath262 . thus , we may choose @xmath263 , @xmath220 and @xmath189 in tsb . this is the variant of the kohn - mller branching constuction ( * ? ? ? * lemma 2.3 ) for unequal volume fractions as given in @xcite . this yields a function @xmath264 such that ( see ) @xmath265 . the deformation is sketched in figure [ figkmub](b ) . * if the minimum in is attained by @xmath266 , then in particular @xmath267 , and hence , we may choose @xmath268 , @xmath189 and @xmath269 . this yields the classical laminate construction @xmath270 , for which @xmath271 , where we used that @xmath247 . the function is sketched in figure [ figkmub](a ) . * finally , in the regime in which the minimum in is attained by @xmath272 , we use tsb with @xmath273 , @xmath189 and @xmath274 . note that @xmath275 as motivated by . this choice is indeed admissible : first , @xmath276 follows from @xmath277 . second , @xmath278 since @xmath279 implies that @xmath280 . finally , @xmath281 since @xmath282 implies that @xmath283 . thus , we obtain a function @xmath191 for which ( see ) @xmath284 this function is sketched in figure [ figkmub](c ) . we proceed similarly to prove . * suppose that the minimum in is attained by @xmath261 . there are two possibilities . if @xmath285 , then the branching construction @xmath264 is admissible ( see ( iii ) above ) otherwise , we have @xmath286 . then we choose @xmath228 , @xmath220 and @xmath259 in tsb . this yields a periodic variant @xmath287 of the truncated branching construction from ( * ? ? * theorem 1 ) with ( see ) @xmath288 . this function is sketched in figure [ figtruncb ] ( right ) . * suppose that the minimum in is attained by @xmath289 . again , there are two possibilities : if @xmath290 , then the laminate construction @xmath270 ( see ( iv ) above ) is admissible , for which @xmath291 . otherwise , we use the single laminate @xmath292 ( which corresponds to the choice @xmath228 , @xmath269 and @xmath189 in tsb ) , for which @xmath293 . * finally , suppose that the minimum in is attained by @xmath294 , and consider again the two possibilities separately . if @xmath295 , we use tsb as in ( v ) above , i.e. , with @xmath273 , @xmath189 and @xmath274 , which yields a function @xmath191 with @xmath296 . note that admissibility of @xmath142 follows verbatim as in the case of neumann boundary conditions . otherwise , we have @xmath297 , and we use tsb with @xmath228 , @xmath298 and @xmath142 as before . note that @xmath177 is chosen to make @xmath228 . in the regime under consideration , we indeed have @xmath227 . then by , this yields a function @xmath299 with @xmath300 . we remark that in the case @xmath301 all functions obey @xmath302 for all @xmath230 . we now turn to the proof of the lower bounds of theorem [ th:1 ] . our main new contribution is the proof in the case @xmath40 , in particular step 2 of the proof of lemma [ lemma1 ] below . we proceed in two steps . [ lemma1 ] there exist @xmath303 $ ] and @xmath37 with the following properties : for all @xmath304 , @xmath305 $ ] , @xmath306 $ ] , @xmath307 $ ] , @xmath308 $ ] such that @xmath309 holds , one has with @xmath310 @xmath311 and @xmath312 _ step 1 . general setting and localization . _ we start with the assertion . let @xmath2 , @xmath1 , @xmath18 , @xmath313 , @xmath177 and @xmath314 be given such that the assumptions of the lemma are satisfied ( without @xmath315 which will be chosen at the end of the proof ) , and let @xmath316 . for ease of notation we shall work for most of the proof with the function @xmath317 given by @xmath318 . then @xmath319 almost everywhere in @xmath13 and @xmath320 for any @xmath321 we estimate @xmath322 ( with the functions @xmath323 understood as traces ) . the first term is controlled by @xmath324 and the second one by @xmath325 where we used hlder s inequality and the trace theorem . therefore @xmath326 choose @xmath327 $ ] which minimizes @xmath328 . such a point @xmath329 exists since @xmath330 is discrete ( by ( * ? ? ? * section 4 ) we also know that we can assume without loss of generality that @xmath331 , this is however not needed here ) . consider the central interval @xmath332 . let @xmath333 and @xmath334 . if @xmath335 then @xmath336 and the assertion is proven otherwise , consider @xmath337 . note that @xmath338 consists of at most @xmath339 intervals . we claim that we only need to treat the case @xmath340 . indeed , for every @xmath341 , @xmath342 since @xmath343 almost everywhere , we have @xmath344 therefore either @xmath345 or at least one of @xmath346 must hold . if ( [ eq : possneumann ] ) holds , the assertion is established , since @xmath347 for all @xmath348 . we thus can focus on the case @xmath345 and observe that if @xmath349 then @xmath350 for all @xmath351 and all @xmath352 , leading to ( recall ) @xmath353 again , either one of the possibilities in ( [ eq : possneumann ] ) holds and the assertion is proven , or @xmath354 . in the rest we hence consider the case @xmath355 . _ step 2 : construction of the test function and proof of for small @xmath356 . _ the set @xmath357 , which is the @xmath358-neighbourhood of @xmath338 , is the union of at most @xmath359 intervals @xmath360 , where @xmath361 are the midpoints of the connected components of @xmath338 . then @xmath362 for every @xmath363 , and @xmath364 . we define @xmath365_+\,.\ ] ] then @xmath366 in @xmath367 , and @xmath368 outside @xmath369 . hence , since @xmath370 and @xmath371 , we deduce that @xmath372 $ ] . note that @xmath373 and therefore @xmath374 further , @xmath375 we next estimate @xmath376_{h^{1/2}_p((0,1))}$ ] . for every @xmath377 let @xmath378 be the radially symmetric extension of @xmath379 to @xmath380 . then @xmath381 and hence , @xmath382_{h^{1/2}_p((0,1))}^2&\leq & \sum_{i=1}^n\int_0^\infty\int_0 ^ 1|d\psi_i(x_1,x_2-y_i)|^2\,{\ , \mathrm{d } x}\lesssim n\ln\frac{1}{m}\leq\frac{1}{\lambda}\ln\frac{1}{m}.\end{aligned}\ ] ] since @xmath383 on @xmath338 , and @xmath384 everywhere , we have by and with @xmath385 , @xmath386_{h_n^{1/2}((0,1))}[\psi]_{h^{1/2}_p((0,1))}+\theta\|\psi\|_{l^1((0,1))}\nonumber\\ & \leq&\frac{c}{m^{1/2}\lambda } |x_1^\ast|^{1/2}f^{1/2}(v)+\frac{cf^{1/2}(v)}{(\lambda\mu)^{1/2}}\ln^{1/2}\frac{1}{m}+c\theta,\end{aligned}\ ] ] where in the last step we used , and . since @xmath345 , there is @xmath387 $ ] such that for @xmath388 the last term of the right - hand side can be absorbed in the left - hand side and @xmath389 ( this defines @xmath390 ) . we deduce that @xmath391 which , combined with and , concludes the proof of if the first option in ( [ lemma1cases ] ) holds . _ step 3 : construction of the test function and proof of for large @xmath392 . _ note that if the second option in ( [ lemma1cases ] ) holds then @xmath393 . in this case , the argument from @xcite can be applied to obtain the lower bound but we present here an alternative proof in the spirit of _ step 2_. at variance with what was done in step 2 , in this case the test function is supported inside @xmath338 . the set @xmath338 is the union of @xmath394 intervals , @xmath395 , and @xmath396 . note that the points @xmath361 are as in _ step 2 _ chosen to be the midpoints of the connected components of @xmath338 , but the radii @xmath397 are different . we set @xmath398 we compute @xmath399 and @xmath400 extending @xmath120 to @xmath127 by @xmath401 , we obtain @xmath402_{h^{1/2}_p((0,1))}^2\le \int_{(0,\infty)\times(0,1)}|d\psi(x)|^2{\ , \mathrm{d } x}= \sum_{i=1}^n \frac12 |b(0 , g_i)| = \frac\pi2 \sum_{i=1}^n g_i^2\,.\ ] ] as in ( [ eq : compn ] ) , we use @xmath120 to test @xmath403 , integrate by parts and use ( [ eq : interpolfourier2 ] ) , @xmath404_{h_n^{1/2}((0,1))}[\psi]_{h^{1/2}_p((0,1))}+\theta \sum_{{{\color{black}}{i=1}}}^{{\color{black}}{n } } g_i^2\,.\end{aligned}\ ] ] since @xmath405 , @xmath406 therefore at least one of @xmath407 must hold and the proof of is concluded . _ step 4 : proof of _ let @xmath99 and choose @xmath119 and @xmath329 as in _ step 1_. by the periodicity assumption , we have @xmath408 . hence , possibly choosing a different cell of periodicity , we may assume that @xmath409 , where @xmath338 is defined as in _ step 1 _ , i.e. , @xmath410 . the rest of the proof is the same as for @xmath411 . we remark that deriving the condition @xmath340 directly from the periodicity allowed us to avoid using ( [ eq : possneumann ] ) , and therefore the option @xmath412 does not appear in the conclusion . we now conclude the proof of the lower bounds of theorem [ th:1 ] . [ prop : kmlb]there is a constant @xmath37 such that for all @xmath2 , @xmath1 , @xmath31 and all @xmath413 $ ] , we have with @xmath414 , @xmath415 and @xmath416 to prove , we use with appropriate choices for the parameters @xmath313 , @xmath417 and @xmath177 . we start fixing @xmath418 , and distinguish several cases . * if @xmath419 , we choose @xmath420 , @xmath421 , and @xmath422 . then ( [ lemma1cases ] ) holds for all @xmath18 , and by we obtain @xmath423 . * if @xmath424 , we choose @xmath420 , @xmath425 and @xmath426 . as above , implies @xmath427 . * it remains to consider @xmath428 . we set @xmath429 if @xmath430 this gives @xmath431 ; if @xmath432 instead @xmath433 . therefore we can use . note that @xmath434 . we distinguish two subcases : * * if @xmath435 , we choose @xmath422 and @xmath436 . then ( since @xmath437 and @xmath438 ) , we have @xmath439 . * * if @xmath440 , we choose @xmath425 and @xmath441 . then @xmath442 . we finally prove using . we fix @xmath443 . by the periodicity assumption we have @xmath444 . as above , we distinguish several cases . * if @xmath419 we choose @xmath420 , @xmath421 , and @xmath422 as in ( i ) above and obtain @xmath445 . * if @xmath446 then @xmath447 and the assertion is proven . * it remains to consider @xmath428 . we define @xmath314 as in ( [ eqdefm ] ) and distinguish the same two subcases as above . * * if @xmath435 , we choose the same parameters as in ( iii)(a ) above and obtain @xmath448 . * * if @xmath440 , then as in ( ii ) we obtain @xmath449 . we now turn to the model for plastic microstructures described in section [ sec : dislocmodel ] , and aim to prove theorem [ th:2 ] . we note that it suffices to prove theorem [ th:2 ] for @xmath39 . the result for general cubes @xmath48 then follows by rescaling @xmath450 and @xmath451 ( see ( * ? ? ? * section 4 ) ) . we start with the upper bound . [ prop : dislocub ] there is a constant @xmath37 such that for all @xmath452}}$ ] , and all @xmath2 , @xmath71 , we have @xmath453 following ( * ? ? ? * section 4.1 ) , test functions for the kohn - mller model with appropriate modifications can be used to construct test functions for the energy given in . we treat the regimes @xmath454 and @xmath455 separately , and afterwards outline the general construction and then consider the various regimes . set @xmath456 _ step 0 : the regimes @xmath454 and @xmath455 . _ choosing @xmath457 and @xmath458 , we find that @xmath459 . choosing @xmath460 and @xmath461 we get @xmath462 . _ step 1 : the general setting . _ suppose that there are @xmath463 and a function @xmath464 such that @xmath465 and @xmath466 we use the signed distance @xmath467 to define @xmath468 and @xmath469 we collect some properties of these functions : since @xmath470 , the function @xmath4 is continuous at @xmath471 . for the rest we only need to consider the stripe @xmath472 . further , @xmath473 , @xmath474 @xmath475 outside @xmath69 . finally , @xmath476 on @xmath477 , @xmath478 on @xmath479 , and @xmath480 . therefore , by ( for the ease of notation we suppress the constraint @xmath472 in all domains of integration ) @xmath481 _ step 2 : construction of @xmath482 . _ it remains to find @xmath482 and @xmath483 such that and hold . we consider the various regimes separately and provide functions @xmath484 and parameters @xmath463 that satisfy and . for that , we build on the functions from section [ secpfkm ] and in particular the proof of proposition [ prop : kmub ] with @xmath485 . in each case it is easy to see that the @xmath486 term is irrelevant . * if @xmath487 we choose @xmath488 and @xmath489 . * if @xmath490 , we use a modification of the classical branching construction @xmath264 , which is a variant of the branching construction given in ( * ? ? ? * section 4.1 ) with unequal volume fractions and corresponds to the finite branching construction given in ( * ? ? ? * pf . of theorem 2 ) . we use the notation of section [ secpfkm ] and set @xmath491 , @xmath492 , @xmath493 and @xmath494 with @xmath495 given in remark [ rem : trunc ] . note that for @xmath496 , we always find an admissible @xmath243 since @xmath497 and @xmath498 . we define @xmath482 via @xmath499 by the computations of section [ secpfkm ] , in particular , it remains to show @xmath500 first , since @xmath501 , we have @xmath502 second , since @xmath503 , @xmath504 using @xmath505 and . * consider now the regime in which @xmath506 . set @xmath496 , @xmath507 and @xmath508 . admissibility of these choices follows as in the case of neumann boundary conditions in the proof of proposition [ prop : kmub ] . choose @xmath243 such that holds and set @xmath509 . we define @xmath510 again , in view of section [ secpfkm ] , it suffices to consider @xmath511 . using that @xmath512 and @xmath513 , we have @xmath514 where we again used that @xmath505 and . we split the proof of the lower bound in several lemmas . we start with the local structure of @xmath8 . [ lemmarect ] let @xmath515 , @xmath516 , and @xmath517 , and set @xmath518 . suppose that @xmath119 , @xmath519 with @xmath520 almost everywhere . assume further that @xmath521 and also @xmath522 then there are @xmath523 pairwise disjoint , possibly degenerate intervals @xmath524 such that @xmath525 for ease of notation we change coordinates for the proof , and assume @xmath526 , @xmath527 . by and fubini , there exists @xmath528 such that @xmath529 define @xmath530 . then by @xmath531 and therefore by @xmath532 further , for @xmath533 one has @xmath534 and therefore by @xmath535 set @xmath536 . then by the assumption @xmath537 and , we have @xmath538 in @xmath539 , and by @xmath540 by the coarea formula , @xmath541 , and therefore there is @xmath542 such that the set @xmath543 is the union of at most @xmath523 intervals @xmath544 . for these intervals holds since they cover @xmath539 and @xmath545 on all @xmath546 . for @xmath547 and @xmath548 , we set @xmath549 + \mu \int_{r\times { { \color{black}}{[}}(-\infty,0)\cup(1,\infty){{\color{black } } { ] } } } |du(x)-du_0(x)|^2{\ , \mathrm{d } x}.\ ] ] we fix a function @xmath550)$ ] such that @xmath551 on @xmath552 and @xmath553 . for @xmath554 and @xmath124 set @xmath555 . [ lemmatest ] there exists a constant @xmath37 with the following property : for every @xmath548 , and every @xmath556 , @xmath180 , @xmath557 , @xmath558 for which @xmath559 there exists @xmath560 such that for all @xmath561 and all @xmath562 , setting @xmath563 , @xmath564\ , \end{gathered}\ ] ] and @xmath565 . \end{aligned}\ ] ] let @xmath548 . choose @xmath560 such that @xmath566 , where @xmath567.\ ] ] since @xmath568 , we have , integrating by parts , @xmath569 by definition of @xmath570 , we have the estimates @xmath571 and @xmath572 . it remains to estimate the first term of the right - hand side . since @xmath573 , and since by symmetry of @xmath570 we have @xmath574 for all @xmath575 , we obtain ( writing for brevity @xmath576 instead of @xmath577 ) @xmath578 which implies that ( recall that @xmath579 depends only on @xmath580 ) @xmath581 putting things together and using that @xmath582 , we obtain the first assertion . the second one follows similarly by performing the same computation in the other direction and using that @xmath583 . [ cor : dischoice ] under the assumptions of lemma [ lemmatest ] , if @xmath584 , then at least one of @xmath585 and @xmath586 holds . . then @xmath588 and @xmath589 . thus we have @xmath590 , or by @xmath591.}\ ] ] rearranging terms and using the assumption @xmath584 yields . we now use the previous estimates to obtain a lower bound on @xmath75 similarly to lemma [ lemma1 ] in the martensite case . [ lem : dislb ] there are @xmath592 and @xmath37 with the following property : for all @xmath593 , all @xmath413 $ ] and all @xmath594 such that @xmath595 holds , and for all @xmath306 $ ] , one has @xmath596 . the marked rectangle is @xmath597 , the inner , light green rectangle is @xmath598 . the blue region is @xmath599 where @xmath600 concentrates.,height=188 ] _ step 1 . _ let @xmath1 , @xmath2 , @xmath18 , @xmath314 , and @xmath313 be given with the properties stated in the lemma . it suffices to consider an arbitrary pair @xmath548 . we use the short - hand notation @xmath601 , and similarly for @xmath602 and @xmath603 . we introduce auxiliary parameters @xmath556 , @xmath180 and @xmath255 . precisely , fix @xmath604 , @xmath605 $ ] ( @xmath606 will do ) , let @xmath168 be such that @xmath607 , and set @xmath608 . choose @xmath557 , @xmath558 such that @xmath609 and @xmath610 ( see figure [ figlambdislr ] ) . let @xmath611 . finally pick @xmath612 as in lemma [ lemmatest ] , i.e. , such that @xmath613 . + we apply corollary [ cor : dischoice ] to @xmath598 which shows that if does not hold then holds with @xmath614 on @xmath615 . since @xmath616 we deduce from lemma [ lemmarect ] , with @xmath617 , that we have @xmath618 or there are disjoint intervals @xmath619 such that holds with the above choices . if ( [ eq : disest1 ] ) holds , the proof is concluded . suppose now that the other option holds , and set @xmath620 and @xmath621 . then the second estimate of yields , since @xmath622 , @xmath623 by lemma [ lemmatest ] , since @xmath624 , @xmath625.}\end{aligned}\ ] ] therefore , if @xmath626 then @xmath627 if ( [ eq : disest2 ] ) holds , then the proof is concluded ( in proving @xmath628 we use that ( [ lemma1casesdisl ] ) implies @xmath629 ) . if instead @xmath630 , we proceed along the lines of the proof of lemma [ lemma1 ] . construction of the test function and conclusion of the proof . _ let @xmath631 , and set for @xmath632 , @xmath633_+\,.\ ] ] note that @xmath120 has compact support in @xmath634 , @xmath635 in @xmath338 , @xmath636 , and @xmath637 . then by the first of and since @xmath600 , @xmath120 and @xmath638 are non - negative , we have , using @xmath583 , @xmath639 by lemma [ lemmah12 ] and lemma [ lem : h1/2average ] , we have @xmath640_{h^{1/2}_n((x_b - b , x_b+2b))}\left[\int_{x_a}^{x_a+a}(u_0-u)(x_\xi,\cdot){\ , \mathrm{d } x}_\xi\right]_{h^{1/2}_n((x_b - b , x_b+2b))}\\ \lesssim & a^{1/2}[\psi]_{h^{1/2}_n((x_b - b , x_b+2b))}[u_0-u]_{h^{1/2}_n(\tilde{r})}.\end{aligned}\ ] ] we use that ( cf . step 2 in the proof of lemma [ lemma1 ] ) @xmath641 , @xmath376_{h^{1/2}_n}\lesssim n^{1/2}\ln^{1/2}\frac{1}{{{\color{black}}{6}}m}$ ] , @xmath642 and @xmath643 , where the norms are taken on @xmath634 . hence , @xmath644 if @xmath315 is small enough , then the last term can be absorbed into the left - hand side and @xmath389 . hence , @xmath645 recalling @xmath604 , @xmath646 , @xmath647 , and @xmath648 the proof is concluded . proceeding along the lines of the proof of proposition [ prop : kmlb ] , we now conclude the proof of the lower bound . [ prop : disloclb ] there is a constant @xmath37 such that for all @xmath593 , and all @xmath649 , we have @xmath650 we start fixing @xmath651 , @xmath1 , @xmath7 and @xmath413 $ ] , and use lemma [ lem : dislb ] with different choices of parameters in different regimes . * if @xmath652 , we choose @xmath420 and @xmath653 . then we obtain @xmath654 . * if @xmath655 , we choose @xmath420 and @xmath425 . then by , @xmath656 . * it remains to consider @xmath657 . we set @xmath658 since @xmath405 we have @xmath659 , therefore ( [ lemma1casesdisl ] ) holds . we distinguish two subcases : * * if @xmath660 , we choose @xmath661 . using @xmath662 in the @xmath663 term , and @xmath664 in the @xmath665 term , we conclude @xmath666 . since @xmath667 , the proof is concluded in this case . * * if @xmath668 , we choose @xmath425 and @xmath669 . then @xmath670 . we thank michael goldman and stefan mller for several interesting discussions . this work was partially supported by the deutsche forschungsgemeinschaft through the sonderforschungsbereich 1060 _ `` the mathematics of emergent effects '' _ , projects a5 and a6 . k. anguige and p. dondl . energy estimates , relaxation , and existence for strain - gradient plasticity with cross - hardening . in s. conti and k. hackl , editors , _ analysis and computation of microstructure in finite plasticity _ , volume 78 , pages 157173 . springer , 2015 . a. chan and s. conti . energy scaling and domain branching in solid - solid phase transitions . in m. griebel , editor , _ singular phenomena and scaling in mathematical models _ , pages 243260 . springer international publishing , 2014 . j. cui , y. chu , o. famodu , y. furuya , j. hattrick - simpers , r. james , a. ludwig , s. thienhaus , m. wuttig , z. zhang , and i. takeuchi . combinatorial search of thermoelastic shape - memory alloys with extremely small hysteresis width . , 5:286290 , 2006 . r. delville , s. kasinathan , z. zhang , j. van humbeeck , r. james , and d. schryvers . transmission electron microscopy study of phase compatibility in low hysteresis shape memory alloys . , 90:177195 , 2010 . r. james and z. zhang . a way to search for multiferroic materials with " unlikely " combinations of physical properties . in l. manosa , a. planes , and a. saxena , editors , _ the interplay of magnetism and structure in functional materials _ , volume 79 . springer , 2005 . s. mller . variational models for microstructure and phase transitions . in f. bethuel et al . , editors , _ calculus of variations and geometric evolution problems _ , springer lecture notes in math . 1713 , pages 85210 . springer - verlag , 1999 . r. zarnetta , r. takahashi , m. young , a. savan , y. furuya , s. thienhaus , b. maa , m. rahim , j. frenzel , h. brunken , y. chu , v. srivastava , r. james , i. takeuchi , g. eggeler , and a. ludwig . identification of quaternary shape memory alloys with near - zero thermal hysteresis and unprecedented functional stability . , 20(12):19171923 , 2010 .

we study extremal systolic problems on riemann surfaces , and specifically dyck s surface and its orientable genus-2 double cover . there are three conformal types of genus-2 surfaces that are significant for various problems of systolic extremality . we will denote them , , and . here is the bolza surface , is a surface first considered in detail by parlier , and is introduced in the present text . we show that ( 1 ) an extremal systolic inequality for nonpositively curved metrics , whose case of equality is attained by a quotient of by a fixedpointfree anticonformal involution , equipped with a singular flat metric with finitely many conical singularities ; and ( 2 ) the conformal classes , , and are distinct . in more detail , w we prove an optimal systolic inequality for nonpositively curved dyck s surfaces . the extremal surface is flat with eight conical singularities , six of angle and two of angle for a suitable with . relying on some delicate capacity estimates , we also show that the extremal surface is not conformally equivalent to the hyperbolic dyck s surface with maximal systole , yielding a first example of systolic extremality with this behavior . primary 53c23 ; secondary 30f10 , 58j60

1205.0188

the systole of a metric space @xmath6 is the least length of a loop in @xmath6 which can not be contracted to a point . we will study extremal systolic problems on riemann surfaces , and specifically dyck s surface and its orientable genus-2 double cover . it turns out that there are three conformal types of genus-2 surfaces that are significant for various problems of systolic extremality . we will denote them @xmath0 , @xmath1 , and @xmath2 . here @xmath0 is the bolza surface , @xmath1 is a surface first considered in detail by parlier , and @xmath2 is the main subject of the current text . our main results are the following : 1 . an extremal systolic inequality for nonpositively curved metrics , whose case of equality is attained by a quotient of @xmath2 by a fixedpointfree antiholomorphic involution , equipped with a singular flat metric with finitely many conical singularities ; and 2 . the conformal classes @xmath0 , @xmath1 , and @xmath2 are distinct . the systole of a metric space @xmath6 is the least length of a loop in @xmath6 which can not be contracted to a point . only a small number of optimal systolic inequalities relating the systole and the volume of @xmath6 are available in the literature . loewner s torus inequality asserts that every riemannian torus @xmath7 satisfies @xmath8 see pu s paper @xcite . pu s inequality asserts that every real projective plane @xmath9 satisfies @xmath10 bavard s inequality @xcite for the klein bottle @xmath11 is the bound @xmath12 the burago - ivanov - gromov inequality relates the stable 1-systole of an @xmath13-torus to its volume : @xmath14 where @xmath15 is the hermite constant ; see burago and ivanov @xcite , @xcite , gromov @xcite , and @xcite . bangert et al . proved the following optimal inequality for orientable riemannian @xmath13-manifolds @xmath6 : @xmath16 where @xmath17 is the first betti number , and @xmath18 is the berg - martinet constant ; see bangert et al . @xcite , @xcite , and @xcite . in the context of nonpositively curved metrics , we proved the following optimal inequality in @xcite : every nonpositively curved genus two surface @xmath19 satisfies the bound @xmath20 furthermore , the equality case is attained by a piecewise flat metric with @xmath21 conical singularities of angle @xmath22 . such a metric can be expressed by a conformal factor vanishing at the singular points , with respect to a smooth metric ( e.g. , the hyperbolic one ) in the conformal class of the bolza surface . the latter is the surface possessing the maximal group of symmetries in genus 2 . the _ hyperbolic _ metric on the bolza surface had also been shown to have maximal systole in this genus by jenni @xcite and others . the combination of these two ( old ) results can be summarized as follows . [ 11 ] the systolically extremal surface of nonpositive curvature and the systolically extremal hyperbolic surface belong to the same conformal class , namely that of the bolza surface . it turns out that such coincidence of conformal classes no longer holds in the case of dyck s surface ( see theorem [ prop : distinct ] ) . in the present article , we develop an analogous optimal systolic inequality for nonpositively curved metrics on dyck s surface . here surface @xmath23 is the nonorientable closed surface of euler characteristic @xmath24 , homeomorphic to the connected sums @xmath25 by convention , a point on dyck s surface will be referred to as a weierstrass point if it is the image of a weierstrass point of the orientable double cover . [ theo : main1 ] each nonpositively curved metric on dyck s surface @xmath26 satisfies the following optimal systolic inequality : @xmath27 furthermore , the equality case is attained by a piecewise flat surface composed of a flat mbius band and three identical symmetric nonregular flat hexagons centered at its weierstrass points . observe that the optimal constant @xmath28 in is less than the one in , which is approximately @xmath29 . this is consistent with the expected monotonicity of the systolic area as a function of the euler characteristic . for general metrics , it is more difficult to obtain estimates for the systolic ratio , but in @xcite we obtained a small improvement ( for dyck s surface ) of a general estimate of gromov s . the extremal surface has eight conical singularities , which correspond to the vertices of he hexagons : two of angle @xmath30 and six of angle @xmath31 , where @xmath32 the two conical singularities of angle @xmath30 correspond to the points where the three hexagons meet , while those of angle @xmath33 correspond to the points where the mbius band meets two hexagons . we will refer to the extremal surface as the _ extremal nonpositively curved dyck s surface _ and denote it @xmath34 . we present two constructions of @xmath34 , one in section [ sec : construction ] and the other in section [ sec : extremal ] . theorem [ theo : main1 ] is proved in section [ six ] . the hyperbolic dyck s surface of maximal systole , denoted by @xmath35 , was described by parlier @xcite and then by gendulphe @xcite . it has the same symmetry group as the extremal nonpositively curved dyck s surface . however , using some recent work of b. muetzel @xcite , we show that they are not conformally equivalent . [ prop : distinct ] the conformal types of the extremal nonpositively curved dyck s surface @xmath34 and the extremal hyperbolic dyck s surface @xmath35 are distinct , and can be distinguished by the capacities of the associated annuli @xmath36 and @xmath37 obtained by cutting open their orientable double covers : @xmath38 ( see section [ nine ] for the definitions of the annuli ) . this yields a first example of systolic extremality with this behavior , in contrast with the situation in genus @xmath39 summarized in theorem [ 11 ] . in genus 3 , the question is open . for genus @xmath40 surfaces , do the extremal hyperbolic surface and the extremal surface of nonpositive curvature belong to the same conformal type ? we also observe that the orientable double cover of @xmath34 is not conformally equivalent to the bolza surface ; proposition [ prop : bolza ] . since the extremal hyperbolic metric and the extremal singular flat metric on dyck s surface are defined explicitly , one might have hoped to find a ( simpler ) indirect proof of the fact that they belong to distinct conformal classes . however , this appears to be difficult . we have therefore relied on the capacity estimate which produces a direct numerical argument distinguishing between the two conformal classes . the proof of the main theorem makes use of an averaging argument by the hyperelliptic involution and the rauch comparison theorem as in @xcite . unlike the situation treated in @xcite , the orientable double cover of the extremal surface does not have a maximal group of symmetries . therefore our approach requires some new features . more specifically , it relies on the following : 1 . an analysis of the lengths of some distance curves to a canonical nonorientable loop ; 2 . an analysis of some optimal flat pieces ( hexagons and cylinder ) of a partition of the extremal surface ; 3 . a study of the combinatorics of these pieces ; 4 . a delicate capacity estimate as explained above . the extremal dyck s surface @xmath34 is flat with eight singularities : six of angle @xmath3 and two of angle @xmath41 , with @xmath42 ( see formula ) . we describe below their positions in terms of ramifications points of some ramified covers . the orientable double cover of @xmath34 is a genus two surface which is a branched double cover over @xmath43 with six branch points . the branch points correspond to the vertices of an inscribed prism on an equilateral triangle . the weierstrass points of the extremal surface are actually smooth , therefore the quotient metric on @xmath43 has conical singularities of angle @xmath44 at the six branch points , in addition to the singular points arising from the extremal metric . the singular points of the extremal metric on the orientable double cover can be described as follows . the centers of the triangular faces of the prism lift to 4 conical singularities of angle @xmath41 . the other 12 conical singularities , of angle @xmath3 , correspond to the vertices of another inscribed prism on an equilateral triangle with the same axis as the branch point prism.can this be explained in more detail ? what is this other prism exactly ? it would be helpful to give a reference to where it is described in the paper . ] in terms of the dual polyhedron of the branch point prism . more precisely , the vertices of the dual polyhedron lift to conical singularities of angle @xmath3 and the centers of the two triangular sides lift to conical singularities of angle @xmath41.the centers of the triangular faces ( not sides ) apparently are also vertices of the dual polyhedron . perhaps this should be clarified . ] in addition to the description below , we will present an alternative construction of the extremal surface at the end section [ sec : extremal ] . we normalize the systole to @xmath45 . consider real numbers @xmath46 and @xmath47 specified by the relations @xmath48 the piecewise flat surface defining the extremal nonpositively curved dyck s surface @xmath34 can be constructed as follows . 1 . take a flat isosceles trapezoid of height @xmath47 and acute angle @xmath46 with the shorter ( internal ) side of length @xmath49 , where parameters @xmath46 and @xmath47 satisfy the relations ; 2 . form a nonplanar hexagonal annulus @xmath50 composed of six identical trapezoids ( see figure [ fig:0 ] ) , where the inner boundary component of @xmath50 is of length @xmath39 ; 3 . [ three ] form the torus with a disk removed , obtained from the hexagonal annulus @xmath50 by identifying the opposite sides of the outer boundary component of the annulus , as in figure [ fig:0 ] ; 4 . attach a flat mbius band of width @xmath51 and boundary length @xmath39 to the torus with a disk removed constructed in step . the values of the parameter @xmath46 and @xmath47 are chosen so that the contribution of the hexagonal annulus and the mbius band to the ( systolic ) area of the dyck s surface is minimal . this optimal configuration forces the systolic loops to decompose into three families , 2.5 . , height=151 ] ( 0,0)(0,0 ) ( 2,134)@xmath47 ( -18,138)@xmath46 ( -4,60)@xmath49 the extremal surface @xmath34 so defined is of nonpositive curvature in the sense of alexandrov , since @xmath52 . more precisely , it is piecewise flat and has eight conical singularities : * two of angle @xmath53 corresponding to the vertices of the outer boundary component of @xmath50 , and * six of angle @xmath54 corresponding to the vertices of the inner boundary component of @xmath50 . the angle @xmath3 of six of the eight conical singularities satisfies @xmath55 , which leads to the following expression : @xmath56 the weierstrass points of @xmath34 correspond to the midpoints of the sides of the outer boundary component of @xmath50 after completion of the steps ( 1 ) through ( 4 ) of the construction of the extremal surface . the weierstrass points of the extremal surface are actually smooth , therefore the quotient metric on @xmath43 from has conical singularities of angle @xmath44 at the six branch points , in addition to the singular points arising from the extremal metric . the optimal configuration for @xmath34 follows from an equilibrium between the systolic area contribution of the mbius band and that of the voronoi cells centered at the weierstrass points . the following proposition provides a description of the automorphism and symmetry groups of @xmath57 viewed as a riemannian klein surface . [ prop : sym ] the automorphism group and the symmetry group of @xmath57 are both isomorphic to @xmath58 the natural homomorphism between the automorphism ( resp . symmetry ) group of @xmath57 and the permutation group @xmath59 of its weierstrass points is surjective . its kernel is composed of dianalytic ( _ i.e . _ locally holomorphic or antiholomorphic ) automorphisms preserving the weierstrass points . the only two automorphisms with this property are the identity map and the hyperelliptic involution ( which is an isometry ) . since they commute with every holomorphic map ( and so every isometry ) , @xcite , we deduce that the automorphism ( resp . symmetry ) group of @xmath57 is isomorphic to @xmath60 . more specifically , it can be described as follows . the orientable double cover of @xmath34 is a genus two riemann surface @xmath19 which can be thought of as a holomorphic ramified double cover @xmath61 whose branch points ( _ i.e . _ the images of the weierstrass points of @xmath19 ) form the set of @xmath62 vertices of a triangular prism inscribed in @xmath43 . an affine form of such a riemann surface can be taken to be the locus in @xmath63 of the equation @xmath64 for a suitable @xmath65 ( cf . equation below ) . the prism can be taken to be a standard right prism on an equilateral triangle inscribed in a unit sphere in @xmath66 . the symmetry group @xmath67 of this prism can be described geometrically as follows . the dihedral group @xmath59 acts by isometries on the prism ( globally ) preserving each triangular face . the second factor @xmath68 is the center of the symmetry group , and is generated by an orientation - reversing involution given by the reflection in the plane ( through the origin ) parallel to the triangular faces . the double cover @xmath19 has an additional automorphism given by the hyperelliptic involution . its group of holomorphic and antiholomorphic automorphisms is of order 24 and is isomorphic to @xmath69 . the surface @xmath19 belongs to the complex family of riemann surfaces with holomorphic automorphism group @xmath70 given by bolza s classification @xcite ( see also the presentation in @xcite ) . note that the extremal nonpositively curved genus two surface admits a description in terms of a uniform _ antiprism _ ( rather than a prism ) , namely , a regular octahedron ( @xcite ) . its holomorphic automorphism group is isomorphic to @xmath71 . the unique riemann surface with this automorphism group is bolza s curve , @xcite , @xcite . in particular , the conformal class of the double cover of the surface @xmath34 is distinct from the bolza curve . there are three types of systolic loops on the extremal surface : * the soul of the flat mbius band , * the loops orthogonal to a short base of one of the trapezoids of @xmath50 , * the loops orthogonal to a leg of one of the trapezoids of @xmath50 . note that the last type of systolic loops contains loops joining any pair of weierstrass points . since the extremal surface admits regions at every point of which exactly one systolic loop passes , it is not extremal for the curvature - free systolic inequality on @xmath23 ; ( * ? ? ? * lemma 2.1 ) . actually , a direct application of the characterization of conformally extremal surfaces established by bavard in @xcite shows that @xmath34 is not even extremal in its conformal class for the curvature - free systolic inequality . thus , an extremal metric for the curvature - free systolic inequality on @xmath23 necessarily admits regions of both positive and negative curvature . we also describe the riemann surface structure of the extremal nonpositively curved dyck s surface . [ theo : main2 ] the algebraic curve representing the orientable double cover of the extremal nonpositively curved dyck s surface is the smooth completion of the following affine curve in @xmath63 : @xmath72 the orientable double cover of the hyperbolic dyck s surface @xmath26 of maximal systole was described by parlier @xcite . silhol @xcite identified a presentation of its affine form : @xmath73 where @xmath74 . see also lelivre & silhol @xcite and gendulphe @xcite . the set of the roots of the equations and are _ not _ equivalent by a mobius transformation of @xmath75 . indeed , the roots of lie on a common circle , while those of lie neither on a common circle , nor on a straight line ( a similar argument shows that the algebraic curve is not conformally equivalent to the bolza surface of equation @xmath76 ) . we will also detect differences at the level of the hyperbolic metrics in section [ last ] . we therefore obtain the following corollary . in this section , we review some conformal constructions and results that we will need for the proofs of theorem [ theo : main1 ] and theorem [ prop : distinct ] . a riemann surface homeomorphic to @xmath23 is the quotient @xmath77 of a genus two riemann surface @xmath19 by a fixed point - free antiholomorphic involution @xmath78 . recall that every genus two riemann surface @xmath19 is hyperelliptic . it admits an affine model @xmath79 in @xmath63 , where @xmath80 is a degree @xmath62 complex polynomial with six distinct roots , which correspond to the weierstrass points of @xmath19 . in this presentation , the hyperelliptic involution @xmath81 is given by the transformation @xmath82 on the affine model . it is the only holomorphic involution of @xmath19 with six fixed points ( which are the six weierstrass points ) . by uniqueness , every holomorphic or antiholomorphic involution of @xmath19 commutes with @xmath83 ; @xcite . in particular , @xmath84 thus , the hyperelliptic involution @xmath83 on @xmath19 descends to an involution , denoted @xmath85 , on @xmath23 : @xmath86 the projection of the locus of the equation to the @xmath87-coordinate induces a holomorphic double cover @xmath88 ramified at the weierstrass points of @xmath19 . the presence of the real structure @xmath78 entails that the polynomial @xmath80 in may be assumed to have real coefficients , and that the involution @xmath89 restricts to the complex conjugation on the affine model , namely @xmath90 since the upperhalf plane in @xmath91 is a fundamental domain for the action of the complex conjugation , the points on @xmath92 can be represented by points in the closure of the upperhalf plane . more precisely , consider the northern hemisphere @xmath93 with the equator included . we will think of the surface @xmath23 as the ramified double cover @xmath94 induced by . such a cover is branched along the equator of the hemisphere @xmath95 as well as at three additional branch points , namely the projections to @xmath95 of the weierstrass points of @xmath19 . [ 31 ] let @xmath96 be the preimage of the northern hemisphere @xmath95 under the double cover @xmath97 of . clearly , @xmath98 is a torus with an open disk removed . let @xmath99 be the restriction of the involution @xmath78 of @xmath19 to the boundary circle @xmath100 . the map @xmath99 is the antipodal map on the boundary circle @xmath101 . the original surface @xmath26 can be viewed as the quotient space of @xmath98 by @xmath99 : @xmath102 by construction , the subsurface @xmath96 is @xmath83-invariant and its boundary @xmath100 is the fixed point set of the involution @xmath103 of on @xmath19 . we will reduce the problem to symmetric metrics , and then establish an optimal lower bound for the area of some collars in @xmath23 . similar considerations for general metrics , albeit with worse constants , can be found in @xcite . in this section , we begin the proof of the sharp systolic inequality of theorem [ theo : main1 ] by decomposing nonpositively curved dyck s surfaces into various regions . the rest of the proof will occupy the next two sections . the average metric @xmath104 of a riemannian metric @xmath105 on @xmath23 by the hyperelliptic involution is defined as @xmath106 we will need the following results regarding the average metric . [ lem:1 ] the average metric @xmath104 of a nonpositively curved riemannian metric @xmath105 on @xmath23 is similarly nonpositively curved . [ lem:2 ] let @xmath105 be a riemannian metric on @xmath23 . the average metric @xmath104 has a better systolic area than @xmath105 , that is , @xmath107 every nonpositively curved metric on dyck s surface @xmath26 satisfies @xmath108 furthermore , the equality case is attained by a nonpositively curved ( in the sense of alexandrov ) piecewise flat metric with one conical singularity of angle @xmath109 . by lemmas [ lem:1 ] and [ lem:2 ] , the nonpositively curved metric on @xmath23 ( and so its lift to @xmath110 ) may be assumed invariant under the hyperelliptic involution . we normalize the metric by rescaling it to unit systole , so that @xmath111 . these assumptions on the metric will be implicit throughout the article . + consider the surface @xmath112 as in definition [ 31 ] . recall that @xmath98 is a torus with an open disk removed , @xmath98 is endowed with the @xmath83- and @xmath78-invariant nonpositively curved metric inherited from @xmath113 . as the fixed - point set of the isometric involution @xmath103 , the boundary @xmath100 of @xmath98 is geodesic . [ lem : level ] relative to the normalisation @xmath111 , the level curves of @xmath98 at distance less than @xmath114 from @xmath100 are loops freely homotopic to @xmath100 . furthermore , they are of length at least @xmath39 . by morse theory , the level curve at distance @xmath115 from @xmath100 deforms to @xmath100 if the function @xmath116 has no singular value between @xmath117 and @xmath115 . let @xmath115 be the least value for which this is not the case . since @xmath98 is nonpositively curved , there exist two length - minimizing paths of length @xmath115 joining @xmath100 to the same critical point of @xmath118 on @xmath119 . furthermore , these two length - minimizing paths form a geodesic arc @xmath120 with endpoints in @xmath100 which induces a nontrivial class in @xmath121 . note that @xmath120 orthogonally meets @xmath100 at its endpoints . now , the hyperelliptic involution @xmath83 on @xmath98 induces the homomorphism @xmath122 on @xmath121 . thus , the arcs @xmath120 and @xmath123 lie in the same relative homotopy class in @xmath121 . from the flat strip theorem , these two geodesic arcs are parallel and bound with some arcs of @xmath100 a @xmath83-invariant flat rectangle . the center @xmath87 of this rectangle is clearly a weierstrass point and @xmath124 now , the segment joining @xmath100 to @xmath87 forms with its image by @xmath83 a geodesic arc @xmath125 with opposite endpoints on @xmath100 . this arc @xmath125 projects to a noncontractible loop of @xmath23 . hence the length of @xmath125 , which is twice @xmath126 , is at least @xmath45 . combined with , we derive the first part of the lemma , namely @xmath127 each of the two arcs of @xmath100 joining a pair of opposite points projects to a noncontractible loop of @xmath26 . therefore , the length of @xmath100 is at least @xmath39 . since @xmath98 is nonpositively curved and @xmath100 is a closed geodesic , every loop of @xmath98 freely homotopic to @xmath100 is of length at least @xmath39 , and so are the level curves of @xmath98 at distance less than @xmath115 from @xmath100 . the coarea inequality yields a first lower bound on the systolic area of a nonpositively curved metric on @xmath23 . [ prop : u ] consider a normalized nonpositively curved metric on @xmath128 invariant by the hyperelliptic involution and let @xmath129 . the @xmath130-tubular neighborhood @xmath131 of @xmath100 in @xmath98 is a topological cylinder which satisfies @xmath132 in particular , @xmath133 we introduce a decomposition of dyck s surface leading to the description of the extremal metric in section [ sec : extremal ] . this decomposition will also allow us to estimate the systolic area of nonpositively curved metrics . the decomposition is motivated by the following observation . if the systolic inequality were optimal , the extremal surface could be defined from a flat cylinder of circumference @xmath39 and height @xmath114 by identifying pairs of opposite points on one of its boundary component ( which leads to a mbius band ) and by gluing the other boundary component onto itself so as to obtain the right topological type for the surface . it turns out this is impossible without decreasing the systole , which shows that the inequality is not optimal . actually , we will see that the extremal surface decomposes into a flat mbius band and a torus with a disk removed . the flat mbius band is defined as previously from a flat cylinder of circumference @xmath39 and height @xmath134 , while the torus with a disk removed is made of three isometric flat hexagons . the conical singularities will correspond to the points where the various flat regions meet . the sizes of the cylinder and the hexagons must be chosen to make the gluing possible while minimizing the systolic area . the comparison with nonpositively curved metrics will be carried out afterwards . let us introduce some quantities related to the sizes of the cylinder and hexagons , whose geometric interpretations will appear in proposition [ prop : hexa ] . fix @xmath135 and @xmath136 such that @xmath137 note that @xmath138 more explicitly , we have @xmath139 let @xmath140 be defined as @xmath141 . we will use the notations and assumptions from the previous section . in particular , we assume that @xmath128 as in is endowed with a normalized nonpositively curved metric invariant under the hyperelliptic involution . we would like to decompose @xmath98 into @xmath142 and three voronoi cells centered at the three weierstrass points of @xmath98 . the main theorem will then follow from a comparison between the areas of these voronoi cells and those of some euclidean polygons . in order to describe the voronoi cells and their comparison euclidean polygons , it is convenient to proceed as follows . since @xmath113 is nonpositively curved , the open collar @xmath143 of width @xmath130 around the closed geodesic @xmath100 of @xmath113 is convex . removing this collar and gluing back the boundary components of @xmath144 yields a new @xmath145 genus two surface @xmath146 . we will identify the regions of @xmath146 with those of @xmath144 . let @xmath147 be the universal cover of @xmath146 . the voronoi cell of @xmath147 centered at a lift of a weierstrass point of @xmath146 is the region of @xmath147 formed of the points closer to this lift than to any other lift of a weierstrass point . the voronoi cells on @xmath147 are polygons whose edges are arcs of the equidistant curves between a pair of lifts of weierstrass points . note that these edges are not necessarily geodesics . since the metric is nonpositively curved , the voronoi cells on @xmath147 are topological disks , while their projections to @xmath113 or @xmath23 , still called voronoi cells , may have more complicated topology . note that the voronoi cells on @xmath147 have the same area as their projections to @xmath113 or @xmath23 . this is because the _ interior _ of a voronoi cell projects injectively . the voronoi cells on @xmath147 have the same area as their projections to @xmath113 or @xmath23 . this is because the _ interior _ of a voronoi cell projects injectively . furthermore , since the metric on @xmath146 is @xmath83-invariant , the voronoi cells on @xmath147 are symmetric with respect to their centers , and since it is also @xmath148-invariant , the projections of their boundaries to @xmath113 contain the boundary components of @xmath143 . + now assume @xmath149 is the lift of a weierstrass point , and let @xmath150 be the lift of another weierstrass point . denote by @xmath151 the preimage of @xmath152 by the exponential map @xmath153 from the tangent plane @xmath154 to @xmath147 at @xmath155 . define @xmath156 as the equidistant line in @xmath154 between the origin @xmath157 and @xmath158 . for every @xmath159 , set @xmath160 . by the rauch comparison theorem , the exponential map @xmath153 does not decrease distances . thus , @xmath161 consider the euclidean polygon in the tangent plane @xmath154 , obtained as the intersection of the halfspaces containing the origin , defined by the lines @xmath156 , as @xmath152 runs over all weierstrass points distinct from @xmath155 . this euclidean polygon will be referred to as the _ comparison euclidean polygon _ corresponding to the voronoi cell centered at @xmath155 . it follows from the inequality that the exponential image of this polygon is contained in the voronoi cell of @xmath155 . since the exponential map does not decrease distances , we obtain the following proposition . [ prop : voronoi ] the area of a voronoi cell is bounded from below by the area of its comparison euclidean polygon . by construction , @xmath98 decomposes into @xmath142 and three voronoi cells centered at the three weierstrass points of @xmath98 . + we conclude this section with some distance estimates on the centers of the voronoi cells of @xmath98 , that is , on the weierstrass points . [ lem : dist ] 1 . the distance between two weierstrass points of @xmath98 is at least @xmath114 . every weierstrass point of @xmath98 is at distance at least @xmath47 from @xmath131 . every minimizing segment between a pair of isolated branch points of the double cover @xmath162 lifts to a noncontractible loop of @xmath23 . thus , the distance between two weierstrass points of @xmath98 is at least @xmath114 . similarly , every minimizing segment between an isolated branch point of @xmath162 and the equator lifts to a noncontractible loop of @xmath23 . thus , every weierstrass point of @xmath98 is at distance at least @xmath114 from @xmath100 and so at distance at least @xmath47 from @xmath131 . in this section , we bound from below the area of the voronoi cells in some special case and describe the extremal nonpositively curved dyck s surface using the constructions and notation defined earlier . as previously , we assume that @xmath23 is endowed with a normalized nonpositively curved metric invariant under the hyperelliptic involution . this metric descends to a singular metric on @xmath163 under the ramified cover . [ 51 ] let @xmath164 on @xmath163 be the connected graph given by the projections of the edges of the voronoi cells to the hemisphere , see proposition [ prop : voronoi ] . denote by @xmath165 the spherical cap of @xmath163 bounded by the level curve @xmath166 of @xmath163 at distance @xmath130 from the equator . from lemma [ lem : level ] , the length of @xmath166 is at least @xmath45 . by construction , the graph @xmath164 lies in @xmath165 , contains @xmath166 and bounds @xmath167 faces . from the formula @xmath168 , where @xmath169 and @xmath170 are the numbers of vertices and edges of @xmath164 , and the well - known inequality @xmath171 , we derive that @xmath164 has at most @xmath62 edges and @xmath172 vertices . suppose that @xmath164 has three vertices lying in @xmath166 and a fourth one in the interior of @xmath165 from which arise three edges connecting the three other vertices on @xmath166 , figure [ fig : a ] . in other words , @xmath164 bounds three triangles in @xmath165 . ( 0,0)(0,0 ) ( 35,-100)@xmath163 ( 23,-58)@xmath166 , height=151 ] by construction , each of these triangles lifts to a hexagonal voronoi cell in @xmath147 whose comparison euclidean polygon is a ( symmetric ) hexagon . furthermore , the center of this euclidean hexagon @xmath173 is at distance at least @xmath174 from two pairs of opposite sides and at distance at least @xmath47 from the other pair of opposite sides , see lemma [ lem : dist ] + the following result provides a sharp lower bound on the area of the comparison euclidean hexagon and therefore on the hexagonal voronoi cells . [ prop : hexa ] let @xmath173 be the symmetric euclidean hexagon which is the comparison hexagon of the voronoi cell as above . then @xmath175 furthermore , the equality case is attained by a symmetric euclidean hexagon composed of six pairwise opposite isosceles triangles based at its center : four of them have height @xmath174 and main angle @xmath176 , and two of them have height @xmath47 and base @xmath49 , figure [ fig : h ] . here , @xmath47 and @xmath176 are defined in . ( 0,0)(0,0 ) ( 57,-32)@xmath177 ( -75,-85)@xmath178 ( -9,1)@xmath179 ( -7,-120)@xmath49 ( -75,-32)@xmath180 ( 57,-85)@xmath181 ( -14,-28)@xmath47 ( 37,-49)@xmath174 ( 13,-53)@xmath176 ( -9,-72)@xmath182 , height=151 ] [ rem : hexa ] note that the angles between two sides of length other than @xmath49 are equal to @xmath183 . let @xmath182 be the center of @xmath173 . denote by @xmath177 , @xmath179 , @xmath180 , @xmath184 , @xmath185 and @xmath186 the sides of @xmath173 ( in this order ) with @xmath187 opposite to @xmath188 such that , @xmath189 and @xmath190 , figure [ fig : h ] . the area of the triangle @xmath191 with vertex @xmath182 and side @xmath188 is bounded from below by @xmath192 where @xmath193 is the angle of @xmath191 at @xmath182 . thus , from the relation , we have @xmath194 where @xmath195 . the second inequality comes from the definition of convexity applied to the convex function @xmath196 between @xmath197 and @xmath198 . using a classical relation between @xmath199 and @xmath200 , we observe that this lower bound is minimal when @xmath201 that is , when @xmath202 from . the minimal lower bound is @xmath203 furthermore , the equality case occurs only if @xmath204 and @xmath205 are isosceles triangles of height @xmath114 and main angle @xmath176 , and @xmath206 is an isosceles triangle of height @xmath47 and main angle @xmath207 . from our choice of @xmath47 and @xmath176 , , the sides arising from the main vertices of these isosceles triangles have the same length , namely @xmath208 this shows that it is possible to put these isosceles triangles together in order to obtain an hexagon satisfying the desired constraints with area . in this case , the base of @xmath206 is of length @xmath209 from proposition [ prop : u ] and proposition [ prop : hexa ] , the nonpositively curved dyck s surface @xmath23 , which decomposes into the @xmath130-tubular neighborhood @xmath131 and three hexagonal voronoi cells , satisfies the following area lower bound @xmath210 this area lower bound is optimal . it is attained by the nonpositively curved ( in alexandrov s sense ) piecewise flat dyck s surface @xmath57 obtained as follows . glue three copies of the optimal flat hexagon described in proposition [ prop : hexa ] and identify the opposite sides of lengths other than @xmath49 , figure [ fig : b ] and remark [ rem : hexa ] . + ( 0,0)(0,0 ) ( 59,-61)@xmath211 ( -45,-1)@xmath211 ( -45,-123)@xmath211 ( -14,-71)@xmath80 the resulting surface is of boundary length @xmath39 . now attach a flat cylinder of circumference @xmath39 and height @xmath130 to it and identify the opposite boundary points . we obtain a nonpositively curved ( in alexandrov s sense ) piecewise flat dyck s surface @xmath57 with two conical singularities @xmath80 and @xmath211 of angle @xmath212 where the three hexagons meet , and six conical singularities of angle @xmath213 at the other vertices of the hexagons . its area satisfies @xmath214 the surface @xmath34 has unit systole . consider one of the two singular points ( corresponding to the north pole " ) . consider its cut locus in the tangent plane " . this is not actually a plane but a cone with total angle @xmath215 . the cut locus should be something like a regular hexagon with the six singular points lying on its boundary . it seems to me that the cut locus can be localized on this cone without any additional singularities ( other than those on the boundary ) . this would help clarify the picture at least as far as the injectivity radius from this particular point is concerned , and clarify the picture with the systole of the extremal surface . by construction , the extremal surface @xmath34 decomposes into an open flat mbius band @xmath216 and six trapezoids ; section [ sec : construction ] . the mbius band has unit systole . hence , the systole of @xmath34 is at most @xmath45 . furthermore , every geodesic arc of @xmath216 with endpoints on @xmath217 is of length at least @xmath218 . similarly , every geodesic arc of the complement @xmath219 with endpoints on @xmath217 is of length at least @xmath220 . thus , a noncontractible geodesic loop of @xmath34 intersecting @xmath216 is of length at least @xmath45 . consider now a systolic loop @xmath120 of @xmath34 which does not meet @xmath216 . denote by @xmath125 its projection to @xmath163 by the ramified cover . by construction , the loop @xmath125 lies in the spherical cap @xmath165 of @xmath163 , definition [ 51 ] . here , the hemisphere @xmath163 is endowed with the quotient metric from @xmath34 . furthermore , the loop @xmath125 surrounds more than one branch point in @xmath163 , otherwise its lift @xmath120 would be contractible in @xmath34 . suppose that @xmath125 is not simple . every arc of @xmath125 forming a simple loop surrounds exactly one branch point of in @xmath163 . indeed , if it surrounded exactly two branch points , it would lift to a noncontractible loop of @xmath34 shorter than @xmath120 , which is impossible . if it surrounded three branch points , it would be double covered by a loop of @xmath221 homotopic to @xmath217 and so of length at least @xmath45 , which is impossible since the systole of @xmath34 is at most @xmath45 . now , since @xmath125 surrounds more than one branch point , there exist two arcs of @xmath125 forming two simple loops surrounding two different branch points . from these two simple loops , we can form with the shorter path of @xmath125 joining them a loop in @xmath163 homotopic to a simple loop surrounding exactly two branch points . by smoothing out its corners , the loop we just formed can be made shorter than @xmath125 . this yields a contradiction as it lifts to a noncontractible loop of @xmath34 . in conclusion , the loop @xmath125 is simple and surrounds at least two branch points . now , if it surrounds three branch points , we already showed that it is of length at least @xmath45 . if it surrounds exactly two branch points , its lift @xmath120 is homotopic to one of the three geodesic loops of length @xmath45 made of the two segments connecting a pair of weierstrass points . in this case , the length of @xmath120 is also equal to @xmath45 since the metric is nonpositively curved . in this section , we complete the proof of theorem [ theo : main1 ] by showing that the other configurations for @xmath164 , definition [ 51 ] , correspond to nonpositively curved dyck s surfaces with larger area . + we start with some area estimates on the voronoi cells of a nonpositively curved dyck s surface @xmath23 with unit systole , whose metric is invariant by its hyperelliptic involution . [ lem : pi ] every voronoi cell of @xmath23 has area at least @xmath222 from lemma [ lem : dist ] , the centers of the voronoi cells are at distance at least @xmath114 from each other . in particular , every voronoi cell contains an embedded disk of radius @xmath223 . since the metric is nonpositively curved the area of this disk is at least @xmath224 . [ lem : x ] a voronoi cell of @xmath23 whose projection to @xmath163 has an edge of length @xmath87 lying in @xmath166 has area at least @xmath225 . the comparison euclidean polygon of such a voronoi cell is a convex polygon of @xmath226 , symmetric with respect to its center @xmath182 , with two opposite sides of length @xmath87 at distance at least @xmath47 from @xmath182 . these two opposite sides span a parallelogram lying in the euclidean polygon , which clearly satisfies the desired area lower bound . [ lem:2edges ] a voronoi cell of @xmath23 whose projection to @xmath163 is bounded by exactly two edges has area at least @xmath47 . the comparison euclidean polygon of such a voronoi cell has four sides . it is a parallelogram with two opposite sides at distance at least @xmath220 from each other ; the other two sides are at distance at least @xmath45 from each other . hence its area is at least @xmath47 . recall that the graph @xmath164 ( see section [ sec : extremal ] , definition [ 51 ] ) has at most four vertices . clearly , the valence of each vertex is at least @xmath40 and at least one of the vertices lies in @xmath166 . we will consider four cases based on the number of vertices lying in @xmath166 . before starting our discussion , we observe that a face of @xmath165 can not be bounded by a single edge of @xmath164 , otherwise its comparison euclidean polygon would be bounded by two halfspaces , which is impossible . this observation and the restriction on the number of faces of @xmath165 will be implicitely used in the description of the different cases below . + suppose that only one vertex lies in @xmath166 . by assumption , there is a voronoi cell of @xmath23 whose projection to @xmath163 has @xmath166 as an edge . from lemma [ lem : x ] , the area of this cell is at least @xmath227 since the area of each of the other two cells of @xmath23 is at least @xmath224 , lemma [ lem : pi ] , we obtain using proposition [ prop : u ] that @xmath228 suppose that exactly two vertices lie in @xmath166 . the two edges of @xmath166 , of length @xmath87 and @xmath155 , are part of two different faces of @xmath165 . from lemma [ lem : x ] , the total area of the two corresponding voronoi cells of @xmath23 is at least @xmath229 . since the area of the third cell of @xmath23 is at least @xmath224 , lemma [ lem : pi ] , we conclude using proposition [ prop : u ] that @xmath230 suppose that exactly three vertices lie in @xmath166 . if a fourth vertex lies in the interior of @xmath165 , then we are in the situation already described in section [ sec : extremal ] . if there is no fourth vertex , two faces of @xmath165 are bounded by exactly two edges . by lemma [ lem:2edges ] , the total area of the two corresponding voronoi cells in @xmath23 is at least @xmath220 . we conclude as in case 2 . + suppose that four vertices lie in @xmath166 , which is the maximal number of vertices of @xmath164 . in this case , two faces of @xmath165 are bounded by exactly two edges and we conclude as in case 3 . + this proves that the optimal systolic inequality for nonpositively curved dyck s surfaces is given by , where the equality case is attained by the surface @xmath34 described at the end of section [ sec : extremal ] . the following proposition provides a description of the automorphism and symmetry groups of @xmath57 viewed as a riemannian klein surface . we also present the corresponding groups for its orientable double cover @xmath231 . [ prop : sym ] 1 . the automorphism group and the symmetry group of @xmath57 are both isomorphic to @xmath58 2 . the holomorphic and antiholomorphic automorphism group and the symmetry group of @xmath231 are both isomorphic to @xmath232 in particular , the orientable double cover of @xmath57 is not conformally equivalent to bolza s curve . the natural homomorphism between the automorphism ( resp . symmetry ) group of @xmath57 and the permutation group @xmath59 of its weierstrass points is surjective . its kernel is composed of dianalytic ( _ i.e . _ locally holomorphic or antiholomorphic ) automorphisms preserving the weierstrass points . the only two automorphisms with this property are the identity map and the hyperelliptic involution ( which is an isometry ) . since they commute with every holomorphic map ( and so every isometry ) , @xcite , we deduce that the automorphism ( resp . symmetry ) group of @xmath57 is isomorphic to @xmath233 ( see subsection [ 24 ] for a discussion of the related automorphism group of the inscribed prism ) . similarly , the symmetry group of @xmath231 acts by isometries on the right triangular prism inscribed in @xmath43 whose vertices are the branch points of the holomorphic ramified double cover over @xmath43 , subsection [ 24 ] . since the isometry group of this prism is isomorphic to @xmath234 , we obtain an epimorphism @xmath235 whose kernel is generated by the hyperelliptic involution , @xcite . as the hyperelliptic involution commutes with every holomorphic and antiholomorphic map , we deduce that the symmetry group of @xmath231 is isomorphic to @xmath232 now , the metric on @xmath231 averaged by the holomorphic and antiholomorphic automorphism group of @xmath231 descends to a metric @xmath105 on @xmath236 . the proofs of lemmas [ lem:1 ] and [ lem:2 ] show that this new metric @xmath105 is nonpositively curved with a systolic ratio at least as good as the one of @xmath57 . since @xmath231 is extremal , we deduce from the equality case between the systolic ratios that @xmath105 agrees with the extremal metric on @xmath57 . thus , the holomorphic and antiholomorphic automorphism group of @xmath231 is contained in its symmetry group . as the opposite inclusion is clear , we derive the desired isomorphism . in this section , we compare the conformal classes of three genus two riemann surfaces that are significant for various systolic extremality problems . more specifically , we consider the bolza surface @xmath237 and the two orientable double covers of the extremal hyperbolic dyck s surface @xmath35 and the extremal nonpositively curved dyck s surface @xmath34 . the following well - known result shows that the bolza surface is not the orientable double cover of any dyck s surface . [ prop : bolza ] the bolza surface @xmath237 admits no fixed point - free antiholomorphic involution . suppose there is a fixed point - free antiholomorphic involution @xmath78 on @xmath237 . recall that @xmath78 commutes with the hyperelliptic involution @xmath83 . both quotients @xmath238 and @xmath239 admit a ramified double cover over the hemisphere @xmath163 , section [ sec : conf ] . the boundary of @xmath163 is double covered by a loop of @xmath237 on which both @xmath83 and @xmath78 act as an antipodal map . in the setting of section [ sec : conf ] , the projection of this loop to @xmath43 is an equator with no branch point lying in it . the quotient map @xmath240 induced by @xmath78 is an anticonformal map which fixes pointwise the equator and switches the two hemispheres . thus , the composition @xmath241 of @xmath242 with the reflexion @xmath243 along the equator conformally acts on each hemisphere fixing pointwise the equator . by applying the cauchy integral formula to the holomorphic map on each hemisphere or simply by complex analytic continuation , we deduce that the composition @xmath241 is the identity map . thus , the map @xmath242 is the reflexion along this equator containing no branch point . since @xmath242 preserves the branch points of the ramified double cover @xmath244 , we derive a contradiction . indeed , the branch points of the bolza surface form a regular octahedron on @xmath43 and all the reflexions of @xmath43 acting on this octahedron fix at least one vertex , which is not the case of @xmath242 . we introduce a conformal invariant which will allow us to distinguish the conformal classes of the extremal nonpositively curved dyck s surface @xmath34 and the extremal hyperbolic dyck s surface @xmath35 . + recall that the extremal hyperbolic dyck s surface @xmath35 described in @xcite and @xcite is obtained by identifying opposite pairs of points on the boundary component of the maximal hyperbolic surface of signature @xmath245 with boundary length @xmath246 @xcite . the term maximal " refers to a hyperbolic surface with fixed geodesic boundary length whose systole is maximal . the systole of the extremal hyperbolic dyck s surface @xmath35 is equal to @xmath247 . furthermore , @xmath35 has the same isometry group @xmath248 as @xmath34 , which is isomorphic to @xmath249 , proposition [ prop : sym ] . consider a @xmath248-invariant conformal structure on dyck s surface @xmath23 ( _ e.g. _ , the conformal structure of @xmath34 or @xmath35 ) . the order two element in the center of @xmath248 is the hyperelliptic involution . the axes of the order two symmetries in the subgroup @xmath59 meet at exactly two points @xmath80 and @xmath211 in @xmath23 , figure [ fig : b ] . cut open @xmath23 along the three segments of these axes with endpoints @xmath80 and @xmath211 which pass through a weierstrass point . the union of the fixed - point sets of the order two automorphisms of @xmath23 defines a graph . by removing the edges of this graph which meet the ramification locus of at non - weierstrass points , we obtain a graph @xmath164 on @xmath236 . the _ collar ( or annulus ) corresponding to the conformal structure of @xmath23 _ is the orientable double cover @xmath250 of the surface obtained by cutting @xmath23 open along @xmath164 . by definition , it only depends on the conformal structure of the dyck s surface . for example , the graph @xmath164 on the extremal nonnegatively curved dyck s surface agrees with the outer boundary component of @xmath50 after identification of the opposite sides , figure [ fig:0 ] . the _ soul _ of the open collar @xmath250 is the simple loop @xmath2 defined as the ramification locus of @xmath251 ( since the collar is open the weierstrass points are excluded ) , see . let us recall the following definition . the _ capacity _ of a riemannian collar @xmath250 , with boundary components @xmath252 and @xmath253 , is defined as @xmath254 where @xmath255 runs over piecewise smooth functions on @xmath250 with @xmath256 on @xmath252 and @xmath257 on @xmath253 . the infimum is attained by the unique harmonic function satisfying the boundary conditions . the capacity is a conformal invariant . [ rem : cap ] by construction , the capacity of the collar corresponding to a @xmath248-invariant conformal structure of @xmath23 is a conformal invariant of the surface . in the rest of the article , we estimate the collar capacities for the extremal nonpositively curved dyck s surface @xmath34 and the extremal hyperbolic dyck s surface @xmath35 in the following proposition , we bound from above the capacity of the collar corresponding to the extremal nonpositively curved dyck s surface . [ prop : cap1 ] let @xmath36 be the collar corresponding to the conformal structure of the extremal nonpositively curved dyck s surface @xmath34 . then @xmath258 the surface @xmath34 is tiled by three flat `` hexagons '' ( flat hexagons with two flat rectangles attached to them ) centered at the weierstrass points , figure [ fig : b ] . by construction , the collar @xmath36 ( composed of a cylinder and half hexagons ) has piecewise geodesic boundary components . the simple loop @xmath2 decomposes the collar @xmath36 into two regions @xmath259 and @xmath260 . we define a piecewise smooth function @xmath255 on @xmath36 as follows @xmath261 since the points on the boundary components of @xmath36 are at distance at least @xmath114 from @xmath2 , the function @xmath255 is a test function for the capacity of @xmath36 , . thus , @xmath262 there is a unique minimizing ray @xmath263 from every point @xmath87 of @xmath36 to @xmath2 . the points @xmath87 of @xmath36 such that @xmath263 passes through a given conical singularity @xmath264 form a symmetric flat quadrilateral @xmath265 , with two right angles from which two edges of length @xmath47 meeting at @xmath264 with an angle @xmath176 arise , figure [ fig : d ] . recall that the constants @xmath47 and @xmath176 are defined in . ( 0,0)(0,0 ) ( -45,-70)@xmath47 ( 30,-70)@xmath47 ( 0,-85)@xmath266 ( -8,-122)@xmath264 , height=151 ] the points of @xmath265 outside the disk of radius @xmath47 centered at @xmath264 are at distance at least @xmath114 from @xmath2 . they form a region of area @xmath267 \ , h^2.\ ] ] furthermore , the quadrilaterals @xmath265 are disjoint as @xmath264 runs over the conical singularities of @xmath36 . continuing with , we obtain the following upper bound for the capacity of @xmath36 using @xmath268 \ , h^2 \\ & \leq & 2.28308.\end{aligned}\ ] ] we will need the following lower bound on the capacity of a collar . this bound was established by b. muetzel ( * ? ? ? * lemma 2.2 ) in a more general form . we include a proof for the reader s convenience . [ lem : capa ] consider a hyperbolic collar @xmath250 around a closed geodesic loop of length @xmath269 parametrized in fermi coordinates by @xmath270 then @xmath271 where @xmath272 . in fermi coordinates , the hyperbolic metric on @xmath250 can be expressed as @xmath273 , @xcite . let @xmath274 be the unit vector field on @xmath250 induced by @xmath275 . for every piecewise smooth function @xmath255 on @xmath250 with @xmath256 on @xmath252 and @xmath257 on @xmath276 , we have @xmath277 given a continuous function @xmath278 \to ( 0,\infty)$ ] ( in our case , @xmath279 ) , we want to minimize the integral @xmath280 where @xmath281 \to { { \mathbb r}}$ ] is a piecewise smooth function with @xmath282 and @xmath283 . let @xmath173 be a primitive of @xmath284 . making the change of variables @xmath285 , we obtain @xmath286 ^ 2 \ , d\tau \nonumber \\ & = \int_{h(a)}^{h(b ) } ( f \circ h^{-1})'(\tau)^2 \ , d\tau \label{eq : fh}\end{aligned}\ ] ] since @xmath287 . by the cauchy - schwarz inequality , we have @xmath288 hence , @xmath289 therefore , @xmath290 in the following proposition , we bound from below the capacity of the collar corresponding to the extremal hyperbolic dyck s surface . [ prop : cap2 ] let @xmath37 be the collar corresponding to the conformal structure of the extremal hyperbolic dyck s surface @xmath35 . then @xmath291 the surface @xmath35 is tiled by three hyperbolic hexagons centered at the weierstrass points , @xcite . by construction , the collar @xmath37 is made of half hexagons with one side lying in @xmath2 and has piecewise geodesic boundary components . each tiling hyperbolic hexagon of @xmath35 decomposes into four isometric trirectangles with an acute angle equal to @xmath292 . the sides @xmath293 and @xmath294 opposite to the acute angle are of length @xmath295 and @xmath296 , with the shorter side @xmath294 lying in @xmath2 . observe that @xmath37 is composed of exactly @xmath297 such trirectangles @xmath298 . from the hyperbolic formula @xcite ) for trirectangles , the geodesic arc of @xmath298 orthogonal to @xmath294 at the point at distance @xmath299 from @xmath293 , figure [ fig : c ] , is of length @xmath300.\ ] ] ( 0,0)(0,0 ) ( -47,-60)@xmath293 ( 15,-112)@xmath294 ( -25,-122)@xmath299 ( 22,-19)@xmath292 , height=151 ] in fermi coordinates , the collars @xmath37 is parametrized by @xmath301 where @xmath302 agrees with for @xmath303 ( the other values can be derived by symmetry ) . from muetzel s lemma [ lem : capa ] , we have @xmath304 where @xmath272 . from proposition [ prop : cap1 ] , proposition [ prop : cap2 ] and remark [ rem : cap ] , we immediately derive the following result . the extremal nonpositively curved dyck s surface @xmath34 is not conformally equivalent to the extremal hyperbolic dyck s surface @xmath35 . simpler bounds on the capacities of the collars can be derived both for @xmath37 and @xmath36 as follows . the collar @xmath37 can be isometrically embedded into the bi - infinite hyperbolic cylinder with @xmath2 as a simple geodesic loop . in this cylinder , the collar @xmath37 is contained in the tubular neighborhood @xmath305 of @xmath2 of width the length of the side opposite to @xmath293 in the trirectangle @xmath298 , figure [ fig : c ] . we deduce that the capacity of the collar @xmath37 is bounded from below by the capacity of @xmath305 for which a formula has been established by buser and sarnak @xcite . even more directly , the capacity of the collar @xmath36 is bounded from above by twice the area of @xmath34 . however , none of these estimates is strong enough for our purpose . this explains why we made use of finer estimates . we are grateful to bjoern muetzel , hugo parlier and robert silhol for helpful comments . ambrosio , l. ; katz , m. : flat currents modulo @xmath80 in metric spaces and filling radius inequalities , _ comm . * 86 * ( 2011 ) , no . 3 , 557591 . see http://dx.doi.org/10.4171/cmh/234 and arxiv:1004.1374 . bangert , v ; croke , c. ; ivanov , s. ; katz , m. : filling area conjecture and ovalless real hyperelliptic surfaces . _ geometric and functional analysis ( gafa ) _ * 15 * ( 2005 ) no . 3 , 577 - 597 . see arxiv : math.dg/0405583 gromov , m. : metric structures for riemannian and non - riemannian spaces . based on the 1981 french original . with appendices by m. katz , p. pansu and s. semmes . translated from the french by sean michael bates . reprint of the 2001 english edition . modern birkhuser classics . birkhuser boston , inc . , boston , ma , 2007 . katz , k. ; katz , m. ; sabourau , s. ; shnider , s. ; weinberger , sh . : relative systoles of relative - essential @xmath39-complexes . _ algebraic and geometric topology _ * 11 * ( 2011 ) , 197 - 217 . see arxiv:0911.4265 . klein , c. ; kokotov , a. ; korotkin , d. : extremal properties of the determinant of the laplacian in the bergman metric on the moduli space of genus two riemann surfaces . z. _ * 261 * ( 2009 ) , no . 1 , 73108 . see arxiv : math/0511217 . muetzel , b. : inequalities for the capacity of non - contractible annuli on cylinders of constant and variable negative curvature . _ geometriae dedicata _ ( online first ) .

we study the statistical nature of primordial fluctuations from an anisotropic inflation which is realized by a vector field coupled to an inflaton . we find a suitable gauge , which we call the canonical gauge , for anisotropic inflation by generalizing the flat slicing gauge in conventional isotropic inflation . using the canonical gauge , we reveal the structure of the couplings between curvature perturbations , vector waves , and gravitational waves . we identify two sources of anisotropy , i.e. the anisotropy due to the anisotropic expansion of the universe and that due to the anisotropic couplings among variables . it turns out that the latter effect is dominant . since the coupling between the curvature perturbations and vector waves is the strongest one , the statistical anisotropy in the curvature perturbations is larger than that in gravitational waves . we find the cross correlation between the curvature perturbations and gravitational waves which never occurs in conventional inflation . we also find the linear polarization of gravitational waves . finally , we discuss cosmological implication of our results .

1003.0056

the primordial fluctuations from inflation is supposed to be statistically isotropic , gaussian , and scale invariant . the nature of fluctuations is associated with the nature of de sitter spacetime . however , since the expansion during inflation is not exactly de sitter , the power spectrum is slightly tilted by the order of the slow roll parameter @xcite which characterizes the deviation of the expansion from the exact de sitter expansion . the deviation from the gaussianity is also known to be related to the slow roll parameter @xcite . on the other hand , the statistical isotropy has been regarded as a robust prediction so far because the cosmic no - hair conjecture is thought to be robust @xcite . from an observational point of view , there are various indications that there exists statistical anisotropy in the cosmic microwave background radiation ( cmb ) @xcite . although the statistical significance of these anomalies is still under debate , the possibility of the statistical anisotropy certainly deserves further theoretical investigation @xcite . recently , breaking the statistical isotropy through the vector fields in an inflationary universe is proposed in the paper @xcite and extended in various ways @xcite . however , if the vector field is relevant to inflation , it may also produce anisotropy in an inflationary universe whatever small it is , which seems to contradict the cosmic no - hair conjecture . from the above perspective , it is interesting to ask if it is possible to have anisotropic inflationary universe @xcite . if possible , it provides a simple mechanism to break the statistical isotropy by breaking the isotropy of the spacetime @xcite . in the light of no - hair conjecture @xcite , one may deny this possibility . in fact , many attempts to construct anisotropic inflationary models suffer from the instability @xcite . however , recently , stable anisotropic inflationary models are found for the first time @xcite . this can be regarded as a counter example to the cosmic no - hair conjecture . the interesting point is that the deviation from isotropy is related to the slow roll parameter , namely , the deviation from the exact desitter expansion . of course , that means the degree of the anisotropy is quite small . from the point of view of precision cosmology , however , it is worth exploring theoretical fine structure in an inflationary scenario . in this paper , we study cosmological perturbations in an anisotropic inflationary scenario we have found . the expected phenomenology of the anisotropic inflation is as follows : * there should be statistical anisotropy in curvature perturbations . * there should be statistical anisotropy in gravitational waves . * there should exist the cross correlation between curvature perturbations and gravitational waves . * there should be linear polarization of gravitational waves . the first item will be tested by the planck @xcite . the second one may be detected through b - mode polarization in the cmb @xcite . the third one will imply t - b correlation in cmb @xcite . the last one could be important for the future direct measurement of gravitational waves through the interferometer @xcite . the purpose of this paper is to calculate the above quantities numerically and analytically and reveal the physics behind them . since the spacetime is anisotropic , the formalism treating perturbations is non - standard . although there are many works treating the cosmological perturbations in an anisotropic universe @xcite , there have been several obstructions in extracting concrete predictions for cmb . the main obstruction was the lack of the concrete anisotropic cosmological models . now , since we have such models , we have succeeded in obtaining concrete results by utilizing the canonical gauge which is a generalization of the flat slicing in the conventional isotropic inflationary scenario . recently , during our slow preparation of this paper , two papers have appeared on the archive @xcite . the first one @xcite studied the primordial perturbations in an anisotropic inflationary universe using a perturbative method . the second one @xcite investigated the same issue numerically . the conclusion is quite similar to ours . the main difference is the gauge used in analysis . our canonical gauge allows us to reveal the nature of primordial fluctuations from anisotropic inflation in a transparent way . interestingly , on the contrary to a naive expectation , all of these works including ours imply that even if the anisotropy of the universe is very small , a large statistical anisotropy in the spectrum of curvature perturbations could be created . the organization of this paper is as follows : in section ii , we review an anisotropic inflation which is caused by the inflaton coupled to the vector field . here , we will see the anisotropy is determined by the slow roll parameter . in section iii , we choose the canonical gauge which is a generalization of the flat slicing in the conventional inflation and classify perturbations in anisotropic universe based on the 2-dimensional rotation symmetry . then , we obtain the quadratic action for perturbed quantities . in section iv , we reduce the action to that for physical variables from which we can read off the structure of couplings between those variables . based on the reduced action , we calculate various statistical quantities numerically and analytically to reveal the nature of primordial fluctuations in anisotropic inflation . in section v , we discuss cosmological implication of our results . the final section is devoted to the conclusion . in the appendix a , we provide a detailed derivation of the action for 2-dimensional scalar sector perturbations . in this section , we review background solutions proposed in @xcite , and see how the anisotropic inflation is realized . we consider the vector field @xmath0 whose kinetic term is coupled to the inflaton field @xmath1 . we note that this kind of model is quite natural in the context of the supergravity @xcite . the action is given by @xmath2 \ , \label{eq : action1}\end{aligned}\ ] ] where @xmath3 is the reduced gravitational constant , @xmath4 is the determinant of the metric , @xmath5 is the ricci scalar , @xmath6 is the inflaton potential , @xmath7 is the field strength of the vector field defined by @xmath8 , and @xmath9 is a coupling function of the vector field . we assume that the background spacetime is given by the bianchi type i metric @xmath10 as for the vector field , it can be reduced into `` electric''(@xmath11 ) and `` magnetic''(@xmath12 ) components , and here we consider only the `` electric '' component for simplicity . it is not hard to prove that the direction of the `` electric '' field does not change in time by solving its evolution equation . without loosing the generality , one can take @xmath13-axis in the direction of the `` electric '' field . using the gauge invariance , we can express the vector field as @xmath14 . thus , there exists the rotational symmetry in the @xmath15-@xmath16 plane . given this configuration , it is convenient to parameterize the metric as follows : @xmath17 \ , \ ] ] where @xmath18 , @xmath19 and @xmath20 are an isotropic scale factor and spatial shears , respectively . we also define the averaged hubble parameter as @xmath21 . here , the lapse function @xmath22 is introduced to obtain the hamiltonian constraint . with the above ansatz , the action becomes @xmath23,\ ] ] where an overdot denotes the derivative with respect to the physical time @xmath24 . first , its variation with respect to @xmath20 yields @xmath25 this gives @xmath26 , hence , the anisotropy in the @xmath15-@xmath16 plane rapidly decays as the universe expands . hereafter , for simplicity , we assume @xmath27 and set the metric to be @xmath28 \ .\end{aligned}\ ] ] next , the equation of motion for @xmath29 is easily solved as @xmath30 where @xmath31 is a constant . taking the variation of the action with respect to @xmath32 and @xmath1 and substituting the solution ( [ eq : ax ] ) , we obtain the following basic equations : @xmath33 \ , \label{eq : hamiltonian}\\ % \ddot{\alpha } & = & -3\dot{\alpha}^2+\kappa ^2 v(\phi ) + \frac{\kappa ^2 p_a^2}{6}f(\phi ) ^{-2}e^{-4\alpha-4\sigma } \ , \label{eq : alpha}\\ % \ddot{\sigma } & = & -3\dot{\alpha}\dot{\sigma}+\frac{\kappa ^2 p_a^2}{3}f(\phi)^{-2}e^{-4\alpha -4\sigma } \ , \label{eq : sigma}\\ % \ddot{\phi } & = & -3\dot{\alpha}\dot{\phi}-v_{\phi}+p_a^2f(\phi)^{-3 } f_{\phi}e^{-4\alpha -4\sigma } \label{eq : inflaton } \ , \end{aligned}\ ] ] where the subscript in @xmath34 denotes derivative with respect to @xmath1 . let us check whether inflation occurs in this model . using eqs . ( [ eq : hamiltonian ] ) and ( [ eq : alpha ] ) , the equation for acceleration of the cosmic expansion is given by @xmath35 \ .\ ] ] we see that the potential energy of the inflaton needs to be dominant and the energy density of the vector field @xmath36 and the shear @xmath37 should be subdominant for inflation to occur . next we want to see if the anisotropy is produced during inflation . here we look at the ratio of the shear to the expansion rate @xmath38 to characterize the anisotropy of the inflationary universe . then , eq.([eq : sigma ] ) reads @xmath39 if the anisotropy converges to a value , i.e. @xmath40 becomes negligible , the terminal value should be given by @xmath41 where we used the slow roll equation @xmath42 derived from eq.([eq : hamiltonian ] ) and defined the energy ratio @xmath43 . in order to realize the above situation , @xmath44 must be almost constant . assuming that the vector field is subdominant in the evolution equation of the inflaton field eq.([eq : inflaton ] ) and the conventional single field slow - roll inflation is realized , one can show the coupling function @xmath9 should be proportional to @xmath45 to keep @xmath44 almost constant . in the slow roll phase , @xmath46-folding number @xmath47 is related to the inflaton field @xmath1 as @xmath48 as usual . then , the functional form of @xmath9 is determined as @xmath49 for the polynomial potential @xmath50 , for example , we have @xmath51 . in this scenario , the anisotropy is restricted from the condition that the vector field is negligible in eq.([eq : inflaton ] ) , that is , @xmath52 substituting eq.([eq : function ] ) into this , we obtain @xmath53 where @xmath54 is the slow - roll parameter defined by @xmath55 . thus , the anisotropy is constrained by @xmath56 . the above case is , in a sense , a critical one , and next we want to consider beyond it . for simplicity , we parameterize @xmath9 by @xmath57 where @xmath58 is a constant parameter . now , we look at what happens when @xmath59 . the basic equations become @xmath60 \ , \\ \ddot{\phi } & = & -3\dot{\alpha}\dot{\phi}-v_\phi \left [ 1-\frac{2c}{\epsilon_v } { \cal r } \right ] \ . \label{eq : inflaton2}\end{aligned}\ ] ] in this case , if the vector field is initially small @xmath61 , then the conventional single field slow - roll inflation is realized . during this stage @xmath62 and the vector field grows as @xmath63 . therefore , the vector field eventually becomes relevant to the inflaton dynamics eq.([eq : inflaton2 ] ) . nevertheless , the accelerating expansion of the universe will continue . the point is that @xmath64 can not exceed @xmath65 . in fact , if @xmath64 exceeds @xmath65 , the inflaton field @xmath1 does not roll down , which makes @xmath66 decrease . hence , @xmath67 always holds . in this sense , there exists an attractor where the inflation continues even when the vector field affects the inflaton evolution . the inflaton dynamics is determined by solving the slow - roll equation : @xmath68 assuming @xmath69 , this yields @xmath70 this can be integrated by neglecting the evolutions of @xmath71 as @xmath72 \ .\end{aligned}\ ] ] substituting this into the slow - roll equation eq.([eq : inflaton4 ] ) , we obtain @xmath73 ^{-1 } \ .\end{aligned}\ ] ] this clearly shows a transition from the conventional single field slow - roll inflationary phase , where @xmath74 holds , to what we refer to as the second inflationary phase , where the vector field is relevant to the inflaton dynamics and the inflaton gets @xmath75 times slower as @xmath76 . in the second inflationary phase , the energy density of the vector field becomes @xmath77 which yields the anisotropy of @xmath78 . from eqs.([eq : hamiltonian ] ) and ( [ eq : alpha ] ) , the slow - roll parameter defined in terms of the scale factor becomes @xmath79 where we neglected the anisotropy and used relations @xmath80 and @xmath69 . thus we have a remarkable result @xmath81 in the next section , we will make a perturbative analysis during this second inflationary phase . since the background is expanding anisotropically , we can not use the standard cosmological perturbation theory . in this section , we classify perturbations under the 2-dimensional rotational symmetry and obtain the quadratic actions for 2-dimensional scalar and vector sectors . in order to grasp the meaning of variables , we start with the isotropic case and make a gauge transformation from the flat slicing gauge to the appropriate gauge for 2-dimensional classification . then , the resultant gauge can be promoted to the anisotropic spacetime . the gauge we have chosen makes the analysis and the interpretation easier . once the gauge is fixed , the quadratic action can be calculated . first , we will start with the spatially homogeneous and isotropic universe . for simplicity , we consider flat space . @xmath82 \ , \end{aligned}\ ] ] where we took a conformal time @xmath83 . in that case , we can use 3-dimensional rotational symmetry to classify the perturbed metric . when we want to have diagonal quadratic action , we take the following gauge @xmath84\end{aligned}\ ] ] where we imposed @xmath85 and @xmath86 . if we ignore vector and tensor perturbations @xmath87 , the above gauge is called the flat slicing gauge . now , let us move on to the fourier space . since there exists 3-dimensional rotation symmetry , we can take a wave number vector to be @xmath88 . then , the perturbed metric has the following components : @xmath89 here , we utilized the special choice @xmath88 to solve the constraints @xmath85 and @xmath86 . with the same reason , only @xmath90 remains . we defined @xmath91 . now , we will pretend that we have only 2-dimensional rotation symmetry in @xmath15-@xmath16 plane . in that case , at best , we can take @xmath92 . hence , we make a rotation in the @xmath93-plane so that the wave number becomes @xmath92 . @xmath94 where we have a relation @xmath95 . under this rotation , the perturbed metric becomes @xmath96 to simplify the perturbations , we can make use of gauge transformation @xmath97 where the semicolon denotes the covariant derivative with respect to the background metric . taking the parameter @xmath98 we obtain @xmath99 it should be noted that we did not change slicing but performed only the spatial coordinate transformation . therefore , we are still working in the flat slicing where the 3-dimensional scalar curvature vanishes . in our anisotropic inflation models , the available symmetry is actually small . the background metric is given by @xmath100 that is , @xmath101 . notice that the conformal time in anisotropic inflation is the conformal time in 2-dimensional part @xmath102 . even in this anisotropic spacetime , as we have done in ( [ gauge - a ] ) , one can take the following gauge @xmath103 where we can impose further conditions so that the perturbed metric goes back to ( [ gauge - a ] ) in the isotropic limit . one can classify the perturbed metric using the rotational symmetry in @xmath104-plane . in 2-dimensional flat space , an arbitrary vector @xmath105 where @xmath106 can be decomposed into the scalar part @xmath107 and the vector part @xmath108 . since there exists no tensor part in 2-dimensions , 2-dimensional tensor can be constructed from the 2-dimensional vector . because of the symmetry , the scalar and vector parts are not mixed in the equations . thus , the metric perturbations can be classified to the scalar sector and the vector sector . thanks to the symmetry in the @xmath104 plane , without loss of generality , we can take the wave number vector to be @xmath109 . hence , the vector sector in 2-dimensional classification can be identified as @xmath110 in the above perturbed metric . the remaining components @xmath111 belong to the scalar sector . thus , the perturbations that belong to 2d vector perturbations , can be written as @xmath112 where we have incorporated the anisotropy while keeping the spatial scalar curvature to be zero . as to the vector field , we can take @xmath113 note that we have no residual gauge transformation and , in particular , @xmath114 is a gauge invariant under abelian gauge transformations . and , as we have seen in ( [ gauge - a ] ) , @xmath115 corresponds to the cross - mode polarization of gravitational waves in the isotropic limit @xmath116 . using this gauge , we can calculate the quadratic action as @xmath117 \ .\end{aligned}\ ] ] since the perturbed shift function @xmath118 does not have a time derivative , it is not dynamical . there are two physical degrees of freedom @xmath115 and @xmath114 in this 2-dimensional vector sector . for the 2-dimensional scalar sector , we define the metric perturbations @xmath119 where we have kept the spatial scalar curvature vanishing . the scalar perturbation will be represented by @xmath120 . the variable @xmath121 and @xmath120 are the gauge invariant variables that correspond to the plus mode of gravitational waves and the scalar perturbations , respectively , in the isotropized limit @xmath116 . and , we set the perturbed vector to be @xmath122 where we have fixed the abelian gauge by putting the longitudinal component to be zero . from these ansatz , we can calculate the quadratic action as @xmath123\end{aligned}\ ] ] here , @xmath124 consists of @xmath125 and @xmath126 . among them , @xmath127 and @xmath128 are non - dynamical and can be eliminated . now , we are in a position to calculate the statistical properties of primordial fluctuations from anisotropic inflation . for this purpose , we need to reduce the action to the one for physical variables . then , we can quantize the system and specify the vacuum state . we analyze the vector sector and the scalar sector separately . first , let us eliminate non - dynamical variables from the action for the 2-dimensional vector sector . in fourier space , the action for 2-dimensional vector sector becomes @xmath129 \ , \end{aligned}\ ] ] where we omitted the fourier indices @xmath130 for simplicity . after making the above action real manifestly and completing the square of @xmath118 , we obtain @xmath131 where @xmath132 is given by @xmath133 which becomes constant in the isotropic limit @xmath116 . the first squared term vanishes after substituting the equation of motion for @xmath118 . now , we define canonically normalized variables as @xmath134 then , using the canonical variables and integrating by parts , we obtain the reduced action for physical variables @xmath135 \ . \label{vec - action}\end{aligned}\ ] ] in the isotropic limit @xmath116 , @xmath136 and @xmath137 represent the cross - mode of gravitational waves and vector waves , respectively . the second line in the action ( [ vec - action ] ) describes how both waves are interacting to each other . next , we use the slow roll approximation to simplify the action . to obtain the homogeneous background metric , we integrate the following equations @xmath138 by assuming @xmath139 . the resultant expressions are @xmath140 in this approximation , the universe shows anisotropic power law inflation . we should recall , in the second inflationary phase , the variable @xmath141 is given by @xmath142 note that the range @xmath143 for @xmath58 corresponds to @xmath144 for @xmath141 . the above approximation gives useful formula for the subsequent calculations @xmath145 \ , \quad \frac{a^{''}}{a}=(-\eta)^{-2}\left[~2 + 3\epsilon _ h~\right ] \ , \quad \frac{a^\prime}{a}\frac{b^\prime}{b } = ( -\eta)^{-2}\left[~1 + 2\epsilon _ h+i\epsilon _ h~\right ] \ , \nonumber\\ \frac{b^{'2}}{b^2 } & = & ( -\eta)^{-2 } \left[~1 + 2\epsilon _ h + 2 i\epsilon _ h~\right ] \ , \quad \frac{b^{''}}{b } = ( -\eta ) ^{-2 } \left[~2 + 3\epsilon _ h + 3 i \epsilon _ h~\right ] \ , \nonumber\\ \frac{k ^{'}}{k } & = & ( -\eta ) ^{-1 } \left[~-i\epsilon _ h~\right ] \frac{k_y^2}{k^2}\frac{a^2}{b^2 } \ , \quad \frac{k ^{''}}{k } = ( -\eta ) ^{-2 } \left[~-i\epsilon _ h~\right ] \frac{k_y^2}{k^2}\frac{a^2}{b^2 } \ . \label{parts}\end{aligned}\ ] ] from the background equations eqs.([eq : hamiltonian])-([eq : inflaton ] ) , it is easy to obtain @xmath146 using the formula ( [ parts ] ) , we obtain @xmath147 from eq . ( [ eq : ax ] ) , the background equation for the vector can be found as @xmath148 ^ { ' } = 0 \ .\end{aligned}\ ] ] from this equation , it is easy to deduce the relation @xmath149 \ , \end{aligned}\ ] ] where @xmath150 is a slow - roll parameter defined by @xmath151 of course , @xmath152 is not related to the conformal time @xmath83 . furthermore , we obtain @xmath153 \ .\end{aligned}\ ] ] substituting these results into the action , we obtain the action in the slow roll approximation @xmath154 |\bar{\gamma}| ^2 \right . \nonumber \\ & & \qquad \qquad + \frac{1}{2}|\bar{d}^{'}|^2 + \frac{1}{2 } \left [ -k^2 + ( -\eta)^{-2 } \left\ { 2 + 9\epsilon _ h -3\eta _ h + 6 i\epsilon \sin^2 \theta \right\ } \right ] |\bar{d}|^2 \nonumber \\ & & \qquad \qquad \left . + \frac{\sqrt{6i\epsilon_h}}{2}(-\eta)^{-1 } \sin \theta ( \bar{\gamma}^{'}\bar{d}^{*}+\bar{\gamma}^{*'}\bar{d } ) -\frac{\sqrt{6i\epsilon_h}}{2}(-\eta)^{-2 } \sin \theta ( \bar{\gamma}\bar{d}^{*}+\bar{\gamma}^{*}\bar{d } ) \right ] \ , \label{eq : lagbegin}\end{aligned}\ ] ] where we have defined @xmath155 . this @xmath156 represents the direction dependence . in the isotropic limit @xmath157 , the lagrangian for @xmath158 becomes the familiar one for gravitational waves in a friedman - lemaitre universe . in a similar way , we can derive the quadratic action for physical variables in the 2-dimensional scalar sector . the details can be found in the appendix a. moreover , it is straightforward to deduce the action in the slow roll approximation . the resultant action is given by @xmath159 \ , \label{sca - action } \\ l^{gg}&=&\frac{1}{2}|\bar{g}^{'}|^2 + \frac{1}{2 } \left[-k^2+(-\eta ) ^{-2 } \left\ { 2 + 3\epsilon _ h+3i\epsilon _ h + 3 i\epsilon _ h \sin^{2 } \theta \right\ } \right ] |\bar{g}|^2 , \label{gg } \\ l^{jj}&= & \frac{1}{2}|\bar{j}^{'}|^2 + \frac{1}{2 } \left [ -k^2 + ( -\eta ) ^{-2 } \left\ { 2 + 9\epsilon _ h -3\eta _ h - 6 i\epsilon _ \sin^2 \theta \right\ } \right ] |\bar{j}|^2 , \label{jj } \\ l^{\phi \phi } & = & \frac{1}{2 } | \delta\bar { \phi}^{'}|^2 + \frac{1}{2 } \left [ -k^2 + ( -\eta ) ^{-2 } \left\ { 2 + 9\epsilon _ h -\frac{3\eta _ h}{1-i}-\frac{12i}{1-i}+\left ( 12i \epsilon _ h + \frac{24i}{1-i } \right ) \sin^2 \theta \right\ } \right ] | \delta\bar{\phi}|^2 , \label{dphi } \\ l^{\phi g } & = & -3 i \sqrt{\frac{\epsilon _ h}{1-i } } ( -\eta ) ^{-2 } \sin^2\theta \left(\bar{g } \delta\bar{\phi}^ * + \bar{g}^ * \delta\bar{\phi } \right ) \ , \label{phi : g } \\ l^{\phi j } & = & \sqrt { \frac{6i}{1-i } } ( -\eta ) ^{-1 } \sin\theta \left(\delta\bar { \phi}^{*'}\bar{j}+\delta\bar{\phi}^{'}\bar{j}^{*}\right ) - \sqrt { \frac{6i}{1-i}}(-\eta ) ^{-2}\sin \theta \left(\delta\bar { \phi}^ { * } \bar{j}+\delta\bar{\phi}\bar{j}^*\right ) \ , \label{phi : j } \\ l ^{jg } & = & -\frac{\sqrt{6i \epsilon _ h}}{2 } ( -\eta ) ^{-1 } \sin\theta \left(\bar{g}^{*'}\bar{j } + \bar{g}^ { ' } \bar{j}^{*}\right ) + \frac{\sqrt{6i \epsilon _ h}}{2 } ( -\eta ) ^{-2}\sin\theta \left(\bar{g}^{*}\bar{j}+\bar{g}\bar{j}^*\right ) \ , \label{j : g}\end{aligned}\ ] ] where we defined canonical variables @xmath160 here , @xmath161 and @xmath162 represent the gravitational waves , the vector waves , and the scalar perturbations , respectively . the above action shows there exist the interaction among these variables . we notice the scalar part ( [ dphi ] ) contains @xmath141 without suppression by a slow - roll parameter @xmath163 . therefore , to obtain the quasi - scale invariant spectrum of curvature perturbation , @xmath141 itself has to be small . from the actions ( [ eq : lagbegin ] ) and ( [ sca - action ] ) , we see there are two sources of statistical anisotropy of fluctuations . first , the statistical anisotropy of fluctuations comes from the anisotropic expansion itself . intuitively , this can be understood from the anisotropic effective hawking temperature @xmath164 , where @xmath165 denotes the effective expansion rate . indeed , the expansion rate in the direction of the background vector is relatively small , hence the effective hawking temperature is low . then , this direction has less fluctuation power compared to the other directions . thus , the effective hawking temperature induces the anisotropy in the power spectrum of fluctuations . this effect is encoded in ( [ gg ] ) , ( [ jj ] ) , and ( [ dphi ] ) . the other source of the statistical anisotropy of fluctuations comes from the couplings ( [ phi : g ] ) , ( [ phi : j ] ) and ( [ j : g ] ) due to the background vector field . the essential structure of couplings can be understood without complicated calculations . take a look at the following term @xmath166 here , we should recall the order of magnitude of background quantities @xmath167 for example , to obtain the @xmath168 coupling , one of @xmath7 have to be replaced by the background quantity @xmath169 . hence , the coefficients in the @xmath168 coupling should be proportional to @xmath170 which is of the order of @xmath171 . this explains the strength of the coupling in ( [ j : g ] ) . similarly , @xmath172 coupling should be proportional to @xmath173 because we have to take the variation with respect to @xmath1 . hence , we can estimate its magnitude to be @xmath174 . this agrees to the interaction term ( [ phi : j ] ) . finally , the coupling @xmath175 has a magnitude of the order of @xmath176 which is proportional to @xmath177 . this shows a good agreement with the coupling ( [ phi : g ] ) . thus , we can understand why there is a hierarchy among the couplings of the gravitational waves , the vector waves and the scalar field . in this section , we will calculate power spectrum of various variables . to set the initial conditions , we need to quantize this system by promoting canonical variables to operators which satisfy the following canonical commutation relations @xmath178 = i \delta _ { ab } \delta({\bf x}-{\bf y } ) \ , \quad \left [ q_a(\eta , { \bf x } ) , q_b(\eta , { \bf y } ) \right ] = \left [ p_a(\eta , { \bf x } ) , p_b(\eta , { \bf y } ) \right ] = 0\end{aligned}\ ] ] where @xmath179 @xmath180 denote the variables @xmath181 in this order , while @xmath182 is their conjugate momentum defined by @xmath183 . the point is that , with a given wave number , the actions ( [ eq : lagbegin ] ) and ( [ sca - action ] ) reduce to those of independent harmonic oscillators in the subhorizon limit @xmath184 @xmath185 \ .\ ] ] we choose the bunch - davis vacuum state @xmath186 by imposing the following initial conditions at an initial time @xmath187 @xmath188 where @xmath189 is an annihilation operator whose commutation relations are given by @xmath190 = \delta_{ab}\delta ^{(3)}({\bf k - k^\prime } ) , \qquad \left [ a_{a , \bf k } , a_{b,\bf k^\prime } \right ] = 0 \ .\end{aligned}\ ] ] it is easy to verify the commutation relations imposed for @xmath191 . the bunch - davis vacuum @xmath186 is defined by @xmath192 . later , the system evolves according to the actions ( [ eq : lagbegin ] ) and ( [ sca - action ] ) and the variables @xmath193 are given by linear combination of @xmath194 as @xmath195 \ , \ ] ] where the transfer matrix @xmath196 depends on the wave number @xmath130 and the time @xmath83 . the time dependence is determined by solving the classical equations of motion with the initial conditions given by the coefficients in eqs.([eq : qini ] ) . and the expectation values of operators in this vacuum state are evaluated as @xmath197 especially , we are interested in the power spectrum of the scalar perturbations , cross mode and plus mode of gravitational waves , the cross correlation between scalar perturbations and the plus mode of gravitational waves , and the linear polarization of gravitational waves . the power spectrum of scalar perturbations is given by @xmath198 where we took into account the fact that the 2-dimensional vector sector is decoupled from the scalar sector . the power spectrum of the cross and plus mode of gravitational waves read @xmath199 here , we used the decoupling of the 2-dimensional scalar sector and the vector sector . we can also calculate the cross correlation between the plus mode of gravitational waves and the scalar perturbations @xmath200 the linear polarization of gravitational waves can be calculated once the power spectrum of the cross and plus mode are calculated . as a function of time . we set @xmath46-folding number to be zero at the time of horizon crossing of the given mode . the both axes are taken in log scale.,width=377 ] , where @xmath201 , as a function of time . the axes are in log scale . as one can see , the difference between two modes is quite small . , width=377 ] as a function of the time . the axes are in log scale.,width=377 ] as a function of time . the axes are in log scale.,width=377 ] now , we numerically calculate the transfer matrix @xmath196 with parameters @xmath202 and evaluate various statistical quantities . first of all , we need to calculate the statistical anisotropy in the curvature perturbations . we depict the time evolution of the anisotropy in the curvature perturbation @xmath203 as a function of time in fig.[fg : curv ] . taking look at fig.[fg : curv ] , we see the anisotropy grows with the squared of the e - folding number . the amplitude is larger than the expected one due to the coupling proportional to @xmath174 which is larger than the expected one @xmath171 . similarly , in fig.[fg : gw ] , we show the power spectrum of the cross and plus mode of gravitational waves @xmath204 where @xmath205 . next , we are interested in the cross correlation between the curvature perturbations and the gravitational waves . in fig.[fg : st ] , we depict the cross correlation between the curvature perturbation and the plus mode of gravitational waves which is normalized by the power spectrum of scalar perturbations @xmath206 in fig.[fg : st ] , we see the cross correlation between the scalar perturbations and the gravitational waves grows quadratically in the e - folding number . the amplitude is proportional to the coupling between @xmath121 and @xmath120 , i.e. , @xmath177 . the difference between the power spectrum of two polarization modes normalized by the power spectrum of the plus mode of gravitational waves @xmath207 characterizes the linear polarization of the gravitational wave . as you can see from fig.[fg : gw ] and fig.[fg : polar ] , the anisotropy of power spectrum grows quadratically in the e - folding number , while the difference between the two polarization modes shows the growth of higher power in e - folding number . in this section , we treat the anisotropy perturbatively and estimate its magnitude by using perturbation in the interaction picture . in the interaction picture , the expectation value for a physical quantity @xmath208 is given by @xmath209 { \cal o } ( \eta ) \left [ t \exp \left ( -i\int ^{\eta}_{\eta_i } h_i(\eta ^ { ' } ) d\eta ^ { ' } \right ) \right ] \right| 0 \right > \ , \ ] ] where @xmath210 is an in vacuum in the interaction picture , @xmath211 and @xmath212 denote a time - ordered and an anti - time - ordered product and @xmath213 denotes the interaction part of hamiltonian in this picture . this is equivalent to the following @xmath214 \cdots \right ] \right ] \right| 0 \right>.\ ] ] the successive approximation is made by truncating the summation at a certain order @xmath215 . in our analysis , we assume the noninteracting part of hamiltonian to be that of free fields in desitter spacetime @xmath216 \ , \ ] ] and the operators in the interaction picture are given by @xmath217 and the rest of the lagrangian ( [ eq : lagbegin])-([j : g ] ) is regarded as the interaction part @xmath218 . to see the leading effect on the anisotropy in the scalar perturbation , which is of the order of @xmath141 , we evaluate the correction due to the interaction given by @xmath219 \nonumber\\ & = & \int d^3k \left [ -\sqrt { \frac{6i}{1-i } } ( -\eta ) ^{-1 } \sin\theta \left(\bar{\delta \phi}^{\dagger'}\bar{j } + \bar{\delta\phi}^{'}\bar{j}^{\dagger}\right ) + \sqrt { \frac{6i}{1-i}}(-\eta ) ^{-2}\sin \theta \left(\bar{\delta \phi}^{\dagger } \bar{j } + \bar{\delta\phi } \bar{j}^{\dagger}\right ) \right ] \ .\end{aligned}\ ] ] note that in the analogy with the slow - roll parameter in the ordinary slow - roll inflation , the term proportional to @xmath220 in ( [ dphi ] ) can be expected to give the anisotropy @xmath221 where @xmath222 is the @xmath46-folding number from the horizon exit . thus , the leading correction comes from the interaction through the term @xmath223 . the leading correction is given by @xmath224 \right ] \right| 0 \right>.\ ] ] using eqs.([eq : nonint ] ) and commutation relations for the creation and annihilation operators , we obtain the anisotropy expressed as follows @xmath225 \nonumber\\ & \ & \ , \times { \rm i m } \left [ u(\eta_1 ) u^{*}(\eta _ 2)\left\ { -(-\eta _ 1)^{-1 } u^{'}(\eta _ 1)u^{*}(\eta ) + ( -\eta_1)^{-2 } u(\eta_1 ) u^{*}(\eta ) \right\ } \right ] \ , \end{aligned}\ ] ] where @xmath226 denotes the imaginary part . substituting the function form of @xmath227 ( [ eq : modefunc ] ) and introducing time variables @xmath228 , @xmath229 and @xmath230 , we have @xmath231 \nonumber\\ & \ & \ , \times \bigg [ \cos(-2\chi_1+\chi+\chi_2 ) \left ( 1+\frac{1}{\chi\chi_1}-\frac{1}{\chi\chi_2}+\frac{1}{\chi_1\chi_2 } \right ) \nonumber\\ & \ & \ , + \sin(-2\chi_1+\chi+\chi_2)\left ( -\frac{1}{\chi\chi_1\chi_2}+\frac{1}{\chi_1}-\frac{1}{\chi}-\frac{1}{\chi_2 } \right ) \bigg].\label{eq : integrand}\end{aligned}\ ] ] the contribution to the integral from the subhorizon @xmath232 is negligible . in the limit of superhorizon @xmath233 , we also have @xmath234 . hence , the integrand in eq.([eq : integrand ] ) approximately becomes @xmath235 . thus , the anisotropy can be evaluated as @xmath236 where @xmath237 is the @xmath46-folding number from the horizon exit . for the anisotropy of both two polarizations of gravitational waves , the similar calculations give @xmath238 where we used the interaction term in the action ( [ eq : lagbegin ] ) for @xmath158 and that in ( [ j : g ] ) for @xmath239 . it is interesting to calculate the cross correlation . the leading contribution comes from @xmath240 and @xmath241 . the result is as follows : @xmath242 as we will soon see , this might give a detectable number . on the other hand , it turns out that the linear polarization of the gravitational wave must be the effect of the order higher than @xmath141 . now , we are in a position to discuss cosmological implication of an anisotropic inflationary scenario . as we have listed up in section i , there are many interesting phenomenology in the anisotropic inflation . here , we summarize our results : * there exists statistical anisotropy in curvature perturbations of the order of @xmath243 . * there exists statistical anisotropy in gravitational waves of the order of @xmath244 . * these exists the cross correlation between curvature perturbations and gravitational waves of the order of @xmath245 . * there is small linear polarization of gravitational waves . due to the interaction on superhorizon scales , there is an enhancement factor @xmath246 in the above quantities . because of this enhancement , even when the anisotropy of the spacetime is quite small , say @xmath247 in our example , the statistical anisotropy imprinted in primordial fluctuations can not be negligible in precision cosmology . let us be more precise . the anisotropy is often parameterized by @xmath248 \ .\end{aligned}\ ] ] we have estimated @xmath249 which could be large . the current observational limit of the statistical anisotropy for the curvature perturbations is characterized by @xmath250 @xcite . let us suppose @xmath251 . then , we have the anisotropy in the gravitational waves with @xmath252 . the linear polarization is of the order of @xmath253 . using the definition of curvature perturbations @xmath254 , one can translate the cross correlation ( [ cross ] ) between the scalar perturbations and gravitational waves to that between the curvature perturbations and gravitational waves normalized by the power spectrum of curvature perturbations : @xmath255 where we used @xmath256 and @xmath257 . since the current constraints on the @xmath258 ratio is of the order of @xmath259 @xcite , we need to improve the accuracy by one more order , which might be achieved by planck . in @xcite , it is pointed out that the sign of @xmath260 is different from the observed one . however , it might be possible to modify the model so that the sign of @xmath260 is flipped . for example , we can consider two vector fields . then , the orthogonal direction to the plane determined by two vectors becomes a preferred direction . in this case , we can expect the sign of @xmath261 becomes opposite . we can also utilize anti - symmetric tensor fields to achieve the same aim . the details will be reported elsewhere . we have studied the statistical nature of primordial fluctuations from an anisotropic inflation which is realized by a vector field coupled to an inflaton . first , we have classified metric fluctuations according to the 2-dimensional rotational symmetry . to choose a convenient gauge in an anisotropic universe , we have started from the flat slicing gauge in an isotropic universe and made an appropriate gauge transformation to get a canonical gauge in an anisotropic universe . this gauge choice has made the subsequent analysis and the interpretation of the variables easier . using the canonical gauge , we have revealed the structure of the couplings between curvature perturbations , vector waves , and gravitational waves . we found that there are two sources for anisotropy , i.e. the anisotropy due to the anisotropic expansion of the universe and that due to the anisotropic couplings among variables . it turned out that the latter effect is dominant . we have numerically obtained power spectra . we also presented analytical formula using in - in formalism . since the coupling between the curvature perturbations and vector waves is the strongest one , the anisotropy in the curvature perturbations is larger than that in gravitational waves . more interestingly , we found the cross correlation between curvature perturbations and gravitational waves which is peculiar to anisotropic inflation . we also found the linear polarization of gravitational waves . although there are several mechanism to produce circular polarization in the primordial gravitational waves @xcite , this is the first example which realized the linear polarization in the primordial gravitational waves . we have only considered power spectrum for simplicity . however , as is pointed out in the paper @xcite , the statistical anisotropy could appear in the non - gaussianity strongly and modify the shape of the bispectrum and trispectrum . hence , it is interesting to study non - gaussianity in anisotropic inflation models . we can extend anisotropic inflation in various ways . although we have investigated a chaotic inflation in this paper , it is easy to extend the analysis to other inflation models . it is possible to incorporate multi - vector fields . from the string theory point of view , it is intriguing to consider anti - symmetric tensor field . it is also interesting to consider other bianchi type models @xcite in the context of anisotropic inflation . sk would like to thank the yitp members in kyoto for warm hospitality . a part of this work was done while sk was visiting yitp supported by jsps grant - in - aid for scientific research ( a ) 21244033 . sk is supported by an stfc rolling grant . js is supported by the japan - u.k . research cooperative program , grant - in - aid for scientific research fund of the ministry of education , science and culture of japan no.18540262 , grant - in - aid for scientific research on innovative area no.21111006 and the grant - in - aid for the global coe program the next generation of physics , spun from universality and emergence " . in this appendix , we present a detailed derivation of the action for the 2-dimensional scalar sector . here , we need to eliminate non - dynamical fields @xmath262 , @xmath263 , @xmath264 and @xmath265 . first , let us see the terms concerning @xmath262 in fourier space @xmath266 + { \rm c.c.}\ . \nonumber \\ \end{aligned}\ ] ] by completing the square as @xmath267 the variable @xmath262 can be eliminated because the first squared term vanishes after substituting the equation of motion for @xmath262 . similarly , the terms containing @xmath263 are given by @xmath268 + { \rm c.c.}\ . \end{aligned}\ ] ] completing the square gives @xmath269 again , we can ignore the first term . taking a look at the terms related to the variable @xmath264 , we see only the linear term in @xmath264 appears as @xmath270 + { \rm c.c . } \ . \nonumber\\\end{aligned}\ ] ] the variation with respect to @xmath264 gives @xmath271 \ . \label{eq : c}\ ] ] here we have defined @xmath272 after substituting this result into the action , we obtain the action for physical variables : @xmath273 \ .\end{aligned}\ ] ] the term @xmath274 is given by @xmath275 using the canonically normalized variable @xmath276 , we finally have @xmath277 |\bar{g}|^2.\end{aligned}\ ] ] the term @xmath278 reads @xmath279 using the canonically normalized variable @xmath280 , we can write @xmath281|\bar{\delta\phi}|^2 \ .\end{aligned}\ ] ] similarly , the term @xmath282 reads @xmath283 using the canonically normalized variable @xmath284 , we obtain @xmath285|\bar{j}|^2 \ .\end{aligned}\ ] ] the mixed term @xmath286 is given by @xmath287 \left(g\delta\phi^*+g^*\delta\phi\right ) \ .\end{aligned}\ ] ] or , using variables @xmath288 , we have @xmath289 \left(\bar{g}\bar{\delta\phi}^*+\bar{g}^*\bar{\delta\phi}\right ) \ .\end{aligned}\ ] ] the terms containing @xmath290 and @xmath291 are given by @xmath292 the above action can be rewritten by using variables @xmath293 and @xmath294 as @xmath295 finally , the terms containing @xmath291 and @xmath121 reads @xmath296 using variables @xmath297 , we obtain @xmath298 thus , we have obtained the action for the physical variables @xmath121 , @xmath291 , and @xmath120 . it is useful to check if @xmath291 , @xmath121 and @xmath120 decouple to each other in the special case @xmath299 for which there exists rotational symmetry in @xmath104 plane . indeed , we find @xmath300 the decoupling occurs because the helicity conserves in this special case . the other parts of action become @xmath301 |\bar{g}|^2 , \\ l^{\phi\phi } & = & \frac{1}{2 } |\bar{\delta\phi}^{\prime}|^2 + \frac{1}{2 } \bigg[-k_x^2+\frac{b^{''}}{b}-a^2v_{\phi\phi } -\frac{v^{\prime^2}}{a^2}\left(3f_{\phi}^2-ff_{\phi\phi}\right ) -\frac{b^2}{b^{\prime2}}\phi ^{'2 } \left(a^2v+\frac{f^2v^{'2}}{2a^2 } \right ) \nonumber\\ & \ & \hspace{6em } -2\frac{b}{b^\prime}\phi ^ { ' } \left(a^2v_{\phi}-\frac{ff_{\phi}v^{'2}}{a^2 } \right ) \bigg ] |\bar{\delta\phi}|^2 \ , \\ l^{jj } & = & \frac{1}{2 } |\bar{j}^{'}|^2 + \frac{1}{2}\left [ -k_x^2+\frac{f^{''}}{f } -2\frac{f^2v^{'2}}{a^2 } \right ] |\bar{j}|^2 \ .\end{aligned}\ ] ] e. komatsu _ et al . _ , arxiv:1001.4538 [ astro-ph.co ] . j. m. maldacena , jhep * 0305 * , 013 ( 2003 ) [ arxiv : astro - ph/0210603 ] . r. m. wald , phys . d * 28 * , 2118 ( 1983 ) . h. k. eriksen , f. k. hansen , a. j. banday , k. m. gorski and p. b. lilje , astrophys . j. * 605 * , 14 ( 2004 ) [ erratum - ibid . * 609 * , 1198 ( 2004 ) ] [ arxiv : astro - ph/0307507 ] ; f. k. hansen , a. j. banday and k. m. gorski , mon . not . soc . * 354 * , 641 ( 2004 ) [ arxiv : astro - ph/0404206 ] ; t. r. jaffe , a. j. banday , h. k. eriksen , k. m. gorski and f. k. hansen , astrophys . j. * 629 * , l1 ( 2005 ) [ arxiv : astro - ph/0503213 ] ; h. k. eriksen , a. j. banday , k. m. gorski , f. k. hansen and p. b. lilje , astrophys . j. * 660 * , l81 ( 2007 ) [ arxiv : astro - ph/0701089 ] ; a. de oliveira - costa , m. tegmark , m. zaldarriaga and a. hamilton , phys . d * 69 * , 063516 ( 2004 ) [ arxiv : astro - ph/0307282 ] ; k. land and j. magueijo , phys . lett . * 95 * , 071301 ( 2005 ) [ arxiv : astro - ph/0502237 ] ; k. land and j. magueijo , mon . not . soc . * 378 * , 153 ( 2007 ) [ arxiv : astro - ph/0611518 ] ; c. copi , d. huterer , d. schwarz and g. starkman , phys . d * 75 * , 023507 ( 2007 ) [ arxiv : astro - ph/0605135 ] ; j. hoftuft , h. k. eriksen , a. j. banday , k. m. gorski , f. k. hansen and p. b. lilje , astrophys . j. * 699 * , 985 ( 2009 ) [ arxiv:0903.1229 [ astro-ph.co ] ] ; p. k. samal , r. saha , p. jain and j. p. ralston , mon . not . soc . * 396 * , 511 ( 2009 ) [ arxiv:0811.1639 [ astro - ph ] ] . c. gordon , w. hu , d. huterer and t. m. crawford , phys . d * 72 * , 103002 ( 2005 ) [ arxiv : astro - ph/0509301 ] ; c. armendariz - picon , jcap * 0709 * , 014 ( 2007 ) [ arxiv:0705.1167 [ astro - ph ] ] ; d. c. rodrigues , phys . d * 77 * , 023534 ( 2008 ) [ arxiv:0708.1168 [ astro - ph ] ] ; c. y. tseng and m. b. wise , phys . d * 80 * , 103512 ( 2009 ) [ arxiv:0908.0543 [ astro-ph.co ] ] ; y. shtanov and h. pyatkovska , phys . d * 80 * , 023521 ( 2009 ) [ arxiv:0904.1887 [ gr - qc ] ] ; x. gao , arxiv:0903.1412 [ astro-ph.co ] ; r. battye and a. moss , phys . d * 80 * , 023531 ( 2009 ) [ arxiv:0905.3403 [ astro-ph.co ] ] . s. yokoyama and j. soda , jcap * 0808 * , 005 ( 2008 ) ; k. dimopoulos , d. h. lyth and y. rodriguez , arxiv:0809.1055 [ astro - ph ] ; t. kahniashvili , g. lavrelashvili and b. ratra , phys . d * 78 * , 063012 ( 2008 ) ; k. dimopoulos , m. karciauskas and j. m. wagstaff , phys . b * 683 * , 298 ( 2010 ) [ arxiv:0909.0475 [ hep - ph ] ] . e. dimastrogiovanni , n. bartolo , s. matarrese and a. riotto , arxiv:1001.4049 [ astro-ph.co ] . c. a. valenzuela - toledo , y. rodriguez and d. h. lyth , phys . rev . d * 80 * , 103519 ( 2009 ) [ arxiv:0909.4064 [ astro-ph.co ] ] . c. a. valenzuela - toledo and y. rodriguez , phys . b * 685 * , 120 ( 2010 ) [ arxiv:0910.4208 [ astro-ph.co ] ] . l. h. ford , phys . d * 40 * , 967 ( 1989 ) . n. kaloper , phys . d * 44 * , 2380 ( 1991 ) . s. kawai and j. soda , phys . d * 59 * , 063506 ( 1999 ) [ arxiv : gr - qc/9807060 ] . j. d. barrow and s. hervik , phys . d * 73 * , 023007 ( 2006 ) [ arxiv : gr - qc/0511127 ] . j. d. barrow and s. hervik , phys . d * 81 * , 023513 ( 2010 ) [ arxiv:0911.3805 [ gr - qc ] ] . l. campanelli , phys . d * 80 * , 063006 ( 2009 ) [ arxiv:0907.3703 [ astro-ph.co ] ] . a. golovnev , v. mukhanov and v. vanchurin , jcap * 0806 * , 009 ( 2008 ) [ arxiv:0802.2068 [ astro - ph ] ] ; s. kanno , m. kimura , j. soda and s. yokoyama , jcap * 0808 * , 034 ( 2008 ) . l. ackerman , s. m. carroll and m. b. wise , phys . d * 75 * , 083502 ( 2007 ) . b. himmetoglu , c. r. contaldi and m. peloso , arxiv:0809.2779 [ astro - ph ] ; b. himmetoglu , c. r. contaldi and m. peloso , arxiv:0812.1231 [ astro - ph ] ; b. himmetoglu , c. r. contaldi and m. peloso , phys . d * 80 * , 123530 ( 2009 ) [ arxiv:0909.3524 [ astro-ph.co ] ] . m. a. watanabe , s. kanno and j. soda , phys . lett . * 102 * , 191302 ( 2009 ) [ arxiv:0902.2833 [ hep - th ] ] . s. kanno , j. soda and m. a. watanabe , jcap * 0912 * , 009 ( 2009 ) [ arxiv:0908.3509 [ astro-ph.co ] ] . a. r. pullen and m. kamionkowski , phys . d * 76 * , 103529 ( 2007 ) ; n. e. groeneboom and h. k. eriksen , astrophys . j. * 690 * , 1807 ( 2009 ) ; c. armendariz - picon and l. pekowsky , arxiv:0807.2687 [ astro - ph ] . d. baumann _ et al . _ [ cmbpol study team collaboration ] , arxiv:0811.3919 [ astro - ph ] ; v. gluscevic and m. kamionkowski , arxiv:1002.1308 [ astro-ph.co ] . n. seto , s. kawamura and t. nakamura , phys . rev . lett . * 87 * , 221103 ( 2001 ) [ arxiv : astro - ph/0108011 ] . k. tomita and m. den , phys . d * 34 * , 3570 ( 1986 ) . k. s. dunsby , phys . d * 48 * , 3562 ( 1993 ) . h. noh and j. c. hwang , phys . d * 52 * , 1970 ( 1995 ) . t. s. pereira , c. pitrou and j. p. uzan , jcap * 0709 * , 006 ( 2007 ) [ arxiv:0707.0736 [ astro - ph ] ] ; c. pitrou , t. s. pereira and j. p. uzan , jcap * 0804 * , 004 ( 2008 ) [ arxiv:0801.3596 [ astro - ph ] ] . a. e. gumrukcuoglu , c. r. contaldi and m. peloso , jcap * 0711 * , 005 ( 2007 ) [ arxiv:0707.4179 [ astro - ph ] ] . b. himmetoglu , arxiv:0910.3235 [ astro-ph.co ] . t. r. dulaney and m. i. gresham , arxiv:1001.2301 [ astro-ph.co ] . a. e. gumrukcuoglu , b. himmetoglu and m. peloso , arxiv:1001.4088 [ astro-ph.co ] . j. martin and j. yokoyama , jcap * 0801 * , 025 ( 2008 ) . a. lue , l. m. wang and m. kamionkowski , phys . * 83 * , 1506 ( 1999 ) [ arxiv : astro - ph/9812088 ] ; s. alexander and j. martin , phys . d * 71 * , 063526 ( 2005 ) [ arxiv : hep - th/0410230 ] ; m. satoh , s. kanno and j. soda , phys . rev . d * 77 * , 023526 ( 2008 ) ; m. satoh and j. soda , jcap * 0809 * , 019 ( 2008 ) ; y. f. cai and y. s. piao , phys . b * 657 * , 1 ( 2007 ) [ arxiv : gr - qc/0701114 ] ; c. r. contaldi , j. magueijo and l. smolin , phys . rev . lett . * 101 * , 141101 ( 2008 ) [ arxiv:0806.3082 [ astro - ph ] ] ; t. takahashi and j. soda , phys . lett . * 102 * , 231301 ( 2009 ) [ arxiv:0904.0554 [ hep - th ] ] . dechant , a. n. lasenby and m. p. hobson , phys . d * 79 * , 043524 ( 2009 ) [ arxiv:0809.4335 [ gr - qc ] ] .

a quantum memory at microwave frequencies , able to store the state of multiple superconducting qubits for long times , is a key element for quantum information processing . electronic and nuclear spins are natural candidates for the storage medium as their coherence time can be well above one second . benefiting from these long coherence times requires to apply the refocusing techniques used in magnetic resonance , a major challenge in the context of hybrid quantum circuits . here we report the first implementation of such a scheme , using ensembles of nitrogen - vacancy ( nv ) centres in diamond coupled to a superconducting resonator , in a setup compatible with superconducting qubit technology . we implement the active reset of the nv spins into their ground state by optical pumping and their refocusing by hahn echo sequences . this enables the storage of multiple microwave pulses at the picowatt level and their retrieval after up tos , a three orders of magnitude improvement compared to previous experiments .

1401.7939

the ability to store a quantum state over long times is a desirable feature in many quantum information protocols . in the optical domain , quantum memories ( qm ) are necessary to implement the quantum repeaters needed for future long - distance quantum networks , and are the object of active research @xcite . quantum memories at microwave frequencies have also become of great interest in recent years because of the development of superconducting qubits which have their resonance frequency in the ghz range , in the perspective of implementing holographic quantum computing @xcite . for such schemes , the memory should act as an ideal multi - qubit register , able to store over long times the state of large numbers of qubits and to retrieve them on - demand . spin ensembles have emerged as promising candidates for a multi - mode microwave quantum memory because of their long coherence time @xcite and of the multiple collective modes that a spin ensemble withstands . existing proposals @xcite ( inspired by optical quantum memory protocols @xcite ) proceed in two distinct steps . first , the microwave field prepared in a well - defined quantum state @xmath1 ( for instance by a superconducting qubit ) is absorbed by the spin ensemble . this generates a transverse magnetisation which decays rapidly in a time @xmath2 due to the spread of resonance frequencies in the ensemble . given the weakness of the coupling constant of a single spin to the microwave field , efficient absorption requires embedding the ensemble in a high - quality factor microwave resonator in order to reach the so - called high - cooperativity regime @xcite . the second step of the memory operation consists in retrieving the initial state , by a series of operations after which the spins emit a microwave pulse in a quantum state as close as possible to @xmath1 . in @xcite , this is achieved by a hahn - echo - like sequence consisting of two consecutive @xmath3 pulses on the spins , combined with dynamical tuning of the resonator frequency and quality factor . the maximum storage time of the memory is approximately the hahn - echo decay time @xmath4 , so that the maximal number of stored quantum states is of order @xmath5 , a figure which can be very large in many spin systems . the first step of this protocol ( quantum state transfer ) has been demonstrated at the single - photon level in recent experiments @xcite ; the remaining obstacle to a microwave quantum memory is therefore the implementation of hahn - echo refocusing sequences at the quantum level in a hybrid quantum circuit . the object of this work is precisely to identify the challenges posed by this task and to demonstrate experimentally that they can be solved . for simplicity , we consider from now on a protocol simpler than the full qm @xcite but which constitutes an essential building block : the two - pulse echo ( 2pe ) . as depicted in fig . [ fig1]a , the 2pe consists in storing weak pulses @xmath6 into the spin ensemble at times @xmath7 , and applying a single refocusing pulse at time @xmath8 which triggers the emission of echo pulses @xmath9 at times @xmath10 ( therefore in reverse order ) in the detection waveguide @xcite . performing the 2pe at the quantum level imposes a number of requirements which represent experimental challenges . for quantum states to be well defined , thermal excitations should be absent from the system . this implies both that the spin ensemble has a high degree of polarisation and that the microwave field is in its ground state with high probability , which can only be achieved if the experiments are performed at millikelvin temperatures . at these temperatures however , spins tend to relax very slowly towards their ground state , and an active spin reset is therefore needed in order to repeat the experimental sequence at a reasonable rate ( @xmath11hz ) as requested by experiments at the single photon level . then , applying refocusing pulses to the spins requires large microwave powers potentially incompatible with the detection of quantum fields . finally , the echo emitted by the spins should faithfully restore the initial field , which implies that the echo recovery efficiency @xmath12 , that we define as the ratio of the energy radiated during the echo to the energy of the incoming pulse , should be close to @xmath13 . to summarise , reaching the quantum regime requires a mean excitation per mode ( both microwave and spin ) @xmath14 , input microwave fields with intra - cavity photon number @xmath15 , and an echo efficiency @xmath12 close to @xmath13 . are stored in the spin ensemble . a refocusing pulse @xmath16 acts as time - reversal for the spins and triggers the retrieval of the stored pulses as echoes @xmath9 in reverse order . top and bottom time lines show the applied , and the reflected and echo signals , respectively . ( b ) setup placed in a dilution refrigerator : the cavity is a lumped - element parallel lc resonator in niobium coupled to a coplanar waveguide by a capacitor @xmath17 . it consists of an interdigitated capacitor @xmath18 and a meander wire inductor @xmath19 creating the ac magnetic field shown in inset , for a @xmath20 incident microwave power at resonance . the spin ensemble consists of nv centres in a diamond monocrystal pressed on top of the inductor . laser pulses can be shone on it through an optical fibre glued to its top face . a tunable dc magnetic field @xmath21 is applied parallel to the @xmath22 $ ] direction of the crystal . ( c ) negatively - charged nv centres in diamond consist of a nitrogen atom next to a vacancy of the diamond lattice , having trapped an electron . their electronic spin @xmath23 is coupled by hyperfine interaction to the nitrogen nuclear spin @xmath24 ( for the @xmath25 isotope ) . half of the electronic spins ( sub - ensemble denoted n - orth in blue ) make an angle @xmath26 with @xmath21 , whereas the other half ( sub - ensemble orth in red ) is orthogonal to the field . ( d ) nv simplified energy diagram ( top ) showing the ground @xmath27 and the excited @xmath28 electronic states as well as the zeeman and hyperfine structure of @xmath27 , with @xmath29ghz the zero - field splitting . ( bottom ) magnetic field dependence of the allowed transitions for both n - orth ( blue ) and orth ( red ) sub - ensembles , showing respectively a linear and quadratic zeeman effect . nvs can be optically repumped in their @xmath30 ground state by application of green ( @xmath31 nm ) laser pulses exciting the @xmath27 - @xmath28 transition.,width=377 ] these stringent requirements have never been met in an experiment , by far . the multi - mode character of the 2pe has been recently benchmarked in the classical regime @xcite with an ensemble of phosphorus donors in silicon at @xmath32k in the three - dimensional microwave cavity of an electron paramagnetic resonance spectrometer . that experiment reached @xmath33 , @xmath34 , and an echo recovery efficiency @xmath35 . here we use negatively - charged nitrogen - vacancy ( nv ) centres in diamond , which are colour centres consisting of a substitutional nitrogen atom sitting next to a vacancy of the lattice ( see fig . [ fig1]c ) with properties suitable for a quantum memory : their spin triplet ( @xmath23 ) electronic ground state has a long coherence time @xcite and can be optically repumped in the spin ground state @xmath36 ( see figs . [ fig1]c and d ) . we re - visit the 2pe protocol with an ensemble of nv centres at @xmath37mk coupled to a planar superconducting resonator , in a setup compatible with hybrid quantum circuits , with active reset of the spin at the beginning of each experimental sequence , and we demonstrate the storage of multiple pulses at the picowatt level for @xmath38 , three orders of magnitude longer than in earlier experiments @xcite . our experiment reaches @xmath39 , @xmath40 , @xmath41 , and @xmath42 , and comes therefore closer to the quantum regime than previous work by several orders of magnitude . we quantitatively identify the present limitations and show that they can be solved in future experiments , opening the way to the implementation of quantum memory protocols . the experimental setup is sketched in fig . [ fig1]b ( see also suppl . a diamond crystal homogeneously doped with nv centres ( @xmath43\approx 2$]ppm ) is glued on top of the inductance of a planar superconducting lc resonator cooled in a dilution refrigerator . for optical pumping , @xmath31 nm laser light is injected through a single - mode optical fibre , glued on top of the crystal , @xmath44 mm above the resonator inductance . a magnetic field @xmath45 is applied parallel to the chip along the @xmath22 $ ] crystalline axis ( see fig . [ fig1]c ) . nv centres in their ground state are described @xcite by the hamiltonian @xmath46 $ ] , with @xmath47 ( resp . @xmath48 ) the spin operator of the @xmath23 nv electronic spin ( resp . the @xmath24 nitrogen nuclear spin ) , @xmath29ghz the zero - field splitting between states @xmath30 and @xmath49 , @xmath50mhz the hyperfine coupling , and @xmath51mhz the nuclear quadrupole momentum @xcite . local electric field and strain couple with strength @xmath12 the spin eigenstates @xmath52 @xcite . the energy eigenstates @xmath53 , shown in fig . [ fig1]d , are thus linear combinations of states @xmath52 ; in particular , at zero magnetic field , states @xmath54 are separated in energy by @xmath55 . in the experiment we use transitions between the spin ground state @xmath36 and the two excited states @xmath53 at frequencies close to the zero - field splitting . showing the resonator resonance at @xmath56ghz with quality factor @xmath57 , when the spins are saturated and do not contribute to the signal . ( b ) measured reflection coefficient modulus @xmath58 around the centre of the resonator line , showing the absorption by the spins for different magnetic fieds . top line ( 6 mt , black ) corresponds to all spins ( orth . and n - orth ) being far detuned and shows no absorption . other lines show several absorption peaks moving with magnetic field ( sub - ensemble n - orth ) or not ( sub - ensemble orth ) . ( c ) optical reset of the nv centre spins . the spins are first saturated by a @xmath59 long microwave pulse with frequency @xmath60 and applied power @xmath61dbm ; they are then optically repumped to their ground state with a laser pulse of power @xmath62 and duration @xmath63 ; after letting the system cool down during @xmath64 , the reflected amplitude of an applied weak ( @xmath65dbm ) @xmath66ms long measurement pulse at @xmath67ghz is measured . ( d ) reflected amplitude for @xmath68mw and different @xmath63 . the curves show the hyperfine split @xmath30 to @xmath69 spectroscopic transitions of the n - orth sub - ensemble , with an amplitude that increases with @xmath63 because of increasing spin re - polarisation . ( e ) corresponding imaginary part @xmath70 of the spin susceptibility . in addition , the dashed and dash - dotted lines show respectively @xmath70 measured at thermal equilibrium ( @xmath71mk , no saturating nor optical pulse ) and calculated ( see supplementary methods and supplementary figs.@xmath72 and @xmath73 ) and rescaled by a global factor to match the experiment at @xmath74s.,width=377 ] the resonator is capacitively coupled to measurement lines through which microwave signals are applied , the amplitude and phase of the reflected field being detected by homodyne demodulation after amplification at @xmath75k . the reflection coefficient @xmath76 , shown in figs . [ fig2]a and b , yields the resonator frequency @xmath77ghz and quality factor @xmath57 . such a low @xmath78 was chosen to avoid spin relaxation by superradiant spontaneous emission after excitation by the refocusing pulse @xcite . dips in @xmath58 are due to absorption by the nvs , as evidenced by their dependence on @xmath21 . to demonstrate optical repumping of the nvs in @xmath36 , we probe the spin polarisation after a laser pulse of power @xmath79 and duration @xmath63 , by measuring the absorption of a microwave pulse at @xmath80ghz . in addition to repumping the spins , the laser generates quasiparticles in the superconductor and carriers in the silicon substrate . we thus introduce a delay of @xmath64 between the two pulses for these excitations to relax . in order to start from a reproducible spin polarisation , a strong microwave pulse is applied before the laser pulse , which saturates all the spins at the beginning of each sequence ( see fig . [ fig2]c ) . for @xmath81mw , and on @xmath79 for @xmath82s . the experimental sequence is shown in fig . ( c ) nv spin relaxation time measurement . a series of @xmath83ms weak microwave pulses ( @xmath84dbm ) at @xmath67ghz , separated by @xmath85s , is applied following optical reset of the spins . blue dots are the average reflected amplitude of each pulse . a bi - exponential fit ( red solid line ) yields @xmath86s and @xmath87s.,width=325 ] the results are shown in fig . [ fig2]d for @xmath68mw . without laser pulse , the reflected pulse amplitude is independent of @xmath21 , proving that the spins are efficiently saturated by the initial microwave pulse . for non - zero @xmath63 , absorption peaks with the triplet shape characteristic of the nv hyperfine structure are observed , indicating sizeable nv polarisation . to quantify the effect , we convert the absorption signal into the imaginary part of the spin susceptibility @xmath88 ( see fig . [ fig2]e and supplementary information ) , which yields the relative spin polarisation @xmath89 , with @xmath90 the maximum repumping time . the polarisation increases with @xmath63 and then saturates ( see figs . [ fig2b]a and b ) , which shows that the spins reach the maximum polarisation allowed by optical pumping at @xmath31 nm , close to @xmath91 according to earlier work @xcite . the refrigerator cold stage was heated up to @xmath37mk due to laser power ; all the following results were obtained under these conditions . better alignment of the fibre with the resonator should reduce the power needed by two orders of magnitude . using the optical pumping , we measure the energy relaxation of the spins . in that goal the spins are first repumped , after which a series of a @xmath66ms resonant probe microwave pulse separated by @xmath32s are applied . the average reflected amplitude of each pulse is plotted in fig . [ fig2b]c and shows a bi - exponential response with time constants @xmath86s and @xmath87s , similar to recent measurements @xcite . these very long values confirm the need of actively resetting the spins for operating a qm . as a first step towards the application of refocusing pulses to the spins , we measure their time - domain response to microwave pulses of varying power . the experiments are performed at @xmath92mt . the zero - field spin susceptibility @xmath93 ( see fig . [ fig3]a ) shows two broad peaks corresponding to the @xmath94 and @xmath95 transitions . the width of these peaks is governed by the inhomogeneity of local electric fields and strain acting on the nvs , which results in a broad distribution of @xmath12 , causing the hyperfine structure to be barely resolved as seen in fig . [ fig3]a . on the @xmath95 transition , the spin absorption reaches a maximum at @xmath96ghz , that we will thus use as the frequency of all microwave pulses in the following . square microwave pulses of varying input power @xmath97 are sent to the sample , and their reflected amplitude @xmath98 is measured . the data are shown in fig . [ fig3]b and c , rescaled by @xmath99 , and compared to the reflected amplitude of the same microwave pulse with the spins initially saturated by a strong pulse . at low power ( the linear regime ) , after an initial transient where resonator and spins exchange energy , @xmath98 reaches half of the saturated value in steady state , indicating that the spins absorb @xmath100 of the incoming power . the steady - state value of @xmath98 increases with incoming power , indicating reduced spin absorption caused by progressive saturation of the ensemble . note that no clear rabi oscillations are observed . this is due to the spatial inhomogeneity of the microwave field generated by the planar resonator ( see fig . [ fig1]b ) , which causes a spread of rabi frequency within the ensemble ; in particular , this prevents the application of precise @xmath3 pulses to all the spins @xcite , which is an issue for hahn echo sequences . of the spin susceptibility at @xmath92mt . the calculated curve ( see suppl . methods ) was rescaled by a global factor to match the experiment . ( b ) reflected field amplitude @xmath98 for a square input microwave pulse of power @xmath97 . solid lines are experimental data with @xmath101 , @xmath102 , @xmath103 , and @xmath104dbm ( blue , green , yellow , and red ) ; dashed lines are simulations . the black curve is obtained when spins have been saturated by an initial strong pulse . the curves have been rescaled by @xmath105 for easier comparison , with @xmath106dbm . ( c ) in - phase ( blue ) and out - of - phase ( green ) quadrature of the reflected field for @xmath101dbm . ( d ) spin - echo sequence . an incoming microwave pulse @xmath107 with power @xmath102dbm is followed by a delay @xmath8 and a @xmath108 long refocusing pulse ( @xmath16 ) with power @xmath109dbm , yielding an echo _ e _ at time @xmath110 . saturation of the amplifiers ( shown in red ) limits the measurable amplitude to about @xmath111v . ( e ) experimental ( crosses ) and simulated and rescaled ( open circles ) area of the echo as a function of the refocusing pulse power @xmath112 . ( f ) measured ( crosses ) decay of the echo maximum amplitude as a function of @xmath8 . dashed and solid lines are an exponential fit yielding a characteristic time @xmath113 and a bi - exponential fit @xmath114 yielding @xmath115 and @xmath116 , respectively.,width=415 ] in order to understand in detail the spin dynamics , we compare the experimental data to the result of numerical simulations . these simulations consist of a number of mean value equations along the lines of @xcite and explained in further detail in the supplementary information . in particular , the inhomogeneity in both spin frequency and coupling strength is taken into account by dividing the ensemble into a sufficiently large set of homogeneous sub - ensembles and integrating the equations of motion for the resonator field and the spin components of all the sub - ensembles . the distribution of spin frequencies follows from the spin susceptibility shown in fig . [ fig4]a , and the distribution of coupling strengths depend on the resonator - field vacuum fluctuations , whose spatial distribution is calculated using the comsol simulation package and exemplified in the inset of fig . the actual distributions used are shown in supplementary fig.s5 . the simulations employed assume an ensemble of spin-1/2 particles , which is an approximation in the case of nv centres having a spin of 1 . however , in the linear , non - saturated regime this description is exact , and for the non - linear , saturated regime we expect the approximation to be justified since the applied pi pulse has a narrow frequency bandwidth and is tuned predominantly to the @xmath95 transition of the nv centres . in fig . [ fig3]b and c the measured and calculated reflected field are compared and show a convincing agreement , without any adjustable parameter . this confirms the validity of the calculations , both in the linear and non - linear regime , and proves in particular that the frequency distribution used is correct . despite the impossibility to apply well - defined @xmath3 pulses to the spins , we implement a spin - echo sequence with an initial microwave pulse creating a transverse magnetisation , followed after @xmath8 by a refocusing pulse . its power @xmath117dbm is chosen such that spin saturation is reached within the pulse duration , as requested for spin - echo . the reflected signal amplitude is shown in fig . [ fig3]d , with the expected spin - echo observed at @xmath110 . we have studied the amplitude of this echo as a function of @xmath112 , and compared this curve to the result of the simulations . the agreement is quantitative , as shown in fig . [ fig3]e ; in particular the power at which the echo amplitude saturates is well predicted by the simulations . this brings further evidence of the validity of calculated coupling strengths and of the spin-1/2 approximation . the dependence of the echo amplitude on @xmath8 is fitted by a bi - exponential function @xmath118 , with two different coherence times @xmath119 and @xmath120 , and @xmath121 and @xmath122 ( see fig . [ fig3]f ) . such a dependence is expected for an ensemble of nv centres in zero magnetic field . indeed , the coherence time of nv centres is limited by dipolar interactions with the surrounding spin bath , either paramagnetic impurities ( p1 centres ) or @xmath123 nuclear spins . this spin bath can be approximated as generating a fluctuating magnetic field that blurs the phase of the nv centre . in zero magnetic field , an interesting situation occurs : the nuclear spin state @xmath124 becomes immune to first order to magnetic fluctuations @xcite because of the strain - induced coupling between states @xmath49 which gives rise to an avoided level crossing , and thus to a transition frequency independent of magnetic field to first order ( see fig . [ fig1]d ) . this was shown in previous work to make the free - induction decay time @xmath2 one order of magnitude longer in zero magnetic field @xcite , and should equally lead to a longer hahn echo time @xmath4 . this is however not true for states with @xmath125 , which should therefore have a shorter decoherence time @xmath4 in zero magnetic field . more details will be given in future work . ( @xmath126 ) with an identical amplitude ( corresponding to @xmath127 photons in the resonator ) and different phases @xmath128 and @xmath129 , and a @xmath130db stronger refocusing pulse @xmath16 with phase @xmath131rad . ( b ) amplitude of the measured ( solid line ) and calculated ( dash - dotted line ) output signal showing the reflected pulses @xmath6 ( after partial absorption by the spins ) and @xmath16 ( its amplitude being trimmed by amplifier saturation , shown in red ) , as well as the six re - emitted echoes @xmath9 ( magnified by a factor 5 ) . inset : the comparison between the energies of the reflected @xmath6 pulses with the spins saturated ( black line ) or reset in their ground state ( blue line ) shows that about 75% of the incident power is absorbed by the spins . ( c ) iq quadratures of the output signal , showing that the @xmath9 pulses ( magnified by 10 ) are recovered with phase @xmath132 , as expected . ( d ) spin - echo @xmath133 of @xmath134 photons in the resonator for a low power incoming @xmath107 pulse populating the resonator with only @xmath135 photons . the refocusing pulse ( dashed line ) was suppressed in the room - temperature detection chain by a microwave switch to avoid saturating the follow - up amplifiers.,width=415 ] we finally implement the multi - mode 2pe protocol with weak microwave pulses . six consecutive microwave pulses with a varying phase and identical amplitude corresponding to @xmath136 photons in the resonator are first absorbed by the spin ensemble ; a strong refocusing pulse is then applied @xmath137 later ( see fig . [ fig4]a ) . the sequence is averaged @xmath138 times at a repetition rate of @xmath13hz , made possible by the active reset of the spins . as shown in fig . [ fig4]b , the six pulses are recovered after the refocusing pulse up to @xmath38 after their storage , with an amplitude reduced by @xmath139 compared to the incoming pulse , corresponding to @xmath140 photon in the resonator . as expected , the pulses are re - emitted in reverse order ( see fig . [ fig4]c ) . note that the strong refocusing pulse ( @xmath141 photons in the cavity ) does not prevent detection of fields at the single - photon level few microseconds later . we were able to detect a measurable spin - echo signal for pulses containing up to @xmath142 times lower energy , thus populating the resonator with @xmath41 photons on average ( see fig . [ fig4]d ) . an important figure of merit is the field retrieval efficiency @xmath12 , defined as discussed in the introduction as the ratio between the energy recovered during the echo and the energy of the incoming pulse . in the data shown in fig . [ fig4]b , @xmath12 is seen to decrease with @xmath8 due to spin decoherence , following approximately the relation @xmath143 , which yields @xmath144 for @xmath145 . coming back to the figures of merit defined in the introduction , our measurements reach @xmath39 , @xmath40 , @xmath41 , and @xmath42 , many orders of magnitude closer to the quantum regime than previous state - of - the - art experiments @xcite . reaching the quantum regime however requires a recovery efficiency @xmath12 close to @xmath13 , and therefore calls for a quantitative understanding of our measurements imperfections . in that goal we have performed simulations of the multi - mode 2pe protocol . as seen in fig . [ fig4]b the measurements are well reproduced , although a seven times higher efficiency @xmath146 is predicted . we attribute the discrepancy between @xmath147 and @xmath148 to the imperfect modelling of decoherence . indeed , our simulations treat spin decoherence in the markov approximation . this is not an adequate treatment since it is well - known that the spin bath environment displays strong memory effects . in particular this markov approximation is expected to describe improperly the dynamics of a spin under the action of a microwave drive , as happens during the refocusing pulse . this non - markovian bath causes the rabi oscillation of a single spin to decay faster than the spin - echo damping time @xmath4 as was observed in @xcite for instance . this effect is not included in our simulations and might explain the remaining discrepancy between theory and measurements . overall we infer from the simulations that @xmath148 would reach @xmath149 for a sample with infinite @xmath4 ; this number quantifies the reduced efficiency caused by refocusing pulse imperfections and finite spin absorption . in the measured efficiency @xmath150 , finite spin coherence causes a further @xmath151 reduction , thus appearing as the main limitation of the field retrieval efficiency in the present experiment . a one order of magnitude increase of the coherence time will thus be necessary to reach the quantum regime . this can be achieved @xcite with samples having a reduced concentration of nitrogen paramagnetic impurities as well as isotopic enrichment of @xmath152 . better refocusing could be obtained either by rapid adiabatic passage @xcite , or by tailoring the spin spatial distribution @xcite . these combined advances should make possible to reach the figures of merit requested for the quantum regime , and therefore to implement a complete quantum memory protocol @xcite at the single photon level , and to explore experimentally its fidelity . optical pumping in a hybrid circuit , as demonstrated here , is also a first step towards the polarisation of the nitrogen nuclear spins @xcite , and in a longer term towards a nuclear - spin based quantum memory . in conclusion we have implemented the multi - mode storage and retrieval of microwave fields in an ensemble of nv centres in diamond at millikelvin temperatures , with active reset by optical pumping and refocusing by a strong microwave pulse . these results demonstrate that complex dynamical control of spin ensembles is compatible with hybrid quantum circuits , thus enabling the long - term storage of quantum information in electronic or nuclear spin ensemble quantum memory . * we acknowledge technical support from p. snat , d. duet , j .- c . tack , p. pari , p. forget , as well as useful discussions within the quantronics group and with a. drau , j .- f . roch , t. chanelire and j. morton . we acknowledge support of the french national research agency ( anr ) with the qinvc project from chistera program , of the european project scaleqit , and of the cnano idf project quantrocryo . y. kubo is supported by the japanese society for the promotion of science ( jsps ) . b. julsgaard and k. mlmer acknowledge support from the villum foundation . the sample we use is a polished @xmath153 plate of dimensions @xmath154 taken from a @xmath155 growth sector of a synthetic type - ib diamond crystal with its edges along @xmath156 $ ] . the synthetic diamond crystal was grown by a temperature gradient method under high pressure and high temperature ( hpht ) of 5.5gpa and @xmath157c . the crystal contained @xmath66ppm of neutral substitutional nitrogen ( the p1 centre ) as measured by ir absorption . irradiation with @xmath111mev electrons was carried out in two steps . first , it was irradiated at rt with a dose of @xmath158e / c@xmath159 and annealed at @xmath160c for @xmath161 hours in vacuum . secondly , it was irradiated at @xmath162c to a dose of @xmath158e / c@xmath159 and annealed at @xmath163c for @xmath111 hours in vacuum . from the measured absorption we deduce the nv centre concentration @xmath164ppm , implying a probable concentration of remaining neutral substitutional nitrogen ( the p1 centre ) of @xmath165ppm . in samples with such large p1 centre concentration , the typical nv centre coherence time is @xmath166 @xcite , in agreement with measurements shown in fig . the niobium resonator was fabricated using optical lithography followed by dry etching . microwave simulations indicates an impedance @xmath167 , corresponding to a total inductance @xmath168nh . this inductance arises from the capacitor fingers , and from the meander wire connecting the two capacitor electrodes on top of which the diamond is pressed by a copper spring . simulations indicate that the meander wire inductance is @xmath169nh . since the diamond crystal covers only this wire , the spin filling factor is @xmath170 . the wire was designed purposely to occupy a small area of @xmath171 in order to minimise the laser power needed to repump the spins . the single - mode fibre , with numerical aperture @xmath172 , was brought into our cryogen - free dilution cryostat through a home - made vacuum feedthrough . a yag laser doubled at @xmath31 nm is injected into the room - temperature end of the fibre . it is pulsed with @xmath173db dynamics by a double - pass acousto - optic modulator . up to @xmath174mw laser power could be injected into the fibre . at low temperatures , the fibre and cladding were stripped over @xmath13 cm . this short bare fibre part was glued to a glass @xmath13 mm thick spacer itself glued to the @xmath175 mm thick diamond , so that the fibre - to - sample distance was @xmath44 mm , corresponding to a nominal beam diameter of @xmath176 at the sample , therefore matching the area covered by the resonator meander wire . prior to being glued , the fibre was positioned on top of this wire , with a precision estimated to be better than @xmath175 mm . the detailed microwave setup is shown in supplementary fig . [ figs11 ] . the incoming microwave pulses are attenuated at low temperatures , routed to the input waveguide of the resonator via a circulator , and the reflected signal is amplified at @xmath75k by a low - noise hemt amplifier , and demodulated at room - temperature , yielding the field quadratures @xmath177 or equivalently the amplitude and phase @xmath178 . note that the attenuation in the input line ( @xmath66db at @xmath75k and @xmath32db at @xmath142mk ) is not sufficient to fully suppress thermal photons in the input waveguide to the resonator , implying that a thermal field with @xmath179 photon is present in the resonator , and causing sizeable thermal excitation of the spin ensemble as shown in fig . 2 of the main text . this was done purposely to apply more conveniently the refocusing pulses that require large microwave powers at the sample input . one difficulty of the experiment is to switch on and off with very high dynamics the strong microwave pulses needed to saturate or refocus the spins . we found that one microwave switch was not sufficient , and we used in all the experiments two switches in series , one internal to the microwave source , and one external ( see supplementary figs . [ figs11 ] and [ figs12 ] ) . , @xmath180 and @xmath181 ghz , respectively . cuni coax is a coaxial cable made of cuni , and ag - cuni coax is a silver - plated cuni coaxial cable . sc coax is a superconducting nbti coaxial cable . flex coax is a low - loss flexible coaxial cable . rectangles represent ports terminated by 50@xmath182 . the cryogenic microwave amplifier is a citcryo 1 - 12 from caltech , with gain @xmath18338 db and noise temperature @xmath1835 k at @xmath184 ghz . a dc magnetic field @xmath21 is applied parallel to the chip by passing a dc current through an outer superconducting coil . the sample box and the coil are surrounded by two magnetic shieldings consisting of a lead cylinder around which permalloy tape is wrapped . the sample box , coil , and the shieldings are thermally anchored at the mixing chamber with base temperature @xmath71mk ( note that in the experiments using active reset of the spins with @xmath44mw laser power , the temperature was @xmath37mk instead ) . ( b ) full configuration of the measurement apparatus at room temperature for spins polarization measurements ( fig . 2 of the main text ) . the saturation pulse is shaped with @xmath185db dynamics by two microwave switches in series . the dc waveform supplied to the external microwave switch is delayed by @xmath186ns to synchronize both switches . lpf is a low - pass filter with cutoff frequency @xmath13 mhz . , width=604 ] db dynamics by two microwave switches in series . the dc waveform supplied to the external microwave switch is delayed by @xmath186ns to synchronize both switches . ( a ) full configuration of the measurement apparatus at room temperature for multimode storage experiment ( fig . 5 b , c of main text ) . lpf is a low - pass filter with cutoff frequency @xmath161 mhz . ( b ) full configuration of the measurement apparatus at room temperature for few - photon storage experiment ( fig . 5 d of main text ) . the local oscillator of the iq demodulator is pulsed to suppress the refocusing pulse from the reflected signal . lpf is a low - pass filter with cutoff frequency @xmath13mhz.,width=604 ] the goal of this section is to give the necessary elements to understand the theoretical curves presented in the main text . after defining the model , we explain 1 ) how the spin susceptibility @xmath187 is extracted from microwave absorption measurements ( figs@xmath111d and @xmath184a in the main text ) , 2 ) how this measured susceptibility can be computed from the spin hamiltonian assuming phenomenological distributions of the various hamiltonian parameters ( again figs@xmath111d and @xmath184a in the main text ) , and 3 ) how the experimental sequences with refocusing pulses were simulated ( figs@xmath75b , c , e and @xmath161b of the main text ) . here we follow the model already described in@xcite . the spin-1 nv centers are approximated by two spin-1/2 particles ( see justification in the main text and below ) . the nv ensemble is thus modelled as an ensemble of @xmath188 spin-1/2 particles of frequency @xmath189 . each spin couples to the cavity field ( described by creation and annihilation operators @xmath190 and @xmath191 ) with a coupling constant @xmath192 and a jaynes - cummings type of interaction . the resonator frequency is @xmath193 , and its field damping rate @xmath194 . the total system hamiltonian is then the dynamics predicted by this model is quite complex ( see below ) . however it becomes simpler in the limit where the number of excitations present in the system is much lower than the total number of spins @xmath188 . indeed , in this regime , spin saturation can be neglected , and the spins behave as weakly excited harmonic oscillators . this is the so - called holstein - primakoff approximation , by which all previous experiments on spins coupled to resonators have been theoretically described so far . the measurements shown in figs . @xmath111 and @xmath75a of the main text are also performed in that limit , which is why we briefly discuss it in the next paragraph . [ [ microwave - absorption - and - spin - susceptibility - in - the - linear - regime ] ] microwave absorption and spin susceptibility in the linear regime ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ @xmath204 being the spin dephasing rate . the spins shift the resonance frequency @xmath193 by @xmath205 and add a damping term @xmath206 to the field damping rate @xmath207 . these quantities can be directly extracted from the microwave measurements as explained in the following . in the experiment , we measure the amplitude and phase of the field reflected on the resonator . we thus want to calculate the reflection coefficient @xmath208 , @xmath209 being the reflected field . from input - output theory we have @xmath210 , so that in the experiment , we measure reflected microwave signals through measurement cables and amplifiers which have a complex frequency - dependent transmission coefficient @xmath212 giving us access to @xmath213 ( the complex conjugate is taken because of a different sign convention between theory and experiment ) . to calibrate @xmath212 , the reflected signal @xmath214 is compared to the steady - state values of the reflected signal with spins saturated which is given by @xmath215 , with @xmath216 the reflextion coefficient of the cavity without spins . in total we obtain that we find it useful to express @xmath218 in terms of the spin susceptibility @xmath219 , defined as the ratio of the induced magnetization @xmath220 and the applied microwave field @xmath221 . more precisely for an applied field @xmath222 , the induced magnetization is @xmath223 , with @xmath224 @xcite . this changes the resonator inductance @xmath19 into @xmath225 @xcite , @xmath226 being the filling factor and @xmath227 the complex spin susceptibility in cgs units . this implies that the resonator frequency is shifted by @xmath228 , and the extra field damping rate is @xmath229 . this yields the following direct link between @xmath218 and @xmath219 : equations ( [ eq : kmeasurements ] ) and ( [ eq : kandchi ] ) explain how the experimental spin susceptibility was derived from the measurements ( figs @xmath111d and @xmath75a of the main text ) . note that the corresponding absorption curves were measured at powers @xmath231dbm corresponding to few intra - cavity photons , thus by far low enough for the holstein - primakoff approximation to be justified . the goal of this section is to demonstrate that it is possible to quantitatively understand from the nv centers hamiltonian the measured susceptibility curves , assuming phenomenological distributions of the parameters entering this hamiltonian . this is how we computed the theory curves in figs.@xmath111e and @xmath75a of the main text . note that this section is to a large extent independent of the rest of the paper : it explains the theory curves in figs2d and inset of 3a , but importantly the numerical simulations of the echo experiments do not rely in any way on the distributions of strain or magnetic field fluctuations obtained phenomenologically in this section . as explained in @xcite this implies that @xmath234 ( this relation holds in the limit where the inhomogeneous frequency spread is much larger than the homogeneous spin linewidth , which is the case here ) . therefore , @xmath235 is proportional to @xmath236 . we assume that the spatial distribution ( which determines the coupling constant @xmath237 ) and the frequency distribution of the spins are uncorrelated , which would be the case if the frequency distribution were only caused by local fields ( magnetic , electric , strain , see below ) , with a spatially independent distribution . one can then write @xmath238 , with @xmath239 and @xmath240 normalized such that @xmath241 . what we are interested in here is to reproduce the frequency distribution @xmath240 observed in the experiment , starting from the nv center hamiltonian , with only one distribution of the hamiltonian parameters ( strain @xmath12 , magnetic field @xmath242 , zero - field splitting @xmath243 ) . with @xmath246 ghz the zero - field splitting , @xmath12 the strain splitting , @xmath247 mhz the nuclear quadrupole momentum , @xmath248 mhz the hyperfine coupling of the nv to the @xmath244 nucleus , and @xmath249 the magnetic field felt by the nv . our ensemble of nv centers has a certain frequency distribution because the hamiltonian parameters have a distribution , that we assume to be static . here we will consider that both @xmath98 and @xmath78 are fixed for all nvs . on the other hand , @xmath250 has evidently a certain distribution characterized by a function @xmath251 such that the number of spins seeing a certain magnetic field between @xmath250 and @xmath252 is given by @xmath253 . this distribution originates from the different magnetic environments due to the local random distribution of @xmath254 centers and @xmath255 nuclei . note that although one can safely assume that @xmath256 , @xmath257 and @xmath250 have the same distribution , we will only consider the @xmath250 distribution because it is the one that couples most strongly to the nv center , a good approximation when @xmath258 as is the case here . in the following we write @xmath259 , and we note that @xmath260 , @xmath21 being the applied magnetic field at an angle @xmath261 from the nv axis and @xmath262 the z component of the field due to the local environment of each nv . what is constant in the problem is the distribution of @xmath262 , @xmath263 the strain parameter @xmath12 has another distribution @xmath264 . and finally , the zero - field splitting @xmath243 is distributed with density @xmath265 , which is validated by recent work @xcite . the hamiltonian diagonalization leads to @xmath266 states , corresponding to the 3 nuclear spin states @xmath267 , and the @xmath184 nv center states due to their spin @xmath23 . this gives @xmath268 transition frequencies @xmath269 $ ] . our goal is now to express @xmath270 as a function of @xmath271 , @xmath272,@xmath273 . we write for @xmath271 and @xmath272 we will assume a lorentzian shape , which at least for @xmath271 has a physical justification ( the linewidth of a dipolar broadened spin ensemble is usually lorentzian ) , with a width that will be `` guessed '' or adapted to fit the data . for @xmath273 we use the @xmath92 dataset ( see fig . @xmath75a of the main text ) to find an appropriate distribution . the formula above is in principle sufficient to compute @xmath270 numerically given @xmath271 , @xmath272,@xmath273 ; however it would lead to very long calculation times and we need to simplify it . the first simplification is that instead of explicitly diagonalizing the hamiltonian to obtain @xmath275 $ ] we use approximate formulas : with @xmath279 , considering the hyperfine interaction with the nuclear spin as a nuclear - spin - state dependent effective magnetic field of modulus @xmath280 . these formulas are valid when @xmath258 , a very good approximation in our case . this allows to very easily invert the formula yielding , for given frequency @xmath281 , strain @xmath12 and zero - field splitting @xmath243 , the local magnetic field @xmath282 needed so that @xmath275=\omega$ ] . this equation has either zero or two solutions depending on @xmath281 . for the @xmath283 transitions there are two solutions if @xmath284 , and zero else ; for the @xmath285 transitions there are two solutions if @xmath286 , and zero elsewhere . @xmath296\right ) \\ & = & \sum_{m_i,\pm , i}\iint dedd\rho_{e}(e)\rho_{d}(d)\frac{\rho_{b}\left(b_{m_i,\pm^{(i)}}[\omega , e , b_{nv},d]\right)}{\left|\frac{\partial\omega_{m_i,\pm}}{\partial b}(b_{m_i,\pm}[\omega , e , b_{nv},d])\right|}.\end{aligned}\ ] ] note that from the previous formulas it is clear that the density of nv centers at a given frequency @xmath281 can have a strong dependence on the nuclear spin state . this might explain in particular why the relative contributions of the @xmath297 and @xmath124 to the spin echo signal at @xmath298ghz were found to be slightly different from the expected @xmath299 and @xmath300 by fitting the decoherence signal ( see main text ) . a difficulty arises when @xmath301 vanishes , giving rise to a divergence . to smoothen this out , we discretize the problem : we choose some small frequency scale @xmath302 and we solve the equation @xmath303-\omega_{m_i,+}[e , b_{nv},d , b]=d\omega_{0}$ ] . this equation has always two solutions , we take the min of the two yielding the quantity @xmath304 $ ] . the new formula is we assume a lorentzian distribution for both @xmath263 and @xmath265 with respective widths @xmath306 and @xmath307 . we use the data at @xmath308 to guess the distribution @xmath264 . we find that a bi - exponential distribution @xmath309 / ( e_1 + a_1 e_2)$ ] yields a computed @xmath310 that reproduces semi - quantitatively the data . in total we use the following parameters : @xmath311gs , @xmath312mhz , @xmath313mhz , @xmath314mhz , @xmath315 . in this way we obtain the @xmath92gs spin susceptibility shown in fig . [ figs3 ] ( the corresponding @xmath264 distribution is shown in inset ) . . open red circles are experimental data , solid blue line is the theory computed with the bi - exponential strain distribution @xmath316 shown in the inset , with a lorentzian @xmath317 and @xmath318 with respective widths @xmath319gs and @xmath320khz.,width=453 ] after having in this way determined the distributions @xmath321 , we compute without further adjustable parameters the rescaled @xmath322 . the experimental distribution includes contributions both from the spins that are orthogonal to @xmath21 and from those that are non - orthogonal , each of those having a very different resonance frequency dependence on @xmath21 as shown in fig . 2 of the main text . each family contains exactly half of the total number of spins contributing to the signal ; however , spins from each family have a different coupling constant to the resonator field due to the angle they make with this field . this difference in coupling constant can be incorporated in a single numerical factor that yields a different ensemble coupling constant for each of the two spin families , @xmath323 and @xmath324 . indeed , the coupling constant of a nv center ensemble to a resonator is given by @xmath325 @xcite , with @xmath326 the ensemble filling factor and @xmath327 the angular factor . the geometrical filling factor is clearly identical for the two families , but the factor @xmath199 differs . we have numerically calculated the ratio @xmath328 , yielding @xmath329 . in this way we are able to compute the total @xmath330 . the rescaled result is shown in fig . [ figs4 ] . as can be seen from figs . [ figs3 ] and [ figs4 ] , the agreement is semi - quantitative . all the features are reproduced but not exactly with the appropriate weight . these remaining discrepancies might be due to the fact that the strain distribution is likely to have some spatial dependence ; in particular it could depend to some extent of the distance to the surface , in which case it would be correlated with the coupling constant distribution making the above analysis an approximation . numerical simulations are performed to describe spin - echo experiments and quantitatively estimate the echo field retrieval efficiency . therefore they clearly do not use the linear approximation developed in the previous paragraphs , in order to account for refocusing effects , but instead use the complete hamiltonian eq . ( [ eq : hamiltonian_working ] ) . to perform the calculations while taking into account the inhomogeneity in both spin resonance frequencies and coupling strengths , the entire inhomogeneous ensemble is divided into @xmath331 homogeneous sub - ensembles , @xmath332 , each of them describing spins having an identical frequency @xmath333 and coupling to the cavity field @xmath334 . the total number of spins in one sub - ensemble is defined as @xmath335 . we define the sub - ensemble spin collective operators : @xmath336 where @xmath337 runs over all spin sub - ensembles . here again , the spin-1 nv center is treated approximately as two spin-@xmath338 particles . by incorporating the effect of resonator leakage and spin decoherence in the markov approximation , the dynamical evolution of mean values in the frame rotating at @xmath339 ( with @xmath339 the mean spin frequency ) is described by ( see ref . @xcite for details ) : @xmath340 here @xmath341 , @xmath342 , @xmath343 ; and @xmath344 and @xmath345 are the cavity field quadratures such that @xmath346 = i$ ] , @xmath347 and @xmath348 are real and imaginary parts of the external driving field with @xmath349 being the incident number of microwave photons per second , @xmath350 is the spin population decay rate . in the experiment the population decay time @xmath351s ( see fig.@xmath184c of main text ) is very long compared to the typical refocusing time scales and we use the excellent approximation @xmath352 . the first step is to determine the size @xmath335 of each sub - ensemble @xmath353 , which requires knowledge of the distribution of coupling constants and resonance frequency within the spin - ensemble . the distribution of coupling constants can be computed from the known resonator geometry and crystalline orientation ( see next section ) ; the distribution of resonance frequencies is not a priori known but is extracted from absorption measurements as explained earlier . we first come back to the interaction hamiltonian between the nv center spin @xmath354 and the quantized resonator magnetic field : @xmath355 , where @xmath356 is the rms fluctuations of the resonator vacuum field . the external field @xmath357 and the effective field generated by the nuclear spin , @xmath358 , can be treated classically . in the rotating wave approximation the spin part of the hamiltonian can be expressed in terms of the energy eigen states @xmath359 at zero bias magnetic field as : @xmath360 \:-\ : ig_y\left[{\ensuremath{\hat{a}_{\mathrm{c}}}}{\ensuremath{|-\rangle}}\!{\ensuremath{\left<0\right| } } \:-\ : { \ensuremath{\hat{a}^{\dagger}_{\mathrm{c}}}}{\ensuremath{|0\rangle}}\!{\ensuremath{\left<-\right|}}\right ] , \end{split}\ ] ] where @xmath361 . we note the energy splitting of @xmath55 between the @xmath362 and @xmath363 states , and we observe that two jaynes - cummings - like interaction terms emerge . under the holstein - primakoff approximation ( i.e. in absence of saturation with all population essentially in the ground state @xmath364 ) , the nv center can be treated accurately as two separate spin-@xmath338 particles . of course , if the resonator field saturates the nv center , and hence depletes the @xmath364 state , this two - particle picture fails . nonetheless , we argue below that it is a reasonable approximation to maintain this picture throughout our simulations . the hamiltonian above is expressed with the quantization axis ( the @xmath365-axis ) along the nv center axis . since this does not coincide with our laboratory coordinate system ( defined in fig . 1b of the main text ) , we must define a local coordinate system for each of the four possible nv center axes . as local @xmath365-axes we choose : @xmath366 i.e. for family 1 we take , @xmath367 , etc . for symmetry reasons the coupling strengths from family 1 and 2 turn out identical , and likewise for families 3 and 4 . we thus proceed with families 1 and 3 only . as local @xmath368- and @xmath369-axes for family 1 we have a selection of choices : @xmath370 @xmath371 being the angle between the nv axis and the direction of non - axial strain in the diamond matrix . for all angles @xmath371 we obtain a local orthogonal coordinate system @xmath372 $ ] with the @xmath365-axis pointing along the nv center axis as required . the angle @xmath371 must be chosen such that the hamiltonian term @xmath373 describes correctly the actual physical strain experienced by the nv center from the surrounding host material . however , since this strain has no preferred direction we shall later average our coupling - strength distribution over this angle . similarly , for family 3 we adopt the local coordinate system : @xmath374 and @xmath375 next , provided that the resonator vacuum field @xmath356 is known , the coupling constants @xmath376 and @xmath377 for the two nv center transitions follow as @xmath378 and @xmath379 for each spin family . the sign ( or in general the phase ) of @xmath380 is irrelevant and we use in the simulations a positive quantity for the coupling constants . for the spin families 1 and 3 we then find : @xmath381 the distribution of @xmath380-parameters thus follows from both an inhomogeneous distribution of magnetic fields @xmath356 and the angular distribution of crystal strain experienced by the nv centers . we note that the distributions of @xmath382 and @xmath383 ( and likewise for @xmath384 and @xmath385 ) are identical when the angular average is taken into account . to proceed , the vacuum field @xmath356 is calculated using the comsol simulation software . the shape of @xmath356 is shown graphically in the inset of fig . 1b in the main text , and the curved sections of the resonator element was neglected such that @xmath356 essentially only consists of an @xmath368- and a @xmath369-component . the correct magnitude for @xmath356 is obtained by scaling the field to the one corresponding to the resonator current set equal to @xmath386 . the resulting distribution [ or rather @xmath387 times the distribution @xmath388 with @xmath389 of @xmath380-parameters is shown in fig . [ fig : extractdist]a , where the effective length of the active crystal was taken to be 100 microns along the @xmath365-axis , the concentration was 2 ppm , and a 0.7 @xmath390 spacing for glue was estimated between the resonator and the diamond crystal . the area under @xmath391 is equal to the squared ensemble coupling constant @xmath392 , and the orthogonal families ( 3 and 4 ) contribute @xmath393 of the total area , while families 1 and 2 contribute the remaining @xmath394 part . the black circles in fig . [ fig : extractdist]a denote the actual discrete distribution , which was used for the simulation results shown in fig . 4b , c , e and 5b in the main text . the frequency distribution @xmath395 is determined as explained earlier , from microwave absorption measurements . note that the directly measured ensemble coupling constant from integrating the curves in fig . [ fig : extractdist]b is @xmath396 mhz . however , these data were obtained under repumping with a laser power of @xmath397 corresponding ( see fig . 3b of main text ) to a spin polarization @xmath398 . in the simulations we thus rescale the number of spins @xmath188 such that the ensemble coupling constant becomes @xmath399 , and when simulating the spin - echo sequences the appropriate initial value of @xmath400 is chosen , where @xmath401 refers to the polarization corresponding to the laser power used in the experiment ( @xmath44mw in most measurements ) . we note that the @xmath380-parameter distribution calculated from first principles as explained above leads to @xmath402 mhz , i.e. @xmath403 70 % of the value stated above . given the measurement accuracy , and the fact that the nv center concentration and the resonator - to - crystal distance are known only approximately , we find the agreement between these numbers satisfactory . the main purpose of our simulations is to understand the measured retrieval efficiency of the echo pulses . hence , we must be quantitative on the echo - pulse magnitudes in the simulations , and accordingly we must adapt the equations of motion to cover the actual case with two spin classes of different coherence time as explained in sec . iv . during the refocusing pulse ( red ) compared to the difference , @xmath404 ( black ) , between the simulations with fast and slow dephasing . the refocusing pulse extends from @xmath405@xmath406 , followed by free - induction decay . ( b ) the red circles denote the field recovery efficiency of the experimental echo pulses versus the delay @xmath110 between storage and readout of the six pulses in the multi - mode echo sequence . the blue dots are simulated efficiencies for a range of driving powers in the vicinity of the experimental value of @xmath407 dbm . the green solid line is the square of the function @xmath408 described in the main text . the remaining solid lines corresponds to a constant pre - factor times @xmath409 ^ 2 $ ] . ( c ) the blue circles correspond to the simulated pre - factors from panel ( b ) with the horizontal blue dashed line denoting the average value of @xmath410 . the red dashed line is the experimental pre - factor equal to @xmath411 . ] the obvious strategy would be doubling the sub - ensemble partitioning into two parts one with coherence time @xmath412 and the other with @xmath413and adapting eqs . ( [ eq : def_sub - ensembles])-([eq : mean_val_eq_sz_kl ] ) accordingly . however , we use a much simpler strategy in the following : the simulations are run first with a single coherence time @xmath412 , and then repeated with the other coherence time @xmath413 . the calculated reflected fields in these two instances are then combined according to the weights given by the fitting parameters @xmath98 and @xmath242 mentioned in the main text . now the two sub - ensembles share in practice a common resonator field @xmath191 which in the simulations is ascribed two different values , and under influence of the refocusing pulse , i.e. in conditions of saturation , linearity will not be applicable . however , for our problem in question ( the multi - mode echo sequence in fig . 5 of the main text ) the refocusing pulse strength is @xmath414 db above saturation , and hence the external driving determines the resonator field to a much larger extent than the reaction field of the spin dipoles . indeed , fig . [ fig : fidelitycalculation]a shows that the resonator field is essentially identical in the two simulation runs with fast and slow coherence time . now , for the multi - mode sequence shown in fig . 5 in the main text , the efficiency is calculated for each of the six pulses in the sequence and plotted as a function of the input - output delay @xmath110 in fig . [ fig : fidelitycalculation ] ( red circles ) . since the efficiency is defined as the energy of the echo pulse relative to the incoming energy , it is expected to behave as : @xmath415 ^ 2 $ ] , where @xmath416 is a constant and @xmath114 was found in the main text as fitting the experimental amplitude - decay of echo pulses [ scaled such that @xmath417 . the function @xmath409 ^ 2 $ ] is shown as the green curve in fig . [ fig : fidelitycalculation]b , and this curve corresponds to the efficiency that one would obtain if spin dephasing were the sole reason for a non - ideal echo protocol . we observe that both simulations ( blue dots ) and experiment lead to a lower efficiency ; however , the overall behavior corresponds indeed to a pre - factor @xmath416 multiplied onto @xmath409 ^ 2 $ ] . before we comment on these pre - factors , there is a little subtlety to mention about the simulations : due to the discrete division of the coupling strengths into @xmath418 bins ( see fig . [ fig : extractdist]a ) , there are artificial oscillations occurring in the echo recovery efficiency as a function of the applied driving power , see the blue circles in fig . [ fig : fidelitycalculation]c . we have checked that an increasing @xmath419 will decrease the magnitude of such oscillations ( not shown ) while the mean value stays essentially fixed . for this reason we compute the mean value of the efficiency pre - factors in fig . [ fig : fidelitycalculation]c , which then amounts to 0.21 ( dotted blue line ) . we thus conclude that the rather low simulated efficiencies ( a few times @xmath151 and decreasing with increasing @xmath110 in fig . [ fig : fidelitycalculation]b ) are caused primarily by spin dephasing . the simulated curve in fig . 4e of the main text was obtained with @xmath420 . next , the experimental retrieval efficiencies follow also the trend of @xmath409 ^ 2 $ ] but with a pre - factor @xmath421 times lower than the one found from simulations . this indicates that there is an additional effect in play , which is not included in our simulations . as explained in the main text , we attribute this discrepancy to the inadequacy of the markov approximation to describe decoherence caused by a spin bath . 36ifxundefined [ 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 10.1038/nature03064 [ * * , ( ) ] link:\doibase 10.1038/nphoton.2009.231 [ * * , ( ) ] link:\doibase 10.1038/nature09662 [ * * , ( ) ] link:\doibase 10.1088/1367 - 2630/13/9/093031 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.101.040501 [ * * , ( ) ] link:\doibase 10.1103/physreva.77.020301 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.103.070502 [ * * , ( ) ] link:\doibase 10.1038/ncomms2771 [ * * , ( ) ] link:\doibase 10.1038/nmat3182 [ * * , ( ) ] link:\doibase 10.1126/science.1217635 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1103/physrevlett.110.250503 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.105.140502 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.105.140501 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.107.060502 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.110.067004 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.110.157001 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.107.220501 [ * * , ( ) ] link:\doibase 10.1038/nature10462 [ * * , ( ) ] link:\doibase 10.1063/1.1721903 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.105.140503 [ * * , ( ) ] link:\doibase 10.1103/physreva.85.012333 [ * * , ( ) ] @noop * * , ( ) link:\doibase 10.1103/physrevb.79.075203 [ * * , ( ) ] link:\doibase 10.1038/nphys1969 [ * * , ( ) ] link:\doibase 10.1103/physreva.86.063810 [ * * , ( ) ] link:\doibase 10.1038/nature10401 [ * * , ( ) ] link:\doibase 10.1063/1.4792205 [ * * , ( ) ] link:\doibase 10.1103/physrevb.74.161203 [ * * , ( ) ] \doibase http://dx.doi.org/10.1016/j.jmr.2013.01.010 [ * * , ( ) ] link:\doibase 10.1103/physrevlett.102.057403 [ * * , ( ) ] http://stacks.iop.org/0022-3727/30/i=12/a=016 [ * * , ( ) ] link:\doibase 10.1103/physreva.84.063810 [ * * , ( ) ] link:\doibase 10.1103/physreva.83.053852 [ * * , ( ) ] @noop _ _ ( , ) @noop

we describe a numerical algorithm which simulates the propagation of light in inhomogeneous universes . this algorithm computes the trajectories of light rays between the observer , located at redshift , and distant sources located at high redshift , using the multiple lens - plane method . the deformation and deflection of light beams as they interact with each lens plane are computed using the filled - beam approximation . we use a particle - particle / particle - mesh ( p m ) n - body numerical code to simulate the formation of large scale structure in the universe . we extend the length resolution of the simulations to sub - megaparsec scales by using a monte - carlo method for locating galaxies inside the computational volume according to the underlying distribution of background matter . the observed galaxy 2-point correlation function is reproduced . this algorithm constitutes a major improvement over previous methods , which either neglected the presence of large - scale structure , neglected the presence of galaxies , neglected the contribution of distant matter ( matter located far from the beam ) , or used the zeldovich approximation for simulating the formation of large - scale structure . in addition , we take into account the observed morphology - density relation when assigning morphological types to galaxies , something that was ignored in all previous studies . to test this algorithm , we perform 1981 simulations , for three different cosmological models : an einstein - de sitter model with density parameter , an open model with , and a flat , low density model with and a cosmological constant . in all models , the initial density fluctuations correspond to a cold dark matter power spectrum normalized to cobe . in each simulation , we compute the shear and magnification resulting from the presence of inhomogeneities . our results are the following : ( 1 ) the magnification is totally dominated by the convergence , with the shear contributing less than one part in . ( 2 ) most of the cumulative shear and magnification is contributed by matter located at intermediate redshifts . ( 3 ) the actual value of the redshift where the largest contribution to shear and magnification occurs depends on the cosmological model . in particular , the lens planes contributing the most are located at larger redshift for models with smaller . ( 4 ) the number of galaxies directly hit by the beam increases with redshift , while the contribution of lens planes to the shear and magnification decrease with increasing lens - plane redshift for , indicating that the bulk of the shear and magnification does not originate from direct hits , but rather from the tidal influence of nearby and more distant galaxies , and background matter . ( 5 ) the average contributions of background matter and nearby galaxies to the shear is comparable for models with small . for the einstein - de sitter model , the contribution of the background matter exceeds the one of nearby galaxies by nearly one order of magnitude .

astro-ph9708129

gravitational lensing systems have relatively simple geometry and most of them are cosmological in nature because of their scale . these two facts have stimulated cosmologists in the last few decades to use gravitational lensing as a tool in physical cosmology . there are four distinct applications of gravitational lenses to cosmology . ( a ) observations of lensed sources provide information about the mass and density structure of the lens ( zwicky 1937a , b ) . ( b ) lensed sources are amplified , and therefore easier to detect and resolve ( zwicky 1937a , b ) . ( c ) lensing of distant sources can be used to measure distances on cosmological scales ( klimov 1963 ; liebes 1964 ; refsdal 1964 ) . ( d ) microlensing can be used to study the stellar composition of the lens ( chang & refsdal 1979 ) . for a review of these various applications of gravitational lensing , we refer the reader to blandford & nayaran ( 1992 ) , and references therein . in this paper , we specifically focus on the problem of determining the nature and structure of the universe , using gravitational lensing of distant sources ( quasars ) . the _ nature _ of the universe is described by friedmann- cosmological models . such models describe idealized universes which are structureless , and obey the weak cosmological principle ( homogeneity and isotropy ) . a matter - dominated friedmann- model is characterized by the present values of the density parameter @xmath7 and the hubble constant @xmath9 . all other cosmological parameters , such as the age of the universe @xmath10 , the matter density @xmath11 , the deceleration parameter @xmath12 , or the curvature parameter @xmath13 , can be expressed in terms of @xmath7 and @xmath9 . if the universe contains additional components , such as radiation , cosmic strings , domain walls , or a nonzero cosmological constant , the model describing this universe has one additional parameter for each component , which measures the contribution of that component to the energy density of the universe . gravitational lensing of distant quasars can be used to measure ( or constrain ) the value of these cosmological parameters , in two different ways . first , the angular diameter distances between the source , the lens , and the observer enter into the lens equation ( schneider , ehlers , & falco 1992 , and references therein ) . therefore , observables such as amplification , time delays , or image splitting will depend on these distances . since the relationship between the angular diameter distance and the redshift is model - dependent ( weinberg 1972 ; fukugita et al . 1992 ) , we can use observations of lensed quasars to estimate @xmath7 and @xmath9 . second , the probability that a distant quasar will be lensed increases with the distance between that quasar and the observer , since a larger distance implies a larger amount of matter along the line of sight . hence , by using the _ fraction _ of lensed quasars , we can probe the distance - redshift relation and estimate the cosmological parameters . the _ structure _ of the universe represents the deviations from homogeneity and isotropy . these deviations are usually described in terms of primordial fluctuations that grow with time as a result of gravitational instability , to eventually form the large - scale structures of the universe , galaxies , clusters , and voids ( peebles 1980 ) . for most cosmological models , the primordial density fluctuations originate from a gaussian random process , and therefore are characterized entirely by a density power spectrum . there are numerous cosmological models describing the formation of large - scale structures in the universe ( cold dark matter , hot dark matter , mixed dark matter , @xmath14 ) , each model having its own power spectrum . since the formation and evolution of large - scale structures are responsible for forming the lenses , observations of lensed sources can be used to measure the amplitude and possibly the shape of the power spectrum , and to ultimately determine the correct cosmological model for large - scale structure formation . to apply these methods , we need to study the propagation of light in inhomogeneous universes described by particular cosmological models . the most common approach consists of using numerical methods to simulate both the formation of large - scale structures in an expanding universe and the propagation of photons through these structures . with the existence of compact inhomogeneities in the universe , it is reasonable to suspect that a light beam from a distant source undergoes a series of perturbations while traveling to the observer . we can simulate the effect of these perturbations by dividing the space between the source and the observer into redshifts intervals , and then projecting the matter inside each interval onto a plane normal to the line of sight , called a lens plane . in this so - called _ thin - lens approximation _ , the deflection of light resulting from each lens plane can be computed using geometrical optics . we can follow the evolution of a light beam propagating through the inhomogeneities , adding successively the contributions of each lens plane to the deflection and deformation of the beam ( blandford & nayaran 1986 ; blandford & kochanek 1987 ; schneider & weiss 1988a , b ; jaroszyski et al . 1990 ; jaroszyski 1991 , 1992 ; babul & lee 1991 ; bartelmann & schneider 1991 ; wambsganss , cen , & ostriker 1996 ; see also kochanek & apostolakis 1988 ; paczyski & wambsganss 1989 ) . this _ multiple lens - plane method _ is discussed in detail in schneider et al . ( 1992 , chap . 9 ) . to apply the multiple lens - plane method , we need to generate the surface density on each lens plane . the simplest method consists of distributing equal - mass objects ( galaxies or dark halo ) randomly in space , schneider & weiss ( 1988b ) , paczyski & wambsganss ( 1989 ) , and lee & paczyski ( 1990 ) have used this method to obtain statistics for shear and amplification caused by gravitational lensing . the obvious drawback of this approach is that it completely ignores the large - scale structure formation models that are responsible for the formation of inhomogeneities , as well as the observed properties of these inhomogeneities , such as the galaxy 2-point correlation function or the morphology - density relation . these simulations are useful for studying the properties of gravitational lenses , and their dependence upon cosmological parameters such as @xmath7 and @xmath9 , but do not provide any information or constraint on the large - scale structure formation scenario . furthermore , since the known properties of galaxy clustering in the universe are ignored , these randomly - generated distributions of deflectors are unrealistic , and the relevance of the results is unclear . several authors have described analytical methods for generating mass distributions that are consistent with particular cosmological models . babul & lee ( 1991 ) have developed an analytical model in which the effect of large - scale structure enters the calculation of light propagation through the density auto - correlation function @xmath15 . since this function is the fourier transform of the density power spectrum , this method effectively distinguishes among different models of structure formation . bartelmann & schneider ( 1991 ) use the semi - analytical model of buchert ( 1989 ) to generate large - scale structure for an einstein - de sitter model with a flat perturbation spectrum . jaroszyski ( 1991 , 1992 ) generates initial density fluctuations that reproduces a particular power spectrum , and then uses the zeldovich approximation to simulate the evolution of these density fluctuations . he then locates galaxies in each lens plane using an empirical method based on the local matter density . by combining the schechter luminosity function with a monte carlo method , he generates a luminosity for each galaxy , chooses a morphological type at random , and then model that galaxy using a non - singular isothermal profile whose parameters are related to the luminosity and morphological type of the galaxy . this constitutes a major improvement over previous work , in three different ways . ( a ) the method takes into account the fact that the large - scale structures in the universe do originate from the growth of primordial fluctuations , and allows for experimentations with different power spectra . ( b ) the galaxies have a spectrum of luminosity and masses that reproduce observations . ( c ) the method takes into account the existence of various galaxy morphological types ( ellipticals , s0 s , and spirals ) , by ascribing different surface density profiles to galaxies of different types . there are still several weaknesses in the approach used by jaroszyski . first , the zeldovich approximation is based on linear perturbation theory , and therefore underestimates the growth of large - scale structures in overdense regions . this can be a serious problem since these overdense regions are the one most likely to affect significantly the evolution of the beam . second , while this method acknowledges the existence of various morphological types , the morphological type of each galaxy is chosen randomly . this ignores the existence of the _ morphology - density relation _ ( dressler 1980 ; postman & geller 1984 ) , which relates the likelihood of any galaxy to have a particular morphological type to the richness of the environment in which that galaxy is located . in an earlier paper ( jaroszyski et al . 1990 ) , a particle - mesh [ pm ] code was used for generating the large - scale structure of the universe . this approach constitutes a significant improvement over using the zeldovich approximation . however , that paper did not include a treatment of the galaxies similar to the one in jaroszyski ( 1991 , 1992 ) . blandford et al . ( 1991 ) also used a pm code for generating the large - scale structure , and also considered the alternative approach of representing galaxies by randomly distributed isothermal spheres . however , they did not take the additional step of combining the results of the pm simulations with the density profiles of galaxies , as we shall do in this paper . there is also a potential problem with some of the methods described above . in several cases , only the matter located near the beam is included in the calculation of the deflection and deformation of the beam . the influence of distant matter is neglected . neglecting the contribution of distant matter is probably correct if the galaxy distributions are generated randomly , but , as we said , these galaxy distribution are unrealistic to start with . the analytical methods and pm simulations described above allow the formation of large - scale structures such as clusters of galaxies . as blandford et al . ( 1991 ) showed , the effect of distant clusters on the evolution of the beam can be important ( this is supported by the analytical work of kaiser [ 1992 ] ) . in this paper , we present a new method which addresses all these various concerns . in designing this method , our main goal was to generate matter distributions that take into account all the known constraints imposed by large - scale structure formations models and by observations of the actual distribution , morphological types , and structure of galaxies in the universe . to achieve this goal , we use a state - of - the - art particle - particle / particle - mesh ( @xmath16 ) code ( hockney & eastwood 1981 ) to simulate the formation and evolution of large - scale structure in the universe . for all simulations presented in this paper , we use a cold dark matter density power spectrum normalized to cobe , but the method can be used with any power spectrum and any normalization . using the particle distributions generated by the p@xmath1 m code , we locate the galaxies ( in the densest regions ) , using a monte carlo method that reproduces the observed galaxy 2-point correlation function fairly well ( within the limitation of cdm model ) . then , instead of randomly choosing the morphological type of each galaxy , a method which ignores the existence of morphological segregation , we determine the morphological type of each galaxy according to the local environment , using the observed morphology - density relation ( martel , premadi , & matzner 1997a , hereafter mpm ) . each galaxy is given a surface density profile which is chosen according to the galaxy luminosity and morphological type , as in jaroszyski ( 1992 ) . by combining the distribution of background matter simulated by the p@xmath1 m algorithm with the distribution and surface densities of galaxies , we are effectively describing the surface density of the lens planes over 9 orders of magnitude in length , from the size of the largest superclusters and voids , @xmath17 , down to the core radii of the smallest galaxies , @xmath18 . this approach for generating the surface density on the lens planes , the key part of any multiple lens - plane algorithm , differs significantly from all the ones that have been published previously . in their early work , schneider & weiss ( 1988a , b ) distributed clumps of equal masses randomly on the lens planes , thus ignoring both the existence of large - scale structure and the mass spectrum and structure of galaxies . jaroszyski et al . ( 1990 ) and wambsganss et al . ( 1996 ) simulated the formation of large - scale structure , but did not take galaxies into account . blandford et al . ( 1991 ) performed n - body simulation of large - scale structure formation , and also computed the deflection of light by randomly distributed galaxies . however , they considered these two approaches separately , and did not choose the location of the galaxies according to the results of the n - body simulations , as we do in this paper . the only algorithm which combines large - scale structure formation with galaxies is the one described by jaroszyski ( 1991 , 1992 ) . however , the evolution of the large - scale structure in that algorithm was simulated using the zeldovich approximation instead of a n - body code . furthermore , the contribution of distant matter to the evolution of the beam was ignored . our algorithm is also the first that takes the morphology - density relation into account . the reason is clear : none of the previous algorithms could have done it , either because galaxies were ignored , or the process of cluster formation was either ignored or approximated , making the morphology - density relation unapplicable . the combination of fully nonlinear large - scale structure formation , galaxy distributions that reproduce the observed 2-point correlation function , morphological type distributions that reproduce the observed morphology - density relation , and galaxy surface density profiles , gives to the matter distribution in our algorithm a level of realism that was not present in any of the previous studies . of course , the algorithms used by previous authors can be perfectly adequate , depending on the particular problem that is being studied , and have produced very interesting results . we feel , however , that for the purpose of determining the correct cosmological model of structure formation in the universe , and the value of the cosmological parameters , it is critical to generate matter distribution that are as realistic as possible , over the largest possible range of length scales . this was our goal in designing this algorithm . we briefly review the theory of gravitational lensing in 2 , mentioning only the aspects which are directly relevant to this paper . in 3 we describe the method for generating the large - scale mass distribution of the universe . the simulation of the light propagation and the resulting statistics are described in 4 . summary and conclusion are presented in 5 . with the existence of compact inhomogeneities in our universe , it is reasonable to suspect that a light bundle from a distant source undergoes a series of perturbations due to those inhomogeneities while traveling to the observer . we attempt to use a series of gravitational lensings to approximate this phenomena . first , we idealize the inhomogeneities as being distributed on thin sheets , called lens planes , which are arranged perpendicular to the line of sight . we assume that lensing only takes place on each of those planes . this way we can analyze the lensing properties of each plane separately , and let the light beam carries the effect of lensing while propagating from one plane to the next . this is known as the multiple lens - plane method ( schneider et al . 1992 ) . consider @xmath19 lens planes located at redshifts @xmath20 , with @xmath21 , and ordered such that @xmath22 for @xmath23 . figure 1 shows an example with @xmath24 . all angles are greatly exaggerated . each lens plane is characterized by its respective surface mass density @xmath25 , where @xmath26 is the impact vector of the ray on the @xmath27-th lens plane . let @xmath28 denote the deflection angle the light ray experiences on the @xmath27-th plane at a position @xmath26 . from this geometry , we can derive the _ lens equation _ , @xmath29 where @xmath30 is source position vector ( on the source plane ) , @xmath26 is the impact vector on the @xmath27-th plane , @xmath31 is the angular diameter distance between the @xmath32-th plane and the observer , and @xmath33 is the angular diameter distance between the @xmath27-th and @xmath32-th planes , with @xmath34 identifying the source plane . knowing the impact vector @xmath35 on the image plane , the impact vector on subsequent planes can be obtained recursively using @xmath36 the deflection angle is related to the surface density by @xmath37 where @xmath38 is the gravitational constant , @xmath39 is the speed of light , and the integral extends over the lens plane . we can rewrite this expression conveniently as @xmath40 it is useful to rewrite these equations in a dimensionless form . we define for each lens plane a critical surface density as @xmath41 and introduce the following dimensionless quantities , @xmath42 equations ( 2 ) , ( 4 ) , and ( 5 ) reduce to @xmath43 where @xmath44 and the gradient is now taken relative to @xmath45 . by using the identity @xmath46 ( where @xmath47 is the two - dimensional delta function ) , we can invert equation ( 11 ) , and get @xmath48 to compute the scaled position @xmath49 of the source on the source plane , we simply set @xmath50 . equation ( 12 ) gives @xmath51 , and equation ( 9 ) becomes @xmath52 this ray - tracing equation is a mapping from the image plane ( @xmath53 ) onto the source plane ( @xmath54 ) . since we are using the filled - beam approximation , the relevant distances to use in the lens equations are the angular diameter distances in an homogeneous friedmann universe ( schneider & weiss 1988a ) . for the cosmological models considered in this paper , all the appropriate distance formulae are given in fukugita et al . for the einstein - de sitter model , the angular diameter distance @xmath55 between redshifts @xmath56 and @xmath57 is @xmath58\,,\ ] ] where @xmath59 is the hubble radius . for the open model with @xmath60 , the distance is @xmath61\,.\ ] ] finally , for the flat model with nonzero cosmological constant , the distance is @xmath62^{-1/2}\,.\ ] ] of course , equation ( 15 ) is a special case of both equations ( 16 ) and ( 17 ) . in all cases , we are assuming @xmath63 , which gives @xmath64 . figure 2 shows the angular diameter distances @xmath65 for all three cosmological models considered in this paper . the effect of each lens plane on the evolution of the beam is described by the following jacobian matrix , @xmath66 where the commas denote differentiation with respect to the components of @xmath45 . since @xmath67 , and equation ( 13 ) gives @xmath68 , we can rewrite equation ( 18 ) as @xmath69 where @xmath70 we now define @xmath71 the determinant and trace of @xmath72 can be expressed entirely in terms of @xmath73 and @xmath74 , as follows : @xmath75 the quantities @xmath76 , @xmath77 , and @xmath74 are called _ magnification _ , _ convergence _ ( or ricci focusing ) , and _ shear _ , respectively . to compute the cumulative effect of all the lens planes , we consider the jacobian matrix of the mapping given by equation ( 14 ) , @xmath78 where @xmath79 is the @xmath80 identity matrix , and @xmath81 and @xmath82 are defined by @xmath83 after substituting equation ( 10 ) into equation ( 26 ) , we get @xmath84 where @xmath72 is given by equation ( 18 ) . hence , @xmath81 describes the effect the @xmath27-th plane would have on the beam if all the other planes were absent , and equation ( 25 ) simply combines the effect of all the planes . to compute the matrices @xmath82 , we differentiate equation ( 9 ) , and get @xmath85 since @xmath86 , we can use equation ( 29 ) to compute all matricies @xmath82 by recurrence . the image of a small circular source is essentially constant across the area of the source . ] is an ellipse with semi - axes @xmath87 and @xmath88 , where @xmath89 is the radius of the image in the absence of lensing . it can be shown that @xmath90 we can solve these equations for @xmath91 and @xmath92 . assuming @xmath93 , we can then compute the aspect ratio of the image . after some algebra , we get @xmath94 ^ 2 -4(\det{\bf b})^2\right\}^{1/2}\over2\det{\bf b}}\,.\ ] ] the magnification is given by @xmath95 light rays coming from distant sources are propagating through the universe while the large - scale structure in the universe is forming and evolving . ultimately , it would be desirable to simulate the evolution of large - scale structure and the light propagation simultaneously . however , such approach would be rather difficult , and is beyond the reach of present computer capabilities . the great advantage of using the multiple lens - plane method is that it allows us to consider the large - scale structure evolution problem and the light propagation problem separately , thus effectively breaking up the problem into two steps . the first step consists of generating the large - scale structure and galaxy distribution in the universe at various redshifts , and projecting these distributions onto lens planes , normal to the optical axis . then , once these lens planes are generated , we can compute numerically the trajectory of light rays through them , using the formalism described in 2 . clearly , many different experiments can be conducted using the same set of lens planes , simply by varying the shape , size , and number of rays in the beam , or the location of the beam on the planes . in 3.2 and 3.3 , we describe the method we use for generating galaxy distributions . several aspects of this method were previously discussed in detail in mpm , so we only give a brief summary . the ray - shooting method is described in 3.4 . all n - body simulations presented in this paper are done using the p@xmath1 m algorithm ( hockney & eastwood 1981 ) . the calculations evolve a system of gravitationally interacting particles in a cubic volume with triply periodic boundary conditions , comoving with hubble flow . the forces on particles are computed by solving poisson equation on a cubic grid using a fast fourier transform method . the resulting force field represents the newtonian interaction between particles down to a separation of a few mesh spacings . at shorter distances the computed force is significantly smaller than the physical force . to increase the dynamical range of the code , the force at short distance is corrected by direct summation over pairs of particles separated by less than some cutoff distance @xmath97 . with the addition of this so - called _ short - range correction _ , the code accurately reproduces the newtonian interaction down to the softening length @xmath98 . in all calculations , @xmath98 and @xmath97 were set equal to 0.3 and 2.7 mesh spacing , respectively , with @xmath99 particles and a @xmath100 grid . with these particular values , the code has a dynamical range of three orders of magnitude in length . the system is evolved forward in time using a second order runge - kutta time - integration scheme with a variable time step . to implement the multiple lens - plane method , we divide the space between @xmath0 and @xmath101 into a chain of cubic boxes of equal comoving size @xmath102 . we first need to determine the redshifts of the interfaces between these cubic boxes . let us assume that the photons that are reaching the observer at present entered a particular box at time @xmath103 , redshift @xmath104 and exited that box at time @xmath105 , redshift @xmath106 . the redshifts @xmath104 and @xmath106 are related by @xmath107c\,dt\,,\ ] ] where @xmath39 is the speed of light . using this equation , with the appropriate relation for @xmath108 , we can find the redshifts of the interfaces . the front side of the box closest the the observer is , by definition , at @xmath0 . plugging this value into equation ( 34 ) gives us the redshift @xmath104 of the back side of the box , which is also the redshift @xmath106 of the front side of the next box . then , by using equation ( 34 ) recursively , we can compute the redshifts of all the interfaces . the derivation of the recurrence relations for the einstein - de sitter model ( @xmath2 , @xmath109 ) , open models ( @xmath60 , @xmath109 ) , and flat models with a nonzero cosmological constant ( @xmath110 ) are presented in detail in premadi ( 1996 ) . next , we need to determine the matter distribution inside each box . however , during the time photons propagate across a particular box , the matter distribution inside that box evolves . in the thin - lens approximation , we need to choose for each box a `` snapshot redshift '' @xmath111 between the redshift @xmath104 when the photons enter the box and the redshift @xmath106 when the photons exit the box , generate the matter distribution at that redshift , and make the approximation that this distribution is valid at all redshifts between @xmath104 and @xmath106 . then , we need to choose a `` projection redshift '' @xmath112 , also between @xmath104 and @xmath106 , which is the redshift of the plane onto which we project the three - dimensional distribution of galaxies . we decided to set the projection redshift @xmath112 equal to the snapshot redshift @xmath111 , as every other author does . schneider & weiss ( 1988b , eq . [ 8 ] ) choose for the snapshot redshift @xmath111 the arithmetic mean @xmath113 . we decided to improve on this , by determining the snapshot redshift as follows : since the deflection angle varies linearly with the surface density of the lens plane , we choose @xmath111 to be the redshift at which the density contrast @xmath114 is equal , at each point , to the time - averaged value @xmath115 of the density contrast at that point between @xmath104 and @xmath106 ( this only makes sense in the context of linear perturbation theory , where the density contrast at any given point evolves independently of the density contrast at other points ) . for the einstein - de sitter model , the linear density contrast @xmath116 , where @xmath10 is the present time and @xmath117 is a constant . the time - averaged linear density contrast between two epochs @xmath103 and @xmath105 is then given by @xmath118\,.\ ] ] we set @xmath119 , and solve for @xmath111 as a function of @xmath104 and @xmath106 . we get @xmath120 - 1\,.\ ] ] computing @xmath111 for the other models is a significantly more complicated procedure , and constitutes an overkill . equation ( 36 ) is valid at high redshift , where all models resemble the einstein - de sitter model . at low redshift , linear theory is inaccurate whether we use the correct model or not . however , equation ( 36 ) reduces to the schneider & weiss formula @xmath121 in the low redshift limit . hence , it is correct to use equation ( 36 ) for all cosmological models , and it constitutes an improvement over the formula used by schneider & weiss ( 1988b ) the method we use for computing the galaxy locations was described in great detail in mpm . in this subsection , we give a brief summary of the method . it consists of three parts . first , we determine the locations of the galaxies at @xmath0 . second , we ascribe to each galaxy a morphological type ( e , s0 , or spiral ) . finally , we trace the galaxies back in time to determine their locations on each lens plane . we consider the large - scale structure at present ( @xmath0 ) resulting from the p@xmath1 m simulations , and design an empirical monte - carlo method for locating galaxies in the computational volume , based on the constraints that ( 1 ) galaxies should be predominantly located in the densest regions , and ( 2 ) the resulting distribution of galaxies should resemble the observed distribution on the sky . our method is the following : we divide the present computational volume into @xmath100 cubic cells of size @xmath122 , and compute the matter density @xmath123 at the center of each cell , using the same mass assignment scheme as in the p@xmath1 m code . we then choose a particular density threshold @xmath124 . we locate @xmath19 galaxies in each cell , where @xmath19 is given by @xmath125 the actual location of each galaxy is chosen to be the center of the cell , plus a random offset of order of the cell size . this eliminates any spurious effect introduced by the use of a grid . we then experiment with various values of the density threshold @xmath124 until the total number of galaxies comes out to be of order 40000 . this gives a number density of @xmath126 . this method bears some similarities with the one used by jaroszyski ( 1991 , 1992 ) . tests showed that the observed galaxy 2-point correlation function is fairly well reproduced ( mpm ) . there is a well - known observed relationship between the distribution of morphological types and the surface density of galaxies ( dressler 1980 ; postman & geller 1984 ) . regions of the sky with high concentration of galaxies contain on average more elliptical and s0 s and less spiral than regions with lower concentration of galaxies . by combining this relation with a monte - carlo method , we can ascribe a morphological type to each galaxy , as follows . we first compute the volume number density of galaxies @xmath127 around each galaxy , using @xmath128 where @xmath129 is a positive integer , and @xmath130 is the distance of the @xmath131 nearest neighboring galaxy . in all cases , we choose @xmath132 . once the densities are computed , we compute the fractions @xmath133 , @xmath134 , and @xmath135 of spirals , s0 s , and ellipticals , respectively , from the morphology - density relation . we then ascribe a morphological type to each galaxy by generating a random number @xmath136 between 0 and 1 ( with uniform probability ) . the galaxy is identified as a spiral if @xmath137 , a s0 if @xmath138 , and an elliptical if @xmath139 . the p@xmath1 m algorithm provides us with the distributions of particles at various intermediate redshifts between the initial redshift and the present , and , in particular , at all snapshots redshifts @xmath140 , @xmath141 . by combining these particle distributions with our simulated galaxy distributions at present , we can trace galaxies back in time and reconstruct their trajectories . to do this , we simply find the nearest particle @xmath142 of each galaxy @xmath143 at present ( where the subscript @xmath13 identifies the galaxy ) . then we `` tie '' the galaxy @xmath143 to that nearest particle . the location of the galaxy @xmath143 at any redshift @xmath106 is then given by : @xmath144+{\bf r}'\,,\ ] ] where @xmath145 is a small random offset , which we introduce to avoid the unfortunate situation of having two galaxies located at the top of each other because they happen to by tied to the same particle . this allows us to construct galaxy distributions at any redshift . to determine the physical parameters of each galaxy , we start by assuming that the present galaxy luminosities follow the schechter luminosity function , @xmath146 where @xmath147 is the number density of galaxies per unit luminosity . the parameters @xmath148 , @xmath149 , and @xmath150 are obtained from observation as follows : @xmath151 , @xmath152 , and @xmath153 , where @xmath154 is the luminosity in the @xmath155 band ( efstathiou , ellis , & peterson 1988 , hereafter eep ) . there is a fourth parameter , the luminosity @xmath156 of the faintest galaxies , which must be introduced to prevent the total number of galaxies from diverging . we now make the assumption that the numerical values of @xmath150 and @xmath149 given in eep are quite reliable , but the numerical values of @xmath148 and @xmath156 might be less reliable , because of the difficulty of detecting galaxies at the low - luminosity end . instead , we shall solve for the values of @xmath148 and @xmath156 . this requires two constraints . we impose that the mean density @xmath157 of galaxies matches the value of @xmath158 that we assume in our simulations , and that the mean luminosity density matches the value @xmath159 given in eep . equation ( 40 ) allows us to directly compute the present number density @xmath157 , and luminosity density @xmath160 , @xmath161\,,\end{aligned}\ ] ] where @xmath162 and @xmath163 . the last equality in equation ( 42 ) was obtained by integrating by part , and then substituting in equation ( 41 ) . we now substitute the numerical values of @xmath149 , @xmath164 , and @xmath150 ( with @xmath165 ) , and get , @xmath166 this system of equations can be solved numerically for @xmath167 and @xmath168 . the solution is @xmath169 since , for all calculations presented in this paper , we assume a hubble constant of @xmath170 , equation ( 45 ) can be rewritten as @xmath171 , which is , within error bars , consistent with the value given in eep . we adopt the galaxy models described in jaroszyski ( 1991,1992 ) . the projected surface density of each galaxy is given by @xmath172 where @xmath89 is the projected distance from the center . the parameters @xmath173 , @xmath174 , and @xmath175 are the core radius , maximum radius , and velocity dispersion , respectively , and are given by @xmath176 @xmath177 @xmath178 we use a monte - carlo method to generate for each galaxy a luminosity @xmath179 , with a probability @xmath180 proportional to @xmath147 . this ensures that the ensemble of @xmath181 galaxies in the computational volume follow the luminosity function given by equation ( 40 ) . then , we compute the galaxy parameters using equations ( 48)(50 ) . since each plane represents a different region of the universe , the large - scale structure inside each plane should be uncorrelated with the large - scale structure inside the neighboring planes . this is clearly a problem if the lens planes originate from one single calculation , since they would then represent the same large - scale structure at various evolutionary stages . to solve this problem , we perform five independent calculations for each cosmological model , by using five different sets of initial conditions . we then choose randomly which calculation will provide each lens plane , making sure that two consecutive lens planes never come from the same calculation . to eliminate correlations even more , we make use of the periodic boundary conditions by giving to the galaxy and background matter distributions in each lens plane a random shift . this is equivalent to choosing randomly on each lens plane the location where the beam will hit . we could eliminate correlations even more by rotating and/or reflecting the galaxy distributions before projecting them onto the lens planes , using the 48-fold symmetry of the cubic computational volume , but we consider this to be an overkill . in this paper , we use the term `` background matter '' to refer to the total matter in the universe , if the presence of galaxies is ignored . hence , the distribution of background matter in the universe at various redshifts is what the p@xmath1 m code simulates . to compute the effect of the background matter on the propagation of the beam , we solve equation ( 13 ) numerically on a two - dimensional grid ( a similar technique was used by blandford et al . [ 1991 ] ) . we compute the right - hand side of equation ( 13 ) on a square grid , using the location of the particles provided by the p@xmath1 m code , and invert equation ( 13 ) using a fast fourier transform method which is essentially the method that the p@xmath1 m algorithm itself uses for solving the three - dimensional poisson equation . the details of the calculation are given in appendix a. for all simulations presented in this paper , the comoving size @xmath102 of the computational volume was @xmath182 , and the grid used for solving equation ( 13 ) was @xmath183 in size . therefore , the grid spacing at redshift @xmath106 was @xmath184 . this has the effect of smoothing out any density fluctuation in the background matter on scales below @xmath185 . this is consistent with the assumption made by jaroszyski ( 1991 ) that the actual background matter distribution in the universe at that scale should be smooth . of course , galaxies contain dark matter halos with are presumably smaller than this , but these dark matter halos are taken into account in the galaxy profiles given by equation ( 47 ) . in this subsection , we are considering the smoother component of the background matter that has not been accumulated into galactic halos . the calculation of the background matter potential described above is sufficient for computing the effect of distant matter on the propagation of the beam . however , at distances less than a few mpc s , we can not ignore the fact that matter has collapsed to form galactic - size objects which are much smaller than the resolution of the p@xmath1 m algorithm or the algorithm used for solving equation ( 13 ) . this is why we added galaxies to the simulations using the method described in 3.3 . each lens plane contains about 40000 galaxies , but only the ones located near the beam can have a significant effect on its evolution . this enables us to greatly reduce the computation time by only including nearby galaxies . we identify one particular ray in the beam as being the `` central ray . '' then , in each lens plane , we only compute the contribution of the galaxies which are within a projected distance @xmath186 of the central beam . in this paper , we chose @xmath187 . hence , the mean number of galaxies per lens plane included in the calculation is @xmath188 . of course , the actual number varies over a wide range because galaxies are clustered . the calculation of the background matter contribution takes the _ total _ matter in the system into account . therefore , for every galaxy we add to the calculation , we must subtract something in order to conserve mass . we make the assumption that each galaxy has formed by accumulating matter that was originally distributed over a region of comoving radius @xmath189 , which is of order the present mean spacing between galaxies . this accumulated matter should be removed from the calculation . this is done by putting on the top of each galaxy a `` hole '' with radius @xmath190 and negative density , so that the combined mass of the galaxy and hole is zero . in this model , each galaxy has formed by accumulating matter for a region of fixed size , the more massive galaxies simply accumulating more matter from that region . this model is of course crude . one might suggest that more massive galaxies accumulate matter from a larger region . however , the relation between the galaxy mass and the size of that region is unknown at present , due to our limited understanding of the galaxy formation process . furthermore , this relation most certainly depends of the environment ( whether the galaxy forms in isolation or in a cluster ) . until better models for galaxy formation are available , it is a reasonable approximation to use a constant value for @xmath191 . to eliminate spurious edge effects , we do not use a hole with a flat negative density profile . instead , we use a gaussian density profile with a fwhm equal to @xmath191 . the calculation of the potential for the galaxies and the holes are given in appendix b. we consider three different cosmological models : an einstein - de sitter model with @xmath2 , @xmath109 , an open model with @xmath3 , @xmath109 , and a flat , low - density model with @xmath3 , @xmath4 , where @xmath7 and @xmath192 are the present values of the density parameter and cosmological constant , respectively . we set the present value @xmath9 of the hubble constant equal to @xmath170 to avoid conflict between the models and the measurements of globular cluster ages . with these parameters , the age of the universe @xmath10 is 13.0 gyr , 16.6 gyr , and 21.04 gyr for the einstein - de sitter , open , and cosmological constant models , respectively . in all cases , the comoving length of the computational volume is @xmath193 ( present length units ) . the total mass of the system is @xmath194 . we use @xmath195 equal mass particles . the mass per particle is therefore @xmath196 for the einstein - de sitter model and @xmath197 for the other two models . for all simulations , we use the cold dark matter ( cdm ) power spectrum of bardeen et al . ( 1986 ) , with the normalization of bunn , scott , & white ( 1995 ) . as mentioned in mpm , we use the same power spectrum for all three models , which is inconsistent , since the cdm power spectrum depends upon @xmath7 and @xmath192 . our motivation for doing this is the following : our goal in this paper is not to find which model fits the observations of the present universe better ( we defer this to a forthcoming paper ) . instead , we want to select cosmological models that will bracket the behavior of the large - scale structure formation process . using the same power spectrum for all models allows us to investigate directly the effects of the growth rate and the age of the universe on the evolution of the beam . in the same spirit , we are considering open models and models with a cosmological constant that are somewhat too extreme to agree with recent observations , which suggests that @xmath7 is more likely to be somewhere in the range 0.250.5 ( ostriker & steinhardt 1995 ; martel , shapiro , & weinberg 1997 , and references therein ) . models with a larger @xmath7 and/or a smaller @xmath192 would reproduce observations better , but would resemble the einstein - de sitter model more than the ones we are considering , thus providing less insight on the effect of the cosmological parameters on the beam evolution . the reader should therefore keep in mind that the power spectra we are using for the open and cosmological constant models are not consistent with a standard cdm model , and are chosen only for practical considerations . we ran 5 simulations for each of the three cosmological models , for a total of 15 simulations . for each model , the 5 simulations differ only in the ensemble of random phases used for generating the initial conditions ( see mpm ) . all simulations start at an initial redshift @xmath198 , and end at @xmath0 . in all experiments , we propagate a beam composed of several light rays backward in time , starting from the image plane , located near the observer , and ending at the source plane , located at @xmath199 . there are 55 lens planes for the einstein - de sitter model , 73 for the open model , ant 96 for the cosmological constant model . in this experiment , the light beams consist of 65 rays arranged in two concentric rings of 32 rays each , plus a center ray . the rings diameters on the image plane are @xmath200 and @xmath201 , corresponding to angular sizes 1.54 and 2.30 arc seconds , respectively . we performed 500 calculations for each of the 3 cosmological models . for each calculation , we used a different seed for the random number generator that computes the random shifts of the lens planes . hence , the calculations within each model differ from one another in the location on the lens planes where the beam hits . figure 3 shows the configuration of the beam on the source plane , located at @xmath199 , was determined by using equation ( 34 ) with @xmath202 , and @xmath203 being the redshift of the last lens plane . the actual redshifts are @xmath204 , @xmath205 , and @xmath206 , for the einstein - de sitter , open , and cosmological constant models , respectively ] for a subset of 27 calculations , 9 for each cosmological models . the labels `` eds , '' `` o , '' and `` @xmath207 '' in this figure and all the subsequent ones identify the einstein - de sitter model , the open model , and the cosmological constant model , respectively . the panels labeled `` null '' show for comparison the configuration of the beam in the absence of lensing , computed using equation ( 2 ) with @xmath208 . the size of the panels is @xmath209 . the beam has a smaller diameter for the eds model than the other two , because of the dependence of the lensing equation upon the angular diameter distance . the deformation of the beam is comparable for the eds and @xmath207 models . this results from the combination of two different effects that partly cancel each other : on one hand , the large - scale structure is more developed in the eds model than in the @xmath207 model . on the other hand , there are almost twice as many lens planes between @xmath0 and @xmath101 in the @xmath207 model than in the eds model . figure 4 shows the individual contribution of each lens plane to the shear as a function of the lens - plane redshift @xmath106 for 3 particular runs , one for each model . these results were obtained by averaging the magnification matrix over all 65 rays in the beam . the sharp variations result from the absence of correlation between neighboring lens planes . the beam may experience a strong shear in one particular plane simply because the beam happens to pass near a large cluster . in order to eliminate this source of noise , we average the shear over all 500 calculations for each model . the results are shown in figure 5 . the lens planes that contribute most to the shear are located at intermediate redshifts , of order @xmath6 , for all three models . the contribution of lens planes located near the source or near the observer is significantly smaller . figure 6 , shows the individual contribution of each lens plane to the magnification , as a function of the lens - plane redshift @xmath106 for 3 particular calculations , one for each model , using again the average magnification matrix . as in figure 4 , the large fluctuations are caused by the absence of correlations between consecutive lens planes . figure 7 shows the result of averaging the magnification over all 500 calculations for each model . as for the shear , the lens planes that contribute the most to the magnification are located at intermediate redshifts . notice that the average magnification is almost always larger than unity . this is not a violation of flux conservation . if we were averaging the magnification matrix over _ sources _ , the average magnification would be exactly unity . but we are instead averaging over _ light rays_. this clearly weights in favor of sources with @xmath210 , since more light rays originate from these sources , which is why they are magnified in the first place . hence , averaging the magnification matrix over light rays instead of sources will gives @xmath211 . the shear is much larger for the einstein - de sitter model than the other two models . even though the large - scale structure is more evolved for the einstein - de sitter model , this can not account for the difference . in particular , the cosmological constant model resembles the einstein - de sitter model much more than the open model at late time ( mpm ) . hence , the differences between the structure in the various models is not sufficient to explain the results shown on figure 5 . the correct explanation is quite simple : though all models contain the same mass _ in galaxies _ , the _ total mass _ is 5 times larger for the einstein - de sitter model than for the other two models , for which @xmath3 . we will prove this affirmation below . notice that even though the contribution of individual lens planes is larger for the einstein - de sitter model , the number of such planes between @xmath0 and @xmath101 is smaller , so we do not necessarily expect the cumulative shear and magnification for distant sources to be larger for this model . for all 1500 calculations , and all lens planes , we computed the ratio @xmath212 , which measures the relative contributions of the shear and convergence to the magnification ( see eq . the largest value was @xmath213 , implying that the contribution of the shear to the magnification is totally negligible , for all models , all calculations , and at all redshifts . this is a well - known result ( lee & paczyski 1990 ; jaroszyski et al . 1990 ) . the most interesting result that comes out of these calculations is the fact that the largest contribution to both shear and magnification comes from matter located at intermediate redshifts . equation ( 6 ) shows that the critical surface density is large for lens planes located near the image plane ( @xmath214 small ) , or near the source plane ( @xmath215 small ) , resulting in a small deflection potential at small and large redshifts , and therefore `` favoring '' the lens planes located at intermediate redshifts . however , large - scale structures grow with time , an effect that favors lens planes located at small redshift . furthermore , since we assume that the physical size of galaxies does not evolve , the total cross section of the galaxies is larger in the past since galaxies are closer to one another . hence , we expect a larger number of direct hits of galaxies by the beam at larger redshift . on figure 8 , we plotted the number of galaxies hit by the beam at each redshift , averaged over all 500 calculations for each model ( which explains why the numbers are not integer ) . effectively , the number of galaxies hit increases monotonically with redshift , with an average of 1 galaxy hit by the beam at @xmath101 for the einstein - de sitter model , and 2 for the other models ( the beams diverge more in the open and cosmological models that in the einstein - de sitter model [ see figs . 2 and 3 ] , so more galaxies get hit in these models ) . this shows that , in spite of the fact that there are more structures at small redshift and more galaxies hit at large redshift , the geometrical factors in equation ( 6 ) dominate , making the individual contribution of lens planes to the shear and magnification larger at intermediate redshift . notice that on figure 5 , the average shear peaks at a redshift @xmath216 for the einstein - de sitter model , while it peaks at redshift of order @xmath217 for the other models . we interpret this result as follows : in model with @xmath60 , the linear growth of the density perturbation `` freezes out '' at redshift @xmath218 , whereas in an @xmath2 model , linear growth persists all the way to the present . hence , in the einstein - de sitter model , there is significant growth taking place at small redshift , giving a `` boost '' to the shear for lens planes located at @xmath216 relative to the ones located near @xmath219 . this effect results in a shift of the peak toward smaller redshift in figure 5 . figures 47 show the individual effect of each lens plane on the propagation of the beam . to get the cumulative effect of all the lens planes , we need to combine them using the formalism described in 2 . figure 9 ( solid lines ) shows the distributions of cumulative aspect ratios computed using equation ( 32 ) . these distributions are very different for the different models . the distribution is narrow for the open model and broad for the other two models . the distributions have a very similar shape for the einstein - de sitter and cosmological constant model , but the mean value of the distribution is significantly larger for the cosmological constant model . figure 10 ( solid lines ) shows the distributions of cumulative magnifications , computed using equation ( 33 ) . as for the distributions of aspect ratios , the distributions of magnifications are broad and similar in shape for the einstein - de sitter and cosmological constant model , the latter one being shifted to larger values , while the distribution is narrow for the open model . these distributions are characterized by a sharp increase on the low side and a more extended tail on the high side . they are qualitatively in agreement with the analytical and numerical results obtained by various authors ( schneider & weiss 1988a ; lee & paczyski 1990 ) we ran an additional 450 calculations , 150 for each of the three cosmological model , with the source plane located at redshift @xmath220 instead of 5 . figures 9 and 10 ( dotted lines ) show the distributions of aspect ratios and magnifications for these calculations ( we multiplied the counts in each bin by @xmath221 to allow a direct comparison ) . they are qualitatively very similar to the distributions for sources located at @xmath222 . the aspects ratios are smaller , and the magnifications are closer to unity , but the relative similarity and differences between the various models are the same . the only difference is in the aspect ratios , where the distribution for the cosmological constant model is not shifted to larger values compared with the einstein - de sitter model . the distributions of aspect ratios peak at values of order 1.1 , which is somewhat large compared with observation of lensed quasars at that redshift . this simply indicates that the standard cdm model normalized to cobe does not reproduce observations well . returning to the 1500 calculations with @xmath222 , we plot in figure 11 the average shear versus the redshift of the lens plane , where the shear is computed by including either the contribution of the background matter only ( solid curves ) or the contribution of the galaxies only ( dashed curves ) . notice that the total shear , shown in figure 5 , is not equal to the sum of these components , since it is the matrix elements @xmath223 and @xmath224 , and not the shear @xmath225 , that add up . this figure shows that the contributions of the background matter and galaxies are nearly identical for the open and cosmological constant model , while the background matter contribution greatly exceeds the galaxies contribution for the einstein - de sitter model . this shows the importance of including the contribution of distant background matter to the evolution of the beam , something that was overlooked in some previous studies ( schneider & weiss 1988a , b ; jaroszyski 1991 , 1992 ) . we should point out , however , than these results are obtained by averaging over 500 calculations for each model . in one individual calculation , the effect of a single massive galaxy might dominate over the effect of the background if the beam hits or nearly hits that galaxy . in this experiment , the beam consists of @xmath226 light rays arranged in a square lattice . the spacing between rays on the source plane is 1 arcsecond , about the size of an extended radio source . we ran 10 simulations for each cosmological model . figure 12 shows the final configuration of the beam , for 6 particular runs , two for each model . the overall deformation of the square array into an irregular , 4-side polygon is caused by the background matter . this deformation is large for the einstein - de sitter and cosmological constant models , and small for the open model , consistent with the results shown in figure 3 . in all cases , there are small regions on the source plane where several rays converge . this convergence results from the presence of galaxies . a source located in one of these regions has in general multiple images . we could use these results to estimate the fraction of high - redshift quasars with multiple images simply by computing the fraction of the source plane which is covered by these regions . from figure 12 , we can estimate that a few percent of high - redshift quasars have multiple images , by counting the number of rays in the regions of convergence . are known . ] to get statistically significant results , however , we need to perform many more calculations . also , in order to estimate the multiplicity ( double , triple , ... ) of each lensing event , we need to improve the resolution of these calculations , by increasing the number of rays per unit solid angle in the beam . hence , the results we are presenting in this subsection are for illustrative purpose only . a detailed study of the statistics of multiple images of quasars will be presented in a forthcoming paper . we redid the calculation shown in the middle left panel of figure 12 with twice the resolution per dimension ( @xmath227 rays ) . the results are shown in figure 13 . increasing the resolution reveals several additional regions of convergence . only two such regions are clearly visible in figure 12 , whereas there are 6 clearly visible regions of convergence in figure 13 . we have developed a numerical algorithm for studying the propagation of light in inhomogeneous universes . this is the first algorithm that combines fully nonlinear n - body simulations of large - scale structure formation with realistic distributions of galaxies , reproduces the 2-point correlation function of galaxies , and takes the morphology - density relation into account when ascribing morphological types to galaxies . as a result , this algorithm reproduces the matter distribution in the universe with a level of realism that is unprecedented . the density structure of the lens planes is simulated over 9 orders of magnitude in length , from the size of superclusters and voids down to the core radii of small galaxies . we use this new algorithm to study the propagation of light beams in inhomogeneous universes , for three different cosmological models : the einstein - de sitter model , an open model with @xmath3 , and a flat cosmological constant model with @xmath3 , @xmath4 . we performed 1981 simulations , propagating light beams back in time , up to a redshift of @xmath228 or 5 . the average magnitude of shear and magnification amongst models are shown to be different , with the values for the cosmological constant model being significantly larger than for the other two models , for sources located at @xmath101 . the contribution of individual lens planes to the shear and magnification is larger for planes located at intermediate redshift , of order @xmath229 , even though structures are more evolved at low redshift and direct hits of galaxies are more frequent at high redshift . the lens planes providing the largest average contribution to the shear are located at lower redshift for the einstein - de sitter model than for the other two models . the contribution of distant background matter to the shear is as important as the contribution of nearby galaxies ( see fig . 11 ) for low @xmath7 models , and significantly more important for the einstein - de sitter model . these results , combined with observations of lensed quasars , might eventually put limits on the value of the cosmological parameters @xmath7 and @xmath192 . this paper has focussed on the description of the method . applications of this method to various problems will be presented in forthcoming papers . these include studying the statistics of multiple imaging of quasars , and their dependence upon the source redshift ( premadi , martel , & matzner 1997a ) , performing a cosmological parameter survey , for several values of the cosmological parameters @xmath9 , @xmath7 , and @xmath192 , and for various density perturbation spectra ( premadi , martel , & matzner 1997b ) , and modifying the algorithm to include the effect of microlensing by stars inside galaxies ( martel , premadi , & matzner 1997b ) this work benefited from stimulating discussions with alan dressler , daniel holz , and inger jrgensen . we are pleased to acknowledge the support of nasa grant nag5 - 2785 , nsf grants phy93 10083 and asc 9504046 , the university of texas high performance computing facility through the office of the vice president for research , and cray research . we compute the potential of the background matter by solving equation ( 13 ) numerically . we first rewrite this equation as @xmath230 where we have introduced the mean surface density @xmath231 to be consistent with the filled - beam approximation , which requires that the mean surface density in each lens plane vanishes ( schneider & weiss 1988a , b ) . to solve this equation , we first use the location of the particles provided by the p@xmath1 m code to compute the source term @xmath232 in equation ( a1 ) on a square grid of size @xmath233 , using the _ triangular shaped cloud _ ( tsc ) assignment scheme ( hockney & eastwood 1981 ) . the values @xmath234 of @xmath232 at each grid point @xmath235 is given by @xmath236 where @xmath237 is the particle mass , @xmath238 is the grid spacing , @xmath239 , @xmath240 are the coordinates of particle @xmath241 , and @xmath242 , @xmath243 are the coordinates of the grid point @xmath235 . for the tsc assignment scheme , the weight function @xmath244 is given by @xmath245 once the function @xmath232 has been computed on the grid , we solve equation ( a1 ) using a finite difference method . the finite - difference form of equation ( a1 ) is @xmath246 where we used the standard 5-point formula for the two - dimensional laplacian . since the grid has periodic boundary conditions , we can easily invert this equation using fourier techniques . we write the potential as @xmath247 where @xmath248 is the discrete fourier transform of @xmath249 ( not to be confused with the dimensional potential ) . we use a similar expression for the source term @xmath232 . we eliminate @xmath249 and @xmath232 in equation ( a4 ) and get , after some algebra , @xmath250 to solve for the potential , we compute @xmath251 by taking the two - dimensional fast fourier transform ( fft ) of @xmath232 on the grid . we then use equation ( a6 ) to compute the function @xmath248 on the grid , and take the two - dimensional inverse fft on the grid to get @xmath249 . once we know the potential on the grid , we can compute its first and second derivatives using standard finite difference formulae , @xmath252 this gives us the derivatives at the grid points . to interpolate these derivatives at the location of the rays , we use again the tsc assignment scheme , @xmath253 where @xmath254 are now the coordinates of the ray , the sum is on the grid points , and @xmath255 stands for any of the derivatives given by equations ( a7)(a11 ) . the deflection potential caused by a galaxy is given by equation ( 5 ) , @xmath256 with @xmath262 being the angle between the two vectors @xmath263 and @xmath264 . equation ( b2 ) then becomes @xmath265ds\nonumber\\ & = & { 4\pi g\over c^2 } \int_0^{r_{\max}}\sigma(s)s\ln\left({r^2+s^2+|r^2-s^2|\over2}\right)ds \,,\end{aligned}\ ] ] where we used the fact the @xmath266 is a function of @xmath267 only to perform the integration over @xmath262 . if the point @xmath268 is located outside the density distribution , then @xmath269 for all @xmath270 , and equation ( b4 ) reduces to @xmath271 where @xmath276 . the first and second derivatives are then given by @xmath277\,,\\ { \partial\hat\psi_{\rm gal}\over\partial y}&=&{2\pi u^2y\over r^2 } \bigl[(r_{\max}^2+r_c^2)^{1/2}-r_c\bigr]\,,\\ { \partial^2\hat\psi_{\rm gal}\over\partial x^2}&=&{2\pi u^2(y^2-x^2)\over r^4 } \bigl[(r_{\max}^2+r_c^2)^{1/2}-r_c\bigr]\,,\\ { \partial^2\hat\psi_{\rm gal}\over\partial x\partial y}&= & -{4\pi u^2xy\over r^4}\bigl[(r_{\max}^2+r_c^2)^{1/2}-r_c\bigr]\,,\\ { \partial^2\hat\psi_{\rm gal}\over\partial y^2}&=&{2\pi u^2(x^2-y^2)\over r^4 } \bigl[(r_{\max}^2+r_c^2)^{1/2}-r_c\bigr]\,,\end{aligned}\ ] ] where @xmath136 and @xmath278 are the components of @xmath263 . for the case @xmath279 ( eq . [ b6 ] ) , we get , after some algebra @xmath280\nonumber\\ & + & ( r^2+r_c^2)^{1/2}-r_c\ln\bigl[r_c+(r^2+r_c^2)^{1/2}\bigr]\biggr\ } \,.\end{aligned}\ ] ] notice that the first two terms are constant and thus do not contribute to the derivatives . the first and second derivatives are given by @xmath281\,,\\ { \partial\hat\psi_{\rm gal}\over\partial y}&=&{2\pi u^2y\over r^2 } \bigl[(r^2+r_c^2)^{1/2}-r_c\bigr]\,,\\ { \partial^2\hat\psi_{\rm gal}\over\partial x^2}&=&{2\pi u^2\over r^4 } \biggl\{(y^2-x^2)\bigl[(r^2+r_c^2)^{1/2}-r_c\bigr ] + { x^2r^2\over(r^2+r_c^2)^{1/2}}\biggr\}\,,\\ { \partial^2\hat\psi_{\rm gal}\over\partial x\partial y}&= & -{4\pi u^2xy\over r^4}\biggl[(r^2+r_c^2)^{1/2}-r_c -{r^2\over2(r^2+r_c^2)^{1/2}}\biggr]\,,\\ { \partial^2\hat\psi_{\rm gal}\over\partial y^2}&=&{2\pi u^2\over r^4 } \biggl\{(x^2-y^2)\bigl[(r^2+r_c^2)^{1/2}-r_c\bigr ] + { y^2r^2\over(r^2+r_c^2)^{1/2}}\biggr\}\,.\end{aligned}\ ] ] to conserve mass , we superpose on the top of each galaxy a `` hole '' of negative density , which represents the matter that has been used up to form the galaxy . this hole has a volume density given by @xmath282 where @xmath283 is the mass of the galaxy , and @xmath284 is the three - dimensional radial distance . the normalization constant in equation ( b20 ) was chosen such that the total mass of the galaxy and hole vanishes . the projected surface density is given by @xmath285 the last integral can not be solved using elementary functions . this is not a problem , since we are only interested in the derivatives of the potential . after differentiation , we get @xmath288\,,\\ { \partial^2\psi_{\rm hole}\over\partial y^2}&= & -{m\over\pi r^4}\left[(x^2-y^2)(1-e^{-w^2})+2y^2w^2e^{-w^2}\right]\,,\\ { \partial^2\psi_{\rm hole}\over\partial x\partial y}&= & { 2mxy\over\pi r^4}\left[1-e^{-w^2}-w^2e^{-w^2}\right]\,,\end{aligned}\ ] ] where @xmath289 . notice that these derivatives are well - behaved at @xmath290 . by combining equations ( b9)(b13 ) with equation ( b23)(b27 ) , we see that the combined potential of the galaxy and hole drops to zero as @xmath89 goes to infinity . in practice , we neglect the combined potential at distances larger than @xmath291 .

we present the first galaxy evolution results from the local cluster substructure survey ( locuss ) , a multi - wavelength survey of 100 x - ray selected galaxy clusters at . locuss combines far - uv through far - ir observations of cluster galaxies with gravitational lensing analysis and x - ray data to investigate the interplay between the hierarchical assembly of clusters and the evolution of cluster galaxies . here we present new panoramic _ spitzer_/mips observations of the merging cluster abell 1758 at spanning and reaching a completeness limit of . we estimate a global cluster star - formation rate of within of the cluster centre , originating from 42 galaxies with . the obscured activity in a1758 is therefore comparable with that in cl0024 , the most active cluster previously studied at . the obscured galaxies faithfully trace the cluster potential as revealed by the weak - lensing mass map of the cluster , including numerous mass peaks atmpcthat are likely associated with infalling galaxy groups and filamentary structures . however the core ( ) of a1758n is more active in the infrared than that of a1758s , likely reflecting differences in the recent dynamical history of the two clusters . the 24 results from a1758 therefore suggest that dust - obscured cluster galaxies are common in merging clusters and suggests that obscured activity in clusters is triggered by both the details of cluster - cluster mergers and processes that operate at larger radii including those within in - falling groups . our ongoing far - uv through far - ir observations of a large sample of clusters should allow us to disentangle the different physical processes responsible for triggering obscured star formation in clusters . [ firstpage ] galaxies : active galaxies : clusters : general galaxies : evolution galaxies : stellar content

0903.4911

unravelling the physics of how gas - rich disk galaxies , recently arrived in galaxy clusters , are transformed into the quiescent elliptical and s0 populations that dominate dense cluster cores at @xmath14 is one of the longest - standing unsolved problems in astrophysics . numerous studies of massive early - type galaxies in clusters have revealed predominately old stellar populations with apparently very little growth in stellar mass in massive cluster galaxies since @xmath15 @xcite . this indicates that the majority of stars in cluster early - type galaxies formed at high redshift , @xmath16 . however @xcite found that the fraction of cluster members with blue optical colours increases from zero in the local universe to @xmath17 by @xmath18 , suggesting that the fraction of cluster galaxies that are actively forming stars increases with increasing lookback time . empirically the star - forming spiral galaxies found by @xcite at @xmath19.20.4 are mostly replaced by s0 galaxies in local clusters @xcite . indeed , s0s may be completely absent at @xmath20 @xcite , although @xcite suggest that the s0 fraction remains constant over @xmath21 ( see also * ? ? ? a simple interpretation is that clusters accrete blue gas - rich star - forming spirals at @xmath22.51 and that these galaxies were transformed somehow into the passive s0s found in local clusters . numerous mechanisms have been proposed to deplete the reservoir of gas in late - type spiral galaxies and thus cut off the fuel supply for further star - formation , leading to transformation into s0 galaxies ( for reviews see * ? ? ? * ; * ? ? ? historically most of the work on this area has concentrated on optical wavelengths , for example identifying the so - called e+a galaxies @xcite . these galaxies exhibit deep balmer absorption lines and lack nebular emission lines indicating that these apparently passive galaxies were actively forming stars in the preceding @xmath23.51.5gyr . indeed , simple evolutionary models have been used to connect , via aging and thus fading , this newly quiescent population with actively star - forming spiral galaxies . mid - infrared observations of cluster galaxies with _ iso _ and _ spitzer _ have challenged the idea that all spiral galaxies that fall into clusters simply fade into passive galaxies via an intermediate e+a phase . for example , _ iso _ revealed a population of luminous infrared galaxies ( @xmath24 lirgs ) and near - lirgs in clusters that imply that the star - formation rates derived from optical diagnostics ( e.g. [ oii ] ) under - estimate the true star formation rate ( sfr ) by @xmath2503@xmath26 ( see * ? ? ? * for a review ) . deep radio observations have also revealed that many e+as harbour significant ongoing star - formation missed by optical diagnostics due to heavy dust obscuration @xcite . the simple optically - derived fading of spiral galaxies therefore seems to be substantially incomplete . radio and mid - infrared observations have also revealed an order of magnitude cluster - to - cluster variations in the amount of obscured star formation in clusters @xcite . these differences have broadly been attributed to differences in the recent dynamical history of the clusters ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ) , however early _ iso _ results had already suggested that a simple one - to - one relationship between cluster - cluster mergers and obscured star formation does not exist . for example , _ iso _ identified 10 lirgs in cl0024 and none in a1689 , a2218 , and a2390 ( coia et al . , 2005b ; see also metcalfe et al . , all four clusters are well known strong lensing clusters , the mass distributions of which are constrained by their strong lensing signal to be multi - modal , i.e. indicative of recent cluster - cluster merging ( e.g. * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? indications of the likely complex relationship between the dynamical state of clusters and their current star - formation rate was also found in the pioneering optical studies . for example , for the coma cluster @xcite found numerous galaxies with post - starburst characteristics ( strong balmer - line absorption ) between the two x - ray peaks corresponding to the main coma cluster and the ngc4839 subcluster , suggestive of starbursts and subsequent quenching resulting from a cluster - cluster merger . in contrast , @xcite found no enhancement in the blue galaxy distribution between the two x - ray peaks for an apparently similar system abell 168 . for this latter case , @xcite suggests that the lack of enhanced star - formation could be due to the galaxies being previously gas - poor and hence unable to be triggered on interaction with their environment . @xcite measured the fraction of cluster galaxies with blue optical colours out to @xmath27 , although their sample was dominated by clusters at @xmath28 . the blue galaxy fractions in these `` intermediate '' redshift clusters span the full range of values found in clusters at @xmath14 and @xmath27 , i.e. @xmath2900.2 . numerous uv / optical , infrared and radio surveys have broadly confirmed that the scatter is large , however none have combined the sample size and multi - wavelength dataset required to relate directly actively star - forming galaxies in clusters to differences in the intracluster medium and recent dynamical history of clusters . the local cluster substructure survey ( locuss ; pi : g. p. smith ; http://www.sr.bham.ac.uk/locuss ) is a systematic multi - wavelength survey of 100 galaxy clusters at @xmath0 drawn from the rosat all sky survey cluster catalogues @xcite . the overall goal of the survey is to probe the relationship between the recent hierarchical infall history of clusters ( as revealed by strong and weak lensing observations ; * ? ? ? * ) and the baryonic properties of the clusters . this results in two complementary aspects of the survey : mass - observable scaling relations @xcite , and evolution of cluster galaxy populations . both of these themes are connected by the aim of measuring and understanding the physical origin of the cluster - to - cluster scatter . this is the first in a series of papers about the first batch of 31 clusters observed by locuss , for which the following data have been gathered : wide - field subaru / suprime - cam imaging , _ hubble space telescope _ wfpc2 and/or acs imaging of the cluster cores , wide - field _ spitzer_/mips @xmath1 maps to match the half degree suprime - cam field of view , _ galex _ near- and far - ultraviolet ( nuv / fuv ) imaging , and wide - field near - infrared ( nir ) imaging obtained via a combination of ukirt / wfcam and kpno-4m / newfirm . we have also been awarded 500ksec on _ herschel _ as an open time key program to observe this sample at @xmath30 and @xmath31 with pacs . the galaxy evolution goals that we will tackle with these data include to measure robustly the scatter in the amount of obscured star formation in clusters at `` low '' redshift @xcite , and to cross - correlate the location of the obscured cluster galaxies with mass over - densities found in the weak - lensing mass maps derived from the subaru data @xcite . full details of the survey design will be given in a future paper , however here we emphasize that at the nominal redshift of our sample , @xmath32 , the suprime - cam field - of - view covers @xmath33 centred on each cluster . a @xmath34 @xcite dark matter halo has a virial radius of @xmath35 . our survey therefore probes out to @xmath36 the virial radius of the clusters . the uv and ir data discussed above all cover at least the same field of view as the subaru data . this survey is therefore sensitive to the infall regions of clusters studied recently by , for example , @xcite and @xcite . in this paper we present an analysis of abell 1758 ( a1758 ) at @xmath2 as a case study on the influence of cluster - cluster mergers on obscured star formation , and to outline the methods that will be applied to the full sample in future papers . a1758 was selected for this study because previous x - ray and lensing studies have identified it as a merging cluster , comprising two gravitationally bound components one to the north and one to the south ( a1758n and a1758s respectively ) both of which are undergoing a merger @xcite . we note that throughout the paper , we describe the star - formation measured from the @xmath1 data as obscured , making no distinction between heavily obscured star - formation or the much less attenuated diffuse emission from normally star - forming spirals as done by @xcite . the most straightforward interpretation @xcite of the @xmath1 emission is that it traces the dust obscured star - formation , while the observed uv or h@xmath37 emission traces the unobscured one @xcite . in future , we will combine uv and ir data to obtain robust measures of both obscured and unobscured star - formation in cluster galaxies , as well as the level of extinction . in [ sec : data ] we describe our observations of a1758 , plus archival data from the xmm - lss field that we use to establish optimal colour selection criteria to identify galaxies at the cluster redshift . in [ sec : analysis ] and [ sec : results ] we present the photometric analysis and the main results respectively . we discuss the results in [ sec : discuss ] and summarize them in [ sec : conc ] . when necessary we assume , and , giving a lookback time of @xmath38 and an angular scale of @xmath39 arcmin@xmath40 at the cluster redshift . all magnitudes are quoted in the vega system unless otherwise stated . the 24@xmath41 m data were obtained with mips @xcite on board the _ spitzer space telescope _ @xcite , and consist of two separate aors observed on january 6 and may 17 , 2008 ( pid:40872 , pi g. p. smith ) . the pointing centres of the two aors are separated by 6@xmath42 , each covering a field of view of 25@xmath43 with a @xmath44 grid . in the earlier aor , the central pointing was excluded by using the fixed cluster mode ( with 400 offsets ) since the area had already been imaged by a guaranteed time observations program 83 to a much deeper depth ( @xmath453000s / pixel ) . however , the second aor uses the raster mode ( i.e. , with no gaps ) , and as a result , it contains the gto area as well . for both the fixed - cluster and raster aors , we performed at each grid point two cycles of the small - field photometry observations with a frame time of 3s , producing a per pixel exposure of 90s for each aor , or 180s in the overlapping region . the 24@xmath41 m data were reduced and combined with the data analysis tool ( dat ) developed by the mips instrument team @xcite . a few additional processing steps were also applied as described in @xcite . the data were resampled and mosaicked with half of the original instrument pixel scale ( 1245 ) to improve the spatial resolution . the clusters were observed with wfcam @xcite on the 3.8-m united kingdom infrared telescope ( ukirt ) on march 10/20 2008 ( i d : u/08a/32 , pi : g. p. smith ) . a total exposure time of 800sec per pixel was accumulated in both @xmath46- and @xmath47-bands using the same macrostepping ( @xmath48 offsets in ra and dec to cover the large gaps between the detectors ) , jittering and @xmath49 microstepping strategy used by the ukidss deep extragalactic survey ( dxs * ? ? ? the data were reduced and processed by the cambridge astronomical survey unit , and photometrically calibrated to the vega magnitude system using 2mass stars in the fields . the final pixel scale of the ukirt data is @xmath50 ; point sources detected in both filters have a full width half maximum ( fwhm ) of @xmath51 . we also make use of wide - field optical imaging from the 8.3-m subaru telescope . these suprime - cam @xcite data ( i d : s05a-159 , pi : n. okabe ) have been used by @xcite to study the mass distribution of a1758 via the cluster s weak lensing signal the weak lensing mass map from okabe & umetsu is reproduced in fig . [ fig : spatial ] . full details of the observations and data reduction are provided by okabe & umetsu . the reduced data consist of stacked @xmath52 and @xmath53 ( hereafter @xmath54 ) exposures totalling 720s in @xmath52 and 2880s in @xmath54 . point sources in both filters have fwhms of @xmath55 . suprime - cam s field of view is @xmath56 with a pixel scale of @xmath57 . both optical frames were registered onto the @xmath47-band frame using the iraf tools geomap and geotran . the transformations were defined by comparison of the positions of 100300 sources in each image , and were found to be accurate to 0.10.@xmath58 across the whole field . in [ sec : colours ] we use broad - band optical / near - ir colours of @xmath1 sources to identify them as likely cluster galaxies or not . these selection criteria are derived from analysis of archival data on the xmm - lss field at @xmath59 . this field was observed at the relevant wavelengths by swire ( * ? ? ? * @xmath1 ) , ukidss dxs ( * ? ? ? * @xmath60 ) , the cfht legacy survey d1 field ( @xmath61 ) , and the vimos vlt deep survey ( * ? ? ? * vvds ) . the properties of the swire data are similar to our own @xmath1 discussed in [ sec : a1758data ] & [ sec : phot ] : an overall exposure time per pixel of 80s , a final pixel scale of @xmath62 , and complete to @xmath63 . in our analysis we use the swire xmm - lss dr2 catalogue from the entire 9deg@xmath64 xmm - lss field . the ukidss dxs @xmath65-band data also match our data well , with total integration times of 640sec , the same macro- and micro - stepping strategy , the same final pixel - scale , and comparable image quality of @xmath66 . the final reduced ukidss dxs @xmath65-band frames were obtained from the ukirt science archive . the cfhtls d1 @xmath67-band frames were obtained from the cfhtls terapix first data release at the canadian astronomy data centre . these data have slightly worse image quality than okabe & umetsu s subaru data ( @xmath68 ) and are @xmath69 deeper . crucially for the calibration of our photometric selection methods described in [ sec : colours ] , a large spectroscopic redshift catalogue is available for the xmm - lss field . the virmos team obtained 8981 redshifts down to a limit of @xmath70 , of which 1543 correspond to @xmath71 galaxies i.e.brighter than @xmath72 at @xmath73 . in [ sec : results ] we consider only those with redshift quality flags of 2 or greater , indicating a @xmath74 confidence in the redshift . the @xmath1 mosaic of a1758 was analysed with sextractor @xcite ; following swire we estimated the flux of objects within an aperture of diameter @xmath75 , and applied an aperture correction of @xmath76 . the flux detection limits and completeness of the mosaic were determined by individually inserting 500 simulated sources into the mosaic for a range of fluxes and determining their detection rate and recovered fluxes , using identical extraction procedures . the sources used in the simulations were formed by extracting isolated , high signal - to - noise and unresolved sources from the mosaic itself . from these simulations , we estimate that the @xmath4 completeness limit of our @xmath1 mosaic is @xmath77 . the optical and near - ir data on both a1758 and xmm - lss were analysed with sextractor @xcite in two frame mode , such that objects were selected in the @xmath47-band . the sextractor mag_auto magnitude was adopted for the @xmath47-band total magnitudes , and colours were measured from the @xmath78-band frames within @xmath79 diameter apertures , after degrading the optical images to have the same fwhms as the @xmath47-band data . note that the xmm - lss @xmath80-band photometry was transformed to @xmath54 using the @xmath81 colour correction of lupton ( 2005 ) . the additional uncertainty introduced by this transformation is 0.0072mag . these optical / near - ir catalogues were then used to identify the optical / near - ir counterpart of each @xmath1 source in the a1758 and xmm - lss catalogues , as the nearest @xmath47-band source ( where one exists ) within @xmath82 of the @xmath1 centroid . from the completeness simulations described above we expect the positional uncertainty of @xmath1 sources to be @xmath83 . in total 571 @xmath84 sources were detected in the mips mosaic of a1758 . of these 455 have galaxy counterparts with @xmath71 , 26 are stars , and 90 have no @xmath71 counterpart within @xmath82 . these latter sources we assume to be at high redshifts . galaxy colours depend on a complex combination of star - formation history , metallicity , redshift and dust extinction , whose effects are often difficult to disentangle , especially if only optical data are available . for example , it is particularly important to consider the reddening effects of dust when considering ir - luminous sources whose spectral energy distributions ( seds ) are likely to be strongly affected by dust obscuration . fortunately , the seds of galaxies are much less affected by both ongoing star - formation history and dust in the near - ir than the optical ; the @xmath85 colour varies by just @xmath86 across the entire hubble sequence . in contrast the @xmath85 colour increases monotonically with redshift to @xmath87 , from @xmath88 at @xmath89 to @xmath90 at @xmath91 . this near - ir redshift - colour relationship can be seen in fig . [ vvds ] galaxies of a particular redshift lie along a single narrow colour - magnitude ( c - m ) relation , the slope of which is due to the mass - metallicity relation whereby galaxies become systematically more metal - rich with mass ( e.g. * ? ? ? note that we do not see a separate red sequence and blue cloud , due to the insensitivity of the @xmath85 colour to current star - formation rate . we exploit the fact that galaxies of a given redshift lie on such a narrow c - m relation to empirically removing the effects of metallicity on galaxy colour by fitting the slope of the c - m relation , and subtracting it out . we optimize this for the cluster redshift , selecting those @xmath71 vvds galaxies in the redshift range 0.2540.304 ( shown as large green symbols in the right panel of fig . [ vvds ] ) and fitting their @xmath92 and @xmath93 c - m relations using the biweight estimator , obtaining the following transforms : @xmath94 we show the resulting @xmath95 colour - colour plot for a1758 and xmm - lss in fig . [ a1758cols ] . the robust separation of stars and galaxies in @xmath95 colour - space is striking ; stars have much bluer @xmath96 colours than galaxies . the only overlap between stellar and galaxy loci is at @xmath97 , @xmath98 , i.e. for galaxies at @xmath99 which are of no interest in this study . star - galaxy separation is therefore achieved by selecting as galaxies objects with _ either _ extended morphologies _ or _ @xmath95 colours outside the stellar locus . using colour information to perform star - galaxy separation instead of a conventional extended versus unresolved morphological classification has the important advantage of removing the potential bias against very compact galaxies . the large green symbols in the right panel of fig . [ a1758cols ] indicate those xmm - lss galaxies at the cluster redshift ( @xmath100 ) , and can be seen to be well confined to a small region in the @xmath95 colour - space , and we use this to define a robust colour - colour selection criterion of @xmath101 as indicated in fig . [ a1758cols ] by the black box . we use the colours of those xmm - lss galaxies with spectroscopic redshifts to estimate the completeness and purity of the galaxy sample selected using these colour cuts . we find that the completeness of the colour - selected sample is @xmath102 i.e. @xmath4 of xmm - lss galaxies at @xmath103 satisfy the colour cuts . the sample has a purity of @xmath104 i.e. @xmath105 of the galaxies selected in this way actually lie at @xmath103 . varying the colour cuts would adjust the trade - off between completeness and purity . these colour cuts were chosen to optimise for completeness , with the impurity being addressed via a statistical contamination subtraction ( [ sec : global ] ) . we find no evidence within the xmm - lss data that we are missing dusty star - forming galaxies at the cluster redshift by not extending to redder @xmath106 or @xmath107 colours . finally we note that objects with blue optical - nir colours ( @xmath108 ) are predominately quasars ( not at the cluster redshift ) , hence necessitating the blue cut in @xmath107 . we therefore apply these colour cuts to the a1758 photometric catalogue , yielding 793 probable cluster members with @xmath109 and within @xmath110 of the cluster centre ( left panel of fig . [ a1758cols ] ) note that the observed concentration of galaxies with @xmath111 , @xmath112 , corresponding to the early - type galaxies in a1758 , provides independent confirmation of the validity of the colour selection criteria developed above from the xmm - lss data . the angular distribution of galaxies selected to be likely cluster members in this manner is over - plotted on the weak - lensing mass map in fig . [ fig : spatial ] . hereafter we refer to galaxies that satisfy the colour cuts developed in this section as `` cluster galaxies '' . we searched ned for galaxies with spectroscopic redshifts within the subaru field of view . out of the 25 galaxies with redshifts , 9 lie at @xmath113 and are therefore classified as cluster members . these 9 include the bcg of both a1758n ( @xmath114 ) and a1758s ( @xmath115 ) , but none of the galaxies that are detected at 24@xmath41 m . nevertheless all 9 spectroscopically confirmed members satisfy our @xmath95 colour selection criteria , confirming its high expected completeness . moreover , all eleven of the galaxies with @xmath116 were excluded from the cluster galaxy catalogue on the basis of having too blue @xmath85 colours , confirming our ability to remove foreground galaxies . we are less able to remove background galaxies , with 5 of 7 galaxies with @xmath117 classed as probable cluster members , an issue also apparent from analysis of the vvds galaxies in the xmm - lss field . in future articles , we will have much more complete redshift information both for this cluster and the other 30 clusters in the locuss sample , through a recently commenced large spectroscopic programme to obtain redshifts for 200400 members per cluster . we now turn to the population of dust - obscured galaxies in a1758 : 82 of the 793 probable cluster members have @xmath84 within 12.5arcmin ( @xmath118 ) of the centre of a1758 , producing a combined @xmath1 flux of @xmath119 . we estimate the predicted contamination of field @xmath1 sources by using the optical / near - infrared photometry of mid - infrared sources from the swire xmm - lss survey . in total we identify 165 @xmath109 sources with @xmath84 lying in the same @xmath120 selection box used to identify cluster members , within a total region of 2025arcmin@xmath64 . correcting for this level of field contamination , we report an excess of 42 mid - infrared sources with @xmath84 , producing a combined flux of @xmath121 . the uncertainty in this last value , and the global measurements that follow , are estimated by assuming each 24@xmath13 source in the a1758 field has a probability of ( 42/82 ) of being a cluster member , and averaging over many monte carlo realizations of randomly selected cluster populations , such that the results are robust against any individual 24@xmath13 source turning out to be a cluster member or not . . the sfrs are from the mid - infrared populations within @xmath122mpc , and these are normalized to the best estimate of the cluster mass , derived by weak lensing estimates or from the x - ray luminosities . the solid point corresponds to the estimate for a1758 . the errors are derived from bootstrap resampling of the mid - infrared distribution . the open points and upper / lower limits for the remaining clusters are taken from @xcite . also plotted is an evolutionary model for the counts of star - forming ulirgs from @xcite.,width=302 ] our @xmath1 flux limit of @xmath77 corresponds to average total infrared luminosities of @xmath123 in a1758 . this translates into a star - formation rate of @xmath124 , assuming the star - formation calibration of @xcite . they consider a range of sed templates to estimate the total infrared luminosity , @xmath12581@xmath126 , of galaxies from their @xmath1 flux , as a function of redshift . this calibration has a significant amount of scatter due to the range of possible infrared seds , which at @xmath2 corresponds to @xmath127dex . integrating over all the @xmath1 sources within @xmath118 of the cluster centre , we estimate a global cluster star - formation rate of @xmath128 , including a 0.1dex uncertainty in the overall sfr calibration . if we assign each individual cluster galaxy to either a1758n or a1758s based on proximity to the respective cluster centres then the estimated star formation rate of each cluster is @xmath129 and @xmath130 respectively . following @xcite , we also divide our global cluster sfr by the cluster mass to estimate the specific star formation . adopting the lensing results of @xcite , we estimate that the mass of a1758 , within the same @xmath131 aperture used in this paper of @xmath132 . this mass translates to a specific sfr of @xmath133 placing a1758 among the more actively star - forming clusters ( fig . [ geach ] ) . in the above estimates we have neglected the possible contribution from agn . we now investigate this by searching for x - ray point sources associated with the @xmath1 sources in archive _ xmm - newton _ and _ chandra _ x - ray images , which cover both a1758n and a1758s @xcite , reaching sensitivities of @xmath1340.510kev@xmath135ergs@xmath40 at the cluster redshift . among our 82 probable cluster members with @xmath136jy we identify just two with coincident x - ray point sources , consistent with the typical x - ray selected agn fractions in low - redshift ( @xmath137 ) clusters of 15% @xcite . we note that this may be an underestimate of the agn contamination , as many mir - selected agn are not detected in x - ray surveys ( e.g. * ? ? ? in the future we will also be able to identify agn from their optical spectra based on the emission - line ratios @xcite . colour - magnitude diagrams for photometrically selected cluster members within 1mpc of abell 1758n ( _ top panel _ ) and 1758s ( _ bottom panel _ ) . black circles indicate those with @xmath138 , the area of each symbol being proportional to the @xmath1 flux . crosses indicate those galaxies not detected by _ spitzer_.,width=302 ] a final global measurement is the fraction of obscured star - forming galaxies in a manner analogous to the pioneering measurements of blue galaxy fractions by @xcite . the advantage of making these measurements in the mid - ir is that identifying galaxies as star - forming solely on their optical colours is prone to under - estimating the amount of star formation due to the obscuring effects of dust in the most actively star - forming galaxies . following @xcite , we estimate the fraction of cluster galaxies brighter than @xmath139 and located within 1mpc of the cluster centre that are detected by mips , @xmath140 , separately for the clusters a1758n and a1758s . for each cluster galaxy we derived rest - frame @xmath141-band absolute magnitudes and @xmath142 colours from the observed subaru @xmath52 and @xmath54-band magnitudes . following @xcite , this was done by deriving linear transformations from observed to rest - frame magnitudes for model galaxies at the cluster redshift covering a range of star - formation histories , using the stellar evolution code of @xcite . in fig . [ a1758 cm ] we show the resulting rest - frame @xmath143 colour - magnitude diagrams for cluster galaxies within 1mpc of a1758n and a1758s . based on a comparison of source counts in these regions and the xmm - lss field , we expect that field contamination to be at the level of one - in - ten for @xmath144 galaxies , rising to one - in - four for those detected by _ spitzer_. we see that most @xmath1 sources would also be classified as blue by the butcher - oemler criterion ( i.e. blueward of the dashed line in each panel ) , but a significant fraction ( @xmath145% ) of the star - formation in a1758 comes from galaxies with colours consistent with those of early - type galaxies . taking into account field galaxy contamination , we estimate the fraction of cluster galaxies brighter than @xmath139 and located within 1mpc of the cluster centre that are detected by _ spitzer _ ( i.e. @xmath84 ) to be @xmath146 for a1758n , and @xmath147 for a1758s ( fig . [ bo ] ) . by comparison to those galaxies in the xmm - lss field satisfying the same colour selection criteria , we estimate that @xmath148 of @xmath144 ( assuming that they are at @xmath149 ) field galaxies at these redshifts have @xmath84 , somewhat higher than observed in the two clusters , as expected from the sf - density relation ( e.g. * ? ? ? * ; * ? ? ? we note that in terms of @xmath140 a1758s appears the more active cluster , yet in terms of the global sfr , a1758n appears twice as active as a1758s , due primarily to a1758n being somewhat richer than a1758s . cluster galaxies within 1mpc of the cluster centre , that are star - forming as revealed by the mips @xmath1 observations , as a function of redshift . the solid symbols correspond to a1758n ( lower ) and a1758s ( upper ) at @xmath2 and are offset slightly in @xmath150 for clarity . the open symbols are taken from @xcite.,width=302 ] sources within 12.5arcmin of the a1758 cluster centre , and whose @xmath95 colours place them at the cluster redshift . the @xmath1 luminosity function is based on statistical subtraction of @xmath1 sources in the xmm - lss field satisfying the same colour criterion . scaled estimates of the bolometric infrared luminosities , based on the calibration of @xcite are indicated along the top axis . , width=302 ] in figure [ fig : lf ] we show the differential number counts ( solid black points ) of @xmath1 sources with @xmath95 colours consistent with the cluster redshift and within @xmath118 of the cluster centre . the open points show the differential 24@xmath13 number counts after correcting for field galaxy contamination on the basis of the @xmath1 sources in the xmm - lss field satisfying the same colour selection , providing an estimate of the @xmath1 luminosity function for a1758 . according to this , almost all of the cluster @xmath1 sources have @xmath151 , which corresponds to a bolometric infrared luminosity at @xmath152 @xmath153 based on the calibration of @xcite . we estimate there to be @xmath154 lirgs in abell1758 , the same number as found for cl0024 , although clearly without any redshift information this number is quite uncertain . we also identify one possible cluster ulirg ( @xmath155 ) , which appears from the optical image to be a merging galaxy system with a combined nir luminosity of @xmath156 . we now turn to the morphology of the dust - obscured galaxies in a1758 . ideally this analysis would be based on wide - field high resolution imaging with the _ hubble space telescope _ , allowing detailed morphological classifications across the full field ( e.g. * ? ? ? * ) . in the absence of such data , we analyse the subaru @xmath54-band frame , using 2dphot @xcite to derive structural parameters for each cluster galaxy . 2dphot is an automated tool to obtain both integrated and surface photometry of galaxies , the main steps of which are : i ) estimation of the fwhm and the robust identification of stars ; ii ) construction of an accurate psf model , taking into account both possible spatial variations as well as non - circularity ; iii ) derivation of structural parameters ( effective radius @xmath157 , mean surface brightness @xmath158 , and sersic index @xmath159 ) by fitting 2d psf - convolved sersic models to the data . the high - quality subaru data , obtained in excellent conditions , allows us to obtain structural parameters accurate to @xmath160 . -band magnitudes of those galaxies within 12.5arcmin of the a1758 cluster centre whose @xmath161 colours place them at the cluster redshift . circles indicate galaxies detected with _ spitzer _ , the area of each symbol is proportional to the @xmath1 flux . crosses indicate those galaxies not detected in @xmath1.,width=302 ] in fig . [ fig : morph ] we plot sersic index @xmath159 versus @xmath47-band magnitude for all cluster members . those galaxies detected by _ spitzer _ are indicated by circles whose area is proportional to the @xmath1 flux , while non - detections ( @xmath162 ) are shown as crosses . sersic indices can be used to achieve a crude morphological classification into bulge- and disk - dominated galaxies ( e.g. * ? ? ? * ; * ? ? ? * ) , and so adopting a nominal dividing line of @xmath163 we classify the _ spitzer_-detected cluster galaxies as being either bulge - dominated ( @xmath164 ) or disk - dominated ( @xmath165 ) . this yields the following fractions : @xmath166 of mid - ir - bright cluster galaxies are bulge - dominated and @xmath167 are disk - dominated where the quoted uncertainties are @xmath168 binomial error bars following @xcite . spitzer_-detected cluster galaxies are therefore dominated by disk galaxies , indeed many have large axis ratios which may suggest that some of the dust obscuration is caused by viewing these galaxies close to edge - on . however , we see no significant difference in the axis ratios between those disk galaxies detected by spitzer and those not , which points against any major inclination effect . although a quarter of mid - infrared bright cluster galaxies are early - types , these galaxies contribute just ( @xmath169 ) of the total 24@xmath13 flux estimated in [ sec : global ] . hence , even if all of the 24@xmath13 emission in these galaxies is due to agn , they do not significantly effect our global sfr estimates in [ sec : global ] . in figure [ fig : spatial ] we over - plot the spatial distribution of cluster galaxies in a1758 on the projected cluster mass distribution from okabe & umetsu ( 2008 ; black contours ) . the spatial distribution of cluster early - type galaxies ( filled red squares ) closely follow the underlying dark matter distribution . the cores of both of the merging components of a1758n ( at @xmath170 , @xmath171 and @xmath172 , @xmath173 see [ sec : discuss ] ) are particularly dominated by early - type galaxies . numerous mass peaks of lower significance e.g. at @xmath174 , @xmath175 ; @xmath176 , @xmath175 ; and @xmath177 , @xmath178 ) , also contain over - densities of early - type galaxies . the distribution of dusty star - forming galaxies ( @xmath179 ; blue circles ) is much less concentrated spatially than the early - types , consistent with being an infalling population . an important caveat is that @xmath180 of these galaxies are probably field contaminants that should be randomly distributed across the field . indeed , these field galaxies may help to explain the population of star - forming galaxies that are not associated with mass - over - densities in the weak - lensing mass map . on the other hand most of the dusty galaxies are indeed associated with over - densities in the mass map . in addition to the population of dusty disk galaxies that are associated with mass over - densities at large radii presumably in - falling galaxy groups we also find dusty disk galaxies occupying the central region ( projected radii of @xmath181 ) of a1758n . in contrast the central region of a1758s is completely devoid of dusty disk galaxies . we also identify a third significant population of galaxies in a1758 passive spiral galaxies , defined empirically by : @xmath182 , @xmath165 and undetected at @xmath1 , i.e. @xmath183 ( yellow circles in fig . [ fig : spatial ] ) . we identify 37 such galaxies all of which lie within 3mpc of the centre of either a1758n or a1758s . this population trace the underlying mass distribution very well , and appears to be more concentrated than the dusty spiral galaxies , but less concentrated than the early - type galaxy population . the apparent concentration of these objects towards the cluster centre , suggests that these are a cluster population . this is confirmed by their relative rarity in field regions we find a density of just @xmath184 such galaxies in the xmm - lss data , corresponding to an expected field contamination of just @xmath185 within a region of projected radius @xmath186 . to examine the relative concentrations of the three galaxy populations discussed above more quantitatively , we show in fig . [ fig : radial ] the radial distributions of early - type galaxies , passive spirals , and dusty star - forming galaxies in a1758 , after correcting for field galaxy contamination . the number density of all three populations decreases with increasing projected cluster - centric radius , confirming that all three are associated with the cluster . the distribution of late - type galaxies ( whether detected or not at @xmath1 ) is less centrally concentrated than the distribution of early - type galaxies ; the radial distributions of @xmath1-detected and undetected late - type galaxies are consistent with the errors out to @xmath187 , however at larger radii there is an excess of @xmath1-detected galaxies over both early - types and passive spirals . this excess corresponds to the galaxies discussed above that are associated with galaxy groups detected in the out - skirts of the clusters in the weak - lensing map . @xcite and @xcite have studied a1758 in detail as a merging cluster using weak lensing and x - ray data respectively . in particular @xcite investigated the thermodynamics of both cluster components . here we compare our results on obscured star - formation in a1758 with the thermodynamics of the intracluster medium , and thus with the likely dynamical state of the cluster . @xcite found no evidence of compressional heating of the intracluster medium ( icm ) in the region between a1758n and a1758s , suggesting that the two clusters have not begun to merge with each other yet , although they may be gravitational bound to each other already . a1758n itself is interpreted as being a large impact parameter merger between two 7kev clusters , one cluster ( to the north - west ) currently moving to the north and the other ( to the south - east ) currently moving to the south - east . in contrast , the merger in a1758s is interpreted as having a much smaller impact parameter with the merger occurring close to the line - of - sight through the cluster . the core radii of the merging components within a1758n are @xmath188 ; in contrast the merging components of a1758s have core radii of @xmath189 and @xmath190 . @xcite therefore suggested that , as a consequence of the differing impact parameters , the galaxies in the merging clusters of a1758s have experienced lower ram pressures than those in a1758n , and that a1758s may be at an earlier stage in its merger than a1758n . however , both a1758n and a1758s do contain regions of hotter gas that have presumably been shock - heated during the respective mergers . the precise timing difference between the two mergers is difficult to ascertain , however @xcite proposed that a1758n is seen significantly after the point of closest approach whilst a1758s may be seen at approximately the epoch of closest approach . okabe & umetsu s weak lensing mass map successfully resolves the two merging components of a1758n , and locates a single mass peak in the core of a1758s , both of which are consistent with david & kempner s interpretation of the x - ray data . how well do our mid - ir results fit with the x - ray and lensing results ? one of the most striking features of the distribution of obscured galaxies in fig . [ fig : spatial ] is their absence from the central @xmath191 region of a1758s , in contrast to the same region in a1758n , especially in the vicinity of the south - east component of a1758n . this difference between the clusters is reminiscent of the large cluster - to - cluster scatter within the core regions of merging clusters found with _ intense star formation can be triggered in cluster - cluster mergers by the passage of gas - rich galaxies through shocked regions of the icm @xcite . in a1758n the merger - induced shock - front(s ) have dissipated @xcite , which helps to explain why shock - fronts are not found adjacent to the dusty galaxies in the core of that cluster . a1758s is claimed to be an earlier stage in the merging process , however @xcite did detect shock - heated gas in this cluster . therefore naively one would expect that star - formation would have been triggered by the shocks as in a1758n . however no @xmath1 sources are found in the central @xmath181 region of a1758s . the merger geometry of a1758s may help to explain the absence of obscured galaxies in its core . for example the @xmath1 sources at @xmath192 may have been scattered to large radii by the line - of - sight merger that is ongoing in the core of the cluster in a manner analogous to the distribution of cluster galaxies in cl0024 , another line - of - sight merger cluster @xcite . assuming that the galaxies are moving at @xmath193 it would take them @xmath194 to travel to a radius of @xmath195 from the centre of the cluster . on the face of it , this is inconsistent with david & kempner s proposal that the sub - clusters in the a1758s merger are approximately at closest approach . an alternative interpretation of the @xmath1 data might be that the shocks in a1758s were not strong enough to trigger detectable star - formation in the cluster galaxies that passed though them . we also note that a1758s is much less optically rich than a1758n , so the absence of obscured activity in a1758s could simply be due to a1758s being a poorer cluster than a1758n . this is supported by fig . [ bo ] because the fraction of star - forming @xmath1-bright galaxies in both clusters is consistent within the errors . in summary , our mid - ir results are qualitatively consistent with the previous x - ray and lensing results . however the absence of obscured galaxies in the core of a1758s remains a puzzle . one possibility is that the galaxies were already gas - poor before the cluster merger , and hence had no available fuel to undergo new star - formation . our results confirm and extend the view formed from _ iso _ that the cluster - to - cluster scatter in merging clusters is large , and likely related to the physical parameters of cluster mergers . the global sfr of a1758 estimated from the @xmath1 flux of photometrically selected members is @xmath128 , qualifying it as one of the most infrared active clusters studied to date . for example it is comparable with the most active cluster studied to date cl0024 at @xmath196 with @xmath197 @xcite . globally this is unsurprising because a1758 contains two ongoing cluster - cluster mergers and one ( between a1758n and a1758s ) that may occur within a few gyr . however the picture is more subtle when considering the specific sfr ( @xmath198 ) and star - forming fraction of galaxies ( @xmath199 see figs . [ geach ] & [ bo ] respectively ) . cl0024 stands out in fig . [ geach ] as the most active cluster by a factor of @xmath200 in @xmath198 because of it s low mass relative to a1758 . on the other hand , cl0024 and a1758 have comparable values of @xmath201 ( fig . [ bo ] ) , indicative of the clusters having comparable optical richness . this further underlines the conclusion that numerous parameters influence the level of obscured star - formation in galaxy clusters : merger geometry , density of the icm , location of shock - heated gas with respect to the cluster galaxies , the available supply of gas in cluster galaxies in which star - formation may be triggered by shock - heating of the icm , the optical richness of the cluster , and cluster mass . we will investigate these issues in detail with the full sample in a future paper . we find a significant population of passive spiral galaxies , defined empirically by @xmath182 , @xmath165 and @xmath202 . the radial distribution of these galaxies is intermediate in radial extent between the cluster early - types and the actively star - forming galaxies detected at @xmath1 ( fig . [ fig : radial ] ) . if the radial extent of each distribution indicates the length of time that each population has on average spent in the cluster potential well , then the passive spirals have been accreted by a1758 less recently than the actively star - forming galaxies . a simple interpretation is therefore that the passive spirals were formerly actively forming stars and thus brighter at @xmath1 , i.e. that there is an evolutionary path linking the dusty star - forming galaxies , passive spirals and cluster early - type galaxies ( see e.g. * ? ? ? @xcite identified a significant population of passive spirals in cl0024 over a wide range of environments , and used a combination of _ galex _ uv photometry and optical spectroscopy to find that their star - formation quenching time - scales are often @xmath203 , suggestive more of starvation , although the rapid quenching expected for ram - pressure stripping was also observed . as the simulations of @xcite show , gas is lost from infalling galaxies over a wide range of environments ( even beyond the virial radius ) on both short and long time - scales , and @xcite estimate that the observed frequency and quenching time - scales of passive spirals in @xmath204 clusters could account for the entire build - up of s0s between @xmath205 and @xmath206 . the relationship between moran et al.s uv / optical selected passive spirals and our optical / ir selected passive spirals is currently unclear . for example , the evolutionary path outlined above may not include the uv / optical passive spirals discussed by @xcite . on the other hand , if the two populations of passive spirals are indeed the same class of object , then the evolutionary path may not be valid . in that case an alternative interpretation of the differing radial distributions of dusty and passive spirals could be that the physical processes that trigger star formation in the former population are more efficient at the largest radii than the processes that act to quench star formation in the latter population . the _ galex _ , _ spitzer _ , _ herschel _ , ground - based near - ir photometry and optical spectroscopy dataset that we are assembling on a large sample of clusters will be very powerful for disentangling the respective galaxy populations . we have presented the first results from our panoramic multi - wavelength survey of 31 clusters at @xmath207 clusters as part of the local cluster substructure survey ( locuss ; pi : g. p.smith ) . for this initial study we chose a1758 , an actively merging cluster at @xmath152 that has been previously studied with high quality weak lensing and x - ray data @xcite . specifically , we have combined a wide - field ( @xmath208 ) _ spitzer_/mips @xmath1 map of the entire a1758n / a1758s complex with panoramic @xmath65-band imaging from ukirt / wfcam and archival optical data from subaru to map out the locations of dusty star - forming galaxies in addition to cluster early - types and cluster spirals that are undetected down to @xmath5 . the main results are as follows : \(i ) we detect 82 probable cluster members in the mid - ir with _ spitzer_/mips at @xmath1 within 3mpc of the cluster centre . after correcting for residual field contamination using archival data from the xmm - lss , we measure an excess of 42 mid - ir sources over the field counts with @xmath84 , responsible for a total flux of @xmath209 . \(ii ) adopting @xcite s ( @xcite ) sed templates , we convert our 90% completeness limit of @xmath5 into a luminosity sensitivity of @xmath210 i.e. sub - lirg luminosities . most of the ir - luminous cluster sources appear to be sub - lirgs . we estimate there to be 10 lirgs within 3mpc of the cluster centre , comparable to the number in cl0024 . we identify just one possible cluster ulirg ( @xmath155 ) , which appears to be a merging galaxy system with a combined nir luminosity of @xmath211 . \(iii ) we also convert the total cluster flux at @xmath1 to a total cluster star formation rate of @xmath212 . this is consistent within the errors with the most active cluster studied to date : cl0024 @xcite . we split the @xmath1 sources between the a1758n and a1758s based the proximity of each galaxy to the respective cluster centres , to obtain @xmath213 and @xmath214 respectively i.e. a1758n is almost twice as active as a1758s . \(iv ) when the level of obscured activity is normalized to the mass and the optical richness of the cluster , then we find that the specific star - formation rate of a1758 is lower than that of cl0024 and that the fraction of cluster galaxies that are detected at @xmath1 is comparable with cl0024 . these results reflect the facts that a1758 is more massive than and of comparable richness to cl0024 . \(v ) dust - obscured galaxies with late - type morphologies are detected in a network of infalling groups and filamentary structures that surround both a1758n and a1758s that is delineated by the reconstructed cluster mass distribution from @xcite . however in the core regions of each cluster we find that the central @xmath11 region of a1758n is comparatively rich in dusty galaxies in contrast to the absence of such objects in the same region of a1758s . this difference may be due to the differing recent dynamical histories of the two clusters , as discussed by @xcite . \(vii ) finally , we identify a population of ir - selected passive spiral galaxies , i.e. they have a late - type morphology and are undetected at @xmath1 . the radial distribution of these galaxies does not extend as far as that of the mips - detected population , suggesting that they were accreted into the cluster potential on average longer ago than the active dusty population . however the difference in radial distribution may also indicate that they are a distinct population in transition from spiral to early - type morphology , and that they are affected by different physical process than the dusty galaxies . comparison of these ir passive spirals with uv - selected passive spirals will help to clarify the status of both populations . the next steps in this study are to repeat this analysis for a large ( @xmath215030 ) sample of clusters in order to measure for the first time the scatter in the infrared properties of clusters . the biggest caveat on our results is the reliance on photometry to select cluster galaxies . we will therefore also significantly enhance the precision of our analysis as spectroscopic redshifts become available from our ground - based optical spectroscopic observations recently commenced with hectospec at the mmt . we also plan to integrate this optical / near - ir / mid - ir analysis with our _ galex _ observations of the same clusters , and _ herschel _ data when they become available . the overall goal is to pinpoint the contribution that the various physical processes make to the transformation of gas rich spiral galaxies into early - type cluster galaxies via a large and detailed study of clusters in a single redshift slice . cph and gps acknowledge financial support from stfc . gps and rse acknowledge support from the royal society . we acknowledge nasa funding for this project under the spitzer program go:40872 . we wish to thank the staff at ukirt , casu and wfau for making the observations and rapidly processing the nir data . 99 bai l. , et al . , 2007 , apj , 664 , 181 baldwin a. , phillips m. m. , terlevich r. , 1981 , pasp , 93 , 5 ball n. m. , loveday j. , brunner r. j. , 2008 , mnras , 383 , 907 barger a. j. , aragon - salamanca a. , ellis r. s. , couch w. j. , smail i. , sharples r. m. , 1996 , mnras , 279 , 1 bertin e. , & arnouts s. , 1996 , a&as , 117 , 393 biviano a. , et al . , 2004 , a&a , 425 , 33 bhringer h. , et al . , 2004 , a&a , 425 , 367 boselli a. , & gavazzi g. , 2006 , pasp , 118 , 517 bruzual g. , & charlot s. , 2003 , mnras , 344 , 1000 butcher h. , & oemler jr . a. , 1978 , apj , 219 , 18 butcher h. , & oemler jr . a. , 1984 , apj , 284 , 426 caldwell n. , rose j. a. , sharples r. m. , ellis r. s. , bower r. g. , 1993 , aj , 106 473 calzetti d. , et al . , 2007 , apj , 666 , 870 casali m. , et al . , 2007 , a&a , 467 , 777 coia d. , et al . , 2005a , a&a , 430 , 59 coia d. , et al . , 2005b , a&a , 431 , 433 cowie l. l. , barger a. j. , fomalont e. b. , capak p. , 2004 , apjl , 603 , 69 crowl h. h. , & kenney j. d. p. , 2008 , aj , 136 , 1623 czoske o. , moore b. , kneib j .- , soucail g. , 2002 , a&a , 386 , 31 david l. p. , kempner j. , 2004 , apj , 613 , 831 de propris r. , stanford s. a. , eisenhardt p. r. , holden b. p. , rosati p. , 2007 , aj , 133 , 2209 desai v. , et al . , 2007 , apj , 660 , 1151 dressler a. , 1980 , apj , 236 , 351 dressler a. , et al . , 1997 , apj , 490 , 577 dressler a. , rigby j. , oemler a. jr . , fritz j. , poggianti b. , rieke g. , bai l. , 2008 , arxiv/0806.2343 driver s. , et al . , 2006 , mnras , 368 , 41 duc p .- a . , et al . , 2004 , in iau colloq . 195 : outskirts of galaxy clusters : intense life in the suburbs , ed . a. diaferio , 347351 duc p .- a . , et al . , 2002 , a&a , 382 , 60 ebeling h. , edge a. c. , bohringer h. , allen s. w. , crawford c. s. , fabian a. c. , voges w. , huchra j. p. , 1998 , mnras , 301 , 881 ebeling h. , edge a. c. , allen s. w. , crawford c. s. , fabian a. c. , huchra j. p. , 2000 , mnras , 318 , 333 egami e. , et al . , 2006 , apj , 647 , 922 ellingson e. , lin h. , yee h. k. c. , carlberg r. g. , 2001 , apj , 547 , 609 fadda d. , elbaz d. , duc p .- a . , flores h. , franceschini a. , cesarsky c. j. , moorwood a. f. m. , 2000 , a&a , 361 , 827 gallazzi a. , et al . , 2009 , apj , 690 , 1883 geach j.e . , smail i. , ellis r. s. , moran s. m. , smith g. p. , treu t. , kneib j .- p . , edge a. c. , kodama a. c. , 2006 , apj , 649 , 661 gehrels n. , 1986 , apj , 303 , 336 gordon k. d. , et al . , 2005 , pasp , 117 , 505 haines c. p. , gargiulo a. , la barbera f. , mercurio a. , merluzzi p. , busarello g. , 2007 , mnras , 381 , 7 haines c. p. , et al . , 2009 , in preparation hickox r. c. , et al . , 2009 , preprint ( arxiv:0901.4121 ) holden b. p. , et al . 2007 , apj , 670 , 190 kauffmann g. , 1995 , mnras , 274 , 153 kennicutt r. c. jr . , et al . 2007 , apj , 671 , 333 kneib j .- p . , ellis r. s. , smail i. , couch w. j. , sharples r. m. , 1996 , apj , 471 , 643 la barbera f. , et al . 2008 , pasp , 120 , 681 lawrence a. et al . 2007 , mnras , 379 , 1599 le fvre o. et al . 2005 , a&a , 439 , 640 le floch e. , et al . 2005 , apj , 632 , 169 lewis i. et al . 2002 , mnras , 334 , 673 limousin m. , et al . , 2007 , apj , 668 , 643 lonsdale c. et al . 2003 , pasp , 115 , 897 marcillac d. , rigby j. r. , rieke g. h. , kelly d. m. , 2007 , apj , 654 , 825 martig m. , & bournaud f. , 2008 , mnras , l38 martini p. , kelson d. d. , kim e. , mulchaey j. s. , athey a. a. , 2006 , apj , 644 , 116 metcalfe l. , et al . , 2003 , a&a , 407 , 791 metcalfe l. , fadda d. , & biviano a. , 2005 , ssrv , 119 , 425 miller n. a. , & owen f. n. , 2003 , aj , 125 , 2427 miller n. a. , oegerle w. r. , & hill j. m. , 2006 , aj , 131 , 2426 miyazaki s. , et al . , 2002 , pasj , 54 , 833 moran s. m. , ellis r. s. , treu t. , smail i. , dressler a. , coil a. , smith g. p. , 2005 , apj , 634 , 977 moran s. m. , ellis r. s. , treu t. , smith g. p. , rich m. r. , smail i. , 2007 , apj , 671 , 1503 moss c. , 2006 , mnras , 373 , 167 muzzin a. , wilson g. , lacy m. , yee h. k. c. , stanford s. a. , 2008 , apj , 686 , 966 navarro j. f. , frenk c. s. , & white s. d. m. , 1997 , apj , 490,493 okabe n. , & umetsu k. , 2008 , pasj , 60 , 3450 okabe n. , takada m. , umetsu k. , futamase t. , smith g. p. , 2009 , preprint ( arxiv:0903.1103 ) owen f. , ledlow m. j. , keel w. c. , morrison g. e. , 1999 , aj , 118 , 633 poggianti b. m. , smail i. , dressler a. , couch w. j. , barger a. j. , butcher h. , ellis r. s. , oemler a. jr . , 1999 , apj , 518 , 576 poggianti b. m. , bressan a. , franceschini a. , 2001 , apj , 550 , 195 postman m. , et al . , 2005 , apj , 623 , 721 rieke g. h. , et al . , 2004 , apjs , 154 , 25 rizza e. , morrison g. e. , owen f. n. , ledlow m. j. , burns j. o. , hill j. , 2003 , aj , 126 , 119 roettiger k. , burns j. o. , & loken c. , 1996 , apj , 473 , 651 saintonge a. , tran k .- v . h. , holden b. p. , 2008 , apjl , 685 , 113 silverman j. d. , et al . , 2008 , preprint ( arxiv:0812.3402 ) sivakoff g. , martini p. , zabludoff a. i. , kelson d. d. , mulchaey j. s. , apj , 682 , 803 smail i. , et al . , 1999 , apj , 525 , 609 smith g. p. , treu t. , ellis r. s. , moran s. m. , dressler a. , 2005 , apj , 620 , 78 smith g. p. , kneib j .- p . , smail i. , mazzotta p. , ebeling h. , czoske o. , 2005 , mnras , 359 , 417 smith g. p. , & taylor j. e. , 2008 , apjl , 682 , 73 swinbank a. m. , bower r. g. , smith g. p. , smail i. , kneib j .- p . , ellis r. s. , stark d. p. , bunker a. j. , 2006 , mnras , 368 , 1631 tanaka m. , kodama t. , arimoto n. , okamura s. , umetsu k. , shimasaku k. , tanaka i. , yamada t. , 2005 , mnras , 362 , 268 tomita a. , nakamura f. e. , takata t. , nakanishi k. , takeuchi t. , ohta k. , 1996 , aj , 111 , 42 tonnesen s. , bryan g. l. , & van gorkum j. h. , 2007 , apj , 671 , 1434 tremonti c. et al . , 2004 , apj , 613 , 898 treu t. , et al . 2003 , apj , 591 , 53 werner m. w. , et al . , 2004 , apjs , 154 , 1 wolf c. , et al . , 2009 , mnras , 393 , 1302 zhang x. , 2008 , pasp , 120 , 121 zhang y .- y . , finoguenov a. , bhringer h. , kneib j .- p . , smith g. p. , kneissl r. , okabe n. , dahle h. , 2008 , a&a , 482 , 451

we revisit the uncertainty in baryon acoustic oscillation ( bao ) forecasts and data analyses . in particular , we study how much the uncertainties on both the measured mean dilation scale and the associated error bar are affected by the non - gaussianity of the non - linear density field . we examine two possible impacts of non - gaussian analysis : ( 1 ) we derive the distance estimators from gaussian theory , but use 1000 n - body simulations to measure the actual errors , and compare this to the gaussian prediction , and ( 2 ) we compute new optimal estimators , which requires the inverse of the non - gaussian covariance matrix of the matter power spectrum . obtaining an accurate and precise inversion is challenging , and we opted for a noise reduction technique applied on the covariance matrices . by measuring the bootstrap error on the inverted matrix , this work quantifies for the first time the significance of the non - gaussian error corrections on the bao dilation scale . we find that the variance ( error squared ) on distance measurements can deviate by up to 12% between both estimators , an effect that requires a large number of simulations to be resolved . we next apply a reconstruction algorithm to recover some of the bao signal that had been smeared by non - linear evolution , and we rerun the analysis . we find that after reconstruction , the rms error on the distance measurement improves by a factor of at low redshift ( consistent with previous results ) , and the variance ( ) shows a change of up to 18% between optimal and sub - optimal cases ( note , however , that these discrepancies may depend in detail on the procedure used to isolate the bao signal ) . we finally discuss the impact of this work on current data analyses . [ firstpage ] cosmology : observations dark energy large - scale structure of the universe distance scale methods : statistical

1106.5548

ever since acoustic peaks were detected in the cosmic microwave background @xcite and galaxy surveys @xcite , much effort has been devoted to use baryon acoustic oscillations ( bao ) as standard rulers to estimate cosmological distances with precision @xcite . bao signals are manifested as a wiggly feature in the matter power spectrum , and precise measurements could shed light on the dynamics of dark energy . the accuracy and precision of such measurement depends directly on the covariance matrix of the matter power spectrum , which is straightforward to compute for a gaussian density field using wick s theorem . however , it is known that non - gaussianities are significant in that covariance matrix @xcite and may have an impact on cosmological distance measurements . a non - linear model is needed for a non - gaussian calculation , hence some criterion for the accuracy of such effects is needed to quantify the precision of the measurement . in particular , an optimal bao measurement should in general incorporate a proper error weighting of the data , which involves the inversion of the full covariance matrix . the accuracy criterion must therefore be based on the confidence we have on the inverted matrix . the estimator constructed in most forecasts and data analyses follow the prescription of @xcite , which is a procedure that constructs both the estimator of the bao scale and the estimator of its uncertainty under the gaussian assumption . the method was originally cross - checked with a @xmath2 analysis of @xmath3 simulations , including a jackknife sub - sampling of the n - body data , and showed good agreement . the actual deviation between this proposed estimator and the optimal estimator constructed in this paper is indeed small , therefore consistent with the work of @xcite . in this paper , we aim at improving on the accuracy of that method , and present a first significant detection of the effect of non - gaussian errors on the measurement of the bao scale ( the effect of non - linear evolution , in the form of erasure of the wiggle signal , has been well - studied of course @xcite ) . some data analyses @xcite also improved on the original method by constructing a non - gaussian estimator for the bao dilation scale uncertainty . this is typically done by modeling the non - linearities in the density fields with mock catalogs , which are produced from log - normal densities @xcite , @xmath4 order perturbation theory @xcite , halo models @xcite , etc . such techniques all attempt to increase the robustness of the analysis by taking into account the coupling between the fourier modes . as mentioned above , an optimal analysis must be based on a reliable inverted covariance matrix , and the accuracy of the inverse matrix constructed from mock catalogs is yet to be demonstrated . the covariance matrix in power spectrum is a four point function that relates pairs of wave numbers . the error on this covariance consists of pairs of these pairs , and is indeed difficult to quantify . without this metric , however , one does not know the significance of a non - gaussian computation . in addition , @xcite have found significant departures between the covariance matrix constructed from lagrangian perturbation theory and that obtained from their @xmath5 simulations . also unknown is the accuracy of log - normal densities at modeling the true covariance matrix and its inverse . other analyses @xcite treat the mode coupling as coming exclusively from the survey selection function , following the widely used fkp @xcite prescription . this specific coupling effect can be reduced with other choices of power spectrum estimators , like that presented in @xcite . in both cases , however , the non - linear mode coupling is not modeled . when it comes to the impact on the bao dilation scale , a gaussian treatment of the data yields a sub - optimal estimation of the mean , and the error bars obtained that way are systematically biased , usually on the low side . in the limit where the sample is large enough , the value of the mean estimated in that fashion does converge to the `` true '' mean , but the estimated error bars never capture the correlation that occurs in the non - linear regime . many analyses attempt to correct for this bias with monte carlo simulations , however this effect is very small and takes a high accuracy and precision to observe . in the non - gaussian case , however , an inversion of the covariance matrix is required , hence it can not be singular . consequently , the convergence of our measured matrix depends on the binning , and the inversion increases the noise even more . for an adequate resolution on all the scales relevant for bao analyses , the number of simulations required to obtain statistically significant conclusions can be large . in the past , different groups used drastically different numbers of simulations : @xcite used 51 , @xcite used 400 , and @xcite used 5000 ; we use 1000 n - body simulations in this work . in order to invert a @xmath6 covariance matrix , one needs _ at least _ @xmath7 simulations to make the matrix non - singular . it is also generally thought that to achieve convergence on each element , we need of the order @xmath8 simulations . even then , the level of accuracy is not clear , and to make a significant claim about non - gaussian effects , one needs to know the uncertainty on the inverted matrix , which we measure from a bootstrap re - sampling , and to propagate the error onto the bao scale . to address this convergence issue , we also apply a noise reduction technique before the inversion : we factorize the covariance matrix with an eigenvector decomposition , and keep only the principal component . this factorization is repeated at each of the bootstrap samplings , which allows us to draw robust conclusions on the convergence of our results . given the fact that the precision of the inverse covariance matrices used in analyses has never been demonstrated , the measurement of the mean and of the error on the mean found on the literature are most likely not optimal . measuring an optimal bao scale in actual data is complicated in many aspects , such as the fact that the universe is not periodic , and that surveys have selection functions . it would nevertheless involve a covariance weighted measurement , which is not included in the prescription given by @xcite . if one could improve the measurement of the power spectrum covariance , however i.e. from n - body simulations , it would be possible to measure a more robust and more accurate _ uncertainty _ on the sub - optimal mean , compared to the original claim . the difference in performance between these two bao dilation scale estimator is still an unmeasured quantity . in this paper , we first attempt to address this question by comparing three different analysis scenarios : 1 . the first case we consider is an attempt at measuring a correct error bar on a sub - optimal mean of the bao dilation scale . even if we know that the universe is non - gaussian , it is still possible to treat it as gaussian , i.e. not use an optimally weighted sum when estimating the mean , even though the power spectrum itself is non - linear . doing so , we must keep in mind that the measurements are non - optimal , and that the naive gaussian error bars are most likely too small compared to the true " error . however , given the fact that we can measure a full covariance matrix from n - body simulations , we can get a better estimate of the error bars on that sub - optimal mean by treating the original covariance matrix as noisy and by performing an appropriate inverse covariance weighting . from now on , we refer to this case as the `` sub - optimal '' estimator of the bao error . this approach is commonly used to obtain `` monte - carlo '' error bars , and exactly what is measured by the forecasting prescriptions mentioned above . we call this approach `` sub - optimal '' because the errors bars could still be further reduced by improving the estimator of the mean bao scale using the non - gaussian model . it provides a quick estimate of the magnitude of non - gaussianity , and a simple way of scaling errors bars obtained from gaussian analysis , which is often used in surveys . the second case , dubbed `` optimal estimator '' , is the best quadratic analysis one can possibly do knowing that the universe is non - gaussian , we treat it as is , measuring a fully non - linear power spectrum covariance matrix and performing an optimally weighted sum to estimate both the means and the uncertainties on the parameters . given the fact that we rely on a large number of n - body simulations , that our volume is periodic , and that we have a high signal to noise , then both our estimators are truly optimal in the least squares sense . is six - dimensional , and non - isotropic in the sense that pairs of mode separated by smaller angles are more correlated . on top of that , the observed quantity is convoluted in six - dimensional with pairs of survey selection function @xcite . for these reasons , we find solutions that apply to periodic volumes , and we leave it for future work the extraction of optimal estimator in more complex data . ] 3 . many times in the literature @xcite , the above case is modified by replacing the non - linear covariance matrix by a gaussian one . this effectively treats the power spectrum measurements as gaussian , even though the data are correlated . under the widely used fkp @xcite approximation , for example , the only source of mode coupling comes from the convolution with the survey selection function ; it thus considers the underlying covariance matrix to be diagonal . for reasons mentioned above , the error bars obtained this way could be systematically underestimated . we did not address the question of shot noise , however , which is also non - gaussian in nature . in the case of surveys that address some of the non - gaussianities with mock catalogs , the error correct bars should lie somewhere between case ( i ) and case ( ii ) ( while some approach case ( iii ) ) , depending on how close to optimal the measured power spectrum is , and how well the catalogs model the non - linear dynamics . @xcite showed that the difference in estimator between cases ( ii ) and ( iii ) is very small ( i.e. , with optimal weighting , errors near the pure - gaussian errors are achieved ) , but this is not the complete story , especially when dealing with current gaussian or sub - optimal data analyses . the question we address is the following by how much does the error bars on the _ least robust _ bao dilation scale ( case ( iii ) ) differ from a correct calculation based on an optimal covariance matrix and properly inverse covariance weighted ( case ( i ) ) ? finally , we go one step further and repeat the measurements of non - gaussian effects on reconstructed density fields . we apply a density reconstruction algorithm @xcite that was developed to improve the bao signal at late times , which is partially lost due to non - linear coupling between fourier modes @xcite . other approaches have been explored to recover some of the fisher information lost in gravitational collapse @xcite , but it was not verified how these methods propagate to constraints on cosmological parameters . the full bao analysis is indeed sensitive to this intermediate stage , but in a non - gaussian treatment , the interplay between the off - diagonal elements of the covariance matrix and the derivatives is quite subtle . we thus set forth to test quantitatively how this reconstruction algorithm affects the bao dilation error , for the three analysis cases mentioned above . the paper is organized as follows . in section [ sec : background ] we briefly review how bao dilation measurements can constrain dark energy . in section [ sec : nbodysimulations ] we describe our set of n - body simulations and the reconstruction algorithm . in section [ sec : powerspectrumanalysis ] we discuss how to extract both the gaussian and the non - gaussian covariance matrices . in section [ sec : eigen ] we describe the eigenvector decomposition , while in section [ sec : parameterestimation ] we present the fisher matrix formalism and the estimators of the bao dilation uncertainty for the three analysis cases . finally , in section [ sec : resultsanddiscussion ] we present and discuss our results . the matter clustering we observe today is the result of tiny inhomogeneities set during inflation in the early universe @xcite . over time , matter collapsed gravitationally into over - dense regions that eventually evolved into large scale structures . as long as the perturbations are small , the structure growth equations can be linearized . we can therefore understand the evolution of inhomogeneities by considering perturbations one at a time . in the early universe , matter and photons are coupled together as a single fluid via thomson scattering . due to radiation pressure , photons begin to disperse away from over - dense regions , pushing the baryons alongside . dark matter , on the other hand , only interacts weakly with that fluid , via gravity , and does not respond efficiently to the photons push . the perturbation eventually grows into a state where the initial clump of dark matter gets surrounded by a spherical ripple of baryon - photon fluid , which expands at the speed of sound @xmath9 . at about @xmath10 , when photons decouple from matter , they no longer push the baryons . the speed of sound in the fluid drops abruptly , and the bao ripple stops moving and freezes out . eventually , dark matter also responds gravitationally to this over - dense region of baryons . the result of this initial point - perturbation is a smooth clump of matter with a spherical shell of density enhancement at the _ sound horizon _ , about 150 mpc away from the center . the complete final field is , by huygens s principle , the superposition of similar spherical ripples of density enhancement from the initial perturbations at all points . therefore , we do not observe these spherical ripples directly , but measure them statistically in the mass auto - correlation function . using galaxies from the sloan digital sky survey as tracers of the matter distribution , an excess correlation at 150 mpc apart has been observed @xcite . although the bao in the correlation function is intuitive , we are interested in the bao power spectrum @xmath11 which is related to @xmath12 by a transformation @xmath13 the correlation peak in real space is manifested as a series of wiggles in fourier space . this oscillatory feature is even more robust against observational contaminations @xcite . the accelerating expansion of the universe is widely blamed on dark energy , a mysterious entity which contributes to more than 70% of the energy content in the universe . dark energy can be described by an equation of state @xmath14 relating its pressure @xmath15 and density @xmath16 . a common parameterization of @xmath17 is @xmath18 where @xmath19 is the value in the present day . the friedmann equation now reads @xmath20 \end{array } \label{eqn : friedmann}\ ] ] where the terms on the right hand side are the energy contributions of matter , radiation , curvature , and dark energy respectively . for an object of a fixed co - moving size @xmath21 , its projections across and along the line of sight are given by @xmath22 @xmath23 respectively . @xmath24 and @xmath25 are the redshift span and angular size of the object on the sky , and @xmath26 is the angular diameter distance to its center . if this object has a fixed co - moving size , it then acts as a standard ruler , where measurements of @xmath24 and @xmath25 give estimates for @xmath27 and @xmath28 , respectively . this , in turn , provides constraints for @xmath29 via equation [ eqn : friedmann ] , given an adequate redshift sampling . in this paper , we restrict ourselves to an idealized isotropic universe with no observational distortions , such that the standard ruler is given by the sound horizon , which has size @xmath30 . therefore , the fractional errors on these quantities can be used to constrain @xmath31 and @xmath32 . to estimate the error on @xmath21 , we construct a fisher matrix that propagates the correlated uncertainty measured in our simulated power spectra . we run a total of 1000 simulations using a particle - particle - particle - mesh ( p@xmath33 m ) n - body code cubep3 m , the successor to pmfast @xcite . each simulation has @xmath34 dark matter particles in a periodic cube of @xmath35mpc on one side . the initial condition of each simulation at @xmath36 is produced by a gaussian random field , which is characterized by an initial transfer function generated by camb . for this work we use the following cosmological parameters : @xmath37 , @xmath38 , @xmath39 , @xmath40 , @xmath41 , @xmath42 . we output the particle positions and velocities at redshifts 0.5 , 1.0 and 2.0 , and then the particles are assigned to a density field @xmath43 on a @xmath44 grid using the cloud - in - cell algorithm @xcite . to understand the density reconstruction algorithm , it is helpful to look at the process by which initial conditions are generated in a typical n - body simulation . a gaussian random field @xmath45 in fourier space can be constructed by generating a field of gaussian random numbers whose variance is determined by an input power spectrum . under the zeldovich approximation @xcite , we then compute a displacement field @xmath46 which , when fourier transformed back into real space , can be applied to displace a uniformly distributed set of particles . in addition , the density field allows us to calculate the gravitational potential , whose gradient gives the particles initial velocities . we solve for these initial conditions at @xmath36 , where our simulations begin . the reconstruction algorithm developed by @xcite essentially uses the @xmath43 output to calculate a displacement field , and subtracts the displacements from the particle positions . this is indeed very similar to the procedure that generates initial conditions described previously , except that the displacements are subtracted from the particles positions , instead of added . the algorithm is the following , as was neatly summarized by @xcite . 1 . calculate the density field @xmath43 using particle positions , and then transform it into fourier space @xmath45 . 2 . calculate the displacement field @xmath47 in fourier space using the zeldovich approximation , where @xmath48 and @xmath49 $ ] is a smoothing function of scale @xmath50 . 3 . transform the displacement field back into real space . subtract this displacement from the positions of the simulation particles and calculate the new density field @xmath51 . 4 . repeat the previous step , but applying the displacement field onto a set of uniformly distributed particles instead of simulation output . calculate the density field @xmath52 from these displaced particles . the `` reconstructed '' density field @xmath53 is given by the difference between the above density fields : @xmath54 the correlation functions @xmath12 before and after reconstruction ( using smoothing scale @xmath55mpc ) are shown in figure [ fig : xi ] . in sections [ sec : effectsofreconstruction ] and [ sec : resultsanddiscussion ] , we quantify the effect that reconstruction has on distance error estimates . at redshifts @xmath56 and @xmath57 , as well as reconstruction ( smoothing scale @xmath55mpc ) at @xmath56 . the functions are rescaled by the square of the growth factors so that the acoustic peak locations match the @xmath56 case.,width=336 ] in the presence of anisotropy ( eg . redshift distortions ) , the power spectrum @xmath58 takes in an angular dependence @xmath59 . in an isotropic universe , however , it is only a function of the scale and is defined as @xmath60 where the angled brackets denote the average of all @xmath61 modes such that @xmath62 . to obtain @xmath11 from our n - body simulations , we assign particles onto a density grid @xmath43 , fourier transform it into @xmath45 , and take the averages over the spherical shells of radius @xmath63 in fourier space , using the nearest - grid - point scheme . because our goal is to measure both a covariance matrix and a fisher matrix from a finite number of realizations , the binning must be chosen carefully . on one hand , we need to be maximally sensitive to the bao signal , hence it is important to resolve many wiggles . otherwise , our analysis would under - sample the rapidly oscillating signal and our results would be less robust . on the other hand , we need to limit the number of matrix elements to address the issue of convergence . we thus opt for mixed binning , which is linear for @xmath64 ( @xmath65 ) where @xmath66 is the box size because the bins have been corrected for averaging with the nearest - grid - point scheme . this correction is significant for small @xmath63 where the number of modes that contributes to the average is small . ] and logarithmic for larger @xmath63 modes ( @xmath67 ) . we end up with 54 bins , which translates into 2916 matrix elements , and we resolve the first seven bao wiggles . figure [ fig : powerspectrum ] shows the average @xmath11 over 1000 simulations . produced by averaging 1000 simulations . _ top panel _ : the simulated dimensionless power spectrum @xmath68 compared against calculations by camb , where `` linear '' is from the linear @xmath69cdm theory , and `` nonlinear '' is based on halofit @xcite numerical calculations . _ bottom panel _ : the simulation power spectrum divided by linear theory and halofit power spectra . the first 47 modes of our spectrum are binned linearly , while the others are binned logarithmically . the error bars in each plot are the errors on the mean . we note the few percent discrepancy between the output and linear theory on large scales , which is caused by overly large time steps taken by the simulator at high redshifts , and will be corrected in further simulations . however we have ran convergence test on the starting redshift of the covariance matrix and found that a starting redshift of @xmath70 produces the most accurate matrix both at high and low redshifts . , title="fig:",width=336 ] produced by averaging 1000 simulations . _ top panel _ : the simulated dimensionless power spectrum @xmath68 compared against calculations by camb , where `` linear '' is from the linear @xmath69cdm theory , and `` nonlinear '' is based on halofit @xcite numerical calculations . _ bottom panel _ : the simulation power spectrum divided by linear theory and halofit power spectra . the first 47 modes of our spectrum are binned linearly , while the others are binned logarithmically . the error bars in each plot are the errors on the mean . we note the few percent discrepancy between the output and linear theory on large scales , which is caused by overly large time steps taken by the simulator at high redshifts , and will be corrected in further simulations . however we have ran convergence test on the starting redshift of the covariance matrix and found that a starting redshift of @xmath70 produces the most accurate matrix both at high and low redshifts . , title="fig:",width=336 ] given our series of @xmath11 measurements , we calculate the covariance matrix between each data point as @xmath71\left[p_i(k^\prime)-\langle p(k^\prime)\rangle\right ] \label{eqn : covariance}\ ] ] where @xmath72 is the power spectrum of the @xmath73th simulation , and @xmath74 is the number of simulations we have . the power spectrum covariance matrix is best visualized as a cross - correlation coefficient matrix @xmath75 this definition normalizes the covariance matrix by the diagonal components so that they are all unity . this is consistent with the fact that a given @xmath63 mode is always perfectly correlated with itself . therefore , @xmath76 ranges from @xmath77 ( perfect anti - correlation ) to + 1 ( perfect correlation ) , where 0 is no correlation at all . figure [ fig : corrmatrix ] shows the correlation matrix corresponding to the 1000 power spectra shown in figure [ fig : powerspectrum ] . we observe that the largest @xmath63-modes in the simulation box exhibits an anti - correlation of about @xmath78 with the small scales , which themselves are highly correlated . the systematic anti - correlation has very little impact on the final measured quantities . we find that neither cutting off these three fundamental modes nor suppressing those negative covariance values to zero at late redshifts affect our results . this is not surprising because , as we show in the following sections , the lowest-@xmath63 modes contain very little fisher information , while the largest-@xmath63 modes usually do not contribute much to our distance errors . therefore , the presence of this anti - correlation has no influence on our conclusions . in the three analysis cases studied in this paper , we work with two kinds of covariance matrices . the _ non - gaussian _ covariance matrices are computed using equation [ eqn : covariance ] , while the _ gaussian _ covariance matrices @xcite are given by @xmath79 where @xmath80 is the number of modes that contributed to the @xmath63 bin ( counting the real and imaginary parts separately ) , and @xmath11 can be calculated from linear theory . indeed the kronecker - delta symbol @xmath81 ensures that the gaussian covariance matrix is diagonal , and assumes that all the @xmath63 modes are uncorrelated . in a real survey , when only one universe can be observed , it is much harder to calculate the correlation between different @xmath63 modes . as shown in figure [ fig : corrmatrix ] , the actual covariance matrix is clearly not diagonal as we approach the non - linear regime . furthermore , as shown in figure [ fig : gaussratio ] , even the diagonal components of the non - gaussian covariance matrix are not gaussian . the fourier modes on small scales are coupled both with neighboring scales , and across different directions on a given scale . in this context , the best estimate of the true covariance matrix is obtained from an ensemble of n - body simulations , and the resulting non - gaussian treatment of the data is our best shot at estimating both the mean and the error on the mean . the replacement of the non - gaussian covariance matrix by a gaussian one is often used as a shortcut , but one should keep in mind that the results in general may not be reliable . ) at @xmath56 produced by averaging 1000 simulation power spectra as shown in figure [ fig : powerspectrum ] . the dark points on the diagonal and at the large-@xmath63 regime indicate positive correlation , as we expected . as described in section [ sec : eigen ] , this figure shows a binning independent version of the correlation matrix.,width=336 ] ) and gaussian ( @xmath82 ) covariance matrices , at @xmath56 . for @xmath83 we evaluated in 2 ways . using equation [ eqn : covariancegauss ] , the @xmath11 can either be averaged over our set of simulations , or simply computed from linear theory using camb . for either case , this plot shows that even the diagonal components of the non - gaussian covariance matrix are not gaussian , as the ratio clearly departs from unity with higher @xmath63 . the sharp increase at @xmath84 is caused by the logarithmic binning , in which each bin suddenly accumulates the departure contributions from many more fourier modes . as we will justify in the results discussion later , we do not consider modes where @xmath85 mpc@xmath86 . , width=336 ] as previously considered by @xcite , a useful quantity that can be derived from the covariance matrix is the cumulative fisher information function @xmath87 where @xmath88 is the @xmath89 sub - matrix of @xmath90 ( or @xmath83 ) with @xmath91 , further normalized by @xmath92 . in other words , @xmath93 here @xmath94 describes the information contained in the amplitude of the power spectrum up to a scale @xmath63 . although the form of @xmath94 seems like a mathematical contrivance , it indeed has physical significance in weak lensing @xcite . figure [ fig : information ] shows the fisher information function using our set of simulations . it also compares the information that can be obtained by a gaussian and a non - gaussian covariance matrix . at small @xmath63 , gravity is still linear , so the non - gaussian information is similar to the gaussian case , as also seen in figure [ fig : corrmatrix ] . at @xmath95 , however , the information flattens out into a so - called `` trans - linear plateau '' , which is consistent with previous studies by @xcite and @xcite . this corresponds to the scale where the @xmath63 modes become correlated . qualitatively , the fisher information function tells us how much information can be extracted from a given @xmath63 mode . at small @xmath63 , each mode is independent , so more information can be extracted by measuring more modes . at large @xmath63 , the fisher information function flattens , meaning very little information can be extracted by measuring extra modes . indeed , as seen from figure [ fig : information ] , one can extract orders of magnitude more information when using a gaussian covariance matrix instead of the true covariance matrix . this information , however , does not exist in the data . consequently , an analysis using a gaussian estimator would most likely underestimate the error bars for any measurements beyond the trans - linear scale ( @xmath95 ) . before proceeding to the construction of the fisher matrix and of the three estimators , we reduce the noise that exists in our covariance matrices , as described in the following section . ) per unit volume at @xmath56 . the filled circles were obtained by using both the reconstructed power spectrum and the reconstructed covariance matrix in equation [ eqn : c_norm ] . _ bottom panel _ : same plots as the top , but with different binning schemes to show that the plateau of information is not a binning artifact . similar plots can also be found in @xcite . , title="fig:",width=336 ] + ) per unit volume at @xmath56 . the filled circles were obtained by using both the reconstructed power spectrum and the reconstructed covariance matrix in equation [ eqn : c_norm ] . _ bottom panel _ : same plots as the top , but with different binning schemes to show that the plateau of information is not a binning artifact . similar plots can also be found in @xcite . , title="fig:",height=142 ] ) per unit volume at @xmath56 . the filled circles were obtained by using both the reconstructed power spectrum and the reconstructed covariance matrix in equation [ eqn : c_norm ] . _ bottom panel _ : same plots as the top , but with different binning schemes to show that the plateau of information is not a binning artifact . similar plots can also be found in @xcite . , title="fig:",height=142 ] as mentioned in the introduction , to achieve convergence on the @xmath8 elements of the covariance matrix is not an easy task . among the possible approaches , the brute force way ( i.e. running a very large number of n - body simulations ) is probably the least affordable in terms of computational resources . in this section , we apply a noise reduction technique that reduces to @xmath96 the number of elements one needs to extract . our technique is based on the assumption that the cross - correlation coefficient matrix ( see figure [ fig : corrmatrix ] ) can be parameterized as the sum of a diagonal and off - diagonal components , and we put a strong prior on the latter : it must be expressed as a sum over the outer products of the principal eigenvectors- and @xmath97-modes , the number of significant eigenvectors is expected to be small . in contrast , a matrix which would be strong when close to the diagonal , but decreasing when moving away , would require a larger number of vectors . ] . the diagonal component is then simply obtained such as each diagonal element in the final matrix is equal to 1 . to extract the eigenvectors , we perform an iterative eigenvalue decomposition of the off - diagonal component and keep only the dominant terms . we start by modeling the full matrix as the sum of an identity matrix @xmath98 and an off - diagonal component , from which we extract the largest eigenvalue @xmath99 and eigenvector @xmath100 . at the next iteration step , we model the diagonal component as @xmath101 , subtract that quantity from the original matrix , and extract the principal eigenvector of the result . we repeat this process until the difference between the original and the model converges four times in this case . at the end of the iterative step @xmath73 , the parameterized matrix is thus given by @xmath102^{2}\right)+ \lambda^{i } u_{\lambda}^{i}(k ) u_{\lambda}^{i}(k ' ) \label{eqn : eigen_init}\ ] ] after the last iteration , we update the first term with the latest @xmath99 and @xmath100 . the diagonal component typically starts off as unity in the largest scales and fades away in the non - linear regime . in this case , the strongest eigenvalue is almost two orders of magnitude larger than all others , hence we keep only a single eigenvector ( see @xcite for the application of this method when the angular dependence of the covariance matrix is preserved ) . the modeled covariance matrix is then recovered from this noise - reduced cross - correlation matrix with the diagonal elements of the original covariance @xmath103 ( not to be confused with the diagonal component of the factorization of @xmath104 ) following equation [ eqn : corrmatrix ] . although we do need @xmath105 matrix elements to start with , the power of this decomposition relies on the fact that the prior is accurate at the few percent level , and that everything that does not fit into this form is considered as noise . we thus recover smooth and accurate covariance matrices , even though the input elements are noisy . the explanation for this is that one needs only to measure the n diagonal elements of the covariance matrix ( which are generally the easiest to resolve ) plus one eigenvector ( which combines the data of a large portion of the matrix into another n elements ) . we present the fractional error between the modeled and the original matrices in figure [ fig : frac_error ] , and observe that individual elements match the model at the ten percent level at @xmath106 . this improves to the few percent level for even smaller scales . we expect these higher @xmath63 modes to be more accurately measured to start with , since they come from the angle averaging of @xmath63-shells that contain much larger number of cells . . as expected , the lower @xmath63 modes are more noisy due to lower statistics . we still achieve an agreement of about ten percent at @xmath107 , and even better for smaller scales . , width=336 ] we then present in figure [ fig : ratio_to_gauss ] the diagonals of the original covariance matrix and of both components , divided by the gaussian prediction . the sum of the two components is equal to the original matrix on the diagonal by construction . the off - diagonal components of the factorization can be tested against the original by comparing the cumulative fisher informations ( equation [ eqn : info ] ) . we observe that the information from the factorized covariance matrices exhibit the same essential features , namely a gaussian increase up to about @xmath108 , followed by a saturation plateau . we present the ratio of this modeled information to the original distribution in figure [ fig : info_ratio_model ] and observe that the factorization reproduces the information content at the few percent level . although the figures presented in this section correspond to the unreconstructed densities at @xmath56 , we achieve comparable performances on reconstructed density fields at all measured redshifts . . the departure from the original is at most @xmath109 , and the agreement is otherwise at the few percent level.,width=336 ] this factorization has one extra advantage , which comes from the fact that @xmath110 is binning independent . the original covariance matrix is binning dependent , since the number of modes entering each element varies as we change from linear to logarithmic bands , for example . this is correct , but it has a major inconvenience the cross - correlation coefficient matrix ( equation [ eqn : covariance ] ) visually changes significantly . it becomes a tedious task to compare figures from different authors . we can therefore attempt to fix this problem by constructing binning independent quantities . the diagonal component , @xmath111 , is close to gaussian , as seen in figure [ fig : ratio_to_gauss ] , hence it is roughly inversely proportional to the number of modes in the bin . we scale it with the ratio between the measured number of modes and the continuous limit case : @xmath112 , and obtain a binning independent quantity . we have also shown in this section that the off - diagonal elements of the covariance matrix are well modeled by @xmath113 , which are binning independent as well . the solution is thus to replace the diagonal of the original covariance matrix such that @xmath114 \frac{n}{n_{cont } } + \lambda u^{2}(k).\ ] ] this bin independent result is made available thanks to the factorization presented above , which allows us to isolate the bin dependence ( that comes exclusively from the diagonal component ) and to apply our correction . the cross - correlation coefficient matrix calculated that way is shown in figure [ fig : corrmatrix ] , and compares well with @xcite , apart from the anti - correlation between the largest and the smallest modes ( see section [ subsec : covmatrix ] , and note that it is not clear that the simulations @xcite were large enough to contradict our anti - correlation result ) . for a set of parameters @xmath115 , the fisher matrix @xmath116 is given by the inverse of a covariance matrix @xmath117 where @xmath118 or @xmath119 , as described in the previous sections . since our @xmath120 is derived from power spectra , the parameters in @xmath116 can be thought of as the power for each @xmath63 mode in the power spectrum . to obtain the fisher matrix for another set of parameters @xmath121 , we can project the power spectrum fisher matrix onto the new parameter space using @xmath122 using @xmath123 , the optimal error estimator for parameter @xmath124 is simply @xmath125 the estimators for cases ( ii ) and ( iii ) are obtained from substituting equations [ eqn : covariance ] and [ eqn : covariancegauss ] respectively in equation [ eqn : fish ] . for our purposes , since we are interested in estimating the fractional errors of @xmath21 ( 1d ) or @xmath126 and @xmath127 ( 2d ) , we set @xmath128 in 1d , and @xmath129 and @xmath130 in 2d . the derivatives in equation [ eqn : fisher ] can be evaluated in a number of ways . we can either produce a variety of @xmath11 using different parameters @xmath21 and take their finite differences , or we can parametrize @xmath11 as a function of @xmath21 , and evaluate the derivatives analytically . in both cases , we decompose the power spectrum into a smooth component and a wiggly component @xmath131 when using bao to measure distances , all information is manifested as `` wiggles '' in the power spectrum . to ensure that all our measurements originate from bao , we subtract @xmath132 @xcite from our full @xmath11 prior to taking derivatives . for the rest of this paper , we refer @xmath133 and @xmath134 to simply @xmath11 and @xmath135 for brevity . the finite differencing of @xmath11 is done by producing many @xmath11 using cmbfast , and dividing their differences by the differences of @xmath21 being used . on the other hand , readers familiar with @xcite would recall that they suggested an analytical form of the wiggles using a damped sync function , and they approximated the square of the sinusoidal components in the fisher matrix as a constant . in our work , since we use non - gaussian ( hence non - diagonal ) covariance matrices , we do not adapt that constant approximation , as the detailed structure of the signal becomes important once off - diagonal errors are present . nevertheless , it is interesting to show what effect this approximation and the off - diagonal elements of the covariance matrix can have on our results . see appendix a for this discussion . in a data analysis , non - linear covariance matrices are usually hard to measure with a high signal to noise , especially in surveys that exhibit complex selection functions . what is often done in that case is to assume gaussianity in the data , while ignoring the fact that the band powers themselves are actually non - gaussian . the values extracted for the mean and the error on the bao dilation scale are not properly weighted , since they assume that all errors in @xmath136bands are uncorrelated . as mentioned earlier , the resulting mean is sub - optimal , while the error bars are most likely off by an unknown amount . since we now have measured a non - linear covariance matrix @xmath137 , we are in a position to compare the correct error bars for existing data analysis ( case ( i ) ) with those quoted in literature , which approach case ( iii ) at various levels . we recall that the difference between sub - optimal and optimal is somewhat reduced for analyses that model the non - linearities in the fields . we derive a `` sub - optimal '' estimator by solving the linear system @xmath138 where @xmath139 is a vector containing a set of cosmological parameters of interest ( here we consider only @xmath140 , hence @xmath139 is a scalar , but this method could be generalized to include @xmath141 and @xmath142 for instance ) , and @xmath143 is a noisy observable , associated with a noisy covariance matrix @xmath144 . in our case , the observable under study is @xmath145 , the deviation from the mean of the power spectrum . with this correspondence , we get @xmath146 a vector in our case . to estimate each component of @xmath139 , we first weight each observed point by the inverse of the covariance matrix associated with the observation of @xmath143 , i.e. @xmath144 , and then proceed to solve for @xmath139 by taking pseudo - inverses such that @xmath147 finally , the errors on the elements of @xmath139 are given by the diagonal components of the covariance matrix @xmath148 . we obtain the following estimator for the error in @xmath149 : @xmath150 where @xmath151 , and @xmath152 is a vector - matrix - vector product similar to equation [ eqn : fisher ] , and where @xmath153 is the improved estimate of the non - linear covariance matrix , which we obtained from our simulations . notice that if the true covariance matrix was indeed gaussian ( i.e. @xmath154 ) , we would recover the optimal estimator where @xmath155 . in other words , cases ( i ) , ( ii ) , and ( iii ) would be identical . conversely , if the original matrix was already the optimal measurement ( i.e. @xmath156 ) , we would get @xmath157 , i.e. cases ( i ) and case ( ii ) would be the same , possibly different from case ( iii ) . we also note that the _ inverse _ of the true covariance matrix is not involved in equation [ eqn : gaussianestimator ] . in the next sections , we consider the case where gaussianity was originally assumed in a bao analysis , we correctly estimate the error of @xmath139 ( with this sub - optimal estimator ) and compare the results produced with an optimal estimator . as seen in figure [ fig : xi ] , density reconstruction sharpens the acoustic peak in the correlation function . this is equivalent to reducing the non - linear damping of the wiggles in the power spectrum . we parametrize the reduction of damping due to reconstruction by an extra factor in front of @xmath158 in the damping factor @xcite such that @xmath159 \rightarrow \exp\left [ - \frac { ( 1-f)^2 k^2 \sigma_{nl}^2}{2 } \right ] \label{eqn : reconfac}\ ] ] where @xmath160 represents 100% reconstruction , canceling any non - linear effects . in reality , such a case is unachievable , as some information has been irreversibly lost . in principle , we could measure @xmath161 by extracting the power spectrum wiggles before and after reconstruction , and find the best fit damping factor . looking at figure [ fig : xi ] , we see that the reconstructed correlation at @xmath56 is very similar to the correlation at @xmath162 . the ratio of growth factors between these two redshifts is 0.55 , and the reconstructed field actually looks slightly better than the @xmath162 field , so we adopt the standard @xmath163 , which reduces non - linear damping by 50% @xcite . we set up hypothetical surveys with volume @xmath164 centered on redshifts @xmath56 and @xmath165 . we then produce fractional distance errors @xmath166 using the three aforementioned estimators : * sub - optimal estimator ( equation [ eqn : gaussianestimator ] ) * optimal estimator ( equation [ eqn : optimalestimator ] ) with a non - gaussian covariance matrix ( equation [ eqn : covariance ] ) , * optimal estimator ( equation [ eqn : optimalestimator ] ) with a gaussian covariance matrix ( equation [ eqn : covariancegauss ] ) . the distance error measurements use only the information up to a limiting @xmath167 in the covariance matrix ; all @xmath168 are marginalized over ( i.e. cut off from the covariance matrix before it is inverted ) . for each error estimate , we also produce bootstrap error bars to show the convergence of our set of 1000 simulations . this is done by picking 1000 random simulations ( allowing repeats ) from our set , and taking the standard deviations of the results using 2000 such random sets . the noise reduction technique is performed for every set . the only redshift dependence in the estimator comes from the non - linear damping scale @xmath169 ( equation [ eqn : reconfac ] ) . in addition , we know from figures [ fig : corrmatrix ] and [ fig : information ] that the covariance matrices in the linear regime are similar to the gaussian covariance matrix . therefore , we expect errors at small @xmath63 to be hardly distinguishable among estimators and redshifts . at large @xmath63 , however , we expect the effects of the covariance matrices and the derivatives to have a more pronounced effect on the estimator . the redshift dependent damping scale @xmath169 becomes important and distinguishes between @xmath56 and @xmath165 . the top panels of figures [ fig : ds_kmax ] and [ fig : ds_kmax_recon ] show the measured distance errors vs limiting @xmath167 , with and without reconstruction , respectively . the bottom panels shows the squares of the ratio of cases ( i ) and ( ii ) to case ( iii ) from the top plot at @xmath56 . in all scenarios , we consider only the modes @xmath170 mpc@xmath86 where our simulations remain reliable ( figure [ fig : powerspectrum ] ) . moreover , we do not expect modes of @xmath85 mpc@xmath86 to carry any bao information , as the wiggles on small scales suffer from silk damping @xcite as well as non - linear damping ( section [ sec : effectsofreconstruction ] ) . figure [ fig : ds_kmax_recon ] shows the distance errors with @xmath163 and covariance matrices from reconstructed simulations . since some of the wiggles are indeed recovered , the optimal distance errors decreased by 50% to 70% compared to the case with no reconstruction ( figure [ fig : ds_kmax ] ) . more importantly , though , the discrepancy between sub - optimal and optimal estimates becomes slightly more severe . in the presence of reconstruction , a naive gaussian assumption can underestimate the variance of the errors by up to 20% near the trans - linear scales ( @xmath171 mpc@xmath86 ) . somewhat disturbingly , at small scales @xmath85 mpc@xmath86 the sub - optimal estimate deviate significantly from the optimal estimate . a sub - optimal estimator certainly is not expected to provide the same result as an optimal estimator . and also , as mentioned above , the regime @xmath85 mpc@xmath86 is irrelevant to bao analysis . therefore , we did not investigate this behavior further . our results from figure [ fig : ds_kmax ] show that the three treatments of the errors have similar constraining power on the bao dilation scale . this conclusion is consistent with that from @xcite , which measured that cases ( ii ) and ( iii ) give almost identical results , without reconstruction . however , we push the envelope further and show that , first , the sub - optimal estimator also behaves similarly , with deviations by up to @xmath172 . second , we show that the reconstruction of density fields improves the constraints on dilation by about 70% at @xmath56 . the improvement is more modest at higher redshifts , where additional bao peaks are still in the linear regime . third , we conclude that with reconstruction , gaussian assumption can underestimate the errors on the dilation scale in a similar level to the case without reconstruction . and @xmath137 respectively . _ sub - optimal _ denotes the correct non - gaussian errors if gaussianity was assumed . the error bars on each line are the 1-@xmath173 bootstrap error bars for the statistical fluctuations in our set of simulations . _ middle panel _ : optimal non - gaussian and sub - optimal errors , each divided by the optimal gaussian errors at @xmath56 . this panel shows that each error estimate deviates by only a few percent from the gaussian estimate . _ bottom panel _ : our @xmath174 template which shows that the regime @xmath85 mpc@xmath86 is no longer relevant for bao analysis , as the wiggles are mostly damped and converged to zero . , title="fig:",width=336 ] and @xmath137 respectively . _ sub - optimal _ denotes the correct non - gaussian errors if gaussianity was assumed . the error bars on each line are the 1-@xmath173 bootstrap error bars for the statistical fluctuations in our set of simulations . _ middle panel _ : optimal non - gaussian and sub - optimal errors , each divided by the optimal gaussian errors at @xmath56 . this panel shows that each error estimate deviates by only a few percent from the gaussian estimate . _ bottom panel _ : our @xmath174 template which shows that the regime @xmath85 mpc@xmath86 is no longer relevant for bao analysis , as the wiggles are mostly damped and converged to zero . , title="fig:",width=336 ] and @xmath137 respectively . _ sub - optimal _ denotes the correct non - gaussian errors if gaussianity was assumed . the error bars on each line are the 1-@xmath173 bootstrap error bars for the statistical fluctuations in our set of simulations . _ middle panel _ : optimal non - gaussian and sub - optimal errors , each divided by the optimal gaussian errors at @xmath56 . this panel shows that each error estimate deviates by only a few percent from the gaussian estimate . _ bottom panel _ : our @xmath174 template which shows that the regime @xmath85 mpc@xmath86 is no longer relevant for bao analysis , as the wiggles are mostly damped and converged to zero . , title="fig:",width=336 ] , this now includes reconstructed covariance matrices and a reconstruction factor of @xmath163 . in the sub - optimal case , the increase at the regime @xmath85 mpc@xmath86 is a statistical curiosity , but irrelevant for a bao analysis . we base our conclusion on the values for @xmath175 mpc@xmath86 . this cut - off coincides with our choice to switch from linear binning to logarithmic binning , but we have checked that the above plots contain no binning artifacts . , title="fig:",width=336 ] , this now includes reconstructed covariance matrices and a reconstruction factor of @xmath163 . in the sub - optimal case , the increase at the regime @xmath85 mpc@xmath86 is a statistical curiosity , but irrelevant for a bao analysis . we base our conclusion on the values for @xmath175 mpc@xmath86 . this cut - off coincides with our choice to switch from linear binning to logarithmic binning , but we have checked that the above plots contain no binning artifacts . , title="fig:",width=336 ] , this now includes reconstructed covariance matrices and a reconstruction factor of @xmath163 . in the sub - optimal case , the increase at the regime @xmath85 mpc@xmath86 is a statistical curiosity , but irrelevant for a bao analysis . we base our conclusion on the values for @xmath175 mpc@xmath86 . this cut - off coincides with our choice to switch from linear binning to logarithmic binning , but we have checked that the above plots contain no binning artifacts . , title="fig:",width=336 ] we have addressed some aspects of the problem of measuring non - gaussian error bars in a bao analysis . we have investigated the full optimal quadratic non - gaussian estimator on reconstructed density field , and quantified the significance of the non - gaussianities on the bao dilation scale error . a major subtlety is that the optimal estimator requires an accurate measurement of the inverse of the covariance matrix of power spectrum , and the actual uncertainty on that inverse had never been measured . the accuracy of the inverse is generally bin dependent , and convergence requires a large number of n - body simulations . we have overcome this problem with a factorization technique that involves an iterative eigenvalue decomposition of the covariance matrix , which we measured from 1000 n - body simulations . we further measured the uncertainty of the inverse of the matrix with bootstrap re - sampling , and were able to achieve convergence at the percent level . having confidence in the inverse matrix , we quantitatively compared the measurement of the error on the bao dilation factor obtained with different estimators . we construct an _ optimal estimator _ , which gives the most accurate measurement of the error achievable on the bao dilation scale . it is derived from a covariance weighted fisher matrix , which is constructed out of the inverse of the non - linear power spectrum covariance matrix . we compared our results with those obtained from the purely gaussian forecast , and we measure significant discrepancies of up to ten percent in the error on the dilation scale . we also measured non - gaussian error bars on the mean bao dilation scale that has been obtained with a gaussian estimator , as is usually encountered in the literature . because we have confidence in the accuracy of our covariance matrix , this _ sub - optimal estimator _ provides a robust estimate of the error bars on the bao dilation scale . to illustrate our point , we considered the case where the original dilation scale was measured under standard gaussian statistics . we found that the variances of those measurements can differ by up to 15% , compared to our optimal estimator . many data analyses did include non - gaussianities in their bao error estimator , hence the discrepancy between these and our optimal estimator is likely to be more modest than that obtained in this work . we note in passing that these results were entirely obtained from n - body simulations , hence the effect of the survey selection function has been factored out of our problem . constructing optimal estimators with actual data will however be more challenging , since it has to include such an effect , which effectively couple fourier modes from different bins , in addition to account for the effect of bias between the sampled tracers ( i.e. galaxies or 21 cm structure ) and the underlying matter density . we have also implemented a density reconstruction algorithm , which recovers some of the lost bao information due to non - linear gravitational collapse at late times . in that case , the error on the dilation scale is reduced by a factor of about 70% at low redshift , but the discrepancy between the sub - optimal and optimal estimates remains similar to the case without reconstruction ( 20% and 15% , with and without reconstruction ) . we mention in conclusion that in a survey , the increase in variance we observed when using a sub - optimal estimator is equivalent to losing about the same percentage of survey volume , because the variance of measurements is inversely proportional to volume . these discrepancies should be taken seriously into account especially when forecasting performances of future telescopes , where the objective is to reach percent level precision on cosmological parameters . the authors thank kiyoshi masui for valuable help and advice , and acknowledge nserc for their financial support . the simulations in this work were produced on the sunnyvale cluster at cita . 99 bassett b.a . , hlozek r. , 2009 , in ruiz - lapuente p. , ed . , dark energy , cambridge university press , new york . blake c. , glazebrook k. , 2003 , apj , 363 , 1329 . blake c. , et al , 2010 , mnras , 406 , 803 . cole s. , percival w. , et al . , 2005 , mnras , 362 , 505 . coles p. , jones b. , 1991 , mnras , 248 , 1 . eisenstein d.j . , hu w. , 1998 , apj , 496 , 605 . eisenstein d.j . , 2005 , new astronomy reviews , 49 , 360 . eisenstein d.j . , zehavi i. , et al . , 2005 , apj , 633 , 560 . eisenstein d.j . , , sirko e. , spergel d.n . , 2007 , apj , 664 , 675 . feldman h. , kaiser n. , peacock j. , 1994 , apj , 426 , 23 . goldberg d. , 2000 , phd thesis , princeton university . guth a. , 2004 , in carnegie observatories astrophysics series , vol . 2 : measuring and modeling the universe , cambridge university press , new york . harnois - draps j. , pen u .- l . , 2011 , mnras submitted . ( arxiv:1109.5746 ) hockney r.w . , eastwood j.w . , 1980 , computer simulation using particles . taylor & francis group , new york . htsi , g. , 2006 , astron . astrophys . , 449 , 891 . lu t. , pen u .- l . , dor o. , 2010 , phys rev d , 81 , 123015 . masui , k.w . , mcdonald p. , pen u .- l . , 2010 , phys rev d , 81 , 103527 . mcdonald p , eisenstein d.j . , 2007 , phys . d 76 , 063009 . merz h. , pen u .- l . , trac h. , 2005 , new astron , 10 , 393 . meiksin , a. , white , s. , 1999 , mnras , 308 , 1179 . miller a.d . , caldwell r. , et al . , 1999 , apj , 524 , l1 . neyrinck m. , szapudi i. , szalay a. , astrophys . 698 ( 2009 ) l90-l93 . noh y. , white m. , padmanabhan n. , 2009 , phys rev d , 80 , 123501 . padmanabhan n. , white m. , cohn j.d . , 2009 , phys rev d , 79 , 063523 . percival w. , baugh c. , et al . , 2001 , mnras , 327 , 1297 . percival w. , beth r. , et al . , 2010 , mnras , 401 , 2148 . rimes c.d . , hamilton a.j.s . , 2005 , mnras , 360 , l82 . rimes c.d . , hamilton a.j.s . , 2006 , mnras , 371 , 1205 . scoccimarro , r. , zaldarriaga , m. , hui , l. , 1999 , apj , 527 , 1 . seo h.j . , eisenstein d.j . , 2003 , apj , 598 , 720 . seo h.j . , eisenstein d.j . , 2005 , apj , 633 , 575 . seo h.j . , eisenstein d.j . , 2007 , apj , 665 , 14 . seo h.j . , eckel j. , eisenstein d.j . , et al . , 2010 , apj , 720 , 1650 . , masanori s. , dodelson s. , jain b. , takada m. , 2011 , apj , 729 , l11 . smith r.e . , peacock j.a . , jenkins a. , et al . , 2003 , mnras , 341 , 1311 . takahashi r. , yoshida n. , takada m. , et al . , 2009 , apj , 700 , 479 . takahashi r. , yoshida n. , takada m. , et al . , 2011 , apj , 726 , 7 . tegmark m. , 1997 , phys rev lett , 79 , 3806 . tegmark m. , eisenstein m. , et al . , 2006 , phys rev d , 74 , 123507 . , harnois - draps , j. , zhang t .- j . , , 2010 , apj submitted . ( arxiv:1012.0444 ) zeldovich , y.b . , 1970 , . , 5 , 84 . zhang t .- j . , yu h .- r . , harnois - draps j. , macdonald i. , pen u .- l . , 2011 , apj , 728 , 35 . interestingly even though the fisher information ( figure [ fig : information ] ) of gaussian and non - gaussian covariance matrices differ by about an order of magnitude near @xmath176 mpc@xmath86 , the measurement errors ( figures [ fig : ds_kmax ] and [ fig : ds_kmax_recon ] ) differ by only a few percent . the reason for this is that when the oscillatory power spectrum derivative @xmath177 is multiplied into the covariance matrix to compute the fisher matrix ( equation [ eqn : fisher ] ) , the off - diagonal elements of the covariance matrix can be canceled out . in @xcite , they considered the regime @xmath178 mpc@xmath86 and approximated a @xmath179 term , which originates from @xmath177 , as simply @xmath180 without any oscillations when computing their fisher matrix . since we are considering a similar regime , we also attempted this approximation . we emphasize that this approximation is not applicable to us , as our non - gaussian covariance matrix is not diagonal . nevertheless , it illustrates the effect that an oscillatory @xmath177 can have on distance measurements . figure [ fig : ds_approx ] shows @xmath181 as a function of limiting @xmath167 , using the approximated derivative . this can be compared to the top panel in figure [ fig : ds_kmax ] , where the only difference is that figure [ fig : ds_approx ] uses the approximated @xmath177 , and figure [ fig : ds_kmax ] computes @xmath177 by finite differences of actual power spectra . in both figures , the optimal estimators follow the inverse of the fisher information ( equation [ eqn : info ] ) . for the optimal gaussian case , the errors using either forms of derivatives give similar results ( other than some oscillations at small @xmath63 ) . this is expected since the gaussian covariance matrix is diagonal ( equation [ eqn : covariancegauss ] ) , so the approximation is valid . for the sub - optimal cases we considered in this paper , however , the discrepancies from the optimal gaussian case reach factors of 2 to 3 , depending on redshift . this can be attributed to the fact that all elements of the covariance matrix are given a uniform weight . when using the finite difference derivatives , however , different elements are weighted according to the values of @xmath177 at their corresponding @xmath63 modes . this effectively cancels most of the contributions from the off - diagonal elements of the covariance matrix , which explains why the optimal and sub - optimal errors differ by only a few percent in that case . ( without reconstruction ) . the only difference here is that the power spectrum derivative in the fisher matrix is an approximation which does not oscillate . compared to figure [ fig : ds_kmax ] , the _ optimal ( gauss ) _ case is similar , but the other cases are clearly different.,title="fig:",width=336 ] [ fig : ds_approx ]

in this paper , we study the approximability of the metric traveling salesman problem ( tsp ) and prove new explicit inapproximability bounds for that problem . the best up to now known hardness of approximation bounds were for the symmetric case ( due to lampis ) and for the asymmetric case ( due to papadimitriou and vempala ) . we construct here two new bounded occurrence csp reductions which improve these bounds to and , respectively . the latter bound is the first improvement in more than a decade for the case of the asymmetric tsp . one of our main tools , which may be of independent interest , is a new construction of a bounded degree wheel amplifier used in the proof of our results .

1303.6437

the * traveling salesman problem ( tsp ) * is one of the best known and most fundamental problems in combinatorial optimization . determining how well it can be approximated in polynomial time is therefore a major open problem , albeit one for which the solution still seems elusive . on the algorithmic side , the best known efficient approximation algorithm for the symmetric case is still a 35-year old algorithm due to christofides @xcite which achieves an approximation ratio of @xmath4 . however , recently there has been a string of improved results for the interesting special case of graphic tsp , improving the ratio to @xmath5 @xcite . for the asymmetric case ( atsp ) , it is not yet known if a constant - factor approximation is even possible , with the best known algorithm achieving a ratio of @xmath6 @xcite . unfortunately , there is still a huge gap between the algorithmic results mentioned above and the best currently known hardness of approximation results for tsp and atsp . for both problems , the known inapproximability thresholds are small constants ( @xmath0 and @xmath1 ( cf . @xcite ) , respectively ) . in this paper , we try to improve this situation somehow by giving modular hardness reductions that slightly improve the hardness bounds for both problems to @xmath2 and @xmath3 , respectively . the latter bound is the first , for more than a decade now , improvement of papadimitriou and vempala bound @xcite for the atsp . the method of our solution differs essentially from that of @xcite and uses some new paradigms of the bounded occurrence optimization which could be also of independent interest in other applications . similarly to @xcite , the hope is that the modularity of our construction , which goes through an intermediate stage of a bounded - occurrence constraint satisfaction problem ( csp ) , will allow an easier analysis and simplify future improvements . indeed , one of the main new ideas we rely on is a certain new variation of the wheel amplifiers first defined by berman and karpinski @xcite to establish inapproximability for 3-regular csps . this construction , which may be of independent interest , allows us to establish inapproximability for a 3-regular csp with a special structure . this special structure then makes it possible to simulate many of the constraints in the produced graph essentially `` for free '' , without using gadgets to represent them . thus , even though for the remaining constraints we mostly reuse gadgets which have already appeared in the literature , we are still able to obtain improved bounds . let us now recall some of the previous work on the hardness of approximation of tsp and atsp . papadimitriou and yannakakis @xcite were the first to construct a reduction that , combined with the pcp theorem @xcite , gave a constant inapproximability threshold , though the constant was not more than @xmath7 for the tsp with distances either one or two . engebretsen @xcite gave the first explicit approximation lower bound of @xmath8 for the problem . the inapproximability factor was improved to @xmath9 by bckenhauer and seibert @xcite , who studied the restricted version of the tsp with distances one , two and three . papadimitriou and vempala @xcite proved that it is -hard to approximate the tsp with a factor better than @xmath10 . presently , the best known approximation lower bound is @xmath0 due to lampis @xcite . the important restriction of the tsp , in which we consider instances with distances between cities being values in @xmath11 , is often referred to as the @xmath12-tsp . the best known efficient approximation algorithm for the @xmath13-tsp has an approximation ratio @xmath14 and is due to berman and karpinski @xcite . as for lower bounds , engebretsen and karpinski @xcite gave inapproximability thresholds for the @xmath12-tsp problem of @xmath15 for @xmath16 and @xmath17 for @xmath18 . more recently , karpinski and schmied @xcite obtained improved inapproximability factors for the @xmath13-tsp and the @xmath19-tsp of @xmath20 and @xmath21 , respectively . for atsp the currently best known approximation lower bound was @xmath1 due to papadimitriou and vempala @xcite . when we restrict the problem to distances with values in @xmath11 , there is a simple approximation algorithm with approximation ratio @xmath22 that constructs an arbitrary tour as solution . blser @xcite gave an efficient approximation algorithm for the @xmath23-atsp with approximation ratio @xmath24 . karpinski and schmied @xcite proved that it is -hard to approximate the @xmath25-atsp and the @xmath19-atsp within any factor less than @xmath26 and @xmath27 , respectively . for the case @xmath18 , engebretsen and karpinski @xcite gave an inapproximability threshold of @xmath28 . * overview : * in this paper we give a hardness proof which proceeds in two steps . first , we start from the max - e3-lin2 problem , in which we are given a system of linear equations mod @xmath29 with exactly three variables in each equation and we want to find an assignment such as to maximize the number of satisfied equations . optimal inapproximability results for this problem were shown by hstad @xcite . we reduce this problem to a special case where variables appear exactly @xmath30 times and the linear equations have a particular structure . the main tool here is a new variant of the wheel amplifier graphs of berman and karpinski @xcite . in the second step , we reduce this @xmath30-regular csp to tsp and atsp . the general construction is similar in both cases , though of course we use different gadgets for the two problems . the gadgets we use are mostly variations of gadgets which have already appeared in previous reductions . nevertheless , we manage to obtain an improvement by exploiting the special properties of the @xmath30-regular csp . in particular , we show that it is only necessary to construct gadgets for roughly one third of the constraints of the csp instance , while the remaining constraints are simulated without additional cost using the consistency properties of our gadgets . this idea may be useful in improving the efficiency of approximation - hardness reductions for other problems . thus , overall we follow an approach unlike that of @xcite , where the reduction is performed in one step , and closer to @xcite . the improvement over @xcite comes mainly from the idea mentioned above , which is made possible using the new wheel amplifiers , as well as several other tweaks . the end result is a more economical reduction which improves the bounds for both tsp and atsp . an interesting question may be whether our techniques can also be used to derive improved inapproximability results for variants of the atsp and tsp ( cf . @xcite,@xcite and @xcite ) or other graph problems , such as the steiner tree problem . in the following , we give some definitions concerning directed ( multi-)graphs and omit the corresponding definitions for undirected ( multi-)graphs if they follow from the directed case . given a directed graph @xmath31 and @xmath32 , for @xmath33 , we define @xmath34 and @xmath35 . for convenience , we abbreviate a sequence of edges @xmath36 , @xmath37 by @xmath38 . in the undirected case , we use sometimes @xmath39 instead of @xmath40 , @xmath41 . given a directed ( multi-)graph @xmath42 , an _ cycle in @xmath42 is a directed cycle that traverses all edges of @xmath42 exactly once . we refer to @xmath42 as _ eulerian _ , if there exists an eulerian cycle in @xmath42 . for a multiset @xmath43 of directed edges and @xmath44 , we define the outdegree ( indegree ) of @xmath45 with respect to @xmath43 , denoted by @xmath46 ( @xmath47 ) , to be the number of edges in @xmath43 that are outgoing of ( incoming to ) @xmath45 . the _ balance _ of a vertex @xmath45 with respect to @xmath43 is defined as @xmath48 . in the case of a multiset @xmath43 of undirected edges , we define the balance @xmath49 of a vertex @xmath44 to be one if the number of incident edges in @xmath43 is odd and zero otherwise . we refer to vertices @xmath44 with @xmath50 as _ balanced _ with respect to @xmath43 . it is well known that a ( directed ) ( multi-)graph @xmath31 is eulerian if and only if all edges are in the same ( weakly ) connected component and all vertices @xmath51 are balanced with respect to @xmath52 . given a multiset of edges @xmath43 , we denote by @xmath53 the number of ( weakly ) connected components in the graph induced by @xmath43 . quasi - tour _ @xmath43 in a ( directed ) graph @xmath42 is a multiset of edges from @xmath52 such that all vertices are balanced with respect to @xmath43 and @xmath54 . we refer to a quasi - tour @xmath43 in @xmath42 as a _ tour _ if @xmath55 . given a cost function @xmath56 , the cost of a quasi - tour @xmath43 in @xmath42 is defined by @xmath57 . in the asymmetric traveling salesman problem ( atsp ) , we are given a directed graph @xmath31 with positive weights on edges and we want to find an ordering @xmath58 of the vertices such as to minimize @xmath59}d_g(v_i , v_{i+1})$ ] , where @xmath60 denotes the shortest path distance in @xmath42 . in this paper , we will use the following equivalent reformulation of the atsp : given a directed graph @xmath42 with weights on edges , we want to find a tour @xmath43 in @xmath42 , that is , a spanning connected multi - set of edges that balances all vertices , with minimum cost . the metric traveling salesman problem ( tsp ) is the special case of the atsp , in which instances are undirected graphs with positive weights on edges . in this section , we define the bi - wheel amplifier graphs which will be our main tool for proving hardness of approximation for a bounded occurrence csp with some special properties . bi - wheel amplifiers are a simple variation of the wheel amplifier graphs given in @xcite . let us first recall some definitions ( see also @xcite ) . if @xmath42 is an undirected graph and @xmath61 a set of vertices , we say that @xmath42 is a _ @xmath62-regular amplifier _ for @xmath63 if the following conditions hold : * all vertices of @xmath63 have degree @xmath64 and all vertices of @xmath65 have degree @xmath62 . * for every non - empty subset @xmath66 , we have the condition that @xmath67 , where @xmath68 is the set of edges with exactly one endpoint in @xmath69 . we refer to the set @xmath63 as the set of _ contact _ vertices and to @xmath65 as the set of _ checker _ vertices . amplifier graphs are useful in proving inapproximability for csps , in which every variable appears a bounded number of times . here , we will rely on @xmath30-regular amplifiers . a probabilistic argument for the existence of such graphs was given in @xcite , with the definition of wheel amplifiers . a wheel amplifier with @xmath70 contact vertices is constructed as follows : first construct a cycle on @xmath71 vertices . number the vertices @xmath72 and select uniformly at random a perfect matching of the vertices whose number is not a multiple of 7 . the matched vertices will be our checker vertices , and the rest our contacts . it is easy to see that the degree requirements are satisfied . berman and karpinski @xcite gave a probabilistic argument to prove that with high probability the above construction indeed produces an amplifier graph , that is , all partitions of the sets of vertices give large cuts . here , we will use a slight variation of this construction , called a bi - wheel . a * bi - wheel amplifier * with @xmath70 contact vertices is constructed as follows : first construct two disjoint cycles , each on @xmath73 vertices and number the vertices of each @xmath74 . the contacts will again be the vertices whose number is a multiple of 7 , while the remaining vertices will be checkers . to complete the construction , select uniformly at random a perfect matching from the checkers of one cycle to the checkers of the other . intuitively , the reason that amplifiers are a suitable tool here is that , given a csp instance , we can use a wheel amplifier to replace a variable that appears @xmath70 times with @xmath71 new variables ( one for each wheel vertex ) each of which appears 3 times . each appearance of the original variable is represented by a contact vertex and for each edge of the wheel we add an equality constraint between the corresponding variables . we can then use the property that all partitions give large cuts to argue that in an optimal assignment all the new vertices take the same value . we use the bi - wheel amplifier in our construction in a similar way . the main difference is that while cycle edges will correspond to equality constraints , matching edges will correspond to inequality constraints . the contacts of one cycle will represent the positive appearances of the original variable , and the contacts of the other the negative ones . the reason we do this is that we can encode inequality constraints more efficiently than equality with a tsp gadget , while the equality constraints that arise from the cycles will be encoded in our construction `` for free '' using the consistency of the inequality gadgets . before we apply the construction however , we have to prove that the bi - wheel amplifiers still have the desired amplification properties . [ thm : biwheel ] with high probability , bi - wheels are 3-regular amplifiers . exploiting the similarity between bi - wheels and the standard wheel amplifiers of @xcite , we will essentially reuse the proof given there . first , some definitions : we say that @xmath69 is a _ bad set _ if the size of its cut is too small , violating the second property of amplifiers . we say that it is a _ minimal bad set _ if @xmath69 is bad but removing any vertex from @xmath69 gives a set that is not bad . recall the strategy of the proof from @xcite : for each partition of the vertices into @xmath69 and @xmath75 , they calculate the probability ( over the random matchings ) that this partition gives a minimal bad set . then , they take the sum of these probabilities over all potentially minimal bad sets and prove that the sum is at most @xmath76 for some constant @xmath77 . it follows by union bound that with high probability , no set is a minimal bad set and therefore , the graph is a proper amplifier . our first observation is the following : consider a wheel amplifier on @xmath71 vertices where , rather than selecting uniformly at random a perfect matching among the checkers , we select uniformly at random a perfect matching from checkers with labels in the set @xmath78 to checkers with labels in the set @xmath79 . this graph is almost isomorphic to a bi - wheel . more specifically , for each bi - wheel , we can obtain a graph of this form by rewiring two edges , and vice - versa . it easily follows that properties that hold for this graph , asymptotically with high probability also hold for the bi - wheel . thus , we just need to prove that a wheel amplifier still has the amplification property if , rather than selecting a random perfect matching , we select a random matching from one half of the checker vertices to the other . we will show this by proving that , for each set of vertices @xmath80 , the probability that @xmath80 is a minimal bad set is roughly the same in both cases . after establishing this fact , we can simply rely on the proof of @xcite . recall that the wheel has @xmath81 checker vertices . given a set @xmath80 with @xmath82 , what is the probability that exactly @xmath83 edges have exactly one endpoint in @xmath80 ? in a standard wheel amplifier the probability is @xmath84 where we denote by @xmath85 the product of all odd natural numbers less than or equal to @xmath86 , and we assume without loss of generality that @xmath87 is even . let us explain this : the probability that exactly @xmath83 edges cross the cut in this graph is equal to the number of ways we can choose their endpoints in @xmath80 and in its complement , times the number of ways we can match the endpoints , times the number of matchings of the remaining vertices , divided by the number of matchings overall . how much does this probability change if we only allow matchings from one half of the checkers to the other ? intuitively , we need to consider two possibilities : one is that @xmath80 is a balanced set , containing an equal number of checkers from each side , while the other is that @xmath80 is unbalanced . it is not hard to see that if @xmath80 is unbalanced , then , we can easily establish that the cut must be large . thus , the main interesting case is the balanced one ( and we will establish this fact more formally ) . suppose that @xmath82 and @xmath80 contains exactly @xmath88 checkers from each side . then the probability that there are exactly @xmath83 edges crossing the cut is @xmath89 let us explain this . if @xmath80 is balanced and there are @xmath83 matching edges with exactly one endpoint in @xmath80 , then , exactly @xmath90 of them must be incident on a vertex of @xmath80 on each side , since the remaining vertices of @xmath80 must have a perfect matching . again , we pick the endpoints on each side , and on the complement of @xmath80 , select a way to match them , select matchings on the remaining vertices and divide by the number of possible perfect matchings . using stirling formulas , it is not hard to see that @xmath91 . also @xmath92 . it follows that @xmath93 is roughly the same as @xmath94 in this case , modulo some polynomial factors which are not significant since the probabilities we are calculating are exponentially small . let us now also show that if @xmath80 is unbalanced , the probability that it is a minimal bad set is even smaller . first , observe that if @xmath80 is a minimal bad set whose cut has @xmath83 edges , we have @xmath95 . the reason for this is that since @xmath80 is bad , then , @xmath83 is smaller than the number of contacts in @xmath80 minus the number of cycle edges cut . it is not hard to see that , in each fragment , that is , each subset of @xmath80 made up of a contiguous part of the cycle , two cycle edges are cut . thus , the extra edges we need for the contacts the fragment contains are at most @xmath96 of its checkers . suppose now that @xmath80 contains @xmath97 checkers on one side and @xmath98 checkers on the other . the probability that @xmath83 matching edges have one endpoint in @xmath80 is @xmath99 the reasoning is the same as before , except we observe that we need to select more endpoints on the side where @xmath80 is larger , since after we remove checkers matched to outside vertices @xmath80 must have a perfect matching . observe that for @xmath100 this gives @xmath93 . we will show that for the range of values we care about @xmath101 achieves a maximum for @xmath100 , and can thus be upper - bounded by ( essentially ) @xmath94 , which is the probability that a set is bad in the standard amplifier . the rest of the proof follows from the argument given in @xcite . in particular , we can assume that @xmath102 , since @xmath103 edges are cut with probability 1 . to show that the maximum is achieved for @xmath100 , we look at @xmath104 . we will show that this is less than 1 . using the identity @xmath105 , we get @xmath106 using the fact that @xmath107 , we end up needing to prove the following . @xmath108 combining that without loss of generality @xmath109 holds with the bounds of @xmath83 and @xmath110 we have already mentioned , the inequality @xmath111 is straightforward to establish . by using the bi - wheel amplifier from the previous section , we are going to prove hardness of approximation for a bounded occurrence csp with very special properties . this particular csp will be well - suited for constructing a reduction to the tsp given in the next section . as the starting point of our reduction , we make use of the inapproximability result due to hstad @xcite for the , which is defined as follows : given a system @xmath112 of linear equations mod 2 , in which each equation is of the form @xmath113 with @xmath114 , we want to find an assignment to the variables of @xmath112 such as to maximize the number of satisfied equations . let @xmath112 be an instance of the and @xmath115 the set of variables , that appear in @xmath112 . we denote by @xmath116 the number of appearances of @xmath117 in @xmath112 . for every @xmath118 , there exists a constant @xmath119 such that given an instance @xmath112 of the with @xmath120 equations and @xmath121 } d(i)\leq b_{\epsilon}$ ] , it is -hard to decide whether there is an assignment that leaves at most @xmath122 equations unsatisfied , or all assignment leave at least @xmath123 equations unsatisfied . similarly to the work by berman and karpinski @xcite ( see also @xcite and @xcite ) , we will reduce the number of occurrences of each variable to @xmath30 . for this , we will use our amplifier construction to create special instances of the hybrid problem , which is defined as follows : given a system @xmath124 of linear equations mod @xmath29 with either three or two variables in each equation , we want to find an assignment such as to maximize the number of satisfied equations . in particular , we are going to prove the following theorem . [ thm : hybrid ] for every constant @xmath125 and @xmath126 , there exist instances of the hybrid problem with @xmath127 equations such that : @xmath128 each variable occurs exactly three times . @xmath130 equations are of the form @xmath131 , @xmath132 equations are of the form @xmath133 and @xmath120 equations are of the form @xmath134 . @xmath135 it is -hard to decide whether there is an assignment to the variables that leaves at most @xmath136 equations unsatisfied , or every assignment to the variables leaves at least @xmath137 equations unsatisfied . let @xmath118 be a constant and @xmath112 an instance of the with @xmath121 } d(i ) \leq b_{\epsilon}$ ] . for a fixed @xmath126 , we can flip some of the literals such that all equations in the instance @xmath112 are of the form @xmath138 , where @xmath139 are variables or negations . by constructing three more copies of each equation , in which all possible pairs of literals appear negated , we may assume that each variable occurs the same number of times negated as unnegated . let us fix a variable @xmath117 in @xmath112 . then , we create @xmath140 new variables @xmath141 . in addition , we construct a bi - wheel amplifier @xmath142 on @xmath143 vertices ( that is , a bi - wheel with @xmath116 contact vertices ) with the properties described in theorem [ thm : biwheel ] . since @xmath144 is a constant , this can be accomplished in constant time . in the remainder , we refer to _ contact _ and _ checker _ variables as the elements in @xmath145 , whose corresponding index is a contact and checker vertex in @xmath142 , respectively . we denote by @xmath146 the associated perfect matching on the set of checker vertices of @xmath142 . in addition , we denote by @xmath147 and @xmath148 the set of edges contained in the first and second cycle of @xmath142 , respectively . let us now define the equations of the corresponding instance of the hybrid problem . for each edge @xmath149 , we create the equation @xmath150 and refer to equations of this form as _ matching equations_. on the other hand , for each edge @xmath151 in the cycle @xmath152 with @xmath153 , we introduce the equation @xmath154 . equations of this form will be called _ cycle equations_. finally , we replace the @xmath155-th unnegated appearance of @xmath117 in @xmath112 by the contact variable @xmath156 with @xmath157 , whereas the @xmath155-th negated appearance is replaced by @xmath158 . the former construction yields @xmath120 equations with three variables in the instance of the hybrid problem , which we will denote by @xmath124 . notice that each variable appears in exactly @xmath30 equations in @xmath124 . clearly , we have @xmath159 equations , thereof @xmath132 matching equations , @xmath130 cycle equations and @xmath120 equations of the form @xmath160 . a consistent assignment to @xmath145 is an assignment with @xmath161 and @xmath162 for all @xmath163 $ ] , where @xmath126 . a consistent assignment to the variables of @xmath124 is an assignment that is consistent to @xmath145 for all @xmath164 $ ] . by standard arguments using the amplifier constructed in theorem [ thm : biwheel ] , it is possible to convert an assignment to a consistent assignment without decreasing the number of satisfied equations and the proof of theorem [ thm : hybrid ] follows . this section is devoted to the proof of the following theorem . [ thm : tspmain ] it is -hard to approximate the tsp to within any constant approximation ratio less than @xmath2 . let us first sketch the high - level idea of the construction . starting with an instance of the hybrid problem , we will construct a graph , where gadgets represent the equations . we will design gadgets for equations of size three ( figure [ fig : eqn3 ] ) and for equations of size two corresponding to _ matching _ edges of the bi - wheel ( figure [ fig : eqn2 ] ) . we will not construct gadgets for the cycle edges of the bi - wheel ; instead , the connections between the matching edge gadgets will be sufficient to encode these extra constraints . this may seem counter - intuitive at first , but the idea here is that if the gadgets for the matching edges are used in a consistent way ( that is , the tour enters and exits in the intended way ) then it follows that the tour is using all edges corresponding to one wheel and none from the other . thus , if we prove consistency for the matching edge gadgets , we implicitly get the cycle edges `` for free '' . this observation , along with an improved gadget for size - three equations and the elimination of the variable part of the graph , are the main sources of improvement over the construction of @xcite . we are going to describe the construction that encodes an instance @xmath124 of the hybrid problem into an instance of the tsp problem . due to theorem [ thm : hybrid ] , we may assume that the equations with three variables in @xmath124 are all of the form @xmath165 . in order to ensure that some edges are to be used at least once in any valid tour , we apply the following simple trick that was already used in the work by lampis @xcite : let @xmath166 be an edge with weight @xmath167 that we want to be traversed by every tour . we remove @xmath166 and replace it with a path of @xmath168 edges and @xmath169 newly created vertices each of degree two , where we think of @xmath168 as a large constant . each of the @xmath168 edges has weight @xmath170 and any tour that fails to traverse at least two newly created edges is not connected . on the other hand , a tour that traverses all but one of those edges can be extended by adding two copies of the unused edge increasing the cost of the underlying tour by a negligible value . in summary , we may assume that our construction contains _ forced _ edges that need to be traversed at least once by any tour . if @xmath171 and @xmath172 are vertices , which are connected by a forced edge @xmath166 , we write @xmath173 or simply @xmath174 . in the following , we refer to unforced edges @xmath166 with @xmath175 as _ simple_. all unforced edges in our construction will be simple . let us start with the description of the corresponding graph @xmath176 : for each bi - wheel @xmath177 , we will construct the subgraph @xmath178 of @xmath176 . for each vertex of the bi - wheel , we create a vertex in the graph and for each cycle equation @xmath179 , we create a simple edge @xmath180 . given a matching equation between two checkers @xmath181 , we connect the vertices @xmath182 and @xmath183 with two forced edges @xmath184 and @xmath185 . we have @xmath186 for each @xmath187 . additionally , we create a central vertex @xmath188 that is connected to gadgets simulating equations with three variables . let @xmath189 be the @xmath155-th equation with three variables in @xmath124 . we now create the graph @xmath190 displayed in figure [ fig : eqn3 ] @xmath191 , where the ( contact ) vertices for @xmath192 have already been constructed in the cycles . the edges @xmath193 with @xmath194 and @xmath195 are all forced edges with @xmath196 . furthermore , we have @xmath197 for all @xmath194 . @xmath198 and @xmath199 are both forced edges , whereas all remaining edges of @xmath190 are simple . this is the whole description of @xmath176 . c|c & + @xmath191 & @xmath200 given an instance @xmath124 of the hybrid problem and an assignment @xmath201 to the variables in @xmath124 , we are going to construct a tour in @xmath176 according to @xmath201 and give the proof of one direction of the reduction . in particular , we are going to prove the following lemma . [ lem : tsp1 ] if there is an assignment to the variables of a given instance @xmath124 of the hybrid problem with @xmath127 equations and @xmath202 bi - wheels , that leaves @xmath110 equations unsatisfied , then , there exists a tour in @xmath176 with cost at most @xmath203 . before we proceed , let us give a useful definition . let @xmath42 be an edge - weighted graph and @xmath43 a multi - set of edges of @xmath52 that defines a quasi - tour . consider a set @xmath204 . the local edge cost of the set @xmath205 is then defined as @xmath206 in words , for each vertex in @xmath205 , we count half the total weight of its incident edges used in the quasi - tour ( including multiplicities ) . observe that this sum contains half the weight of edges with one endpoint in @xmath205 but the full weight for edges with both endpoints in @xmath205 ( since we count both endpoints in the sum ) . also note that for two sets @xmath207 , we have @xmath208 ( with equality for disjoint sets ) and that @xmath209 . first , note that it is sufficient to prove that we can construct a quasi - tour of the promised cost which uses all forced edges exactly once . since all unforced edges have cost 1 , if we are given a quasi - tour we can connect two disconnected components by using an unforced edge that connects them twice ( this is always possible since the underlying graph we constructed is connected ) . this does not increase the cost , since we added two unit - weight edges and decreased the number of components . repeating this results in a connected tour . let @xmath210 be the associated set of bi - wheels of @xmath124 . for a fixed bi - wheel @xmath177 , let @xmath211 be its associated set of variables . due to the construction of instances of the hybrid problem in section [ sec : hybrid ] , we may assume that all equations with two variables are satisfied by the given assignment . thus , we have @xmath212 , @xmath213 and @xmath214 for all @xmath215 $ ] . assuming @xmath216 for some @xmath217 , we use once all simple edges @xmath218 with @xmath219 $ ] and the edge @xmath220 . we also use all forced edges corresponding to matching equations once . in other words , for each biwheel we select the cycle that corresponds to the assignment 1 and use all the simple edges from that cycle . this creates a component that contains all checker vertices from both cycles and all contacts from one cycle . c|c & + @xmath191 & @xmath200 as for the next step , we are going to describe the tour traversing @xmath190 with @xmath221 $ ] given an assignment to contact variables . let us assume that @xmath190 simulates @xmath222 . according to the assignment to @xmath171 , @xmath172 and @xmath223 , we will traverse @xmath190 as follows : in all cases , we will use all forced edges once . + * case ( @xmath224 ) : * then , we use @xmath225 for all @xmath226 and @xmath227 with @xmath228 . for @xmath229 with @xmath230 , we use @xmath231 for all @xmath226 . + * case ( @xmath232 with @xmath126 ) : * in both cases , we traverse @xmath233 for all @xmath234 and @xmath235 . + * case ( @xmath236 ) : * we use @xmath237 with @xmath238 . furthermore , we include @xmath239 for both @xmath226 . let us now analyze the cost of the edges of our quasi - tour given an assignment . for each matching edge @xmath240 consider the set of vertices made up of its endpoints . its local cost is 5 : we pay 4 for the forced edges and there are two used simple edges with one endpoint in the set . let us also consider the local cost for a size - three equation gadget , where we consider the set to contain the contact vertices @xmath241 as well the other 8 vertices of the gadget . the local cost here is 9.5 for the forced edges . we also pay 6 more ( for a total of 15.5 ) when the assignment satisfies the equation or 7 more when it does not . thus , we have given a covering of the vertices of the graph by @xmath132 sets of size two , @xmath120 sets of size 11 and @xmath242 . the total edge cost is thus at most @xmath243 . to obtain an upper bound on the cost of the quasi - tour , we observe that the tour has at most @xmath244 components ( one for each bi - wheel and one containing @xmath188 ) . the lemma follows . in this section , we are going prove the other direction of our reduction . given a tour in @xmath176 , we are going to define an assignment to the variables of the associated instance of the hybrid problem and give the proof of the following lemma . [ lem : tsp2 ] if there is a tour in @xmath176 with cost @xmath245 , then , there is an assignment to the variables of the corresponding instance of the hybrid problem that leaves at most @xmath110 equations unsatisfied . again , let us give a useful definition . consider a quasi - tour @xmath43 and a set @xmath204 . let @xmath246 be the number of connected components induced by @xmath43 which are fully contained in @xmath205 . then , the full local cost of the set @xmath205 is defined as @xmath247 . by the definition , the full local cost of @xmath248 is equal to the cost of the quasi - tour ( plus 2 ) . intuitively , @xmath249 captures the cost of the quasi - tour restricted to @xmath205 : it includes the cost of edges and the cost of added connected components . note that now for two disjoint sets @xmath207 we have @xmath250 since @xmath251 could contain more connected components than @xmath207 together . if we know that the total cost of the quasi - tour is small , then @xmath252 is small ( less than @xmath253 ) . we can use this to infer that the sum of the local full costs of all gadgets is small . the high - level idea of the proof is the following : we will use roughly the same partition of @xmath248 into sets as in the proof of lemma [ lem : tsp1 ] . for each set , we will give a lower bound on its full local cost for any quasi - tour , which will be equal to what the tour we constructed in lemma [ lem : tsp1 ] pays . if a given quasi - tour behaves differently its local cost will be higher . the difference between the actual local cost and the lower bound is called the credit of that part of the graph . we construct an assignment for @xmath124 and show that the total sum of credits is higher that the number of unsatisfied equations . but using the reasoning of the previous paragraph , the total sum of credits will be at most @xmath110 . we are going to prove a slightly stronger statement and show that if there exists a quasi - tour in @xmath176 with cost @xmath245 , then , there exists an assignment leaving at most @xmath110 equations unsatisfied . recall that the existence of a tour in @xmath176 with cost @xmath254 implies the existence of a quasi - tour in @xmath176 with cost at most @xmath254 . we may assume that simple edges are contained only once in @xmath43 due to the following preprocessing step : if @xmath43 contains two copies of the same simple edge , we remove them without increasing the cost , since the number of components can only increase by one . in the following , given a quasi - tour @xmath43 in @xmath176 , we are going to define an assignment @xmath255 and analyze the number of satisfied equations by @xmath255 compared to the cost of the quasi - tour . the general idea is that each vertex of @xmath176 that corresponds to a variable of @xmath124 has exactly two forced and exactly two simple edges incident to it . if the forced edges are used once each , the variable is called honest . we set it to 1 if the simple edges are both used once and to 0 otherwise . it is not hard to see that , because simple cycle edges connect vertices that represent the variables , this procedure will satisfy all cycle equations involving honest variables . we then argue that if other equations are unsatisfied the tour is also paying extra , and the same happens if a variable is dishonest . let us give more details . first , we concentrate on the assignment for checker variables . let us consider the following equations with two variables @xmath256 , @xmath257 , @xmath258 , @xmath259 and @xmath260 . we are going to analyze the cost of a quasi - tour traversing the gadget displayed in figure [ fig : eqn2 ] @xmath191 and define an assignment according to @xmath43 . let us first assume that our quasi - tour is honest , that is , the underlying quasi - tour traverses forced edges only once . * honest tours : * for @xmath261 , we set @xmath262 if the quasi - tour traverses both simple edges incident on @xmath171 and @xmath263 , otherwise . since we removed all copies of the same simple edge , we may assume that cycle equations are always satisfied . if the tour uses @xmath264 , we get @xmath265 , @xmath266 and @xmath267 satisfied equations . given @xmath268 , we obtain @xmath267 satisfied equations as well . let us define @xmath269 . notice that in both cases , we have local cost @xmath270 . we claim that @xmath271 for a valid quasi - tour . in order to obtain a valid quasi - tour , we need to traverse both forced edges in @xmath272 and use at least two simple edges , as otherwise , it implies @xmath273 . given a quasi - tour @xmath43 , we introduce a local credit function defined by @xmath274 . if @xmath275 forms a connected component , we get @xmath276 satisfied equations and @xmath277 , which is sufficient to pay for the unsatisfied equation @xmath278 . on the other hand , assuming @xmath265 and @xmath279 , we get @xmath280 and @xmath281 unsatisfied equation . + * dishonest tours : * we are going to analyze quasi - tours , which are using one of the forced edges twice . by setting @xmath282 , we are able to find an assignment that always satisfies @xmath278 and two other equations out of the five that involve these dishonest variables . the local cost in this case is at least @xmath283 . hence , the credit @xmath284 is sufficient to pay for the two unsatisfied equations . again , we will distinguish between honest tours ( which use forced edges exactly once ) and dishonest tours . this time we are interested in seven equations : the size - three equation @xmath285 and the six cycle equations containing the three contacts . observe that the local cost of @xmath286 is at least 15.5 . the local edge cost of any quasi - tour is 9.5 for the forced edges . for each component @xmath287 with @xmath288 , we need to pay at least 2 more because there are two vertices with odd degree ( @xmath289 ) and we also need to connect the component to the rest of the graph ( otherwise the component already costs 2 more ) . let us define the credit of @xmath290 with respect to @xmath43 by @xmath291 . + + * honest tours : * for each @xmath234 , we set @xmath228 if the tour uses both simple edges incident on @xmath292 and @xmath293 , otherwise . notice that in the case @xmath294 with @xmath295 , this satisfies all seven equations and the tour has local cost at least @xmath296 . case @xmath297 the assignment now failed to satisfy the size - three equation , so we need to prove that the quasi - tour has local cost at least 16.5 . since all vertices are balanced with respect to @xmath43 , the quasi - tour has to use at least one edge incident on @xmath298 and @xmath299 besides @xmath300 and @xmath301 . if the quasi - tour takes @xmath302 for a @xmath303 and all @xmath304 , since all simple edges incident on @xmath192 are used , we get at total cost of at least 16.5 , which gives a credit of 1 . case @xmath305 without loss of generality , we assume that @xmath306 holds . again , only the size - three equation is unsatisfied , so we must show that the local cost is at least @xmath307 . we will discuss two subcases . @xmath128 there is a connected component @xmath308 for some @xmath309 . we obtain that @xmath310 and therefore , a lower bound on the total cost of @xmath307 . @xmath129 since we may assume that @xmath311 , @xmath312 , @xmath313 and @xmath314 are balanced with respect to @xmath43 , we have that @xmath315 for all @xmath316 and @xmath317 . because @xmath318 are also balanced , we obtain @xmath319 for all @xmath316 , which implies a total cost of @xmath307 . + * dishonest tours : * let us assume that the quasi - tour uses both of the forced edges @xmath320 and @xmath321 for some @xmath322 twice . we delete both copies and add @xmath323 instead which reduces the cost of the quasi - tour . hence , we may assume that only one of the two incident forced edges is used twice . first , observe that if all forced edges were used once , then there would be eight vertices in the gadget with odd degree : @xmath324 . if exactly one forced edge is used twice , then seven of these vertices have odd degree . thus , it is impossible for the tour to make the degrees of all seven even using only the simple edges that connect them . we can therefore assume that if a forced edge is used twice , there exists another forced edge used twice . we will now take cases , depending on how many of the vertices @xmath192 are incident on forced edges used twice . note that if one of the forced edges incident on @xmath171 is used twice , then exactly one of the simple edges incident on @xmath171 is used once . so , first suppose all three of @xmath192 have forced edges used twice . the local cost from forced edges is at least @xmath325 . furthermore , there are three vertices of the form @xmath326 , for @xmath288 and @xmath327 with odd degree . these have no simple edges connecting them , thus the quasi - tour will use three simple edges to balance their degrees . finally , the used simple edges incident on @xmath192 each contribute @xmath328 to the local cost . thus , the total local cost is at least @xmath329 , giving us a credit of @xmath30 . it is not hard to see that there is always an assignment satisfying four out of the seven affected equations , so this case is done . second , suppose exactly two of @xmath192 have incident forced edges used twice , say , @xmath330 . for @xmath223 , we select the honest assignment ( 1 if the incident simple edges are used , 0 otherwise ) and this satisfies the cycle equations for this variable . we can select assignments for @xmath330 that satisfy three of the remaining five equations , so we need to show that the cost in this case is at least @xmath331 . the cost of forced edges is at least @xmath332 , and the cost of simple edges incident on @xmath330 adds @xmath281 to the local cost . one of the vertices @xmath333 and one of @xmath334 have odd degree , therefore the cost uses two simple edges to balance them . finally , the vertices @xmath335 have odd degree . if two simple edges incident to them are used , we have a total local cost of 17.5 . if the edge connecting them is used , then the two simple edges incident on @xmath223 must be used , again pushing the local cost to 17.5 . finally , suppose only @xmath171 has an incident forced edge used twice . by the parity argument given above , this means that one of the forced edges incident on @xmath188 is used twice . we can satisfy the cycle equations for @xmath336 by giving them their honest assignment , and out of the three remaining equations some assignment to @xmath171 satisfies two . therefore , we need to show that the cost is at least 16.5 . the local cost from forced edges is @xmath337 and the simple edge incident on @xmath171 contributes @xmath328 . also , at least one simple edge incident on @xmath312 or @xmath311 is used , since one of them has odd degree . for @xmath334 , either two simple edges are used , or if the edge connecting them is used the simple edges incident on @xmath172 contribute @xmath281 more . with similar reasoning for @xmath335 , we get that the total local cost is at least @xmath338 . + let us now conclude our analysis . consider the following partition of @xmath339 : we have a singleton set @xmath242 , @xmath132 sets of size @xmath29 containing the matching edge gadgets and @xmath120 sets of size @xmath340 containing the gadgets for size - three equations ( except @xmath188 ) . the sum of their local costs is at most @xmath341 . but the sum of their local costs is ( using the preceding analysis ) equal to @xmath342 . thus , the sum of all credits is at most @xmath110 . since we have already argued that the sum of all credits is enough to cover all equations unsatisfied by our assignment , this concludes the proof . we are ready to give the proof of theorem [ thm : tspmain ] . we are given an instance @xmath112 of the with @xmath202 variables and @xmath120 equations . for all @xmath343 , there exists a @xmath110 such that if we repeat each equation @xmath110 time we get an instance @xmath344 with @xmath345 equations and @xmath202 variables such that @xmath346 . then , from @xmath344 , we generate an instance @xmath124 of the hybrid problem and the corresponding graph @xmath176 . due to lemmata [ lem : tsp1 ] , [ lem : tsp2 ] and theorem [ thm : hybrid ] , we know that for all @xmath347 , it is -hard to tell whether there is a tour with cost at most @xmath348 or all tours have cost at least @xmath349 . the ratio between these two cases can get arbitrarily close to @xmath2 by appropriate choices for @xmath350 . in this section , we prove the following theorem . [ thm : atspmain ] it is -hard to approximate the atsp to within any constant approximation ratio less than @xmath3 . let us describe the construction that encodes an instance @xmath124 of the hybrid problem into an instance of the atsp . again , it will be useful to have the ability to force some edges to be used , that is , we would like to have bidirected forced edges . a bidirected forced edge of weight @xmath167 between two vertices @xmath171 and @xmath172 will be created in a similar way as undirected forced edges in the previous section : construct @xmath169 new vertices and connect @xmath171 to @xmath172 through these new vertices , making a bidirected path with all edges having weight @xmath170 . it is not hard to see that without loss of generality we may assume that all edges of the path are used in at least one direction , though we should note that the direction is not prescribed . in the remainder , we denote a directed forced edge consisting of vertices @xmath171 and @xmath172 by @xmath351 , or @xmath352 . let @xmath124 consist of the collection @xmath353 of bi - wheels . recall that the bi - wheel consists of two cycles and a perfect matching between their checkers . let @xmath354 be the associated set of variables of @xmath177 . we write @xmath355 to denote the function which , given the index of a checker variable @xmath182 returns the index @xmath155 of the checker variable @xmath183 to which it is matched ( that is , the function @xmath356 is a permutation function encoding the matching ) . we write @xmath357 to denote the inverse function @xmath358 . now , for each bi - wheel @xmath177 , we are going to construct the corresponding directed graph @xmath359 as follows . first , construct a vertex for each checker variable of the wheel . for each matching equation @xmath360 , we create a bidirected forced edge @xmath361 with @xmath362 . for each contact variable @xmath363 , we create two corresponding vertices @xmath364 and @xmath365 , which are joined by the bidirected forced edge @xmath366 with @xmath367 . next , we will construct two directed cycles @xmath368 and @xmath369 . note that we are doing arithmetic on the cycle indices here , so the index @xmath370 should be read as equal to @xmath281 . for @xmath368 , for any two consecutive checker vertices @xmath371 on the un - negated side of the bi - wheel , we add a simple directed edge @xmath372 . if the checker @xmath182 is followed by a contact @xmath373 in the cycle , then we add two simple directed edges @xmath374 and @xmath375 . observe that by traversing the simple edges we have just added , the forced matching edges in the direction @xmath376 and the forced contact edges for the un - negated part in the direction @xmath377 we obtain a cycle that covers all checkers and all the contacts of the un - negated part . we now add simple edges to create a second cycle @xmath369 . this cycle will require using the forced matching edges in the opposite direction and , thus , truth assignments will be encoded by the direction of traversal of these edges . first , for any two consecutive checker vertices @xmath378 on the un - negated side of the bi - wheel , we add the simple directed edge @xmath379 . then , if the checker @xmath380 is followed by a contact @xmath381 in the cycle then we add the simple directed edges @xmath382 and @xmath383 . now by traversing the edges we have just added , the forced matching edges in the direction @xmath384 and the forced contact edges for the negated part in the direction @xmath385 , we obtain a cycle that covers all checkers and all the contacts of the negated part , that is , a cycle of direction opposite to @xmath368 . what is left is to encode the equations of size three . again , we have a central vertex @xmath188 that is connected to gadgets simulating equations with three variables . for every equation with three variables , we create the gadget displayed in figure [ fig : eqn3 ] @xmath200 , which is a variant of the gadget used by papadimitriou and vempala @xcite . let us assume that the @xmath155-th equation with three variables in @xmath386 is of the form @xmath387 . this equation is simulated by @xmath388 . the vertices used are the contact vertices @xmath389 , which we have already introduced , as well as the vertices @xmath390 \}$ ] . for notational simplicity , we define @xmath391 , \gamma\in \{x , y , z\ } , { \alpha}\in \{r , l\ } \big\}$ ] . all directed non - forced edges are simple . the vertices @xmath392 and @xmath393 are connected to @xmath188 by forced edges with @xmath394 , where @xmath395 is a small fixed constant . to simplify things , we also force them to be used in the displayed direction by deleting the edges that make up the path of the opposite direction . this is the whole description of the graph @xmath396 . we are going to construct a tour in @xmath396 given an assignment to the variables of @xmath124 and prove the following lemma . [ lem : atsp1 ] given an instance @xmath124 of the hybrid problem with @xmath202 bi - wheels and an assignment that leaves @xmath110 equations in @xmath124 unsatisfied , then , there exists a tour in @xmath396 with cost at most @xmath397 . before we proceed , let us again give a definition for a local edge cost function . let @xmath42 be an edge - weighted digraph and @xmath43 a multi - set of edges of @xmath52 that defines a tour . consider a set @xmath204 . the local edge cost of the set @xmath205 is then defined as @xmath398 in words , for each vertex in @xmath205 we count the total weight of its outgoing edges used in the quasi - tour ( including multiplicities ) . thus , that this sum contains the full weight for edges with their source in @xmath205 , regardless of where their other endpoint is . also note that again for two sets @xmath207 we have @xmath208 ( with equality for disjoint sets ) and that @xmath209 . let @xmath177 be a bi - wheel with variables @xmath399 . given an assignment to the variables of @xmath124 , due to theorem [ thm : hybrid ] , we may assume that either @xmath400 for all @xmath215 $ ] or @xmath401 for all @xmath215 $ ] . we traverse the cycle @xmath368 if @xmath402 and the cycle @xmath369 otherwise . this creates @xmath202 strongly connected components . each contains all the checkers of a bi - wheel and the contacts from one side . for each matching edge gadget , the local edge cost is @xmath30 . we pay two for the forced edge and @xmath281 for the outgoing simple edge . we will account for the cost of edges incident on contacts when we analyze the size - three equation gadget below . let us describe the part of the tour traversing the graph @xmath388 , which simulates @xmath403 . recall that if @xmath171 is set to true in the assignment we have traversed the bi - wheel gadgets in such a way that the forced edge @xmath404 is used , and the simple edge coming out of @xmath312 is used . according to the assignment to @xmath171 , @xmath172 and @xmath223 , we traverse @xmath388 as follows : * case @xmath405 : * let us assume that @xmath406 holds . then , we use @xmath407 . the cost is @xmath408 for the forced edges , @xmath409 for the simple edges inside the gadget , plus @xmath281 for the simple edge going out of @xmath312 . total local edge cost cost : @xmath410 . * case @xmath411 : * then , we use @xmath412 . again we pay @xmath408 for the forced edges , @xmath276 for the simple edges inside the gadget and @xmath30 for the outgoing edges incident on @xmath413 . total local edge cost : @xmath414 . * case @xmath415 : * let us assume that @xmath416 holds . then , we use @xmath417 with total local edge cost @xmath418 . * case @xmath419 : * we use @xmath420 with @xmath418 . the total edge cost of the quasi - tour we constructed is @xmath421 . we have at most @xmath422 strongly connected components : one for each bi - wheel and one containing @xmath188 . a component representing a bi - wheel can be connected to @xmath188 as follows : let @xmath333 be two contact vertices in the component . add one copy of each edge from the cycle @xmath423 . this increases the cost by @xmath424 but decreases the number of components by one . in this section , we are going to prove the other direction of the reduction . [ lem : atsp2 ] if there is a tour with cost @xmath425 , then , there is an assignment that leaves at most @xmath110 equations unsatisfied . given a tour @xmath43 in @xmath396 , we are going to define an assignment to checker and contact variables . as in lemma [ lem : tsp2 ] , we will show that any tour must locally spend on each gadget at least the same amount as the tour we constructed in lemma [ lem : atsp1 ] . if the tour spends more , we use that credit to satisfy possible unsatisfied equations . let us consider the following equations with two variables @xmath426 , @xmath427 , @xmath428 , @xmath429 , @xmath430 and the corresponding situation displayed in figure [ fig : eqn2 ] @xmath200 . since @xmath43 is a valid tour in @xmath396 , we know that @xmath431 is traversed and due to the degree condition , for each @xmath261 , the tour uses another incident edge @xmath166 on @xmath171 with @xmath432 . therefore , we have that @xmath433 . the credit assigned to a gadget is defined as @xmath434 . let us define the assignment for @xmath182 and @xmath183 . a variable @xmath182 is honestly traversed if either both the simple edge going into @xmath182 is used and the simple edge coming out of @xmath183 is used , or neither of these two edges is used . in the first case , we set @xmath182 to @xmath281 , otherwise to @xmath293 . similarly , @xmath183 is honest if both the edge going into @xmath183 and the edge out of @xmath182 are used , and we set it to @xmath281 in the first case and @xmath293 otherwise . * honest tours : * first , suppose that both @xmath182 and @xmath183 are honest . we need to show that the credit is at least as high as the number of unsatisfied equations out of the five equations that contain them . it is not hard to see that if we have set @xmath282 all equations are satisfied . if we have set both to @xmath281 , then the forced edge must be used twice , making the local edge cost at least @xmath409 , giving a credit of @xmath30 , which is more than sufficient . * dishonest tours : * if both @xmath182 and @xmath183 are dishonest the tour must be using the forced edge in both directions . thus , the local cost is @xmath267 or more , giving a credit of @xmath29 . there is always an assignment that satisfies three out of the five equations , so this case is done . if one of them is dishonest , the other must be set to @xmath281 to ensure strong connectivity . thus , there are two simple edges used leaving the gadget , making the local cost @xmath276 ( perhaps the same edge is used twice ) . we can set the honest variable to @xmath281 ( satisfying its two cycle equations ) , and the other to @xmath293 , leaving at most one equation unsatisfied . first , we note that for any valid tour , we have @xmath435 . this is because the two forced edges of weight @xmath436 must be used , and there exist @xmath437 vertices in the gadget for which all outgoing edges have weight @xmath281 . let us define the credit @xmath438 . * honest traversals : * we assume that the underlying tour is honest , that is , forced edges are traversed only in one direction . we set @xmath171 to @xmath281 if the forced edge is used in the direction @xmath404 and @xmath293 otherwise . in the first case we know that the simple edges going into @xmath311 and out of @xmath312 are used . in the second , the edges @xmath439 and @xmath440 are used . we do similarly for @xmath336 . we are interested in the equation @xmath441 and the six cycle equations involving @xmath192 . the assignment we pick for honest variables satisfies the cycle equations , so if it also satisfies the size - three equation we are done . if not , we have to prove that the tour pays at least @xmath442 . * case @xmath443 : * due to our assumption , we know that @xmath444 is a part of the tour . since @xmath43 is a tour , there exists a vertex in @xmath445 that is visited twice and we get @xmath446 . thus , we can spend the credit @xmath447 on the unsatisfied equation @xmath441 . * case @xmath415 : * without loss of generality , let us assume that @xmath416 holds . then , we know that @xmath448 is a part of the tour . but , this implies that there is a vertex in @xmath449 that is visited twice . hence , we have that @xmath450 . * dishonest traversals : * consider the situation , in which some forced edges @xmath451 are traversed in both directions for some variables @xmath227 . for the honest variables , we set them to the appropriate value as before , and this satisfies their cycle equations . observe now that if a forced edge @xmath452 is also used in the opposite direction , then there must be another edge used to leave the set @xmath225 . thus the local edge cost of this set is at least @xmath30 . it follows that the credit we have for the gadget is at least as large as the number of dishonest variables . we can give appropriate values to them so each satisfies one cycle equation and the size - three equation is satisfied . thus , the number of unsatisfied equations is not larger than our credit . in summary , for every tour @xmath43 in @xmath396 , we can find an assignment to the variables of @xmath124 such that all unsatisfied equations are paid by the credit induced by @xmath43 . we are ready to give the proof of theorem [ thm : atspmain ] . we are again given an instance @xmath112 of the with @xmath202 variables and @xmath120 equations . for all @xmath343 , there exists a @xmath110 such that if we repeat each equation @xmath110 time we get an instance @xmath344 with @xmath345 equations and @xmath202 variables such that @xmath453 . then , from @xmath344 , we generate an instance @xmath124 of the hybrid problem and the corresponding directed graph @xmath396 . due to lemmata [ lem : atsp1 ] , [ lem : atsp2 ] and theorem [ thm : hybrid ] , we know that for all @xmath347 , it is -hard to tell whether there is a tour with cost at most @xmath454 or all tours have cost at least @xmath455 , for some @xmath456 depending only on @xmath457 . the ratio between these two cases can get arbitrarily close to @xmath3 by appropriate choices for @xmath458 . in this paper , we proved that it is hard to approximate the atsp and the tsp within any constant factor less than @xmath3 and @xmath2 , respectively . the proof method required essentially new ideas and constructions from the ones used before in that context . since the best known upper bound on the approximability is @xmath459 for atsp and @xmath4 for tsp , there is certainly room for improvements . especially , in the asymmetric version of the tsp , there is a large gap between the approximation lower and upper bound , and it remains a major open problem on the existence of an efficient constant factor approximation algorithm for that problem . furthermore , it would be nice to investigate if some of the ideas of this paper , and in particular the bi - wheel amplifiers , can be used to offer improved hardness results for other optimization problems , such as the steiner tree problem . c. papadimitriou and s. vempala , _ on the approximability of the traveling salesman problem _ , in proc . 32nd acm stoc ( 2000 ) , pp . 126133 , 2000 ; see also a corrected version in combinatorica 26 , pp . 101120 , 2006 .

we present optical echelle spectra of four gamma - ray burst ( grb ) afterglows ( grb 050730 , grb 050820 , grb 051111 , and grb 060418 ) discovered during the first 1.5 years of operation of the _ swift _ satellite and localized by either the _ swift _ telescope or follow - up ground - based imaging . we analyze the spectra to derive accurate column density measurements for the transitions arising in the interstellar medium ( ism ) of the grb host galaxies . these measurements can be used to constrain the physical properties of the ism including the metallicity , dust - to - gas ratio , ionization state , and chemical abundances of the gas . we also present measurements of the strong systems in the grb afterglow spectra . with the publication of this paper , we provide the first data release of echelle afterglow spectra by the graasp collaboration to the general community .

astro-ph0611092

the launch of the _ swift _ satellite @xcite has revolutionized the study of the interstellar medium of gamma - ray burst ( grb ) host galaxies and fostered new studies on the intervening intergalactic medium @xcite . while previous missions alerted the community to the existence of very bright grb afterglows ( e.g. * ? ? ? * ) , only a few events were identified in time to allow echelle observations @xcite . the rapid localization by _ swift _ of the hard x - ray emission to several arcminutes @xcite and the soft x - ray component to a few arcseconds @xcite has enabled observers to obtain high signal - to - noise , high - resolution spectroscopy of grbs with 10m - class telescopes . in this paper , we present a uniform dataset of echelle spectra of four grb afterglows by our grb afterglows as probes ( graasp ) collaboration . among other applications , the data presented here can be used to derive constraints on the physical conditions of the ism of grb host galaxies : the metallicity , dust - to - gas ratio , chemical abundance patterns , ionization state , the distance of the grb to the absorbing gas , etc . future papers in this series will examine these properties in greater detail and provide comparison with analogous observations along quasar sightlines . in addition to the data presented here , we have collected several low - resolution spectra of these and other grbs and will present that data in a companion paper @xcite along with ism abundance measurements obtained by different analysis techniques more suitable to lower resolution data @xcite . we also present measurements of the strong systems in the grb afterglow spectra . in @xmath0 [ sec : obs ] we summarize the observations and data reduction procedures . the methodology is briefly described in @xmath0 [ sec : colm ] and the line - profiles and column density values are given in @xmath0 [ sec : grb ] and @xmath0 [ sec : mgii ] . the grb afterglow spectra presented here were acquired with either the high resolution echelle spectrometer ( hires ; * ? ? ? * ) on the kecki 10 m telescope or the magellan inamori kycoera echelle spectrometer ( mike ; * ? ? ? * ) on the magellan 6.5 m clay telescope at las campanas observatory . table [ tab : obslog ] presents a summary of the grb events , the afterglow observations , instrumental configurations , and data quality of the spectra obtained for the four afterglows for which we have obtained high - resolution spectra : grb 050730 , grb 050820 , grb 051111 , and grb 060418 . the data were reduced with the hires redux and mike reduxburles / www / mike/ ] pipelines @xcite . briefly , the 2d images were bias subtracted , flat - fielded , and wavelength calibrated using standard thar and quartz lamp calibration images . the data were optimally extracted using a non - parametric spatial profile and coadded after weighting by the global snr@xmath1 of each echelle order . a final , normalized 1d spectrum was produced by fitting high - order legendre polynomials to each echelle order ( hires ) or the fluxed 1d spectrum ( mike ) . we provide additional details related to the specific observations in the following section . following our standard practice for the analysis of interstellar lines observed in the damped ly@xmath2 systems ( quasar absorption systems with @xmath3 * ? ? ? * ; * ? ? ? * ) , we have measured column densities using the apparent optical depth method ( aodm ; * ? ? ? * ) . this technique leads to non - parametric , accurate results provided the profiles are resolved . in cases where one measures multiple transitions for a single ion ( e.g. 1741 , 1751 ) , one can test for line saturation by examining the column density values derived as a function of the line - strength . if there is no trend of increasing column density with decreasing line - strength , then the profiles are free of significant ` hidden ' saturation @xcite . we find no indication of hidden saturation in the grb studied here . we report column density values for all transitions detected at greater than @xmath4 statistical significance and report @xmath5 statistical uncertainty in the sections that follow . for ions or atoms where multiple transitions are measured , we report the weighted mean column density and combined statistical error in that mean . while the data may formally yield errors of less than @xmath6dex , we recommend that one adopt a minimum statistical uncertainty of 5% the column density . saturated line - profiles , i.e. , those where the minimum normalized flux of the line - profile is less than the @xmath5 statistical error , are reported as lower limits . finally , undetected or blended transitions are reported as @xmath4 upper limits assuming statistical uncertainty alone . accounting for continuum error , these limits are more likely @xmath7 values . we note that uncertainties in weak lines are likely to be dominated by systematics associated with the continuum placement , and therefore adopt a minimum uncertainty of @xmath8 in our own analysis . the aodm technique can not be applied to transitions which are blended with coincident features . in cases of particular importance , we have derived line - profile fits to the data using the software package vpfit , kindly provided by r. carswell . finally , when the ly@xmath2 profile is covered by the spectroscopic observations , we have fitted this line using the _ x_fitdla _ algorithm within the xidl packagexavier / idl ] ( e.g. * ? ? ? the following sections present velocity profiles of transitions observed along the grb sightlines . the transitions are separated into two groups : those arising from the ground state ( termed resonance lines ) and those arising from excited levels ( ambiguously termed fine - structure lines ) . in all of the figures , line - blends are indicated by dotted lines . in cases where the line - blending is extremely severe , we have not shown the velocity profile . the atomic data used here has been taken from recent laboratory measurements and theoretical calculations when lab measurements are not available : ( see * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? * ; * ? ? ? the _ swift _ bat instrument triggered on grb 050730 at ut 2005 july 30 19:58:23 and shortly after @xcite reported the detection of a bright optical afterglow ( @xmath9 at 199s after the trigger ) at ra=@xmath10 , dec=@xmath11 . after waiting for the sun to set at las campanas observatory , we began acquiring data with the magellan telescope under good conditions . @xcite presented a preliminary analysis of this dataset ; we present the full set of transitions observed . figure [ fig:050730lya ] presents the ly@xmath2 and ly@xmath12profiles for the host galaxy , and the best voigt profile fit which gives @xmath13dex . the uncertainty in this fit is dominated by the fluxing error associated with echelle spectrometers and continuum fitting . a lower resolution , well calibrated , single - order spectrum would likely give a more precise result . figure [ fig:050730res ] shows the velocity profiles for the resonance lines associated with the ism of the host galaxy for grb 050730 . figure [ fig:050730fine ] shows the fine - structure transitions and a resonance line for comparison . finally , table [ tab:050730 ] lists the column density measurements for all of the transitions . the _ swift _ bat instrument triggered on grb 050820 at ut 2005 august 20 06:34:53 and the _ swift _ team issued an xrt position within a few minutes @xcite . @xcite reported a bright ( @xmath14 ) , optical afterglow at ra=@xmath15 , dec=@xmath16 . we began acquiring data with the hires echelle spectrometer on the kecki telescope shortly after . our instrumental configuration followed the standard setup for the uc planet hunters ( e.g. * ? ? ? * ) . the short slit length ( @xmath17 ) complicated sky subtraction although the seeing ( @xmath18 ) was sufficiently good to give fair results . in figure [ fig:050820lya ] , we present the echelle orders covering the ly@xmath2 profile for the host galaxy of grb 050820 . overplotted is a fit to the data which yields @xmath19dex . the uncertainty in this fit is dominated by continuum error which is related to difficulties in fluxing this echelle data . figure [ fig:050820res ] shows the velocity profiles for the resonance lines associated with the ism of the host galaxy . figure [ fig:050820fine ] shows the fine - structure transitions and a resonance line for comparison . finally , table [ tab:050820 ] lists the column density measurements for all of the transitions . the _ swift _ bat instrument triggered on grb 051111 at ut 2005 november 11 05:59:39 @xcite . owing to the moon constraint , xrt observations were delayed until 5000s after the trigger @xcite . @xcite reported an optical afterglow at ra=@xmath20 , dec=@xmath21 . although keck i was scheduled for engineering on this night , the observatory kindly agreed to switch instruments to hires and obtain echelle spectroscopy of grb 050820 @xcite . figure [ fig:051111res ] shows the velocity profiles for the resonance lines associated with the ism of the host galaxy . figure [ fig:051111fine ] shows the fine - structure transitions and a resonance line for comparison . finally , table [ tab:051111 ] lists the column density measurements for all of the transitions . @xcite have also presented an analysis of the same hires data acquired for this sightline . although our central values are generally in good agreement with their results , we report significantly smaller uncertainties , perhaps because of differences in coadding the data . the _ swift _ bat instrument triggered on grb 060418 at ut 2006 april 18 03:06:08 @xcite . the satellite reported an immediate xrt position and @xcite reported a bright , optical afterglow at ra=@xmath22 , dec=@xmath23 . we began observations shortly after this announcement . figure [ fig:060418res ] shows the velocity profiles for the resonance lines associated with the ism of the host galaxy . figure [ fig:060418fine ] shows the fine - structure transitions and a resonance line for comparison . finally , table [ tab:060418 ] lists the column density measurements for all of the transitions . table [ tab : summ ] presents a summary of the gas - phase abundances observed along the four sightlines . in those cases where the @xmath24 value was not measured ( grb 051111 , grb 060418 ) , we have assumed a conservative upper limit of @xmath25 . the relative gas - phase abundances [ x / fe ] and [ x/@xmath2 ] assume solar relative abundances @xcite and do not include ionization corrections nor differential depletion corrections . in a companion paper @xcite , we have shown that there is an enhancement of strong absorbers along grb sightlines in comparison with qso sightlines . here , we present the full set of velocity profiles and column density measurements for the absorbers along the four grb sightlines considered in that paper . in figures [ fig:050730mg1]-[fig:060418mg3 ] we present the absorption systems with @xmath27 . the ionic column densities and equivalent widths are reported in tables [ tab:050730mgii1]-[tab:060418mgii3 ] . we have measured the @xmath28-statistics @xcite for each profile and find @xmath29 in every case . this suggests that a significant fraction @xmath30 ) of the absorbers are damped ly@xmath2 systems . in table [ tab : mgsumm ] we present a summary of these quantities . this includes the minimum @xmath31 value implied by the observed ionic column densities . in general , the values further support the conclusion that the majority of these absorbers are dla systems . the authors wish to recognize and acknowledge the very significant cultural role and reverence that the summit of mauna kea has always had within the indigenous hawaiian community . we are most fortunate to have the opportunity to conduct observations from this mountain . we acknowledge the efforts of s. vogt , g. marcy , j. wright and k. hurley in obtaining the observations of grb 050820 . is partially supported by nasa / swift grant nng05gf55 g . , r. _ et al . _ 2003 , in instrument design and performance for optical / infrared ground - based telescopes . edited by iye , masanori ; moorwood , alan f. m. proceedings of the spie , volume 4841 , pp . 1694 - 1704 ( 2003 ) . , 1694 . , j. p. _ et al . _ 2005 , in uv , x - ray , and gamma - ray space instrumentation for astronomy xiv . edited by siegmund , oswald h. w. proceedings of the spie , volume 5898 , pp . 352 - 359 ( 2005 ) . , o. h. w. siegmund , 352 . lcccccccccc 050730 & 19:58:23 & 3.969 & magellan / clay & mike & 12 & 2005 jul 31 00:00:04 & 2005 jul 31 00:30:04 & 1.23 & 3400 - 9400 & 11 + & & & & & 12 & 2005 jul 31 00:32:52 & 2005 jul 31 01:02:52 & 1.33 & 3400 - 9400 & 10 + & & & & & 12 & 2005 jul 31 01:05:11 & 2005 jul 31 01:35:11 & 1.48 & 3400 - 9400 & 8 + 050820 & 06:34:53 & 2.615 & kecki & hiresr & 7.5 & 2005 aug 20 07:28:59 & 2005 aug 20 07:43:59 & 1.51 & 3800 - 8000 & 14 + & & & & & 7.5 & 2005 aug 20 07:44:42 & 2005 aug 20 07:59:43 & 1.41 & 3800 - 8000 & 12 + 051111 & 05:59:39 & 1.549 & kecki & hiresr & 5 & 2005 aug 11 07:02:58 & 2005 aug 11 07:32:59 & 1.02 & 4160 - 8720 & 22 + & & & & & 5 & 2005 aug 11 07:34:23 & 2005 aug 11 08:04:23 & 1.06 & 4160 - 8720 & 18 + & & & & & 5 & 2005 aug 11 08:05:22 & 2005 aug 11 08:35:23 & 1.12 & 4160 - 8720 & 14 + 060418 & 03:06:08 & 4.048 & magellan / clay & mike & 12 & 2006 apr 18 03:34:02 & 2006 apr 18 03:54:02 & 1.61 & 3400 - 9400 & 21 + & & & & & 12 & 2006 apr 18 04:01:08 & 2006 apr 18 04:31:08 & 1.44 & 3400 - 9400 & 17 + & & & & & 12 & 2006 apr 18 04:54:31 & 2006 apr 18 05:24:31 & 1.24 & 3400 - 9400 & 9 + lccccccccc + & 0 & 0.00 & 1277.2450 & @xmath32&@xmath33$]&@xmath34&@xmath35&@xmath36 + & 0 & 0.00 & 1656.9283 & @xmath37&@xmath38$]&@xmath39&@xmath40 & + + & 1/2 & 0.00 & 1334.5323 & @xmath41&@xmath42$]&@xmath43&@xmath44&@xmath45 + & 3/2 & 63.40 & 1335.7077 & @xmath46&@xmath47$]&@xmath48&@xmath49&@xmath50 + + & & 0.00 & 1548.1950 & @xmath51&@xmath52$]&@xmath53&@xmath54&@xmath55 + & & 0.00 & 1550.7700 & @xmath56&@xmath52$]&@xmath57&@xmath58 & + + & & 0.00 & 1134.1653 & @xmath59&@xmath60$]&@xmath61&@xmath62&@xmath63 + & & 0.00 & 1134.4149 & @xmath64&@xmath60$]&@xmath65&@xmath66 & + + & & 0.00 & 1238.8210 & @xmath67&@xmath68$]&@xmath69&@xmath70&@xmath71 + & & 0.00 & 1242.8040 & @xmath72&@xmath68$]&@xmath73&@xmath74 & + + & 2 & 0.00 & 1302.1685 & @xmath75&@xmath76$]&@xmath77&@xmath78&@xmath79 + & 2 & 0.00 & 1355.5977 & @xmath80&@xmath33$]&@xmath81&@xmath82 & + & 1 & 158.26 & 1304.8576 & @xmath83&@xmath84$]&@xmath85&@xmath86&@xmath87 + & 0 & 226.98 & 1306.0286 & @xmath88&@xmath84$]&@xmath89&@xmath90&@xmath91 + + & & 0.00 & 1239.9253 & @xmath92&@xmath33$]&@xmath93&@xmath94&@xmath95 + & & 0.00 & 1240.3947 & @xmath96&@xmath33$]&@xmath97&@xmath98 & + + & & 0.00 & 1670.7874 & @xmath99&@xmath100$]&@xmath101&@xmath102&@xmath103 + + & 1/2 & 0.00 & 1260.4221 & @xmath104&@xmath105$]&@xmath106&@xmath107&@xmath108 + & 1/2 & 0.00 & 1304.3702 & @xmath109&@xmath110$]&@xmath111&@xmath112 & + & 1/2 & 0.00 & 1526.7066 & @xmath113&@xmath114$]&@xmath115&@xmath116 & + & 1/2 & 0.00 & 1808.0126 & @xmath117&@xmath118$]&@xmath119&@xmath120 & + & 3/2 & 287.24 & 1264.7377 & @xmath121&@xmath122$]&@xmath123&@xmath124&@xmath125 + & 3/2 & 287.24 & 1265.0020 & @xmath126&@xmath127$]&@xmath128&@xmath129 & + & 3/2 & 287.24 & 1309.2757 & @xmath130&@xmath131$]&@xmath132&@xmath133 & + & 3/2 & 287.24 & 1817.4512 & @xmath134&@xmath33$]&@xmath135&@xmath136 & + + & & 0.00 & 1393.7550 & @xmath137&@xmath42$]&@xmath138&@xmath139&@xmath140 + & & 0.00 & 1402.7700 & @xmath141&@xmath42$]&@xmath142&@xmath143 & + + & 1/2 & 0.00 & 1301.8740 & @xmath144&@xmath145$]&@xmath146&@xmath147&@xmath148 + + & & 0.00 & 1250.5840 & @xmath149&@xmath150$]&@xmath151&@xmath152&@xmath153 + & & 0.00 & 1253.8110 & @xmath154&@xmath150$]&@xmath155&@xmath156 & + & & 0.00 & 1259.5190 & @xmath157&@xmath150$]&@xmath158&@xmath159 & + + & & 0.00 & 1066.6600 & @xmath160&@xmath60$]&@xmath161&@xmath162&@xmath163 + + & 9/2 & 0.00 & 1608.4511 & @xmath164&@xmath47$]&@xmath165&@xmath166&@xmath167 + & 9/2 & 0.00 & 1611.2005 & @xmath168&@xmath60$]&@xmath169&@xmath170 & + & 7/2 & 384.79 & 1618.4680 & @xmath171&@xmath60$]&@xmath172&@xmath173&@xmath174 + & 7/2 & 384.79 & 1621.6856 & @xmath175&@xmath60$]&@xmath176&@xmath177 & + & 5/2 & 667.68 & 1629.1596 & @xmath178&@xmath60$]&@xmath179&@xmath180&@xmath181 + & 3/2 & 862.61 & 1634.3498 & @xmath182&@xmath60$]&@xmath183&@xmath184&@xmath185 + & 3/2 & 862.61 & 1636.3313 & @xmath186&@xmath60$]&@xmath187&@xmath188 & + & 1/2 & 977.05 & 1639.4012 & @xmath189&@xmath60$]&@xmath190&@xmath191&@xmath192 + + & & 0.00 & 1317.2170 & @xmath164&@xmath33$]&@xmath193&@xmath194&@xmath195 + & & 0.00 & 1370.1310 & @xmath196&@xmath197$]&@xmath198&@xmath199 & + & & 0.00 & 1741.5490 & @xmath200&@xmath33$]&@xmath201&@xmath202 & + lccccccccc + & 0 & 0.00 & 1277.2450 & @xmath32&@xmath33$]&@xmath203&@xmath204&@xmath205 + & 0 & 0.00 & 1560.3092 & @xmath206&@xmath33$]&@xmath207&@xmath208 & + & 0 & 0.00 & 1656.9283 & @xmath37&@xmath38$]&@xmath209&@xmath210 & + + & 1/2 & 0.00 & 1334.5323 & @xmath41&@xmath211$]&@xmath212&@xmath213&@xmath214 + & 3/2 & 63.40 & 1335.7077 & @xmath46&@xmath215$]&@xmath216&@xmath217&@xmath218 + + & & 0.00 & 1548.1950 & @xmath51&@xmath219$]&@xmath220&@xmath221&@xmath222 + & & 0.00 & 1550.7700 & @xmath56&@xmath219$]&@xmath223&@xmath224 & + + & & 0.00 & 1159.8170 & @xmath225&@xmath226$]&@xmath227&@xmath228&@xmath229 + & & 0.00 & 1160.9370 & @xmath230&@xmath226$]&@xmath146&@xmath231 & + + & 1/2 & 0.00 & 1083.9900 & @xmath232&@xmath233$]&@xmath234&@xmath235&@xmath236 + + & & 0.00 & 1238.8210 & @xmath67&@xmath226$]&@xmath237&@xmath238&@xmath239 + & & 0.00 & 1242.8040 & @xmath72&@xmath68$]&@xmath240&@xmath241 & + + & 2 & 0.00 & 1302.1685 & @xmath75&@xmath242$]&@xmath243&@xmath244&@xmath245 + & 0 & 226.98 & 1306.0286 & @xmath88&@xmath226$]&@xmath246&@xmath247&@xmath248 + + & & 0.00 & 1827.9351 & @xmath249&@xmath33$]&@xmath246&@xmath35&@xmath250 + & & 0.00 & 2026.4768 & @xmath251&@xmath252$]&@xmath207&@xmath253 & + + & & 0.00 & 1239.9253 & @xmath92&@xmath33$]&@xmath254&@xmath255&@xmath256 + & & 0.00 & 1240.3947 & @xmath96&@xmath33$]&@xmath257&@xmath258 & + + & & 0.00 & 1670.7874 & @xmath99&@xmath211$]&@xmath259&@xmath260&@xmath261 + + & & 0.00 & 1854.7164 & @xmath262&@xmath211$]&@xmath263&@xmath264&@xmath265 + & & 0.00 & 1862.7895 & @xmath266&@xmath211$]&@xmath267&@xmath268 & + + & & 0.00 & 1845.5200 & @xmath269&@xmath33$]&@xmath270&@xmath271&@xmath272 + + & 1/2 & 0.00 & 1260.4221 & @xmath104&@xmath242$]&@xmath273&@xmath274&@xmath275 + & 1/2 & 0.00 & 1304.3702 & @xmath109&@xmath242$]&@xmath276&@xmath277 & + & 1/2 & 0.00 & 1526.7066 & @xmath113&@xmath211$]&@xmath278&@xmath279 & + & 3/2 & 287.24 & 1264.7377 & @xmath121&@xmath280$]&@xmath281&@xmath282&@xmath283 + & 3/2 & 287.24 & 1265.0020 & @xmath126&@xmath252$]&@xmath284&@xmath285 & + & 3/2 & 287.24 & 1533.4312 & @xmath286&@xmath287$]&@xmath288&@xmath289 & + & 3/2 & 287.24 & 1817.4512 & @xmath134&@xmath33$]&@xmath246&@xmath290 & + + & & 0.00 & 1892.0300 & @xmath291&@xmath33$]&@xmath292&@xmath293&@xmath294 + + & & 0.00 & 1393.7550 & @xmath137&@xmath295$]&@xmath296&@xmath297&@xmath298 + & & 0.00 & 1402.7700 & @xmath141&@xmath299$]&@xmath300&@xmath301 & + + & 1/2 & 0.00 & 1152.8180 & @xmath302&@xmath303$]&@xmath304&@xmath305&@xmath306 + & 1/2 & 0.00 & 1532.5330 & @xmath307&@xmath308$]&@xmath309&@xmath310 & + + & & 0.00 & 1250.5840 & @xmath149&@xmath311$]&@xmath312&@xmath313&@xmath314 + & & 0.00 & 1253.8110 & @xmath154&@xmath311$]&@xmath315&@xmath316 & + + & & 0.00 & 1066.6600 & @xmath160&@xmath226$]&@xmath317&@xmath90&@xmath91 + + & & 0.00 & 2056.2539 & @xmath318&@xmath226$]&@xmath319&@xmath320&@xmath321 + & & 0.00 & 2062.2340 & @xmath322&@xmath323$]&@xmath324&@xmath325 & + & & 0.00 & 2066.1610 & @xmath326&@xmath33$]&@xmath327&@xmath328 & + + & 9/2 & 0.00 & 1608.4511 & @xmath164&@xmath329$]&@xmath330&@xmath331&@xmath332 + & 9/2 & 0.00 & 1611.2005 & @xmath168&@xmath333$]&@xmath334&@xmath335 & + & 7/2 & 384.79 & 1618.4680 & @xmath171&@xmath60$]&@xmath336&@xmath337&@xmath338 + & 7/2 & 384.79 & 1621.6856 & @xmath175&@xmath60$]&@xmath339&@xmath340 & + & 5/2 & 667.68 & 1629.1596 & @xmath178&@xmath60$]&@xmath246&@xmath341&@xmath342 + & 3/2 & 862.61 & 1634.3498 & @xmath182&@xmath60$]&@xmath336&@xmath343&@xmath344 + & 3/2 & 862.61 & 1636.3313 & @xmath186&@xmath60$]&@xmath284&@xmath345 & + & 1/2 & 977.05 & 1639.4012 & @xmath189&@xmath60$]&@xmath346&@xmath347&@xmath348 + + & & 0.00 & 1370.1310 & @xmath196&@xmath349$]&@xmath350&@xmath351&@xmath352 + & & 0.00 & 1703.4050 & @xmath353&@xmath226$]&@xmath354&@xmath355 & + & & 0.00 & 1709.6042 & @xmath356&@xmath226$]&@xmath357&@xmath358 & + & & 0.00 & 1741.5490 & @xmath200&@xmath33$]&@xmath359&@xmath360 & + & & 0.00 & 1751.9157 & @xmath361&@xmath226$]&@xmath362&@xmath363 & + + & & 0.00 & 2026.1360 & @xmath364&@xmath226$]&@xmath365&@xmath366&@xmath367 + & & 0.00 & 2062.6640 & @xmath368&@xmath38$]&@xmath369&@xmath370 & + lccccccccc + & 0 & 0.00 & 1656.9283 & @xmath37&@xmath33$]&@xmath371&@xmath372&@xmath373 + + & 1/2 & 0.00 & 2325.4030 & @xmath374&@xmath33$]&@xmath375&@xmath376&@xmath377 + + & & 0.00 & 1747.7937 & @xmath378&@xmath33$]&@xmath379&@xmath380&@xmath381 + & & 0.00 & 1827.9351 & @xmath249&@xmath33$]&@xmath382&@xmath383 & + & & 0.00 & 2026.4768 & @xmath251&@xmath33$]&@xmath384&@xmath385 & + + & & 0.00 & 2796.3520 & @xmath386&@xmath387$]&@xmath388&@xmath389&@xmath390 + & & 0.00 & 2803.5310 & @xmath391&@xmath387$]&@xmath392&@xmath393 & + + & & 0.00 & 1670.7874 & @xmath99&@xmath394$]&@xmath395&@xmath396&@xmath397 + + & & 0.00 & 1854.7164 & @xmath262&@xmath398$]&@xmath399&@xmath400&@xmath401 + & & 0.00 & 1862.7895 & @xmath266&@xmath398$]&@xmath402&@xmath403 & + + & & 0.00 & 1845.5200 & @xmath269&@xmath33$]&@xmath404&@xmath405&@xmath406 + & & 0.00 & 2515.0730 & @xmath407&@xmath33$]&@xmath209&@xmath408 & + + & 1/2 & 0.00 & 1808.0130 & @xmath117&@xmath122$]&@xmath409&@xmath410&@xmath411 + & 1/2 & 0.00 & 2335.1230 & @xmath412&@xmath33$]&@xmath284&@xmath413 & + & 3/2 & 287.24 & 1816.9285 & @xmath414&@xmath33$]&@xmath415&@xmath416&@xmath417 + + & & 0.00 & 1892.0300 & @xmath291&@xmath33$]&@xmath418&@xmath419&@xmath420 + + & 1/2 & 0.00 & 1910.6000 & @xmath421&@xmath33$]&@xmath422&@xmath405&@xmath406 + + & & 0.00 & 2056.2539 & @xmath318&@xmath33$]&@xmath423&@xmath424&@xmath425 + & & 0.00 & 2062.2340 & @xmath322&@xmath33$]&@xmath426&@xmath427 & + & & 0.00 & 2066.1610 & @xmath326&@xmath33$]&@xmath428&@xmath429 & + + & & 0.00 & 2576.8770 & @xmath430&@xmath33$]&@xmath431&@xmath432&@xmath433 + & & 0.00 & 2594.4990 & @xmath434&@xmath33$]&@xmath435&@xmath436 & + & & 0.00 & 2606.4620 & @xmath437&@xmath33$]&@xmath438&@xmath436 & + + & & 0.00 & 2484.0210 & @xmath439&@xmath33$]&@xmath440&@xmath441&@xmath442 + + & 9/2 & 0.00 & 2249.8768 & @xmath443&@xmath122$]&@xmath444&@xmath445&@xmath446 + & 9/2 & 0.00 & 2260.7805 & @xmath447&@xmath448$]&@xmath449&@xmath450 & + & 9/2 & 0.00 & 2344.2140 & @xmath451&@xmath452$]&@xmath453&@xmath454 & + & 9/2 & 0.00 & 2374.4612 & @xmath455&@xmath394$]&@xmath456&@xmath457 & + & 9/2 & 0.00 & 2382.7650 & @xmath458&@xmath452$]&@xmath459&@xmath460 & + & 9/2 & 0.00 & 2586.6499 & @xmath461&@xmath462$]&@xmath463&@xmath464 & + & 9/2 & 0.00 & 2600.1729 & @xmath465&@xmath466$]&@xmath467&@xmath468 & + & 7/2 & 384.79 & 2333.5160 & @xmath469&@xmath33$]&@xmath470&@xmath471&@xmath472 + & 7/2 & 384.79 & 2365.5520 & @xmath473&@xmath33$]&@xmath474&@xmath475 & + & 7/2 & 384.79 & 2383.7884 & @xmath476&@xmath33$]&@xmath477&@xmath478 & + & 7/2 & 384.79 & 2389.3582 & @xmath479&@xmath145$]&@xmath480&@xmath107 & + & 7/2 & 384.79 & 2396.3559 & @xmath481&@xmath33$]&@xmath482&@xmath483 & + & 7/2 & 384.79 & 2626.4511 & @xmath484&@xmath33$]&@xmath485&@xmath486 & + & 5/2 & 667.68 & 2328.1110 & @xmath487&@xmath33$]&@xmath488&@xmath489&@xmath490 + & 5/2 & 667.68 & 2607.8664 & @xmath491&@xmath33$]&@xmath492&@xmath489 & + & 5/2 & 667.68 & 2618.3991 & @xmath493&@xmath33$]&@xmath494&@xmath489 & + & 3/2 & 862.61 & 2338.7250 & @xmath495&@xmath33$]&@xmath496&@xmath497&@xmath498 + & 3/2 & 862.61 & 2359.8280 & @xmath499&@xmath33$]&@xmath500&@xmath501 & + & 3/2 & 862.61 & 2621.1912 & @xmath502&@xmath33$]&@xmath503&@xmath504 & + & 1/2 & 977.05 & 2345.0010 & @xmath505&@xmath506$]&@xmath507&@xmath508&@xmath509 + & 1/2 & 977.05 & 2414.0450 & @xmath510&@xmath33$]&@xmath511&@xmath512 & + & 1/2 & 977.05 & 2622.4518 & @xmath513&@xmath33$]&@xmath514&@xmath515 & + & 1/2 & 977.05 & 2629.0777 & @xmath516&@xmath33$]&@xmath517&@xmath518 & + + & & 0.00 & 1941.2852 & @xmath519&@xmath33$]&@xmath520&@xmath521&@xmath522 + + & & 0.00 & 1703.4050 & @xmath353&@xmath33$]&@xmath146&@xmath523&@xmath524 + & & 0.00 & 1709.6042 & @xmath356&@xmath33$]&@xmath525&@xmath526 & + & & 0.00 & 1741.5531 & @xmath200&@xmath33$]&@xmath527&@xmath528 & + & & 0.00 & 1751.9157 & @xmath361&@xmath33$]&@xmath529&@xmath530 & + + & & 0.00 & 2026.1360 & @xmath364&@xmath33$]&@xmath531&@xmath532&@xmath533 + & & 0.00 & 2062.6640 & @xmath368&@xmath33$]&@xmath534&@xmath535 & + lccccccccc + & 0 & 0.00 & 1656.9283 & @xmath37&@xmath226$]&@xmath536&@xmath537&@xmath538 + + & 1/2 & 0.00 & 2325.4030 & @xmath374&@xmath226$]&@xmath539&@xmath540&@xmath541 + + & & 0.00 & 1548.1950 & @xmath51&@xmath542$]&@xmath543&@xmath544&@xmath545 + & & 0.00 & 1550.7700 & @xmath56&@xmath542$]&@xmath546&@xmath221 & + + & & 0.00 & 1747.7937 & @xmath378&@xmath33$]&@xmath547&@xmath548&@xmath549 + & & 0.00 & 1827.9351 & @xmath249&@xmath550$]&@xmath551&@xmath552 & + & & 0.00 & 2026.4768 & @xmath251&@xmath553$]&@xmath554&@xmath555 & + & & 0.00 & 2852.9642 & @xmath556&@xmath557$]&@xmath558&@xmath559 & + + & & 0.00 & 2796.3520 & @xmath386&@xmath466$]&@xmath560&@xmath561&@xmath562 + & & 0.00 & 2803.5310 & @xmath391&@xmath466$]&@xmath563&@xmath564 & + + & & 0.00 & 1670.7874 & @xmath99&@xmath394$]&@xmath565&@xmath566&@xmath567 + + & & 0.00 & 1854.7164 & @xmath262&@xmath226$]&@xmath568&@xmath569&@xmath570 + & & 0.00 & 1862.7895 & @xmath266&@xmath226$]&@xmath571&@xmath572 & + + & & 0.00 & 1845.5200 & @xmath269&@xmath33$]&@xmath573&@xmath271&@xmath272 + & & 0.00 & 2515.0730 & @xmath407&@xmath33$]&@xmath574&@xmath575 & + + & 1/2 & 0.00 & 1526.7066 & @xmath113&@xmath542$]&@xmath576&@xmath464&@xmath577 + & 1/2 & 0.00 & 1808.0130 & @xmath117&@xmath226$]&@xmath578&@xmath579 & + & 1/2 & 0.00 & 2335.1230 & @xmath412&@xmath33$]&@xmath193&@xmath580 & + & 3/2 & 287.24 & 1533.4312 & @xmath286&@xmath323$]&@xmath581&@xmath582&@xmath583 + & 3/2 & 287.24 & 1816.9285 & @xmath414&@xmath33$]&@xmath547&@xmath584 & + + & & 0.00 & 1892.0300 & @xmath291&@xmath33$]&@xmath585&@xmath586&@xmath587 + + & 1/2 & 0.00 & 1910.6000 & @xmath421&@xmath33$]&@xmath503&@xmath405&@xmath406 + + & & 0.00 & 2056.2539 & @xmath318&@xmath226$]&@xmath588&@xmath589&@xmath590 + & & 0.00 & 2062.2340 & @xmath322&@xmath226$]&@xmath591&@xmath592 & + & & 0.00 & 2066.1610 & @xmath326&@xmath33$]&@xmath593&@xmath594 & + + & & 0.00 & 2576.8770 & @xmath430&@xmath226$]&@xmath595&@xmath596&@xmath597 + & & 0.00 & 2594.4990 & @xmath434&@xmath226$]&@xmath598&@xmath599 & + & & 0.00 & 2606.4620 & @xmath437&@xmath226$]&@xmath600&@xmath601 & + + & & 0.00 & 1941.2852 & @xmath519&@xmath33$]&@xmath602&@xmath603&@xmath604 + + & & 0.00 & 2484.0210 & @xmath439&@xmath33$]&@xmath605&@xmath606&@xmath607 + + & 9/2 & 0.00 & 1608.4511 & @xmath164&@xmath608$]&@xmath609&@xmath610&@xmath611 + & 9/2 & 0.00 & 1611.2005 & @xmath168&@xmath226$]&@xmath612&@xmath613 & + & 9/2 & 0.00 & 2249.8768 & @xmath443&@xmath226$]&@xmath614&@xmath615 & + & 9/2 & 0.00 & 2260.7805 & @xmath447&@xmath226$]&@xmath616&@xmath617 & + & 9/2 & 0.00 & 2344.2140 & @xmath451&@xmath618$]&@xmath619&@xmath274 & + & 9/2 & 0.00 & 2374.4612 & @xmath455&@xmath557$]&@xmath620&@xmath621 & + & 9/2 & 0.00 & 2382.7650 & @xmath458&@xmath557$]&@xmath622&@xmath623 & + & 9/2 & 0.00 & 2586.6499 & @xmath461&@xmath624$]&@xmath625&@xmath626 & + & 9/2 & 0.00 & 2600.1729 & @xmath465&@xmath627$]&@xmath628&@xmath162 & + & 7/2 & 384.79 & 2333.5160 & @xmath469&@xmath629$]&@xmath630&@xmath631&@xmath632 + & 7/2 & 384.79 & 2383.7884 & @xmath476&@xmath33$]&@xmath633&@xmath634 & + & 7/2 & 384.79 & 2389.3582 & @xmath479&@xmath226$]&@xmath635&@xmath636 & + & 7/2 & 384.79 & 2396.3559 & @xmath481&@xmath637$]&@xmath638&@xmath639 & + & 7/2 & 384.79 & 2626.4511 & @xmath484&@xmath197$]&@xmath640&@xmath641 & + & 5/2 & 667.68 & 2328.1110 & @xmath487&@xmath197$]&@xmath642&@xmath643&@xmath644 + & 5/2 & 667.68 & 2349.0220 & @xmath645&@xmath308$]&@xmath646&@xmath647 & + & 5/2 & 667.68 & 2399.9728 & @xmath648&@xmath197$]&@xmath649&@xmath650 & + & 5/2 & 667.68 & 2607.8664 & @xmath491&@xmath33$]&@xmath651&@xmath497 & + & 5/2 & 667.68 & 2618.3991 & @xmath493&@xmath33$]&@xmath652&@xmath653 & + & 3/2 & 862.61 & 2338.7250 & @xmath495&@xmath127$]&@xmath654&@xmath655&@xmath321 + & 3/2 & 862.61 & 2359.8280 & @xmath499&@xmath33$]&@xmath656&@xmath657 & + & 3/2 & 862.61 & 2411.2533 & @xmath658&@xmath33$]&@xmath659&@xmath660 & + & 3/2 & 862.61 & 2621.1912 & @xmath502&@xmath33$]&@xmath661&@xmath662 & + & 1/2 & 977.05 & 2345.0010 & @xmath505&@xmath506$]&@xmath663&@xmath664&@xmath665 + & 1/2 & 977.05 & 2414.0450 & @xmath510&@xmath33$]&@xmath666&@xmath667 & + & 1/2 & 977.05 & 2622.4518 & @xmath513&@xmath33$]&@xmath668&@xmath669 & + & 1/2 & 977.05 & 2629.0777 & @xmath516&@xmath197$]&@xmath670&@xmath671 & + + & & 0.00 & 1703.4050 & @xmath353&@xmath33$]&@xmath672&@xmath310&@xmath673 + & & 0.00 & 1709.6042 & @xmath356&@xmath33$]&@xmath674&@xmath675 & + & & 0.00 & 1741.5531 & @xmath200&@xmath226$]&@xmath676&@xmath677 & + & & 0.00 & 1751.9157 & @xmath361&@xmath33$]&@xmath678&@xmath679 & + + & & 0.00 & 2026.1360 & @xmath364&@xmath226$]&@xmath680&@xmath681&@xmath682 + & & 0.00 & 2062.6640 & @xmath368&@xmath683$]&@xmath684&@xmath685 & + lcccc @xmath686 & 3.9686 & 2.6147 & 1.5495 & 1.4900 + @xmath687 & @xmath688&@xmath689&@xmath690&@xmath690 + @xmath691m / h]@xmath692&@xmath693&@xmath694&@xmath695&@xmath696 + @xmath697&15.15&14.90&15.36&15.26 + @xmath691c / fe ] & @xmath698&@xmath699&@xmath700&@xmath700 + @xmath691n / fe ] & @xmath701&@xmath702&@xmath700&@xmath700 + @xmath691o / fe ] & @xmath703&@xmath704&@xmath700&@xmath700 + @xmath691mg / fe ] & @xmath705&@xmath706&@xmath707&@xmath708 + @xmath691al / fe ] & @xmath709&@xmath710&@xmath711&@xmath712 + @xmath691si / fe ] & @xmath713&@xmath714&@xmath715&@xmath716 + @xmath691p / fe ] & @xmath717&@xmath718&@xmath700&@xmath700 + @xmath691s / fe ] & @xmath719&@xmath720&@xmath700&@xmath700 + @xmath691ti / fe ] & @xmath700&@xmath700&@xmath721&@xmath722 + @xmath691cr / fe ] & @xmath700&@xmath723&@xmath724&@xmath725 + @xmath691mn / fe ] & @xmath700&@xmath700&@xmath726&@xmath727 + @xmath691co / fe ] & @xmath700&@xmath700&@xmath728&@xmath729 + @xmath691ni / fe ] & @xmath730&@xmath731&@xmath732&@xmath733 + @xmath691zn / fe ] & @xmath700&@xmath734&@xmath735&@xmath736 + @xmath737&@xmath87&@xmath738&@xmath739 + @xmath740&@xmath91&@xmath248&@xmath738 + @xmath741&@xmath125&@xmath283&@xmath417&@xmath583 + @xmath742&@xmath174&@xmath338&@xmath472&@xmath632 + @xmath743&@xmath181&@xmath342&@xmath490&@xmath644 + @xmath744&@xmath185&@xmath344&@xmath498&@xmath321 + @xmath745&@xmath192&@xmath348&@xmath509&@xmath665 + @xmath746&@xmath739&@xmath250&@xmath381&@xmath549 + lcccccc + & 2852.9642 & @xmath556&@xmath747$]&@xmath748&@xmath749&@xmath750 + + & 2796.3520 & @xmath386&@xmath747$]&@xmath751&@xmath752&@xmath753 + & 2803.5310 & @xmath391&@xmath747$]&@xmath754&@xmath753 & + + & 2260.7805 & @xmath447&@xmath747$]&@xmath755&@xmath756&@xmath757 + & 2344.2140 & @xmath451&@xmath747$]&@xmath758&@xmath759 & + & 2374.4612 & @xmath455&@xmath747$]&@xmath760&@xmath761 & + & 2382.7650 & @xmath458&@xmath747$]&@xmath762&@xmath763 & + & 2586.6500 & @xmath461&@xmath747$]&@xmath764&@xmath765 & + & 2600.1729 & @xmath465&@xmath747$]&@xmath766&@xmath767 & + lcccccc + & 2852.9642 & @xmath556&@xmath768$]&@xmath769&@xmath770&@xmath770 + + & 2796.3520 & @xmath386&@xmath768$]&@xmath771&@xmath772&@xmath773 + & 2803.5310 & @xmath391&@xmath774$]&@xmath775&@xmath773 & + + & 2260.7805 & @xmath447&@xmath776$]&@xmath777&@xmath778&@xmath779 + & 2374.4612 & @xmath455&@xmath776$]&@xmath780&@xmath779 & + & 2382.7650 & @xmath458&@xmath776$]&@xmath781&@xmath782 & + & 2586.6500 & @xmath461&@xmath776$]&@xmath783&@xmath163 & + & 2600.1729 & @xmath465&@xmath784$]&@xmath785&@xmath786 & + lcccccc + & 2796.3520 & @xmath386&@xmath787$]&@xmath788&@xmath789&@xmath790 + & 2803.5310 & @xmath391&@xmath787$]&@xmath791&@xmath790 & + + & 3934.7770 & @xmath792&@xmath793$]&@xmath794&@xmath795&@xmath796 + & 3969.5910 & @xmath797&@xmath793$]&@xmath798&@xmath799 & + + & 2344.2140 & @xmath451&@xmath800$]&@xmath801&@xmath802&@xmath802 + & 2382.7650 & @xmath458&@xmath787$]&@xmath803&@xmath804 & + lcccccc + & 2852.9642 & @xmath556&@xmath329$]&@xmath805&@xmath806&@xmath806 + + & 2796.3520 & @xmath386&@xmath807$]&@xmath808&@xmath809&@xmath810 + & 2803.5310 & @xmath391&@xmath811$]&@xmath812&@xmath810 & + + & 2260.7805 & @xmath447&@xmath466$]&@xmath813&@xmath814&@xmath815 + & 2344.2140 & @xmath451&@xmath329$]&@xmath816&@xmath817 & + & 2374.4612 & @xmath455&@xmath329$]&@xmath818&@xmath815 & + & 2382.7650 & @xmath458&@xmath329$]&@xmath819&@xmath820 & + & 2586.6500 & @xmath461&@xmath329$]&@xmath821&@xmath822 & + & 2600.1729 & @xmath465&@xmath329$]&@xmath823&@xmath824 & + lcccccc + & 2852.9652 & @xmath556&@xmath825$]&@xmath826&@xmath827&@xmath827 + + & 2796.3520 & @xmath386&@xmath825$]&@xmath828&@xmath829&@xmath830 + & 2803.5310 & @xmath391&@xmath825$]&@xmath831&@xmath830 & + + & 3242.9290 & @xmath832&@xmath833$]&@xmath834&@xmath835&@xmath836 + + & 2260.7805 & @xmath447&@xmath825$]&@xmath837&@xmath778&@xmath838 + & 2344.2140 & @xmath451&@xmath825$]&@xmath839&@xmath840 & + & 2374.4612 & @xmath455&@xmath825$]&@xmath841&@xmath838 & + & 2382.7650 & @xmath458&@xmath825$]&@xmath842&@xmath822 & + & 2600.1729 & @xmath465&@xmath825$]&@xmath843&@xmath844 & + + & 2026.1360 & @xmath364&@xmath833$]&@xmath845&@xmath846&@xmath846 + lcccccc + & 2852.9652 & @xmath556&@xmath847$]&@xmath848&@xmath849&@xmath849 + + & 2796.3520 & @xmath386&@xmath847$]&@xmath850&@xmath824&@xmath851 + & 2803.5310 & @xmath391&@xmath847$]&@xmath852&@xmath851 & + + & 2576.8770 & @xmath430&@xmath853$]&@xmath854&@xmath855&@xmath855 + + & 2260.7805 & @xmath447&@xmath847$]&@xmath856&@xmath857&@xmath857 + & 2344.2140 & @xmath451&@xmath847$]&@xmath858&@xmath859 & + & 2374.4612 & @xmath455&@xmath847$]&@xmath860&@xmath861 & + & 2382.7650 & @xmath458&@xmath847$]&@xmath862&@xmath863 & + & 2586.6499 & @xmath461&@xmath847$]&@xmath864&@xmath865 & + & 2600.1729 & @xmath465&@xmath847$]&@xmath866&@xmath804 & + lcccccc + & 2852.9652 & @xmath556&@xmath867$]&@xmath868&@xmath869&@xmath869 + + & 2796.3520 & @xmath386&@xmath867$]&@xmath870&@xmath871&@xmath820 + & 2803.5310 & @xmath391&@xmath867$]&@xmath872&@xmath820 & + + & 2260.7805 & @xmath447&@xmath867$]&@xmath873&@xmath874&@xmath875 + & 2344.2140 & @xmath451&@xmath867$]&@xmath876&@xmath877 & + & 2374.4612 & @xmath455&@xmath867$]&@xmath878&@xmath879 & + & 2382.7650 & @xmath458&@xmath867$]&@xmath880&@xmath881 & + & 2586.6499 & @xmath461&@xmath867$]&@xmath882&@xmath883 & + & 2600.1729 & @xmath465&@xmath867$]&@xmath884&@xmath885 & + lccccccccc + & & 0.00 & 2852.9642 & @xmath556&@xmath886$]&@xmath887&@xmath888&@xmath889 + + & & 0.00 & 2796.3520 & @xmath386&@xmath890$]&@xmath891&@xmath260&@xmath892 + & & 0.00 & 2803.5310 & @xmath391&@xmath890$]&@xmath893&@xmath894 & + + & & 0.00 & 1670.7874 & @xmath99&@xmath895$]&@xmath896&@xmath897&@xmath898 + + & & 0.00 & 1854.7164 & @xmath262&@xmath899$]&@xmath900&@xmath901&@xmath902 + & & 0.00 & 1862.7895 & @xmath266&@xmath899$]&@xmath903&@xmath904 & + + & 1/2 & 0.00 & 1808.0130 & @xmath117&@xmath905$]&@xmath906&@xmath907&@xmath908 + + & & 0.00 & 3934.7770 & @xmath792&@xmath226$]&@xmath909&@xmath910&@xmath911 + + & 1/2 & 0.00 & 3242.9290 & @xmath832&@xmath226$]&@xmath912&@xmath913&@xmath914 + & 1/2 & 0.00 & 3384.7400 & @xmath915&@xmath226$]&@xmath916&@xmath917 & + + & & 0.00 & 2056.2539 & @xmath318&@xmath226$]&@xmath918&@xmath919&@xmath920 + + & & 0.00 & 2576.8770 & @xmath430&@xmath226$]&@xmath921&@xmath922&@xmath923 + & & 0.00 & 2594.4990 & @xmath434&@xmath226$]&@xmath924&@xmath925 & + + & 9/2 & 0.00 & 2260.7805 & @xmath447&@xmath226$]&@xmath926&@xmath927&@xmath928 + & 9/2 & 0.00 & 2344.2140 & @xmath451&@xmath929$]&@xmath930&@xmath129 & + & 9/2 & 0.00 & 2374.4612 & @xmath455&@xmath929$]&@xmath931&@xmath932 & + & 9/2 & 0.00 & 2382.7650 & @xmath458&@xmath929$]&@xmath933&@xmath934 & + & 9/2 & 0.00 & 2600.1729 & @xmath465&@xmath929$]&@xmath935&@xmath623 & + + & & 0.00 & 1741.5531 & @xmath200&@xmath226$]&@xmath936&@xmath937&@xmath938 + + & & 0.00 & 2026.1360 & @xmath364&@xmath226$]&@xmath939&@xmath370&@xmath940 + & & 0.00 & 2062.6640 & @xmath368&@xmath941$]&@xmath942&@xmath943 & + lccccccc 050730 & 1.773 & 943.6&@xmath757&@xmath739&@xmath750&18.5&7.8 + 050730 & 2.253 & 886.4&@xmath779&@xmath739&@xmath770&19.0&7.2 + 050820 & 0.692 & 2991.9&@xmath802&@xmath738&19.1&7.8 + 050820 & 1.430 & 1935.7&@xmath815&@xmath739&@xmath806&18.8&8.1 + 051111 & 1.189 & 2027.2&@xmath838&@xmath846&@xmath827&19.0&8.5 + 060418 & 0.603 & 1270.4&@xmath857&@xmath739&@xmath849&20.2&8.2 + 060418 & 0.656 & 972.5&@xmath875&@xmath739&@xmath869&18.4&7.2 + 060418 & 1.107 & 1841.4&@xmath928&@xmath940&@xmath889&20.3&7.5 +

precise calculations of core properties in heavy - atom systems which are described by the operators heavily concentrated in atomic cores , like to hyperfine structure and p , t - parity nonconservation effects , usually require accounting for relativistic effects . unfortunately , completely relativistic treatment of molecules containing heavy elements is very consuming already at the stages of calculation and transformation of two - electron integrals with a basis set of four - component spinors . in turn , the relativistic effective core potential ( recp ) calculations of valence ( spectroscopic , chemical etc . ) properties of molecules are very popular because the recp method allows one to treat quite satisfactory the correlation and relativistic effects for the valence electrons of a molecule and to reduce significantly the computational efforts . the valence molecular spinors are usually smoothed in atomic cores and , as a result , direct calculation of electronic densities near heavy nuclei is impossible . in the paper , the methods of nonvariational and variational one - center restoration of correct shapes of four - component spinors in atomic cores after a two - component recp calculation of a molecule are discussed . their efficiency is illustrated in correlation calculations of hyperfine structure and parity nonconservation effects in heavy - atom molecules ybf , baf , tlf , and pbo . * short name : * studying core properties in molecules * keywords for indexing : * electronic structure , atom in a molecule , molecules with heavy atoms , method of ab initio molecular calculation , relativistic effective core potential , one - center restoration .

physics0409071

[ [ spthist ] ] historical background of study p- and t - parity nonconservation effects in heavy - atom molecules . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + after discovery of the combined charge ( c ) and space ( p ) parity violation , or cp - violation , in @xmath0-meson decay @xcite , the search for the electric dipole moments ( edms ) of elementary particles has become one of the most fundamental problems in physics @xcite . permanent edm is induced by the weak interaction that breaks both the space symmetry inversion and time - reversal invariance ( t ) @xcite . considerable experimental effort has been invested in measuring atomic edms induced by the edms of proton , neutron and electron and by p , t - odd interactions between them . the best available restriction for the electron edm , @xmath1 , was obtained in the atomic tl experiment @xcite , which established an upper limit of @xmath2 cm , where @xmath3 is the charge of the electron . the benchmark upper limit on a nucleus edm is obtained in atomic experiment on @xmath4hg @xcite , @xmath5 cm , from which the best restriction on the proton edm , @xmath6 cm , was also recently obtained by dmitriev & senkov @xcite . since 1967 , when sandars suggested to use polar heavy - atom molecules in the experimental search for the proton edm @xcite , the molecules are considered as the most promising objects for such experiments . sandars also noticed earlier @xcite that the p- and p , t - parity nonconservation ( pnc ) effects are strongly enhanced in heavy - atom systems due to relativistic and other effects . for example , in paramagnetic atoms the enhancement factor for an electron edm , @xmath7 , is roughly proportional to @xmath8 , where @xmath9 is the fine structure constant , @xmath10 is the nuclear charge and @xmath11 is the atomic polarisability . it can be of order 100 or greater for highly polarizable heavy atoms ( @xmath12 ) . furthermore , the effective inner molecular electric field @xmath13 acting on electrons in polar molecules can be a few orders of magnitude higher than the maximal field @xmath14 accessible in a laboratory , @xmath15 . the first molecular edm experiment ( on search for the proton edm and other nuclear p , t - odd effects ) was performed on tlf by sandars _ _ @xcite ( oxford , uk ) . in 1991 , in the last series of the @xmath16tlf experiments by hinds _ et al . _ @xcite ( yale , usa ) , the restriction @xmath17 was obtained ( that was recalculated in 2002 by petrov _ et al . _ @xcite as @xmath18 ) . in 1978 the experimental investigation of the electron edm and other pnc effects was further stimulated by labzowsky _ _ @xcite and sushkov & flambaum @xcite who clarified the possibilities of additional enhancement of these effects in diatomic radicals like bis and pbf due to the closeness of levels of opposite parity in @xmath19-doublets having the @xmath20 ground state . then sushkov _ et al . _ @xcite and flambaum & khriplovich @xcite have suggested to use @xmath19-doubling in diatomic radicals with the @xmath21 ground state for such experiments and the hgf , hgh and baf molecules were first studied semiempirically by kozlov @xcite . at the same time , the first two - step _ ab initio _ calculation of pnc effects in pbf initiated by labzowsky was finished in st .- petersburg ( russia ) @xcite . a few years later , hinds started experimental search for the electron edm on the ybf molecule , on which the first result was obtained by his group in 2002 ( sussex , uk ) @xcite , @xmath22 . though that restriction is worse than the best available now @xmath1 datum calculated from the tl experiment ( see above ) , nevertheless , it is limited by only counting statistics , as hinds _ _ pointed out in @xcite . new series of the electron edm experiments on ybf by hinds group ( london , uk ) are in progress and new generation of the electron edm experiments using a vapor cell , on the metastable @xmath23 state of pbo , is prepared by group of demille ( yale , usa ) . the unique suitability of pbo for searching the elusive @xmath1 is demonstrated by the very high statistical sensitivity of the yale experiment to the electron edm . in prospect , it allows one to detect @xmath1 of order of @xmath24 cm @xcite , three four orders of magnitude lower than the current limit quoted above . some other candidates for the edm experiments , hgh , hgf , and teo@xmath25 , are yet discussed and the experiment on pbf is planned in oklahoma , usa . in order to interpret the data measured in such molecular experiments in terms of the electron edm or other fundamental constants of p- and p , t - odd interactions , high - precision calculation of the electronic wave function near a heavy nucleus is required . moreover , _ ab initio _ calculations of some molecular properties are usually required already prior to the stage of preparation of the experimental setup . [ [ shamcalc ] ] heavy - atom molecules , computational strategies . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the most straightforward way for calculation of electronic structure of molecules containing heavy atoms is solution of the eigenvalue problem using the dirac - coulomb or dirac - coulomb - breit hamiltonian when some approximation for the four - component wave function is chosen . however , even applying the relativistic scf approximation to heavy - atom molecules when solving the dirac - fock or dirac - fock - breit equations followed by the transformation of the two - electron integrals is not always an easy task because much larger sets of primitive basis functions are required @xcite for such all - electron four - component calculations as compared to the nonrelativistic case . starting from the pauli approximation and foldy & wouthuysen transformation , many different two - component approaches were developed in which only large components are treated explicitly ( e.g. , see and references ) . the most computational savings are yet achieved when the two - component relativistic effective core potential ( recp ) approximation proposed by lee _ et al . _ @xcite is used . there are the following reasons for it . the recp approach allows one to exclude large number of chemically inactive electrons from molecular calculations and to treat explicitly the valence electrons only . then , the oscillations of the valence spinors are usually smoothed in heavy - atom cores . as a result , the number of primitive basis functions can be reduced dramatically . this approach is based on a well - developed nonrelativistic technique of calculations , however , an effective spin - orbit interaction and other scalar - relativistic effects are taken into account usually by means of radially - local or separable potentials @xcite . many complications of the dirac - coulomb(-breit ) molecular calculations @xcite are avoided when employing recps . the radially - local recp approaches for `` shape - consistent '' ( or `` norm - conserving '' ) pseudoorbitals are the most widely applied in calculations of molecules with heavy elements though `` energy - adjusted / consistent '' pseudopotentials @xcite by stuttgart team and huzinaga - type `` _ ab initio _ model potentials '' @xcite are also actively used . ( in plane wave calculations of many - atomic systems and in molecular dynamics , the separable pseudopotentials are more popular now because they provide linear scaling of computational effort on the basis set size . ) the nonrelativistic shape - consistent effective core potential has been first proposed by durand & barthelat @xcite and then a modified scheme of the pseudoorbital constructing was suggested by christiansen _ _ @xcite and by hamann _ et al . _ @xcite . however , inaccuracy of the conventional radially - local recp approaches reaches 10003000 @xmath26 for the transition and dissociation energies that may be insufficient in practice . in deep reorganization of electronic configuration structure of molecules containing , in particular , transition metals , lanthanides , actinides , and superheavy elements the conventional radially - local recp approaches can not provide reliable results for a wide variety of properties ( as it was shown both in many calculations and theoretically , see @xcite and references ) though otherwise is sometime stated ( e.g. , see @xcite ) such problems can be efficiently overcomed by applying the generalized recp ( grecp ) approach , that involves both radially - local , separable and huzinaga - type potentials as its components and as particular cases . in the grecp concept , the inner core , outer core and valence electrons are first treated employing different approximations for each . nevertheless , calculation of such properties as electronic densities near nuclei , hyperfine structure constants , p , t - parity nonconservation effects , chemical and isotopic shifts etc . with the help of the two - component pseudospinors smoothed in cores is impossible . for evaluation the matrix elements of the operators singular on nuclei , proper shapes of the valence molecular four - component spinors must be restored in atomic core regions after the recp calculation of that molecule performed . in 1959 , a nonrelativistic procedure of restoration of the orbitals from smoothed phillips kleinman pseudoorbitals was proposed @xcite based on the orthogonalization of the latters to the original atomic core orbitals . in 1985 , pacios & christiansen @xcite suggested a modified orthogonalization scheme in the case of shape - consistent pseudospinors . at the same time , a simple procedure of one - center restoration employing the idea of generation of equivalent basis sets in four - component dirac - fock and two - component recp / scf calculations was proposed in @xcite ( i.e. nocr procedure , see below ) and first applied to evaluation of the p , t - odd spin - rotational hamiltonian parameters in the pbf molecule . in 1994 , similar procedure was used by blchl inside the augmentation regions @xcite in solids to construct the transformation operator between pseudoorbitals ( `` ps '' ) and original orbitals ( `` ae '' ) in his projector augmented - wave method . all the above restoration schemes can be called by `` nonvariational '' as compared to the `` variational '' one - center restoration ( vocr , see below ) procedure proposed in @xcite . proper behavior of the molecular orbitals ( four - component spinors in the relativistic case ) in atomic cores of molecules can be restored in the scope of a variational procedure if the molecular pseudoorbitals ( two - component pseudospinors ) match correctly the original orbitals ( large components of bispinors ) in the valence region after the molecular recp calculation . this condition is rather correct when the shape - consistent recp is involved to the molecular calculation with explicitly treated outermost core orbitals and , especially , when the grecp operator is used as is demonstrated in @xcite . at the restoration stage , a one - center expansion on the spherical harmonics with numerical radial parts is most appropriate both for orbitals ( spinors ) and for the description of `` external '' interactions with respect to the core regions of a considered molecule . in the scope of the discussed two - step methods of the electronic structure calculation of a molecule , finite nucleus models and quantum electrodynamic terms including , in particular , two - electron breit interaction @xcite may be used without problems . one - center expansion had been applied first to whole molecules by desclaux & pyykk in relativistic and nonrelativistic hartree - fock calculations for the series ch@xmath27 to pbh@xmath27 @xcite and then in the dirac - fock calculations of cuh , agh and auh @xcite and other molecules @xcite . a large bond length contraction due to the relativistic effects has been estimated . however , the accuracy of such calculations is limited in practice because the orbitals of the hydrogen atom are reexpanded on a heavy nucleus in all the coordinate space . it is worth to note that the recp and one - center expansion approaches were considered earlier as alternatives to each other @xcite . the applicability of the proposed two - step algorithms for calculation of wave functions of molecules with heavy atoms is a consequence of the fact that the valence and core electrons may be considered as two subsystems , interaction between which is described mainly by some integrated and not by detailed properties of these subsystems . the methods for consequent calculation of the valence and core parts of electronic structure of molecules give us a way to combine relative simplicity and accessibility of both molecular recp calculations in gaussian basis set and relativistic finite - difference one - center calculations inside a sphere with the atomic core radius . the first two - step calculations of the p , t - odd spin - rotational hamiltonian parameters were performed for the pbf radical about 20 years ago @xcite with the semiempirical accounting for the spin - orbit interaction . before , only nonrelativistic scf calculation of the tlf molecule using the relativistic scaling was carried out @xcite , in which the p , t - odd values were almost three times underestimated as compared to the relativistic df calculations . the latter were first performed only in 1997 by laerdahl _ et al . _ @xcite and by parpia @xcite . the next two - step calculation , for pbf and hgf molecules @xcite , was carried out with the spin - orbit recp part taken into account using the method suggested in @xcite . later we performed correlation grecp / nocr calculations of the core properties in ybf @xcite , baf @xcite , again in ybf @xcite and in tlf @xcite . in 1998 , first all - electron dirac - fock calculations of the ybf molecule were also performed by quiney _ _ @xcite and by parpia @xcite . very recently we finished extensive two - step calculations of the p , t - odd properties and hyperfine structure of the excited states of the pbo molecule @xcite . in the paper , the main features of the used two - step method are presented and only the last series of the two - step calculations are discussed , in which electron correlations are taken into account by a combined method of the second - order perturbation theory ( pt2 ) and configuration interaction ( ci ) , or `` pt2/ci '' @xcite ( for baf and ybf ) , by the relativistic coupled cluster ( rcc ) method @xcite ( for tlf and pbo ) , and by the spin - orbit direct - ci method ( for pbo ) . in the discussed _ ab initio _ calculations the best up to - date accuracy was attained for the hyperfine constants and p , t - odd parameters regarding the molecules containing heavy atoms . when core electrons of a heavy - atom molecule do not play an active role , the effective hamiltonian with recp can be presented in the form @xmath28 + \sum_{i_v > j_v } \frac{1}{r_{i_v j_v}}\ . \label{heff2}\ ] ] hamiltonian ( [ heff2 ] ) is written only for a valence subspace of electrons which are treated explicitly and denoted by indices @xmath29 and @xmath30 . in practice , this subspace is often extended by inclusion of some outermost core shells for better accuracy but we will consider them as the valence shells below if these outermost core and valence shells are not treated using different approximations . @xmath31 is an recp ( or relativistic pseudopotential ) operator that is usually written in the separable ( e.g. , see @xcite and references ) or radially - local ( semi - local ) @xcite approximations when the valence pseudospinors are smoothed in the heavy - atom cores . the recp operator simulates , in particular , interactions of the explicitly treated electrons with those which are excluded from the recp calculations . besides , the generalized recp operator @xcite described below can be used that includes the radially - local , separable and huzinaga - type relativistic pseudopotentials as special cases . additionally , the grecp operator can include terms of other types which are important first of all for economical but precise treatment of transition metals , lanthanides and actinides . with these terms , accuracy provided by grecps can be even higher than the accuracy of the frozen core approximation employing the same separation on subspaces of valence and core electrons @xcite . in eq . ( [ heff2 ] ) , @xmath32 is the one - electron schrdinger hamiltonian @xmath33 where @xmath34 is the charge of the nucleus decreased by the number of inner core electrons . contrary to the four - component wave function used in dirac - coulomb(-breit ) calculations , the pseudo - wave function in the ( g)recp case can be both two- and one - component . the use of the effective hamiltonian ( [ heff2 ] ) instead of the all - electron relativistic hamiltonians leads to the question about its accuracy . it was shown both theoretically and in many calculations ( see @xcite and references ) that the usual accuracy of the radially - local recp versions is within 10003000 @xmath26 for transition energies between low - lying states . the grecp concept is introduced and developed in a series of papers ( see @xcite and references ) . in contrast to other recp methods , grecp employs the idea of separating the space around a heavy atom into three regions : inner core , outer core and valence , which are treated differently . it allows one to attain practically any desired accuracy , while requiring moderate computational efforts . the main steps of the scheme of generating the grecp version with the separable correction taken into account are : 1 . the numerical all - electron relativistic calculation of a generator state is carried out for an atom under consideration . for this purpose , we use the atomic hfdb code @xcite . the numerical pseudospinors @xmath35 are constructed of the large components @xmath36 of the outer core ( oc ) and valence ( v ) hfd(b ) spinors so that the innermost pseudospinors of them ( for each @xmath37 and @xmath38 ) are nodeless , the next pseudospinors have one node , and so forth . these pseudospinors satisfy the following conditions : @xmath39 @xmath40 where @xmath41 are principal quantum numbers of the valence and outer core spinors , @xmath42 is one more than the highest orbital angular momentum of the inner core ( ic ) spinors . the leading power @xmath43 in the polynomial is typically chosen to be close to @xmath44 in order to ensure sufficient ejection of the valence and outer core electrons from the inner core region . the @xmath45 coefficients are determined by the following requirements : * @xmath46 set is orthonormalized , * @xmath47 and its first four derivatives match @xmath48 and its derivatives , * @xmath47 is a smooth and nodeless function , and * @xmath49 ensures a sufficiently smooth shape of the corresponding potential . + @xmath50 is chosen near the extremum of the spinor so that the corresponding pseudospinor has the defined above number of nodes . in practice , the @xmath50 radii for the different spinors should be chosen close to each other to generate smooth potentials . the @xmath51 potentials are derived for each @xmath52 and @xmath53 for the valence and outer core pseudospinors so that the @xmath49 are solutions of the nonrelativistic - type hartree - fock equations in the _ jj_-coupling scheme for a `` pseudoatom '' with the removed inner core electrons . @xmath54 , \label{u_nlj}\end{aligned}\ ] ] where @xmath55 and @xmath56 are the coulomb and exchange operators calculated with the @xmath49 pseudospinors , @xmath57 are the one - electron energies of the corresponding spinors , and @xmath58 are off - diagonal lagrange multipliers ( which are , in general , slightly different for the original bispinors and pseudospinors ) . 4 . the grecp operator with the separable correction written in the spinor representation @xcite is as @xmath59 { \bf p}_{lj } \nonumber\\ & + & \sum_{n_{oc } } \bigl[u_{n_{oc},lj}(r ) - u_{n_vlj}(r)\bigr ] \widetilde{\bf p}_{n_{oc},lj } + \sum_{n_{oc } } \widetilde{\bf p}_{n_{oc},lj } \bigl[u_{n_{oc},lj}(r ) - u_{n_vlj}(r)\bigr ] \\ & - & \sum_{n_{oc},n_{oc } ' } \widetilde{\bf p}_{n_{oc},lj } \biggl[\frac{u_{n_{oc},lj}(r)+u_{n_{oc}',lj}(r)}{2 } - u_{n_vlj}(r)\biggr ] \widetilde{\bf p}_{n_{oc}',lj}\bigr\ } , \nonumber\end{aligned}\ ] ] where @xmath60 @xmath61 is the projector on the two - component spin - angular function @xmath62 , @xmath63 is the projector on the outer core pseudospinors @xmath64 , and @xmath65 . two of the major features of the grecp version with the separable correction described here are generating of the effective potential components for the pseudospinors which may have nodes , and addition of non - local separable terms with projectors on the outer core pseudospinors ( the second and third lines in eq . ( [ ugrecp ] ) ) to the standard semi - local recp operator ( the first line in eq . ( [ ugrecp ] ) ) . some other grecp versions are described and discussed in papers @xcite in details . the grecp operator in spinor representation ( [ ugrecp ] ) is mainly used in our atomic calculations . the spin - orbit representation of this operator which can be found in @xcite is more efficient in practice being applied to molecular calculations . despite the complexity of expression ( [ ugrecp ] ) for the grecp operator , the calculation of its one - electron integrals is not notably more expensive than that for the case of the conventional radially - local recp operator . the electronic density evaluated from the two - component pseudo - wave function ( that , in turn , is obtained in the ( g)recp calculation accounting for correlation ) very accurately reproduces the corresponding all - electron four - component density in the valence and outer core regions . in the inner core region , the pseudospinors are smoothed , so that the electronic density with the pseudo - wave function is not correct . when operators describing properties of interest are heavily concentrated near the nucleus , their mean values are strongly affected by the wave function in the inner region . the four - component molecular spinors must , therefore , be restored in the heavy - atom cores . all molecular spinors @xmath66 can be restored as one - center expansions in the cores using the nonvariational one - center restoration ( nocr ) scheme that consists of the following steps : * generation of _ equivalent _ basis sets of one - center four - component spinors @xmath67 and of smoothed two - component pseudospinors @xmath68 in finite - difference all - electron dirac - fock(-breit ) and grecp / scf calculations of the same configurations of a considered atom and its ions . the nucleus is usually modeled as a uniform charge distribution within a sphere . the all - electron four - component hfdb and two - component grecp / hfj codes are employed by us to generate two equivalent numerical basis sets used at the restoration . these sets , describing mainly the atomic core region , are generated independently of the basis set used for the molecular grecp calculations . * the molecular _ pseudospinorbitals _ are then expanded in the basis set of one - center two - component atomic pseudospinors ( for @xmath69 , where @xmath70 is the radius of restoration that should be sufficiently large for calculating core properties with required accuracy ) , @xmath71 where @xmath72 denotes spatial and spin variables . note that for linear molecules only one value of @xmath73 survives in the sum for each @xmath74 . * finally , the atomic two - component pseudospinors in the molecular basis are replaced by equivalent four - component spinors and the expansion coefficients from eq . ( [ expansion ] ) are preserved : @xmath75 the molecular four - component spinors constructed this way are orthogonal to the inner core spinors of the heavy atom , because the atomic basis functions used in eq . ( [ restoration ] ) are generated with the inner core electrons treated as frozen . in the variational technique of the one - center restoration ( vocr ) @xcite , the proper behavior of the four - component molecular spinors in the core regions of heavy atoms can be restored as an expansion on the spherical harmonics inside the sphere with a restoration radius , @xmath76 , that should not be smaller than the matching radius , @xmath77 , used at the recp generation . the outer parts of spinors are treated as frozen after the recp calculation of a considered molecule . this method enables one to combine the advantages of two well - developed approaches , molecular recp calculation on gaussian basis set and atomic - type one - center calculation on numerical basis functions , by the most optimal way . it is considered theoretically in @xcite and some results concerning the efficiency of the one - center reexpansion of orbitals on another atom can be found in @xcite . the vocr scheme can be used for constructing the core parts of the molecular spinors ( orbitals ) instead of using equivalent basis sets as is in the nocr technique . some disadvantage of the nocr scheme is that molecular pseudoorbitals are usually computed in a spin - averaged grecp / scf molecular calculation ( i.e. without accounting for the effective spin - orbit interaction ) and the reexpansion of molecular pseudospinorbitals on one - center atomic pseudospinors is yet restricted by accuracy , as it was seen in our grecp / rcc / nocr calculations @xcite of the tlf molecule ( see below ) . with the restored molecular bispinors , the two - electron integrals on molecular bispinors can be easily calculated when following the scheme presented in eqs . ( [ 12.1])([12.4 ] ) of the next section ( see @xcite for more details ) . thus , the four - component transfomation from the atomic basis that is now the most time - consuming stage of four - component calculations of heavy - atom molecules can be avoided . however , the most interesting direction in development of the two - step method is the possibility to `` split '' the correlation structure calculation of a molecule on two consequent correlation calculations : first , in the valence region , when the outer core and valence electrons are explicitly involved into the grecp calculation , and then , in the core region , when the outer and inner core space regions are only treated at the restoration stage . correlation effects in the valence and outer core regions ( not only valence parts of molecular orbitals but also configuration coefficients ) can be calculated , for example , by the grecp / rcc method with very high accuracy . then correlation effects in the inner and outer core regions can be taken into account using the dirac - coulomb(-breit ) hamiltonian and the one - center expansion . increasing the one - center restoration radius @xmath78 one can take into account correlation effects in the intermediate region ( outer core in our case ) with the required accuracy . roughly speaking , the computational efforts for the correlation structure calculations in the core and valence regions are added when using the two - step approach , whereas in the conventional one - step scheme , they have multiplicative nature . to evaluate one - electron core properties ( hyperfine structure , p , t - odd effects etc . ) employing the above restoraton schemes it is sufficient to obtain the one - particle density matrix , @xmath79 , after the molecular recp calculation on the basis of pseudospinors @xmath80 . at the same time , the matrix elements @xmath81 of the property operator @xmath82 should be calculated on the basis of equivalent four - component spinors @xmath83 . the mean value for this operator can be then evaluated as @xmath84 the only explicitly treated valence shells are used for evaluating a core property when applying eq . ( [ < w > ] ) since the atomic frozen core ( closed ) shells do not usually contibute to the properties of practical interest . however , correlations with the core electrons explicitly excluded from the recp calculation can be also taken into account if the effective operator technique @xcite is applied to calculate the property matrix elements @xmath85 on the basis set of bispinors @xmath83 . then , in expression ( [ < w > ] ) one should only replace @xmath81 by @xmath85 . alternatively , the correlations with the inner core electrons can be , in principle , considered within the variational restoration scheme for electronic structure in the heavy atom cores . strictly speaking , the use of the effective operators is correct when the molecular calculation is carried out with the `` correlated '' grecp ( see @xcite ) , in which the same correlations with the excluded core electrons are taken into account at the grecp generation as they are in constructing @xmath85 . nevertheless , the use of the ( g)recp that does not account for the core correlations at the molecular calculation stage can be justified if they do not influence dramatically on the electronic structure in the valence region . the latter approximation was applied in calculations of the ybf and baf molecules described in the following section . when contributions to @xmath86 are important both in the core and valence regions , the scheme for evaluating the mean value of @xmath87 can be as follows : * calculation of matrix elements on the molecular pseudospinorbitals ( which are usually linear combination of atomic gaussians ) over all the space region using conventional codes for molecular recp calculations ( though , the operator @xmath88 can be presented in different forms in calculations with the recp and dirac - coulomb(-breit ) hamiltonians ) , @xmath89 * calculation of the inner part of the matrix element of the operator with the same molecular pseudospinorbitals using the one - center expansion for @xmath80 ( @xmath90 stays for @xmath70 or @xmath78 below , @xmath91 ) : @xmath92 * calculation of the inner part of the matrix element of the operator with the restored molecular four - component spinors using the one - center expansion for @xmath83 : @xmath93 the matrix element @xmath94 of the @xmath87 operator is evaluated as @xmath95 obviously , that the one - center basis functions can be numerical ( finite - difference ) for better flexibility . the mean values of the majority of operators for the valence properties can be calculated with good accuracy without accounting for the inner parts of the matrix elements ( [ 12.2 ] ) and ( [ 12.3 ] ) . as is discussed above , for calculating the mean values of the operators singular near nuclei it is sufficient to account only for the inner parts , ( [ 12.3 ] ) , of the matrix elements of @xmath87 on the restored functions @xmath96 , whereas the other contributions , ( [ 12.1 ] ) and ( [ 12.2 ] ) , can be neglected . calculation of properties using the finite - field method @xcite and eq . ( [ < w > ] ) within the coupled - cluster approach is described in section [ stlf ] . [ [ sbafybfheff ] ] effective hamiltonian . + + + + + + + + + + + + + + + + + + + + + + the molecular spin - rotational degrees of freedom of ybf and baf for the @xmath97yb and @xmath98ba isotopes with nuclear spin @xmath99 are described by hamiltonian @xcite @xmath100 where @xmath101 is the rotational angular momentum , @xmath102 is the rotational constant , @xmath103 is the effective spin of the electron @xcite , @xmath104 is the spin of the @xmath97yb ( @xmath98ba ) nucleus , @xmath105 is the unit vector directed along the molecular axis from the heavy nucleus to fluorine . the spin - doubling constant @xmath43 characterizes the spin - rotational interaction . @xmath106 and @xmath107 are the molecular dipole moment and the external electric field . the axial tensor @xmath108 describes magnetic hyperfine structure on the ytterbium ( barium ) nucleus : @xmath109 which can be also expressed in the notations of the isotropic @xmath110 and dipole @xmath111 hyperfine structure constants . the hyperfine interaction with the @xmath112f nucleus is significantly smaller and is not of interest for the considered effects . the last three terms in eq . ( [ eybf1 ] ) account for the p- and p , t - odd effects : _ the first _ of them describes the electromagnetic interaction of the electron with the anapole moment of the nucleus with the constant @xmath113 @xcite ; _ the second _ term accounts for the scalar p , t - odd electron - nucleon interaction characterized by the dimensionless constant @xmath114 ; _ the third _ term corresponds to interaction of the electron edm @xmath115 with the internal molecular field @xmath116 : @xmath117 the constant @xmath118 characterizes an effective electric field on the unpaired electron . all the p- and p , t - odd constants @xmath119 depend on the electron spin density in the vicinity of the heavy nucleus . the reliability of their calculation can be tested by comparison of the calculated and experimental values for the hyperfine constants @xmath120 and @xmath121 . [ [ sbafybfcalc ] ] details of calculations . + + + + + + + + + + + + + + + + + + + + + + + + let us consider the scheme of calculation for the above molecules on example of the ybf calculation carried out mainly in paper @xcite . the stage of the grecp calculation of the valence structure for the ground state @xmath122 of ybf was performed by analogy with the previous ybf calculation @xcite . the main difference from the latter was the use of technique @xcite for freezing the @xmath123 and @xmath124 pseudospinors derived from the calculation on the yb@xmath125 cation . it was necessary to freeze these shells because their polarization was taken into account at the stage of calculating the effective operator ( eo , see @xcite for details ) . the spin - averaged grecp calculations in the framework of the rasscf method @xcite with 5284 configurations were performed for 11 electrons distributed in ras-1@xmath126(2,0,0,0 ) , ras-2@xmath126(2,1,1,0 ) and ras-3@xmath126(6,4,4,2 ) active orbital subspaces according to the @xmath127 and @xmath128 irreducible representations ( irreps ) of the @xmath129 symmetry group used in the calculation ( see paper @xcite for details on the rasscf method ) . similar calculation was also carried out for the baf molecule @xcite . expressions for the @xmath130 parameter which characterizes the internal molecular field are presented in eqs . ( [ eybf1a ] ) and ( [ eybf1b ] ) , whereas expressions for other electronic matrix elements which correspond to the parameters @xmath120 , @xmath131 and @xmath119 of operator ( [ eybf1 ] ) can be found in papers @xcite . all the radial integrals and atomic four - component spinors were calculated for the finite nucleus @xmath97yb using a model of a uniform charge distribution within a sphere . as is mentioned in the introduction , atomic matrix elements of operator ( [ eybf1a ] ) are proportional to @xmath132 . the same scaling is applicable to the constant @xmath133 , while the matrix elements which contribute to the constants @xmath134 are proportional to @xmath135 . the nuclear charge of fluorine is eight times smaller than that of ytterbium , therefore , the contributions to the @xmath119 parameters from the vicinity of the fluorine nucleus can be neglected . the additional argument in favour of this approximation is connected with the circumstance that the state of the unpaired electron in the ybf molecule is described mainly by the ytterbium orbitals and thus the spin density is also localized near ytterbium . [ [ sbafybfres ] ] results . + + + + + + + + results of the calculation for the spin - rotational hamiltonian parameters are presented in table [ tbafybf ] . comparison of the results of the grecp / rasscf calculations @xcite with the results of the grecp / rasscf / eo calculations @xcite confirms the conclusion made in paper @xcite that the core - valence correlations of the unpaired ybf electron occupying mainly the hybridized @xmath136 orbital of yb with the @xmath123 and then with @xmath124 shells of the yb core ( first of all , spin polarization of the @xmath137 shells ) contribute very essentially to the hyperfine and the p- and p , t - odd constants . our final results for the hyperfine structure constant @xmath120 differ by less than 3% from the experimental value @xcite . this means that the space - isotropic part of the unpaired spin density in the vicinity of the yb nucleus in our calculations is obtained with rather good accuracy . however , it is also important to reproduce the anisotropic ( dipole ) part of the spin density which contributes to the dipole constant @xmath131 . the value for @xmath131 obtained in the rasscf / eo calculation is in slightly better agreement with the experimental datum than the value from paper @xcite but is still underestimated by 23% . about one half of this difference can be explained by the fact that the @xmath138 shell of yb was frozen in our molecular calculation ( it was first pointed out by khriplovich that accounting for excitations from the @xmath138-shell can be important ) . in particular , these excitations can explain the small value of the spin - doubling constant @xmath43 @xcite . it was shown in semiempirical calculation @xcite that the contribution of the @xmath138-shell excitations to the spin density can result in a significant correction to the constant @xmath131 . using eqs . ( 19 ) and ( 31 ) from paper @xcite , one can go to the following estimates for the contributions from the @xmath138-shell excitations to @xmath120 and @xmath131 : @xmath139 note that the semiempirical corrections obtained in @xcite arise from the admixture to the molecular wavefunction of the configuration with the hole in the @xmath138-shell . the weight of this configuration was estimated in paper @xcite to be of the order of 4% . this admixture is a purely molecular effect and is not described by the effective operator technique . thus , one can conclude that the @xmath140 value can be added to the rasscf / eo value for @xmath131 , resulting in @xmath141 mhz which is essentially closer to the experimental value of 102 mhz . note also that _ ab initio _ correlation calculation of ybf with , at least , @xmath138 shell treated explicitly should be performed to check the above described mechanism of increasing the calculated @xmath131 value . it is likely that the real situation is more complicated and indirect contributions to @xmath142 caused by the @xmath138 shell correlation relaxation effects and by other higher order contributions ( which are not accounted for in the considered calculation ) are mainly responsible for the obtained underestimation of @xmath131 . it is essential that similar contribution of the @xmath138-shell excitations to the constant @xmath130 is strongly suppressed . indeed , operator ( [ eybf1a ] ) mixes @xmath143- and @xmath144-waves . the @xmath145-shell is localized rather deeply in the ytterbium core and its mixture with the @xmath138-orbitals by the molecular field is very small while the @xmath146 electrons are relatively weakly - bound and does not penetrate essentially into the region of the @xmath138-shell localization . similar contributions to other constants @xmath119 are negligible due to the contact character of the corresponding interactions . one can see that the values of the @xmath130 constant from the unrestricted dhf calculation ( accounting for the spin - polarization of core shells ) @xcite , the latest semiempirical calculation @xcite and the grecp / rasscf / nocr / eo calculation are in very good agreement . it is also important for reliability of the carried out calculations that the contribution from the valence electron to @xmath130 in paper @xcite differs only by 7.4% from the corresponding grecp / scf / nocr calculation @xcite ( see table [ tbafybf ] ) . another dhf calculation @xcite gives a value that is about two times smaller ( that is rather explained by different definitions used for @xmath130 ) . similar increase in the values for the hyperfine constants and parameters of the p , t - odd interactions when the correlations with the core shells ( first of all , @xmath137 ) are taken into account is also observed for the baf molecule @xcite as one can see in table [ tbafybf ] ) . of course , the corrections on the @xmath138-electron excitations are not required for this molecule . the enhancement factor for the p , t - odd effects in the baf molecule is three times smaller than in ybf mainly because of smaller nuclear charge of ba . therefore , barium fluoride can be considered as a less promising molecule for a search for the pnc effects . lcclcl + & @xmath120 & @xmath147 & @xmath130 & @xmath148 & @xmath149 + & ( mhz ) & ( mhz ) & ( @xmath150 @xmath151 ) & ( khz ) & ( hz ) + + + + experiment @xcite & 7617 & 102 & & & + semiempirical @xcite & & & @xmath1521.5 & 730 & @xmath15248 + semiempirical @xcite ( with @xmath138-correction ) & & & @xmath1521.26 & & @xmath15243 + grecp / scf / nocr @xcite & 4932 & 59 & @xmath1520.91 & 484 & @xmath15233 + grecp / rasscf / nocr @xcite & 4854 & 60 & @xmath1520.91 & 486 & @xmath15233 + restricted dhf ( quiney , 1998 ) @xcite & 5918 & 35 & @xmath1520.31 & 163 & @xmath15211 + restricted dhf + core polarization & 7865 & 60 & @xmath1520.60 & 310 & @xmath15221 + rescaled restricted dhf & & & @xmath1520.62 & 326 & @xmath15222 + unrestricted dhf ( parpia , 1998 ) @xcite + ( unpaired valence electron ) & & & @xmath1520.962 & & + unrestricted dhf @xcite & & & @xmath1521.203 & & @xmath15222 + grecp / rasscf / nocr / eo @xcite & 7842 & 79 & @xmath1521.206 & 634 & + grecp / rasscf / nocr / eo @xcite + ( with @xmath138-correction ) & 7839 & 94 & @xmath1521.206 & 634 & + + + + experiment @xcite & 2326 & 25 & & & + semiempirical @xcite & & & @xmath1520.41 & 240 & @xmath15213 + experiment @xcite & 2418 & 17 & & & + semiempirical @xcite & & & @xmath1520.35 & 210 & @xmath15211 + + + grecp / scf / nocr @xcite & 1457 & 11 & @xmath1520.230 & 111 & @xmath1526.1 + grecp / rasscf / nocr @xcite & 1466 & 11 & @xmath1520.224 & 107 & @xmath1525.9 + grecp / scf / nocr / eo @xcite & 2212 & 26 & @xmath1520.375 & 181 & + grecp / rasscf / nocr / eo @xcite & 2224 & 24 & @xmath1520.364 & 175 & + [ [ stlfheff ] ] effective hamiltonian . + + + + + + + + + + + + + + + + + + + + + + here we consider the interaction of the edm of unpaired proton in @xmath16tl with the internal electromagnetic field of the @xmath16tlf molecule @xcite . this molecule is one of the best objects for the proton edm measurements . the effective interaction with the proton edm in tlf is written in the form @xmath153 where @xmath154 is the tl nuclear spin operator , @xmath105 is the unit vector along @xmath155 ( from tl to f ) , @xmath156 and @xmath157 are _ volume _ and _ magnetic _ constants @xcite @xmath158 @xmath159 is the nuclear schiff moment , @xmath160 is the proton edm , @xmath161 _ { x , y , z=0}\ , \label{x}\ ] ] @xmath162 is the electronic density calculated from the electronic wave function @xmath163 ; @xmath164 where @xmath165 is the nuclear edm arising due to p , t - odd nuclear forces ; @xmath166 , @xmath73 and @xmath10 are the magnetic moment , mass and charge of the tl nucleus ; @xmath167 is the velocity of light , @xmath168 @xmath169 is the orbital momentum for electron @xmath170 ; @xmath171 are its dirac matrices . @xmath172 and @xmath173 are factors determined by the nuclear structure of @xmath16tl : @xmath174_{\vec{i}}\ , \nonumber \label{q}\end{aligned}\ ] ] where @xmath175 is the nuclear wave function . accounting for @xmath176 leads to different hyperfine splitting of tlf in parallel and antiparallel electric and magnetic fields . the level shift @xmath177 is measured experimentally ( for the latest results on tlf see @xcite ) . the parameters @xmath178 of eq . ( [ x ] ) and @xmath179 of eq . ( [ m ] ) are determined by the electronic structure of the molecule . they were calculated recently for the ground @xmath180 state of tlf by parpia @xcite and by laerdahl , saue , faegri jr . , and quiney @xcite , using the dirac fock method with gaussian basis sets of large sizes ( see table [ result ] ) . ( below we refer to paper @xcite with the calculations presented in details and not to the brief communication @xcite of the same authors . ) in paper @xcite the first calculation of @xmath16tlf that accounts for correlation effects was performed ( see also @xcite where the limit on the schiff moment of @xmath16tl was recalculated ) . [ [ stlfcalc ] ] details of calculations . + + + + + + + + + + + + + + + + + + + + + + + + in paper @xcite 21-electron grecp @xcite for tl was used . the correlation spin - orbital basis set consisted of 26@xmath181 , 25@xmath182 , 18@xmath144 , 12@xmath143 , and 10@xmath183 gaussian - type orbitals on tl , contracted to @xmath184 . the basis was optimized in a series of atomic two - component grecp calculations , with correlation included by the relativistic coupled cluster method @xcite with single and double cluster amplitudes ; the average energy of the two lowest states of the atom was minimized . the basis set generation procedure is described in refs . the basis set was designed to describe correlation in the outer core @xmath123 and @xmath124 shells of tl , in addition to the @xmath146 and valence shells . while correlating the @xmath123 and @xmath124 shells may not be important for a majority of chemical and physical properties of the atom , it is essential for describing properties coming from the inner tl region , including p , t - odd effects . the ( @xmath185)/basis set from the ano - l library @xcite is used for fluorine . first , one - component scf calculation of the @xmath186 ground state of tlf is performed , using the spin - averaged grecp for tl which simulates the interactions of the valence and outer core ( @xmath187 ) electrons with the inner core [ kr]@xmath188 . this is followed by two - component rcc calculations , with only single ( rcc - s ) or with single and double ( rcc - sd ) cluster amplitudes . the rcc - s calculations with the spin - dependent grecp operator take into account effects of spin - orbit interaction at the level of the one - configurational scf - type method . the rcc - sd calculations include , in addition , the most important electron correlation effects . at the restoration stage , the nucleus was modeled as a uniform charge distribution within a sphere with radius @xmath189 a.u . , whereas previous calculations @xcite employed a spherical gaussian nuclear charge distribution ( the root mean square radius in all calculations is 5.5 fm , in accord with the parametrization of johnson and soff @xcite , and agrees with the experimental value 5.483 fm for the @xmath16tl nucleus @xcite ) . the two equivalent [ @xmath190 numerical basis sets were generated for the restoration . the quality of the approximation for the two - center molecular spinors and , consequently , of the calculated properties increases with the value of @xmath191 . a series of calculations of the @xmath179 parameter was performed using eq . ( [ restoration ] ) with basis functions including up to @xmath182 , @xmath144 and @xmath143 harmonics . it is found ( see table [ result ] ) that accounting for only @xmath181 and @xmath182 functions in the expansion determines @xmath179 with 90% accuracy . because the contribution of @xmath143 functions is only about 0.3% and amplitudes of higher harmonics on the nucleus are suppressed by the leading term @xmath192 , the error due to the neglect of spherical harmonics beyond @xmath143 is estimated to be below 0.1% . calculation of the @xmath178 parameter requires @xmath181 and @xmath182 harmonics ( see ref . @xcite ) , although , strictly speaking , @xmath144 harmonics also give nonzero contributions . in the rcc calculations , the @xmath178 and @xmath179 parameters were calculated by the finite field method ( e.g. , see refs . @xcite ) . the operator corresponding to a desired property ( eqs . ( [ x ] ) and ( [ m ] ) ) is multiplied by a small parameter @xmath193 and added to the hamiltonian . the derivative of the energy with respect to @xmath193 gives the computed property . this is strictly correct only at the limit of vanishing @xmath193 , but it is usually possible to find a range of @xmath193 values where the energy is almost linear with @xmath193 and energy changes are large enough to allow sufficient precision . the quadratic dependence on @xmath193 is eliminated in the present calculations by averaging absolute energy changes obtained with @xmath193 of opposite signs . [ [ stlfres ] ] results . + + + + + + + + calculations were carried out at two internuclear separations , the equilibrium @xmath194 as in ref . @xcite , and 2.1 , for comparison with ref . the results are collected in table [ result ] . the first point to notice is the difference between spin - averaged scf values and rcc - s values , the latter include spin - orbit interaction effects . these effects increase @xmath178 by 9% and decrease @xmath179 by 21% . the rcc - s function can be written as a single determinant , and results may therefore be compared with dhf values , even though the rcc - s function is not variational . grecp / rcc - s / nocr values of the @xmath179 parameter are indeed within 3% and 1% of difference with the corresponding dhf values @xcite ( table [ result ] ) . this agreement confirms the validity of the used approximations . in particular , freezing the inner core shells is justified , as inner core relaxation effects have little influence on the properties calculated here , a conclusion already drawn by quiney et al . @xcite . much larger differences occur for the @xmath178 parameter . there are also large differences between the two dhf calculations for @xmath178 , which can not be explained by the small change in internuclear separation . the value of @xmath178 may be expected to be less stable than @xmath179 , because it is determined by the derivative of the electronic density at the tl nucleus and involves large cancellations @xcite between contributions of large and small components , each of them is about 20 times larger than their sum . thus , a strong dependence of @xmath178 on the basis used may be expected . the dhf values collected in table [ result ] indeed show such dependence . results obtained in refs . @xcite and @xcite with comparable even - tempered basis sets , @xmath195 and @xmath196 , are rather close , differing by 340 a.u . improving the tl basis to @xmath197 @xcite increases @xmath178 by 650 a.u . or 8% . further improvement of the basis may be expected to yield even higher @xmath178 values . the numerical basis functions obtained in atomic dhf calculations and used for the restoration are highly accurate near the nucleus , so that the grecp / rcc - s / nocr value for @xmath178 , which is higher than that of quiney _ et al . _ @xcite , seems reasonable . the different nuclear models used in the present and dhf @xcite calculations may also contribute to the disagreement in @xmath178 , which is determined by the derivative of the electronic charge density at a single point , the tl origin . @xmath179 is affected by @xmath163 in a broader region , and is therefore far less sensitive to the nuclear model . the electron correlation effects are calculated by the rcc - sd method at the molecular equilibrium internuclear distance @xmath198 . a major correlation contribution is observed , decreasing @xmath179 by 17% and @xmath178 by 22% . the hyperfine structure constants of tl @xmath199 and tl@xmath125 @xmath200 , which ( like @xmath178 and @xmath179 ) depend on operators concentrated near the tl nucleus , were also calculated . the errors in the df values are 1015% ; the rcc - sd results are within 14% of experiment . the improvement in @xmath178 and @xmath179 upon inclusion of correlation is expected to be similar . ll|ddd|r|dd|r & & & + & @xmath201 & @xmath202 & @xmath203 & @xmath201 & @xmath201 & @xmath203 & @xmath201 + & & @xmath178 & & @xmath178 + & 19.67 & 17.56 & 17.51 & 8967 & 19.52 & 17.43 & 8897 + & 16.12 & & 13.84 & 9813 & 16.02 & 13.82 & 9726 + dhf @xcite&tl:(@xmath204 ) & 15.61 & & & 7743 & & & + dhf @xcite&tl:(@xmath205 ) & & & & & & 13.64 & 8098 + & tl:(@xmath206 ) & & & & & & 13.62 & 8089 + & tl:(@xmath207 ) & & & & & & 13.66 & 8492 + & tl:(@xmath208 ) & & & & & & 13.63 & 8747 + & & & 11.50 & 7635 & & & [ result ] it was noted previously that the experiments on the excited @xmath23 @xcite or @xmath210 @xcite states of pbo having nonzero projection of total electronic momentum on the internuclear axis can be , in principle , sensitive enough to detect @xmath1 three or even four orders of magnitude lower than the current limit . the knowledge of the effective electric field , @xmath130 , seen by an unpaired electron is required for extracting @xmath1 from the measurements . in papers @xcite , @xmath130 for the @xmath23 and @xmath210 states of the pbo molecule was calculated by the rcc - sd @xcite and ci methods . to provide an accuracy check in calculation of the electronic structure near the pb nucleus the hyperfine constant , @xmath211 , was also calculated . [ [ spbocalc ] ] details of calculations . + + + + + + + + + + + + + + + + + + + + + + + + the 22-electron grecp for pb @xcite is used at the first stage of the two - step calculations of pbo : the inner shells of the pb atom ( @xmath212 to @xmath138 ) are absorbed into the grecp , and the @xmath213 electrons and all the oxygen electrons are treated explicitly at the integral preparation part . two series of rcc - sd calculations with 10 ( when freezing the @xmath187 shells of lead employing the level shift technique @xcite and freezing the @xmath212 shell of oxigen after the rasscf stage ) and all 30 electrons treated explicitly were performed . only 10 electrons were treated explicitly in the ci calculations . for each series of calculations , correlation spin - orbital basis sets are optimized in atomic two - component grecp / rcc calculations of pb . the generated basis sets on pb are ( @xmath214)/[@xmath215 for 30-electron and ( @xmath216)/[@xmath217 for 10-electron calculations . the correlation - consistent ( @xmath218)/[@xmath219 basis listed in the molcas 4.1 library @xcite is used for oxygen . we found that the @xmath143 basis function of oxygen has little effect on the core properties calculated here and it was not included into the rcc calculations to reduce expenses . the leading @xmath220 coupling terms and configurations for the @xmath23 and @xmath210 states are @xmath221 @xmath222 and @xmath223 @xmath224 , correspondingly . the molecular orbitals used in the ci and rcc - sd calculations are obtained by the rasscf method @xcite with the spin - averaged grecp part @xcite , i.e. only scalar - relativistic interactions are taken into account in the rasscf calculation . the @xmath23 state is of practical interest first of all and , therefore , the lowest @xmath221 state was calculated at this stage . using the @xmath129 point group classification scheme , two a@xmath225 orbitals are in ras1 , six orbitals ( two a@xmath225 , two b@xmath225 , and two b@xmath226 ) in ras2 , and 50 ( 20 a@xmath225 , 6 a@xmath226 , 12 b@xmath225 , and 12 b@xmath226 ) in ras3 for 10-electron calculation . 2 a@xmath225 orbitals are in ras1 , 6 orbitals ( 2 a@xmath225 , 2 b@xmath225 , and 2 b@xmath226 ) in ras2 , and 41 ( 16 a@xmath225 , 5 a@xmath226 , 10 b@xmath225 , and 10 b@xmath226 ) in ras3 for 30-electron calculation . no more than two holes in ras1 and two particles in ras3 are allowed . at the restoration stage the nucleus is modeled by a uniform charge distribution within a sphere of radius @xmath227 . the root mean square radius of the nucleus is 5.52 fm , in accord with the parametrization of johnson and soff @xcite , and agrees with the experimental value of 5.497 fm @xcite for the @xmath209pb nucleus . taking this value for the root mean square radius and a fermi distribution for the nuclear charge changes @xmath211 and @xmath130 by 0.1@xmath228 or less . the equivalent basis sets generated are [ @xmath229 for 10 and [ @xmath230 for 30-electron calculation correspondingly . then the rcc - sd method and the spin - orbit ci approach with the selected single- and double - excitations from some multiconfigurational reference states ( `` mains '' ) @xcite are employed in the sets of different @xmath231s many - electron spin- and space - symmetry adapted basis functions [ [ spbores ] ] results . + + + + + + + + ci calculations @xcite were performed at two internuclear distances , @xmath232 a.u . , as is in rcc calculations , and @xmath233 a.u . ( in the rcc calculations @xcite the only internuclear distance @xmath232 a.u . was used because of problem with convergence at larger distances ) . the calculated values with the one - center expansion of the molecular spinors in the pb core on either @xmath181 , @xmath234 or @xmath235 partial waves are collected in table [ res1 ] . let us consider the @xmath236 orbitals of lead and @xmath212 orbital of oxigen as the outercore ones and @xmath237 , @xmath238 , @xmath239 , @xmath240 , @xmath241 orbitals of the pbo ( consisting mainly of @xmath242 orbitals of pb and @xmath243 orbitals of o ) as valence . although in the ci calculations we take into account only the correlation between valence electrons , the accuracy attained in the ci calculation of @xmath211 is much better than in the rcc - sd calculation . the main problem with the rcc calculation was that the used fock - space rcc - sd version was not optimal in accounting for the nondynamic correlations ( see @xcite for details of rcc - sd and ci calculations of the pb atom ) , though the potential of the rcc approach for electronic structure calculations is very high in prospect , especially , in the framework of the intermediate hamiltonian formulation @xcite . then we estimate the contribution from correlations of valence electrons with outercore ones ( which also account for correlations between outercore pairs of electrons ) as difference in the results of the corresponding 10- and 30-electron grecp / rcc calculations ( see also @xcite where this correction is applied to the pb atom ) . we designate such correlations in table [ res1 ] as `` outercore correlations '' . when taking into account outercore contributions at the point @xmath233 a.u . we used the results of the rcc calculation at the point @xmath232 a.u . since these contributions are relatively small and because the correlations with the outercore electrons are more stable than correlations in the valence region when the internuclear distance is changed the used approximation seems us reasonable . at last , the linear extrapolation of the results of calculations to the experimental equilibrium distances , @xmath244 a.u . for @xmath23 @xcite and @xmath245 a.u . for @xmath210 @xcite is applied . the final results are : @xmath246 mhz , @xmath247 for @xmath23 and @xmath248 mhz , @xmath249 for @xmath210 states . the estimated error for the final @xmath130 parameter is 20% for @xmath210 state . for @xmath23 the estimated error bounds put the actual @xmath130 value between 90% and 130% of the final value ( for details see @xcite ) . lrrrrrrrrrrrrr state & & & + parameters & & & & & & & + expansion & s & s , p & s , p , d & & s , p & s , p , d & & s & s , p & s , p , d & & s , p & s , p , d + + + + 10e - rasscf & -894 & -1505 & -1503 & & 0.73 & 0.70 & & & & & & 0.0 & 0.0 + 30e - rasscf & -759 & -1705 & -1699 & & 0.96 & 0.91 & & & & 1900 & & 0.0 & 0.0 + 10e - rcc - sd @xcite & & & -2635 & & -2.93 & -3.05 & & & & 3878 & & -11.10 & -10.10 + 30e - rcc - sd @xcite & & & -2698 & & & -4.10 & & & & 4081 & & -9.10 & -9.70 + _ outercore ( 30e - rcc - sd - 10e - rcc - sd ) _ & & & _ -63 & & & _ -1.05 & & & & _ 203 & & & _ 0.40 + 10e - ci @xcite & & & -3446 & & & -4.13 & & & & 4582 & & & -10.64 + * final @xcite & & & & & & & & & & & & & + ( 10e - ci + outercore ) & & & * -3509 & & & * -5.18 & & & & * 4785 & & & * -10.24 + + + + 10e - rasscf * * * * * _ _ _ _ & -770 & -1384 & -1383 & & 1.05 & 1.00 & & & & & & 0.0 & 0.0 + 10e - ci @xcite & & & -3689 & & & -4.81 & & & & 4762 & & & -7.18 + * final * @xcite & & & & & & & & & & & & & + ( 10e - ci + outercore ) & & & * -3752 & & & * -5.86 & & & & * 4965 & & & * -6.78 + * * * * [ res1 ] the p , t - parity nonconservation parameters and hyperfine constants are calculated for the heavy - atom molecules which are of primary interest for modern experiments on search for the pnc effects . it is found that accounting for the electron correlations is necessary for precise calculation of these properties . the developed two - step ( grecp / nocr ) scheme of calculation of the properties heavily concentrated in atomic cores is a very efficient way to take account of these correlations with moderate efforts . the results of the two - step calculations for the hyperfine constants differ less than 10% from the corresponding experimental data . the comparable level of accuracy is expected for the p , t - odd spin - rotational hamiltonian parameters which can not be obtained experimentally . the precision of the discussed calculations is limited by the current possibilities of the correlation methods and codes and not by the grecp and nocr approximations , despite the grecp / nocr method allows one to reduce seriously the expenses of the correlation treatment as compared to conventional dirac - coulomb(-breit ) approaches when using basis of molecular spin - orbitals instead of spinors etc . in turn , the attained accuracy is sufficient for a reliable interpretation of the results measured in the molecular pnc experiments . the present work is supported by the u.s . crdf grant rp22339ga02 and rfbr grant 030332335 . is grateful to ministry of education of russian federation ( grant pd021.3236 ) and to st .- petersburg committee of science ( grant pd03 - 1.3 - 60 ) . is also supported by the grants of russian science support foundation and the governor of leningrad district .

the large branching ratios for pure annihilation and decays reported by cdf and lhcb collaborations recently and the so - called and puzzles indicate that spectator scattering and annihilation contributions are important to the penguin - dominated , color - suppressed tree dominated , and pure annihilation nonleptonic decays . combining the available experimental data for , and decays , we do a global fit on the spectator scattering and annihilation parameters , , and , which are used to parameterize the endpoint singularity in amplitudes of spectator scattering , nonfactorizable and factorizable annihilation topologies within the qcd factorization framework , in three scenarios for different purpose . numerically , in scenario ii , we get)=(2.88^{+1.52}_{-1.30},-103^{+33}_{-40})$ ] and)=(1.21^{+0.22}_{-0.25},-40^{+12}_{-8})$ ] at the confidence level , which are mainly demanded by resolving puzzle and confirm the presupposition that . in addition , correspondingly , the-meson wave function parameter is also fitted to be , which plays an important role for resolving both and puzzles . with the fitted parameters , the qcdf results for observables of , and decays are in good agreement with experimental measurements . much more experimental and theoretical efforts are expected to understand the underlying qcd dynamics of spectator scattering and annihilation contributions .

1409.1322

charmless hadronic @xmath7-meson decays provide a fertile ground for testing the standard model ( sm ) and exploring the source of @xmath24 violation , which attract much attention in the past years . thanks to the fruitful accomplishment of babar and belle , the constraints on the sides and interior angles of the unitarity triangle significantly reduce the allowed ranges of some of the ckm elements , and many rare @xmath7 decays are well measured . with the successful running of lhc and the advent of belle ii at superkekb , heavy flavour physics has entered a new exciting era and more @xmath7 decay modes will be measured precisely soon . recently , the evidence of pure annihilation decays @xmath25 @xmath9 @xmath26 and @xmath27 @xmath9 @xmath28 are firstly reported by cdf collaboration @xcite , and soon confirmed by lhcb collaboration @xcite . the heavy flavor averaging group ( hfag ) presents their branching ratios @xcite @xmath29 @xmath30 such results , if confirmed , imply unexpectedly large annihilation contributions in @xmath7 decays and significant flavour symmetry breaking effects between the annihilation amplitudes of @xmath8 and @xmath31 decays , which attract much attention recently , for instance refs . @xcite . theoretically , as noticed already in refs . @xcite , even though the annihilation contributions are formally @xmath32 power suppressed , they are very important and indispensable for charmless @xmath7 decays . by introducing the parton transverse momentum and the sudakov factor to regulate the endpoint divergence , there is a large complex annihilation contribution within the perturbative qcd ( pqcd ) approach @xcite . the latest renewed pqcd estimations @xmath33 @xmath34 @xmath35 and @xmath36 @xmath34 @xmath37 @xcite give an appropriate account of the cdf and lhcb measurements within uncertainties . in the qcd factorization ( qcdf ) framework @xcite , the endpoint divergence in annihilation amplitudes is usually parameterized by @xmath38 ( see eq.([xa ] ) ) . the parameters @xmath39 @xmath40 @xmath41 and @xmath42 @xmath40 @xmath43 ( scenario s4 ) @xcite are adopted conservatively in evaluating the amplitudes of @xmath7 @xmath1 @xmath44 decays , which lead to the predictions @xmath33 @xmath34 @xmath45 and @xmath36 @xmath34 @xmath46 @xcite . it is obvious that the qcdf prediction of @xmath36 agrees well with the data eq.([hfagkk ] ) , but the one of @xmath33 is much smaller than the present experimental measurement eq.([hfagpipi ] ) . this discrepancy kindles the passions of restudy on annihilation contributions @xcite . at present , there are two major issues among the well - concerning focus on the annihilation contributions within the qcdf framework , one is whether @xmath47 is universal for @xmath7 decays , and the other is what its value should be . as to the first issue , there is no an imperative reason for the annihilation parameters @xmath39 and @xmath42 to be the same for different @xmath48 decays , even for different annihilation topologies , although they were usually taken to be universal in the previous numerical calculation for simplicity @xcite . phenomenologically , it is almost impossible to account for all of the well - measured two - body charmless @xmath7 decays with the universal values of @xmath39 and @xmath42 based on the qcdf approach @xcite . in addition , the pqcd study on @xmath7 meson decays also indicate that the annihilation parameters @xmath39 and @xmath42 should be process - dependent . in fact , in the practical qcdf application to the @xmath7 @xmath9 @xmath44 , @xmath49 decays ( where @xmath50 and @xmath51 denote the light pseudoscalar and vector @xmath52 meson nonet , respectively ) , the non - universal values of annihilation phase @xmath42 with respect to pp and pv final states are favored ( scenario s4 ) @xcite ; the process - dependent values of @xmath39 and @xmath42 are given based on an educated guess @xcite or the comparison with the updated measurements @xcite ; the flavour - dependent values of @xmath39 and @xmath42 are suggested recently in the nonfactorizable annihilation contributions @xcite . in principle the value of @xmath39 and @xmath42 should differ from each other for different topologies with different flavours , but we hope that the qcdf approach can accommodate and predict much more hadronic @xmath7 decays with less input parameters . so much attention in phenomenological analysis on the weak annihilation @xmath7 decays is devoted to what the appropriate values of the parameters @xmath39 and @xmath42 should be . this is the second issue . in principle , a large value of @xmath39 is unexpected by the power counting rules and the self - consistency validation within the qcdf framework . the original proposal is that @xmath39 @xmath53 @xmath41 and an arbitrary strong interaction phase @xmath42 are universal for all decay processes , and that a fine - tuning of the phase @xmath42 is required to be reconciled with experimental data when @xmath39 is significantly larger than 1 @xcite . the recent study on the annihilation contributions show that @xmath39 @xmath54 @xmath55 and @xmath56 @xmath57 @xmath58 are acceptable , even necessary , to reproduce the data for some two - body nonleptonic @xmath48 decay modes @xcite . in this paper , we will perform a fitting on the parameters @xmath39 and @xmath42 by considering @xmath7 @xmath9 @xmath6 , @xmath5 and @xmath11 decay modes , on one hand , to investigate the strength of annihilation contribution , on the other hand , to study their effects on the anomalies in @xmath7 physics , such as the well - known @xmath5 and @xmath6 puzzles . the so - called @xmath5 puzzle is reflected by the difference between the direct @xmath24 asymmetries for @xmath59 @xmath9 @xmath60 and @xmath61 @xmath9 @xmath62 decays . with the up - to - date hfag results @xcite , we get @xmath63 which differs from zero by about @xmath64 . however , the direct @xmath24 asymmetries of @xmath65 and @xmath66 are expected to be approximately equal with the isospin symmetry in the sm , numerically for instance @xmath67 in the s4 scenario of qcdf @xcite . the so - called @xmath6 puzzle is reflected by the following two ratios of the @xmath24-averaged branching fractions @xcite : @xmath68 \frac { \tau_{b^0 } } { \tau_{b^+ } } , \qquad r_{00}^{\pi \pi } \equiv 2 \big [ \frac { { \cal b } ( \bar{b}^{0 } \to \pi^0 \pi^0 ) } { { \cal b } ( \bar{b}^{0 } \to \pi^{+ } \pi^{- } ) } \big ] \label{pipipuzzle}.\ ] ] it is generally expected that branching ratio @xmath69 and @xmath70 within the sm . to date , the agreement of @xmath71 between the s4 scenario qcdf @xmath72 @xmath34 @xmath73 @xcite and the refined experimental data @xmath74 @xmath34 @xmath75 @xcite can be achieved consistently within experimental error , while the discrepancy in @xmath76 between the s4 scenario qcdf @xmath77 @xmath34 @xmath78 ( where theoretical uncertainties are unenclosed ) @xcite and the progressive experimental data @xmath74 @xmath34 @xmath75 @xcite is unexpectedly large . it is claimed @xcite that the so - called @xmath6 puzzle could be accommodated by the nonfactorizable contributions in sm . it is argued @xcite that to solve the so - called @xmath5 puzzle , a large complex color - suppressed tree amplitude @xmath79 or a large complex electroweak penguin contribution @xmath80 or a combination of them are essential . an enhanced complex @xmath80 with a nontrivial strong phase can be obtained from new physics effects @xcite . to get a large complex @xmath79 , one can resort to spectator scattering and final state interactions @xcite . recently , the annihilation amplitudes with large parameters @xmath39 is suggested to conciliate the recent measurements eq.([hfagpipi ] ) and eq.([hfagkk ] ) , so surprisingly , the @xmath5 puzzle is also resolved simultaneously @xcite . theoretically , the power corrections , such as spectator scattering at the twist-3 order and annihilation amplitudes , are important to account for the large branching ratios and @xmath24 asymmetries of penguin - dominated and/or color - suppressed tree - dominated @xmath7 decays . so , before claiming a new physics signal , it is essential to examine whether power corrections could retrieve `` problematic '' deviations from the sm expectations . interestingly , our study show that with appropriate parameters , the annihilation and spectator scattering contributions could provide some possible solutions to the @xmath23 and @xmath10 puzzles . our paper is organized as following . in section [ sec02 ] , we give a brief overview of the hard spectator and annihilation calculations and recent studies within qcdf . in section [ sec03 ] , focusing on @xmath23 and @xmath10 puzzles , the effects of spectator scattering and annihilation contributions on @xmath7 @xmath1 @xmath10 , @xmath23 and @xmath11 decays are studied in detail in bluethree scenarios . in each scenario , a fitting on relevant parameters are performed . our conclusions are summarized in section [ sec04 ] . appendix [ app01 ] recapitulates the building blocks of annihilation and spectator scattering amplitudes . the input parameters and our fitting approach are given in appendix [ app02 ] and [ app03 ] , respectively . the effective hamiltonian for nonleptonic @xmath7 weak decays is @xcite @xmath81 where @xmath82 ( @xmath83 @xmath34 @xmath84 , @xmath85 and @xmath86 @xmath34 @xmath87 , @xmath88 ) is the product of the cabibbo - kobayashi - maskawa ( ckm ) matrix elements ; @xmath89 is the wilson coefficient corresponding to the local four - quark operator @xmath90 ; @xmath91 and @xmath92 are the electromagnetic and chromomagnetic dipole operators . + with the effective hamiltonian eq.([eq : eff ] ) , the qcdf method has been fully developed and extensively employed to calculate the hadronic two - body @xmath7 decays , for example , see @xcite . the spectator scattering and annihilation amplitudes ( see fig.[diag ] ) are expressed as the convolution of scattering functions with the light - cone wave functions of the participating mesons @xcite . the explicit expressions for the basic building blocks of spectator scattering and annihilation amplitudes have been given by ref . @xcite , which are also listed in the appendix [ app01 ] for convenience . with the asymptotic light - cone distribution amplitudes , the building blocks for annihilation amplitudes of eq.([ai1]-[af3 ] ) could be simplified as @xcite @xmath93 \label{xai12 } , \\ a_3^i & \simeq & 6 \pi \alpha_s \big(r_\chi^{m_1 } - r_\chi^{m_2 } \big ) \big ( x_a^2 - 2 x_a + \frac{\pi^2}{3 } \big ) \label{xai3 } , \\ a_3^f & \simeq & 6 \pi \alpha_s ( r_\chi^{m_1 } + r_\chi^{m_2 } ) ( 2 x_a^2 - x_a ) \label{xaf3 } , \end{aligned}\ ] ] where the superscripts @xmath94 ( or @xmath95 ) refers to gluon emission from the initial ( or final ) state quarks , respectively ( see fig.[diag ] ) . for the @xmath10 , @xmath23 and @xmath11 final - state , @xmath96 is numerically negligible due to @xmath97 @xmath98 @xmath99 . the model - dependent parameter @xmath100 is used to estimate the endpoint contributions , and expressed as @xmath101 where @xmath102 @xmath34 @xmath103 gev . for spectator scattering contributions , the calculation of twist-3 distribution amplitudes also suffers from endpoint divergence , which is usually dealt with the same manner as eq.([xa ] ) and labelled by @xmath104 @xcite . moreover , a quantity @xmath20 is used to parameterize our ignorance about @xmath7-meson distribution amplitude [ see eq.([hardblock ] ) ] through @xcite @xmath105 the qcdf approach itself can not give information or / and constraint on the phenomenological parameters of @xmath100 , @xmath104 and @xmath106 . these parameters should be determined from experimental data . to conform with measurements of nonleptonic @xmath7 @xmath9 @xmath44 decays , we will adopt a similar method used in ref.@xcite to deal with the contributions from weak annihilation and spectator scattering . focusing on the flavor dependence , without consideration of theoretical uncertainties , annihilation contributions are reevaluated in detail @xcite to explain the @xmath5 puzzle and the recent measurements on pure annihilation decays @xmath25 @xmath9 @xmath26 and @xmath27 @xmath9 @xmath28 [ see eq.([hfagpipi],[hfagkk ] ) ] . the authors of ref . @xcite find that the flavour symmetry breaking effects should be carefully considered for @xmath48 decays , and suggest that the parameters of @xmath39 and @xmath42 in nonfactorizable annihilation topologies @xmath107 [ see eq.([xai12],[xai3 ] ) ] should be different from those in factorizable annihilation topologies @xmath108 [ see eq.([xaf3 ] ) ] . ( 1 ) for factorizable annihilation topologies , i.e. , the gluon emission from the final states fig.[diag](c , d ) , the flavor symmetry breaking effects are embodied in the decay constants , because the asymptotic light - cone distribution amplitudes of final states are the same . in addition , all decay constants have been factorized outside from the hadronic matrix elements of factorizable annihilation topologies . so @xmath109 is independent of the initial state , and is the same for @xmath48 annihilation decays to two light pseudoscalar mesons , that is to say , @xmath110 and @xmath111 should be universal for @xmath48 @xmath1 @xmath44 decays . ( 2 ) for nonfactorizable annihilation topologies , i.e. , the gluon emission from the initial @xmath7 meson fig.[diag](a , b ) , besides the factorized decay constants and the same asymptotic light - cone distribution amplitudes , @xmath7 meson wave functions @xmath112 are involved in the convolution integrals of hadronic matrix elements . hence , @xmath113 should depend on the initial state and be different for @xmath8 from @xmath31 meosn due to flavor symmetry breaking effects , i.e. , parameters of @xmath114 and @xmath115 should be non - universal for @xmath116 and @xmath8 meson decays , and be different from parameters of @xmath110 and @xmath111 for @xmath109 . in fact , the symmetry breaking effects have been considered in pervious qcdf study on two - body hadronic @xmath7 decays @xcite , but with parameters of @xmath110 @xmath34 @xmath114 and @xmath111 @xmath34 @xmath115 . so , it is essential to systematically reevaluate factorizable and nonfactorizable annihilation contributions and preform a global fit on the annihilation parameters with the current available experimental data . in this paper , we will pay much attention to @xmath8 @xmath9 @xmath117 , @xmath5 , @xmath6 decays and the aforementioned @xmath5 , @xmath6 puzzles with a distinction between ( @xmath110 , @xmath111 ) and ( @xmath114 , @xmath115 ) , i.e. , @xmath118 @xmath119 @xmath120 . as aforesaid @xcite , the nonfactorizable spectator scattering amplitudes contribute to a large complex @xmath79 , which is important to resolve the @xmath5 , @xmath6 puzzles . from the building block eq.([hardblock ] ) , it can be easily seen that @xmath7 meson wave functions @xmath112 appear in the spectator scattering amplitudes . therefore , the symmetry breaking effects should also be considered for the quantity @xmath104 that is introduced to parameterize the endpoint singularity in the twist-3 level spectator scattering corrections . similar to @xmath118 , the quantity @xmath104 is related to the topologies that gluon emit from the initial @xmath7 meson . so , for simplicity , the approximation @xmath104 @xmath34 @xmath118 is assumed in our coming numerical evaluation ( scenarios i and ii , see the next section for detail ) . of course , this approximation is neither based on solid ground or from some underlying principle , and should be carefully studied and deserve much research . in fact , our coming phenomenological study ( scenarios iii ) shows that the approximation @xmath104 @xmath34 @xmath118 is allowable with the up - to - date measurement on @xmath8 @xmath9 @xmath117 , @xmath5 , @xmath6 decays . in addition , it can be seen from eq.([hardblock ] ) that the spectator scattering corrections depend strongly on the inverse moment parameter @xmath121 given in eq.([lamdef ] ) . recently , the value of @xmath121 is an increasing concern of theoretical and experimental physicists @xcite . a scrutiny of parameter @xmath121 becomes imperative . in this paper , we will give some information on @xmath121 required by present experimental data of @xmath8 @xmath9 @xmath11 , @xmath5 , @xmath6 decays . with the conventions in ref . @xcite , the decay amplitudes for @xmath8 @xmath9 @xmath23 , @xmath11 , @xmath122 decays within the qcdf framework can be written as @xmath123 \nonumber \\ & & + a _ { k \pi } \big [ \delta_{pu } \alpha_2 + \frac{3}{2 } \alpha_{3,{\rm ew}}^p \big ] \big\ } \label{amp2 } , \\ % % % % % % % % % % % % % % % % b0 - > pi+ k- { \cal a } _ { \bar{b}^0 \to \pi^+ k^- } & = & \sum\limits_{p = u , c } v_{pb}v_{ps}^{\ast } a _ { \pi k } \big\ { \delta_{pu } \alpha_1 + \alpha_4^p + \alpha_{4,{\rm ew}}^p + \beta_3^p - \frac{1}{2 } \beta_{3,{\rm ew}}^p \big\ } \label{amp3 } , \\ % % % % % % % % % % % % % % % % b0 - > pi0 k0 \sqrt{2 } { \cal a } _ { \bar{b}^0 \to \pi^0 \bar{k}^0 } & = & \sum\limits_{p = u , c } v_{pb}v_{ps}^{\ast } \big\ { a _ { \pi k } \big [ - \alpha_4^p + \frac{1}{2 } \alpha_{4,{\rm ew}}^p - \beta_3^p + \frac{1}{2 } \beta_{3,{\rm ew}}^p \big ] \nonumber \\ & & + a _ { k \pi } \big [ \delta_{pu } \alpha_2 + \frac{3}{2 } \alpha_{3,{\rm ew}}^p \big ] \big\ } \label{b02pi0k0 } , \\ % % % % % % % % % % % % % % % % b- - > k- k0 { \cal a } _ { b^- \to k^0 \bar{k}^0 } & = & \sum\limits_{p = u , c } v_{pb}v_{pd}^{\ast } a _ { k k } \big\ { \alpha_4^p - \frac{1}{2 } \alpha_{4,{\rm ew}}^p + \delta_{pu } \beta_2 + \beta_{3}^{p } + \beta_{3,{\rm ew}}^p \big\ } \label{bm2kk } , \\ % % % % % % % % % % % % % % % % b0 - > k+ k- { \cal a } _ { \bar{b}^0 \to k^- k^+ } & = & \sum\limits_{p = u , c } v_{pb}v_{pd}^{\ast } \big\ { b _ { \bar{k } k } \big [ \delta_{pu } b_1 + b_4^p + b_{4,{\rm ew}}^p \big ] + b _ { k \bar{k } } \big [ b_4^p - \frac{1}{2 } b_{4,{\rm ew}}^p \big ] \big\ } \label{amp4 } , \\ % % % % % % % % % % % % % % % % b0 - > k0 k0 { \cal a } _ { \bar{b}^0 \to \bar{k}^0 k^0 } & = & \sum\limits_{p = u , c } v_{pb}v_{pd}^{\ast } \big\ { a _ { \bar{k } k } \big [ \alpha_4^p - \frac{1}{2 } \alpha_{4,{\rm ew}}^p + \beta_3^p + \beta_4^p - \frac{1}{2 } \beta_{3,{\rm ew}}^p - \frac{1}{2 } \beta_{4,{\rm ew}}^p \big ] \nonumber \\ & & + b _ { k \bar{k } } \big [ b_4^p - \frac{1}{2 } b_{4,{\rm ew}}^p \big ] \big\ } \label{b02kzkz } , \\ % % % % % % % % % % % % % % % % b- - > pi- pi0 \sqrt{2 } { \cal a } _ { b^- \to \pi^- \pi^0 } & = & \sum\limits_{p = u , c } v_{pb}v_{pd}^{\ast } a _ { \pi \pi } \big\ { \delta_{pu } ( \alpha_1 + \alpha_2 ) + \frac{3}{2 } ( \alpha_{3,{\rm ew}}^p + \alpha_{4,{\rm ew}}^p ) \big\ } \label{amp5 } , \\ % % % % % % % % % % % % % % % % b0 - > pi+ pi- { \cal a } _ { \bar{b}^0 \to \pi^+ \pi^- } & = & \sum\limits_{p = u , c } v_{pb}v_{pd}^{\ast } a _ { \pi \pi } \big\ { \delta_{pu } ( \alpha_1 + \beta_1 ) + \alpha_{4}^p + \alpha_{4,{\rm ew}}^p + \beta_3^p + 2 \beta_4^p \nonumber \\ & & - \frac{1}{2 } \beta_{3,{\rm ew}}^p + \frac{1}{2 } \beta_{4,{\rm ew}}^p \big\ } \label{amp6 } , \\ % % % % % % % % % % % % % % % % b0 - > pi0 pi0 -{\cal a } _ { \bar{b}^0 \to \pi^0 \pi^0 } & = & \sum\limits_{p = u , c } v_{pb}v_{pd}^{\ast } a _ { \pi \pi } \big\ { \delta_{pu } ( \alpha_2 - \beta_1 ) - \alpha_{4}^p + \frac{3}{2}\alpha_{3,{\rm ew}}^p + \frac{1}{2 } \alpha_{4,{\rm ew}}^p \nonumber \\ & & - \beta_3^p -2 \beta_4^p + \frac{1}{2 } \beta_{3,{\rm ew}}^p - \frac{1}{2 } \beta_{4,{\rm ew}}^p ) \big\ } \label{amp7}. \end{aligned}\ ] ] for the sake for convenient discussion , we reiterate the expressions of the annihilation coefficients @xcite , @xmath124 \label{b3 } , \\ b_{4}^{p } & = & \frac{c_{f}}{n_{c}^{2}}\ , \big [ c_{4 } a_{1}^{i } + c_6 a_2^i \big ] \label{b4 } , \\ b_{3,\rm ew}^p & = & \frac{c_{f}}{n_{c}^{2}}\ , \big [ c_9 a_1^i + c_7 ( a_3^i + a_3^f ) + n_c c_8 a_3^f \big ] \label{b3ew } , \\ b_{4,\rm ew}^p & = & \frac{c_{f}}{n_{c}^{2}}\ , \big [ c_{10 } a_1^i + c_8 a_2^i \big ] \label{b4ew}. \end{aligned}\ ] ] numerically , coefficients of @xmath125 and @xmath126 are negligible compared with the other effective coefficients due to the small electroweak wilson coefficients , and so their effects would be not discussed in this paper . in order to illustrate the contributions of annihilation and spectator scattering , we explore three parameter scenarios in which certain parameters are changed freely . * scenario i : @xmath8 @xmath9 @xmath23 and @xmath11 decays , including the @xmath23 puzzle and pure annihilation decay @xmath127 @xmath1 @xmath128 , are studied in detail . combining the latest experimental data on the @xmath24-averaged branching ratios , direct and mixing - induced @xmath24-asymmetries , total 14 observables ( see table.[pikbr ] , [ pikdcp ] , [ pikmcp ] ) for seven @xmath8 @xmath9 @xmath23 , @xmath11 decay modes [ see eq.([bm2pimkz][b02kzkz ] ) ] , the fit on four parameters ( @xmath110 , @xmath111 ) and ( @xmath114 , @xmath115 ) is performed with the fixed value @xmath20 @xmath34 0.2 gev and the approximation ( @xmath129 , @xmath130 ) = ( @xmath114 , @xmath115 ) , where ( @xmath110 , @xmath111 ) , ( @xmath114 , @xmath115 ) and ( @xmath129 , @xmath130 ) are assumed to be universal for factorizable annihilation amplitudes , nonfactorizable annihilation amplitudes and spectator scattering corrections , respectively . * scenario ii : @xmath8 @xmath9 @xmath23 , @xmath11 and @xmath10 decays , including @xmath10 puzzle , are studied . combining the latest experimental data on the @xmath24-averaged branching ratios , direct and mixing - induced @xmath24-asymmetries , total 21 observables ( see table.[pikbr ] , [ pikdcp ] , [ pikmcp ] ) for ten @xmath8 @xmath9 @xmath23 , @xmath11 , @xmath10 decay modes [ see eq.([bm2pimkz][amp7 ] ) ] , the fit on five parameters ( @xmath110 , @xmath111 ) , ( @xmath114 , @xmath115 ) and @xmath20 is performed with the approximation ( @xmath129 , @xmath130 ) = ( @xmath114 , @xmath115 ) . * scenario iii : as a general scenario , to clarify the relative strength among ( @xmath110 , @xmath111 ) , ( @xmath114 , @xmath115 ) and ( @xmath129 , @xmath130 ) , and check whether the approximation ( @xmath129 , @xmath130 ) = ( @xmath114 , @xmath115 ) is allowed or not , a fit on such six free parameters is performed . other input parameters used in our evaluation are summarized in appendix [ app02 ] . our fit approach is illustrated in detail in appendix [ app03 ] . comparing eq.([amp2 ] ) with eq.([amp3 ] ) , it can be clearly seen that @xmath131 @xmath98 @xmath132 if @xmath133 @xmath134 @xmath135 is negligible compared with @xmath136 @xmath134 @xmath137 . hence it is expected @xmath138 @xmath98 0 in sm , which significantly disagrees with the current experimental data in eq.([acppi ] ) , this is the so - called @xmath23 puzzle . to resolve the @xmath23 puzzle , one possible solution is that there is a large complex contributions from @xmath133 @xmath134 @xmath135 . many proposals have been offered , such as the enhancement of color - suppressed tree amplitude @xmath139 in ref.@xcite , significant new physics corrections to the electroweak penguin coefficient @xmath140 in ref.@xcite , and so on . indeed , it has been shown @xcite that the coefficients @xmath139 and @xmath140 are seriously affected by spectator scattering corrections within qcdf framework . consequently , the nonfactorizable spectator scattering parameters @xmath104 or ( @xmath129 , @xmath130 ) will have great influence on the observable @xmath138 . furthermore , a scrutiny of difference between eq.([amp2 ] ) and eq.([amp3 ] ) , another possible resolution to the @xmath23 puzzle might be provided by annihilation contributions , such as coefficient @xmath141 , as suggested in ref.@xcite . if so , then @xmath138 will depend strongly on the nonfactorizable annihilation parameters ( @xmath142 , @xmath143 ) because @xmath141 is proportional to @xmath144 in eq.([b12 ] ) . additionally , it can be seen from eq.([amp2 ] ) and eq.([amp3 ] ) that annihilation coefficient @xmath145 contributes to amplitudes both @xmath146 and @xmath132 . if @xmath145 could offer a large strong phase , then its effect should contribute to the direct @xmath24 asymmetries @xmath147 and @xmath148 rather than @xmath138 . due to the fact that the lion s share of @xmath145 comes from @xmath149 in eq.([b3 ] ) , the direct @xmath24 asymmetries @xmath147 and @xmath148 should vary greatly with the factorizable annihilation parameters @xmath120 , while @xmath138 should be insensitive to variation of parameters ( @xmath150 , @xmath151 ) . the above analysis and speculations are confirmed by fig.[cpanni ] . from eq.([amp4 ] ) , it is seen that the amplitude @xmath152 depends heavily on coefficients @xmath153 and @xmath154 , which are closely associated with the nonfactorizable annihilation parameter @xmath118 only . the factorizable annihilation contributions vanish due to the isospin symmetry , which is consistent with the pqcd calculation @xcite . the large branching ratio eq.([hfagkk ] ) would appeal for large nonfactorizable annihilation parameter @xmath118 or @xmath142 . the dependence of branching ratio @xmath155 on the parameters ( @xmath142 , @xmath143 ) is displayed in fig.[branni ] . on nonfactorizable annihilation parameters ( @xmath142 , @xmath156 ) . the notes are the same as fig.[cpanni].,scaledwidth=40.0% ] .numerical results of annihilation parameters in scenario i. [ cols="<,^,^,^,^",options="header " , ] to explicitly show the effects of spectator scattering contributions on @xmath23 puzzle , dependance of @xmath147 , @xmath148 and their difference @xmath138 on parameter @xmath20 are displayed in fig.[lbpik ] . it is found that ( 1 ) observables of @xmath147 and @xmath138 are more sensitive to variation of @xmath20 than @xmath148 in the region of @xmath20 @xmath57 100 mev . the reason is aforementioned fact that coefficient @xmath139 in amplitude @xmath146 [ see eq.([amp2 ] ) ] receives significant spectator scattering corrections . a noticeable change of observables is easily seen in the low region of @xmath20 because spectator scattering corrections are inversely proportional to @xmath20 [ see eq.([lamdef ] ) and eq.([hardblock ] ) ] . ( 2 ) a relative small value of @xmath20 @xmath157 [ 150 mev , 220 mev ] , as expected in @xcite , is required to confront with available measurements . especially , the value @xmath20 @xmath158 190 mev provides a perfect description of the experimental data on @xmath147 , @xmath148 and @xmath138 simultaneously . for @xmath7 @xmath1 @xmath10 decays , from eqs.([amp5]-[amp7 ] ) , it is easily seen that amplitude @xmath159 @xmath160 @xmath161 + @xmath139 , @xmath162 @xmath160 @xmath161 , @xmath163 @xmath160 @xmath139 . the coefficient @xmath139 , corresponding to the color - suppressed tree contribution , its value is small relative to @xmath161 , so the experimental data on @xmath71 can be well explained with scenario s4 qcdf where @xmath118 @xmath34 @xmath120 and @xmath164 = 1 ( see table [ pikbr ] ) . but as to observable @xmath76 or / and branching ratio @xmath165 , an enhanced @xmath139 is desirable . hence , the nonfactorizable spectator scattering contributions , which have significant effects on @xmath139 , would play an important role in studying the color - suppressed tree @xmath7 decays , and possibly provide a solution to the @xmath10 puzzle . the dependencies of the branching fractions of @xmath7 @xmath1 @xmath22 decays and ratios @xmath166 , @xmath167 on @xmath20 are shown in fig.[lbpipi ] where the fitted parameters of part a in table [ pikfit ] is used . it is interesting that beside a large value @xmath129 , a small value of @xmath20 @xmath40 200 mev is also required to confront with experimental data on @xmath168 , @xmath166 and @xmath167 . with the available experimental data on @xmath7 @xmath1 @xmath10 , @xmath23 and @xmath11 decays , we perform a comprehensive fit on both annihilation parameters ( @xmath169 , @xmath170 ) and @xmath7-meson wave function parameter @xmath20 . the allowed parameter spaces are shown in fig.[paraspacii ] , and the corresponding numerical results are summarized in table [ pipikfit ] . like scenario i , there are two allowed spaces which are labelled by part a and b. it is easily found that ( 1 ) parameters @xmath171 @xmath34 @xmath172 are still required to have large values ( see table [ pipikfit ] ) , that is to say , it is necessary for penguin - dominated or color - suppressed tree @xmath7 decays to own large corrections from nonfactorizable annihilation and spectator scattering topologies . ( 2 ) there is still no overlap between the regions of @xmath173 and @xmath171 at the 95% confidence level . ( 3 ) the cental values of @xmath174 are a little larger than those in scenario i. the uncertainties on @xmath171 are a little smaller than those in scenario i , because more processes from @xmath7 @xmath1 @xmath10 decays are considered in fitting and the amplitudes for @xmath7 @xmath1 @xmath10 decays are sensitive to @xmath118 and @xmath104 rather than @xmath120 . ( 4 ) a small value of parameter @xmath20 @xmath175 350 mev at the 95% confidence level is strongly required to reconcile discrepancies between results of qcdf approach and available experimental data on @xmath7 @xmath1 @xmath10 , @xmath23 and @xmath11 decays . the two solutions of scenario ii , part a and b , will give similar results , as discussed before . with the best fit parameters of part a in table [ pipikfit ] , we present our evaluations on branching ratios , direct and mixing - induced @xmath24 asymmetries for @xmath8 @xmath9 @xmath23 , @xmath11 , @xmath10 decays in the `` scenario ii '' column of table [ pikbr ] , [ pikdcp ] and [ pikmcp ] , respectively . it is found that the central values of branching ratios for @xmath7 @xmath1 @xmath10 , @xmath23 and @xmath11 decays , expect @xmath176 @xmath1 @xmath2 decay , with the part a parameters of scenario ii , are a little larger than those of scenario i ( see table [ pikbr ] ) , because a bit larger values of @xmath174 and a bit smaller value of @xmath20 than those of scenario i are taken in scenario ii . compared with results of scenario s4 qcdf , agreement between theoretical results within two scenarios and experimental measurements is improved , especially for the observables @xmath138 , @xmath76 and @xmath177 . the above analyses and results are based on the assumption that @xmath178 @xmath34 @xmath179 ( i.e. @xmath180 @xmath34 @xmath181 ) for simplicity . while , there is no compellent requirement for such simplification , except for the fact that wave functions of @xmath7 mesons are involved in the convolution integrals of both spectator scattering and nonfactorizable annihilation corrections , but are irrelevant to the factorable annihilation amplitudes . so , as a general scenario ( named scenario iii ) , we would reevaluate the strength of annihilation and hard - spectator contributions without any simplification for the parameters @xmath180 , @xmath180 and @xmath181 . , @xmath182 ) , ( @xmath183 , @xmath156 ) and ( @xmath129 , @xmath130 ) at @xmath18 c.l .. the two solutions of ( @xmath184 , @xmath182 ) and ( @xmath183 , @xmath156 ) are labeled as part a , b and @xmath185 , @xmath186 , respectively.,scaledwidth=35.0% ] considering the constraints from observables of @xmath8 @xmath9 @xmath11 , @xmath5 and @xmath6 decays , a fit for the annihilation and hard - spectator parameters is performed again . in this fit , @xmath187 , @xmath180 and @xmath181 are treated as six free parameters . moreover , from the hard - spectator corrections illustrated by eq . ( [ hardblock ] ) , it can be seen that @xmath20 and @xmath104 are always combined together . although the inverse moment @xmath20 of @xmath7 wave function could be determined or constricted by further experiments @xcite , @xmath20 is more like a free parameter for the moment due to loose limitation on it . so it is impossible to strictly bound on @xmath20 and @xmath104 simultaneously due to the interference effects between them . in our following fit , we will fix @xmath188 . our fitting results at @xmath18 c.l . are presented in fig . [ specfig ] , where the range of @xmath189 @xmath190 @xmath191 $ ] is assigned to illustrate their relative magnitude . numerically , we get @xmath192)=\left\ { \begin{array}{ll } & ( 1.18^{+0.26}_{-0.23 } , -40^{+12}_{-8})\qquad \text{part a } \\ & ( 2.79^{+0.26}_{-0.20 } , -196^{+5}_{-3})\qquad \text{part b } \end{array } \right . \\ & & ( \rho_a^{i } , \phi_a^{i}[^{\circ}])=\left\ { \begin{array}{ll } & ( 2.85^{+2.18}_{-1.92 } , -103^{+52}_{-63})\qquad \text{part $ { \rm a}^{\prime}$ } \\ & ( 6.54^{+1.81}_{-3.30 } , -206^{+23}_{-24})\qquad \text{part $ { \rm b}^{\prime}$ } \end{array } \right . \\ & & ( \rho_{h } , \phi_{h}[^{\circ}])=(3.09^{+1.64}_{-1.53 } , -102^{+40}_{-31})\ , . \label{solusiii } \end{aligned}\ ] ] it can be easily seen from fig . [ specfig ] that : ( 1 ) for factorizable annihilation parameters @xmath187 , similar to scenarios i and ii , there are two allowed regions ( labelled by part a and b ) ; ( 2 ) for nonfactorizable annihilation parameters @xmath180 , besides the solution similar to scenarios i and ii ( labelled by part @xmath193 ) , another solution ( labelled by part @xmath194 ) with a very large value of @xmath183 is gotten . ( 3 ) it is very intersting that the allowed space of @xmath195 overlaps almost entirely with the `` part @xmath193 '' allowed space of @xmath180 . moreover , their best - fit points @xmath180 @xmath34 @xmath196 of `` part @xmath193 '' and @xmath195 @xmath34 @xmath197 are very close to each other . it might imply that the assumption @xmath178 @xmath180 @xmath34 @xmath104 @xmath181 used in scenarios i and ii is a good simplification . with the best fit parameters in scenarios iii , either the small value of @xmath183 in `` part @xmath193 '' or the large value in `` part @xmath194 '' , our evaluations on branching ratios , direct and mixing - induced @xmath24 asymmetries for @xmath8 @xmath9 @xmath23 , @xmath11 , @xmath10 decays are similar to those given in our scenarios i and ii , so no longer listed here . for the two solutions @xmath193 and @xmath194 of @xmath180 , it is expected by qcdf approach @xcite that the parameter @xmath183 should have a small value , which is also favored by our scenarios i and ii fit . in fact , such two solutions lead to the same results of @xmath198 , but the different ones of @xmath199 , which principally provides an opportunity to refute one of them . however , because @xmath199 is numerically trivial due to @xmath200 for the light mesons , such way is practically unfeasible for current accuracies of theoretical calculation and experimentally measurement . the recent cdf and lhcb measurements of large branching ratios for pure annihilation @xmath0 @xmath1 @xmath2 and @xmath3 @xmath1 @xmath4 decays imply possible large annihilation contributions , which induce us to modify the traditional qcdf treatment for annihilation parameters . following the suggestion of ref.@xcite , two sets of annihilation parameters @xmath118 and @xmath120 are used to parameterize the endpoint singularity in nonfactorizable and factorizable annihilation amplitudes , respectively . besides annihilation effects , the resolution of so - called @xmath5 and @xmath6 puzzles also expect constructive contributions from spectator scattering topologies . with the approximation of @xmath118 @xmath34 @xmath104 , we perform a global fit on both annihilation parameters ( @xmath169 , @xmath170 ) and @xmath7-meson wave function parameter @xmath20 based on available experimental data for @xmath7 @xmath1 @xmath10 , @xmath23 and @xmath11 decays . our main conclusions and findings are summarized as : * the @xmath201 c.l . allowed region of @xmath171 is entirely different from that of @xmath173 . this fact means that the traditional qcdf treatment @xmath202 as universal parameters for different annihilation topologies might be unapplicable to hadronic @xmath7 decays . * the current experimental data on @xmath7 @xmath1 @xmath10 , @xmath23 and @xmath11 decays seems to favor a large value of @xmath142 @xmath40 2.9 , which corresponds to a sizable nonfactorizable annihilation contributions . but the range of @xmath171 is still very large , because the measurement precision of @xmath24 asymmetries is low now . * there are two possible choices for parameters @xmath173 . one is @xmath173 @xmath40 @xmath203 , the other is @xmath173 @xmath40 @xmath204 . these two choices correspond to similar factorizable annihilation contributions , although the qcdf approach tends to have a small value of @xmath150 @xcite . the space for @xmath173 is relatively tight due to the well measured branching ratios for @xmath7 @xmath1 @xmath10 , @xmath23 and @xmath11 decays . * the spectator scattering corrections play an important role in resolving both @xmath23 and @xmath10 puzzles . within qcdf approach , the spectator scattering amplitudes depend on parameters @xmath172 and @xmath7-meson wave function parameter @xmath20 . in our analysis , the approximation @xmath172 @xmath34 @xmath171 is assumed , which is proven to be a good simplification by a global fit in scenario iii . a small value of @xmath20 @xmath175 350 mev at the 95% c.l . is obtained by the global fit on @xmath7 @xmath1 @xmath10 , @xmath23 and @xmath11 decays , which needs to be further tested by future improved measurement on @xmath7 @xmath1 @xmath205 decays . an enhanced color - suppressed tree coefficient @xmath139 , which is supported by both large value of @xmath129 @xmath40 2.9 and small value of @xmath20 @xmath40 200 mev , is helpful to reconcile discrepancies on @xmath138 and @xmath76 between qcdf approach and experiments . the spectator scattering and annihilation contributions can offer significant corrections to observables of hadronic @xmath7 decays , and deserve intensive research especially when we apply the qcdf approach to the penguin - dominated , color - suppressed tree , and pure annihilation nonleptonic @xmath7 decays . as suggested in ref.@xcite and proofed by the pqcd approach @xcite , different parameters corresponding to different topologies should be introduced to regulate the endpoint divergences in spectator scattering and annihilation amplitudes within qcdf approach , even parameters reflecting the flavor symmetry - breaking effects should be considered for @xmath48 decays @xcite . this treatment might provide possible solution to `` problematic '' discrepancies between qcdf results and available measurements . of course , a fine - tuning of these parameters is required to be compatible with the experimental constraints . with the running lhcb and the upcoming superkekb experiments , more refined measurements on @xmath7-meson decays can be obtained , which will provide more powerful grounds to test various approach and confirm or refute some theoretical hypotheses . this work is supported by national natural science foundation of china under grant nos . 11147008 , 11105043 and u1232101 , 11475055 . q. chang is also supported by foundation for the author of national excellent doctoral dissertation of p. r. china under grant no . 201317 , research fund for the doctoral program of higher education of china under grant no . 20114104120002 and program for science and technology innovation talents in universities of henan province ( grant no . 14hastit036 ) . the annihilation amplitudes for two - body nonleptonic @xmath7 @xmath1 @xmath206 decays ( here @xmath207 denotes the light pseudoscalar meson ) can be expressed as the following building blocks @xcite , @xmath208 + r_\chi^{m_1 } r_\chi^{m_2 } \frac{2 \phi_{m_2}^{p}(x ) \phi_{m_1}^{p}(y ) } { \bar{x}y } \big\ } \label{ai1 } , \\ % % % % % % % % % % % % % % % a_2^i & = & \pi \alpha_s \int_0 ^ 1 dx dy \big\ { \phi_{m_2}^{a}(x ) \phi_{m_1}^{a}(y ) \big [ \frac{1}{\bar x(1-x \bar{y } ) } + \frac{1}{\bar{x } y^2 } \big ] + r_\chi^{m_1 } r_\chi^{m_2 } \frac{2 \phi_{m_2}^{p}(x ) \phi_{m_1}^{p}(y ) } { \bar{x } y } \big\ } \label{ai2 } , \\ % % % % % % % % % % % % % % % a_3^i & = & \pi \alpha_s \int_0 ^ 1 dx dy \big\ { r_\chi^{m_1 } \frac{2 \bar{y}\ \phi_{m_2}^{a}(x ) \phi_{m_1}^{p}(y ) } { \bar{x}y(1-x\bar{y } ) } - r_\chi^{m_2 } \frac{2x\ \phi_{m_1}^{a}(y ) \phi_{m_2}^{p}(x ) } { \bar{x}y(1-x\bar{y } ) } \big\ } \label{ai3 } , \\ % % % % % % % % % % % % % % % a_1^f & = & a_2^f = 0 \label{af12 } , \\ % % % % % % % % % % % % % % % a_3^f & = & \pi \alpha_s \int_0 ^ 1 dx dy \big\ { r_\chi^{m_1 } \frac{2(1+\bar x)\ \phi_{m_2}^{a}(x ) \phi_{m_1}^{p}(y ) } { \bar{x}^2 y } + r_\chi^{m_2 } \frac{2(1+y)\ \phi_{m_1}^{a}(y ) \phi_{m_2}^{p}(x ) } { \bar{x } y^2 } \big\ } \label{af3 } , \end{aligned}\ ] ] where the subscripts @xmath209 on @xmath210 correspond to three possible dirac current structures , namely , @xmath209 @xmath34 @xmath41 , @xmath55 , @xmath211 for @xmath212 , @xmath213 , @xmath214 , respectively . @xmath215 @xmath34 @xmath216 , where @xmath217 are the current quark mass of the pseudoscalar meson with mass @xmath218 . @xmath219 and @xmath220 are the twist-2 and twist-3 light - cone distribution amplitudes , respectively . their asymptotic forms are @xmath221 @xmath34 @xmath222 and @xmath223 @xmath34 @xmath41 . the spectator scattering corrections are given by @xcite @xmath224 , \\ \qquad \qquad \qquad \text{for } \ , i=1,2,3,4,9,10 \\ \displaystyle -\frac{b_{m_1 m_2}}{a_{m_1 m_2 } } { \int}_{0}^{1}d{\xi } \frac { \phi_{b}(\xi ) } { \xi } \int_0 ^ 1 dx dy \big [ \frac { \phi_{m_2}^{a}(x ) \phi_{m_1}^{a}(y ) } { x \bar{y } } + r_\chi^{m_1 } \frac { \phi_{m_2}^{a}(x ) \phi_{m_1}^{p}(y ) } { \bar{x } \bar{y } } \big ] , \\ \qquad \qquad \qquad \text{for } \ , i=5,7 \\ 0 , \qquad \qquad \quad \text{for } \ , i=6,8 \end{array } \right . \label{hardblock}\ ] ] where the factorized matrix elements are parameterized as @xcite @xmath225 for the ckm matrix elements , we adopt the fitting results for the wolfenstein parameters given by the ckmfitter group @xcite @xmath226 the pole masses of quarks are @xcite @xmath227 the running masses of quarks are @xcite @xmath228 the decay constants of @xmath7-meson and light mesons are @xcite @xmath229 we take the following heavy - to - light transition form factors @xcite @xmath230 moreover , for the gegenbauer coefficients , we take @xcite @xmath231 for the other inputs , such as the masses and lifetimes of mesons and so on , we take their central values given by pdg @xcite . our fit is performed in a simple way , which is similar to the one adopted in ref.@xcite based on the frequentist framework . considering a set of @xmath232 observables @xmath233 , the experimental measurements are assumed to be gaussian distributed with the mean value @xmath234 and error @xmath235 . the theoretical prediction @xmath236 for each observable could be treated as a function of a set of `` unknown '' free parameters @xmath237 ( here @xmath238 @xmath34 @xmath169 , @xmath170 and @xmath20 in this paper ) . to estimate the values of `` unknown '' parameters @xmath237 and compare the theoretical results @xmath236 with the experimental data @xmath234 , typically , it is need to construct a @xmath239 function as @xmath240 in the evaluation of @xmath236 for hadronic b decays , ones always encounter theoretical uncertainties induced by input parameters , like form factor and decay constant , whose probability distribution is unknown . following the treatment of rfit scheme @xcite that input values are treated on an equal footing , irrespective of how close they are from the edge of the allowed range , the @xmath239 function is modified as @xcite @xmath241 -f_{j\,\rm exp})^2}{\sigma_{j\,\rm exp}^2 } & \quad \text{if } f_{j\,\rm exp } < [ f_{j\,\rm theo}-\delta_{j\,\rm theo,\,sub } ] , \\ \displaystyle \frac{(f_{j\,\rm exp}- [ f_{j\,\rm theo}+\delta_{j\,\rm theo,\,sup}])^2 } { \sigma_{j\,\rm exp}^2 } & \quad \text{if } f_{j\,\rm exp } > [ f_{j\,\rm theo}+\delta_{j\,\rm theo,\,sup } ] , \\ 0 & \quad \text{otherwise } \end{array } \right . \label{chi2ii}\ ] ] where @xmath242 and @xmath243 denote asymmetric theoretical uncertainties , and are defined as @xmath244 . as to the asymmetric experimental errors , we choose the larger one as weighting factor . correspondingly , the confidence levels are defined by the function @xmath245 with @xmath246 @xmath34 @xmath239 @xmath247 @xmath248 and @xmath249 the number of degrees of freedom of free parameters . with the input parameters summarized in appendix [ app02 ] , we scan the space of the parameters @xmath238 and calculate the theoretical results @xmath236 . the @xmath250 could be obtained with eq.([chi2ii ] ) . the numerical results at @xmath251 and @xmath252 confidence levels are gotten from eq.([clfun ] ) by taking @xmath253 @xmath34 @xmath254 and @xmath253 @xmath34 @xmath255 , respectively . y. y. keum , h. n. li and a.i . sanda , phys . b * 504 * ( 2001 ) 6 [ arxiv : hep - ph/0004004 ] ; y. y. keum , h. n. li and a.i . sanda , phys . d * 63 * ( 2001 ) 054008 [ arxiv : hep - ph/0004173 ] ; c. d. lu , k. ukai and m .- z . yang , phys . d * 63 * ( 2001 ) 074009 [ arxiv : hep - ph/0004213 ] . a. ali , g. kramer , y. li , c. d. lu , y. l. shen , w. wang and y. m. wang , phys . d * 76 * ( 2007 ) 074018 [ arxiv : hep - ph/0703162 ] ; y. li and c. d. lu , commun . * 44 * ( 2005 ) 659 [ arxiv : hep - ph/0502038 ] . d. du , h , gong , j. sun , d. yang , and g. zhu , phys . d * 65 * ( 2002 ) 074001 [ arxiv : hep - ph/0108141 ] ; phys . d * 65 * ( 2002 ) 094025 erratum , ibid . * 66 * ( 2002 ) 079904 [ arxiv : hep - ph/0201253 ] ; j. sun , g. zhu and d. du , phys . d * 68 * ( 2003 ) 054003 [ arxiv : hep - ph/0211154 ] . m. beneke , g. buchalla , m. neubert and c. t. sachrajda , nucl . b * 606 * ( 2001 ) 245 [ arxiv : hep - ph/0104110 ] ; m. beneke and m. neubert , nucl . b * 651 * ( 2003 ) 225 [ arxiv : hep - ph/0210085 ] ; m. beneke and m. neubert , nucl . b * 675 * ( 2003 ) 333 [ arxiv : hep - ph/0308039 ] . m. beneke , s. jger , nucl . b * 751 * ( 2006 ) 160 [ arxiv : hep - ph/0512351 ] ; g. bell , v. pilipp , phys . d * 80 * ( 2009 ) 054024 [ arxiv:0907.1016 ] ; m. beneke , t. huber , x. q. li , nucl . phys . b * 832 * ( 2010 ) 109 [ arxiv:0911.3655 ] .

black hole mimickers are possible alternatives to black holes , they would look observationally almost like black holes but would have no horizon . the properties in the near - horizon region where gravity is strong can be quite different for both type of objects , but at infinity it could be difficult to discern black holes from their mimickers . to disentangle this possible confusion , we examine the near - horizon properties , and their connection with far away asymptotic properties , of some candidates to black mimickers . we study spherically symmetric uncharged or charged but non - extremal objects , as well as spherically symmetric charged extremal objects . within the uncharged or charged but non - extremal black hole mimickers , we study non - extremal-wormholes on the threshold of the formation of an event horizon , of which a subclass are called black foils , and gravastars . within the charged extremal black hole mimickers we study extremal-wormholes on the threshold of the formation of an event horizon , quasi - black holes , and wormholes on the basis of quasi - black holes from bonnor stars . we elucidate , whether or not the objects belonging to these two classes remain regular in the near - horizon limit . the requirement of full regularity , i.e. , finite curvature and absence of naked behavior , up to an arbitrary neighborhood of the gravitational radius of the object enables one to rule out potential mimickers in most of the cases . a list ranking the best black hole mimickers up to the worse , both non - extremal and extremal , is as follows : wormholes on the basis of extremal black holes or on the basis of quasi - black holes , quasi - black holes , wormholes on the basis of non - extremal black holes ( black foils ) , and gravastars . since , in observational astrophysics it is difficult to find extremal configurations ( the best mimickers in the ranking ) , whereas non - extremal configurations are really bad mimickers , the task of distinguishing black holes from their mimickers seems to be less difficult than one could think of it .

0806.0845

in recent years , it has been debated in the literature about possible alternatives to black holes , the black hole mimickers , which would look observationally almost like black holes but would have no horizon . the existence of such objects can , in principle , put in doubt astrophysical data which otherwise are considered as observational confirmation in favor of black holes @xcite . on one hand , it is clear that the properties in the near - horizon region where gravity is strong can be quite different for both type of objects . on the other hand , the statements about the difficulties in discerning black holes from their mimickers are usually related to measurements at spatial infinity . thus , one should insist on the question : can an observer at infinity catch the difference between both type of objects in some indirect way , or even rule out some possible mimicker ? in our view , the answer is positive and is connected with key properties , namely , regularity or singularity , of the corresponding geometries . it turns out that the requirement of full regularity up to an arbitrary neighborhood of the gravitational radius of the object enables one to rule out the potential mimickers in most of the cases . the goal of the present work is to examine the near - horizon properties , and their connection with far away asymptotic properties , of some candidates to black mimickers . we study spherically symmetric configurations , and make , two major divisions , or classes , on those candidates . first , uncharged or charged but non - extremal objects , and second extremal objects . within the uncharged or charged but non - extremal one can invoke as black hole mimickers , non - extremal @xmath0-wormholes on the threshold of the formation of an event horizon , some of which are called black foils @xcite ( see @xcite for the construction with other purposes of @xmath0-wormholes , which actually can also act as mimickers ) , and gravastars @xcite . within the extremal charged class one can invoke extremal @xmath2-wormholes on the threshold of the formation of an event horizon , quasi - black holes @xcite ( see also @xcite ) , and wormholes on the basis of quasi - black holes from bonnor stars , to name a few . we want to elucidate , whether or not the objects belonging to these two classes remain regular in the near - horizon limit . the arguments of @xcite which rule out non - extremal limiting configurations as becoming singular do not apply to the wormhole case @xcite . thus , we carry out the corresponding analysis anew for both classes of objects . for our purposes we write a generic spherically symmetric metric as @xmath3 where @xmath4 is the radial coordinate , and the @xmath5 are generic parameters , that depend on each situation we are studying . for instance when treating wormholes one has @xmath6 , and @xmath7 , such that @xmath8 and @xmath9 . here @xmath10 is the radius of the would be horizon when @xmath11 is zero , @xmath12 is the radius of the other possible horizon , @xmath11 is in principle a small quantity , and @xmath13 is the radius of a possible matter shell , satisfying @xmath14 . in other situations , e.g. , for gravastars , one has @xmath15 and , possibly , the energy density and the pressure should be somehow included . for the metric ( [ 1 ] ) the components of the riemann tensor in an orthonormal frame , the hat frame , are equal to , @xmath16 @xmath17 @xmath18 @xmath19 where a prime denotes derivative with respect to @xmath4 . here these components of the riemann tensor have a simple physical meaning . the @xmath20 component in eq . ( [ k(r ) ] ) yields the radial geodesic deviation , the @xmath21 component in eq . ( [ n(r ) ] ) yields the angular deviation , and analogously for the @xmath22 and @xmath23 components in eqs . ( [ f(r ) ] ) and ( [ h(r ) ] ) . in summary , they describe the deviation of geodesics in the corresponding directions . in general , forcing a matching at @xmath13 , with @xmath24 , surface stresses @xmath25 appear , which , in a coordinate frame , are equal to @xcite @xmath26 \text { , } \label{00}\ ] ] @xmath27 + \left ( \frac { d\phi } { dl}\right ) _ { + } -\left ( \frac{d\phi } { dl}\right ) _ { -}\ , , \label{22}\ ] ] @xmath28 being equal to @xmath29 , and @xmath30 being the proper radial distance . now , if metric ( [ 1 ] ) represents a wormhole , than the areal radius @xmath31 should have a local minimum at the throat . thus , we have two branches emerging out of the minimum radius , one with @xmath32 and the other with @xmath33 . then , @xmath34 @xmath35 there are also the bulk stress - energy components , but those do not interest us here and do not need computation . in some situations one has to deal here with naked behavior . this means there are cases in which the kretschmann scalar and other curvature quantities are finite on the horizon in a static coordinate system , but some of those quantities may blow up in a freely falling frame . such a kind of behavior is called naked behavior , and many instances of it have been found @xcite . another example is with quasi - black holes @xcite . one of the features typical of quasi - black holes , consists in precisely showing naked behavior on and beyond the quasihorizon surface . in addition , metrics obtained by gluing two spacetimes , can have a similar behavior , but now the surface stresses , which are finite in a static coordinate frame , blow up in a free - falling frame . thus , since we have found in eqs . ( [ k(r)])-([22 ] ) the curvature and surface stresses in a static frame for the spacetime in question , we now examine the behavior of the same quantities for a free - falling frame . consider then a radial local boost from a static frame with four - velocity @xmath36 to a free - falling frame with the velocity @xmath37 . under a boost the four - velocity transforms according to @xmath38 , where the orthonormal vector @xmath39 is pointing in the radial outward direction , and @xmath40 is the velocity boost parameter . in relation to the tidal forces in the bulk , the curvature components ( [ k(r)])-([n(r ) ] ) in the orthonormal basis responsible for tidal forces transform according to @xmath41 @xmath42 where a bar means a quantity evaluated in the freely falling frame , @xmath43 ( see equation ( [ h(r ) ] ) for the definition of @xmath44 ) , @xmath45 , and @xmath46 is the energy of the particle frame ( see e.g. @xcite for more details ) . the most interesting situation arises when @xmath47 is finite ( so , the kretschmann scalar is also finite ) but @xmath48 diverges . the corresponding horizons can be called truly naked @xcite . in relation to the surface stresses , it is useful to define the quantity @xmath49 which represents the energy density of the shell as observed by the observer with the four - velocity @xmath50 . in a static frame @xmath51 . then , considering a boosted motion along a radial geodesic with energy @xmath46 , one obtains @xmath52 this is a useful expression for analyzing naked behavior of wormholes and other objects . for the wormhole case it reduces to @xmath53 where @xmath13 is the radius at which the shell is located , and we took into account eq . ( [ s00 ] ) . there are many ways of making wormholes @xcite . in this section we are interested in making wormholes from charged metrics , more general than the reissner - nordstrm metric , but for certain choices the metrics can be reduced to the reissner - nordstrm metric . even for this metric one probably can think of many manners of making wormholes . we are interested in two different ways , that easily lead to the threshold of black hole formation , and the discussion of how they mimic black holes . then we compound both ways into one single way . the first way is the surgery approach , see ( see in particular section 15.2.1 of @xcite ) . pick up a spherically symmetric metric of the form @xmath55 , where @xmath10 is the radius of the would be horizon , @xmath12 is the radius of the other possible horizon , and @xmath13 is the radius of a possible matter shell , satisfying @xmath56 . take for instance the non - extremal reissner - nordstrm metric , where @xmath57 , with @xmath58 and @xmath59 , @xmath60 and @xmath61 being the mass and electrical charge of the object , respectively , and @xmath62 is newton s constant ( we use @xmath63 ) . cut the metric at some @xmath13 and join the resulting spacetime with a symmetric branch . this is a non - extremal reissner - nordstrm surgery ( the schwarzschild surgery , with no charge and so @xmath64 , being a particular case of this ) , resulting in a non - extremal wormhole with a thin shell of matter at @xmath13 , the throat . in brief , one places at some radius , @xmath13 , a thin shell which separates two regions , with non - extremal geometries , with @xmath13 also defining the throat . then , one introduces another radial coordinate @xmath30 , such that @xmath65 with @xmath66 , and which covers the whole of the manifold , @xmath67 . the function @xmath68 is monotonically decreasing for the branch @xmath69 , which we call the @xmath70 branch , and monotonically increasing for @xmath71 , the + " branch . in general , giving this construction , surface stresses @xmath25 appear . a second way , i.e. , another approach , to build wormholes , is through metrics of the type @xmath72 , where @xmath0 is a small quantity , @xmath73 is the radius of the would be horizon when @xmath0 is zero , and @xmath74 is the radius of the other possible horizon @xcite . metrics of this type , depending on the parameter @xmath0 , can be generically call @xmath0-spacetimes , which in special cases can become wormholes , i.e. , @xmath0-wormholes . in @xcite the model with metric @xmath75 , was considered , where thus @xmath76 , with @xmath77 being chosen appropriately . in turn , in @xcite the model with metric @xmath78 , was considered , where thus @xmath79 , with @xmath77 being chosen appropriately , and with @xmath80 being an additional parameter . thus , a generic @xmath0-metric of the type given above yields a generic spacetime that comprehends the two cited models , one model studied in @xcite , the other in @xcite . one can calculate the riemann tensor for the @xmath0-metric , and of course , since spacetime is not empty , there is a smooth energy - momentum tensor associated to the metric but we do not need to calculate it here . all these @xmath0-spacetimes are smooth . now , take the @xmath0-metric to construct a wormhole , i.e , an @xmath0-wormhole . since for the construction we need to impose some more conditions , in particular on the the potentials @xmath81 and @xmath77 of the @xmath0-metric , let us adopt the following approach . first , as above , one introduces the radial coordinate @xmath30 , such that @xmath65 and @xmath82 . the function @xmath68 is monotonically decreasing for the @xmath83 branch , @xmath69 , and monotonically increasing for the @xmath84 branch , @xmath71 . second , the dependence of the function @xmath81 on the parameter @xmath0 , @xmath85 , which can be of the type of the models considered above @xcite , is such that @xmath86 , and the dependence of the function @xmath77 on the parameter @xmath0 is also such that @xmath87 , so in the limit @xmath88 the original wormhole configuration indeed approaches a black hole . third , if the first derivative @xmath89 is continuous at the throat , we have @xmath90 ( see @xcite ) . when @xmath77 does not depend on @xmath0 at all , the throat is situated on the possible would be horizon . this is the approach used in @xcite to build a wormhole . this approach is smooth as long as @xmath91 . as a particular instance of this approach , one can choose @xmath77 as being reissner - nordstrm , @xmath92 , as usual . for @xmath93 one has a non - extremal choice for @xmath77 , the case @xmath94 , i.e. , @xmath95 , yielding the schwarzschild potential @xmath77 as a particular case , the one chosen in @xcite . such @xmath2-wormholes have been called foils in @xcite . when @xmath96 we have the full non - extremal reissner - nordstrm metric . so , let us compound both approaches , the surgery approach of @xcite , and the @xmath0 approach of @xcite . write then a generic @xmath2-metric with surgery as @xmath97 i.e. , from ( [ 1 ] ) one chooses @xmath98 where @xmath0 is a small quantity , @xmath10 is the radius of the would - be horizon ( if @xmath99 ) , and @xmath13 is the radius of a possible matter shell , satisfying @xmath56 . since we are studying here non - extremal metrics we have , when @xmath100 , that @xmath101 and @xmath102 . essentially , what we have done is a surgery on @xmath11-metrics , non - extremal reissner - nordstrm ( with schwarzschild included ) metrics being particular @xmath100 instances . this can also be thought of as a one - parametric deformation of the original @xmath11-wormhole metric by gluing two branches at the throat @xmath13 , with @xmath24 . now , given the general @xmath11-metric with surgery , ( [ 1 ] ) and ( [ epsilonwormpot ] ) , and the correspondent wormhole construction , we are interested in getting a spacetime that mimics a black hole . it is then not hard to understand that there are two distinct situations to obtain spacetimes on the threshold of being black holes . if there is no shell , then wormholes approach black holes when @xmath103 . if there is a shell but @xmath104 then the wormhole throat approaches the horizon when @xmath105 . therefore , there is a play of two small parameters @xmath11 and @xmath106 and the limiting procedure should be considered with great care . it gives rise to two distinct situations , depending on the order one takes the limiting procedures . situation bt : this situation is achieved by , in the end turning the wormhole metric into the metric of a black hole ( b ) ( i.e. , taking @xmath103 as the last operation ) , after first having moved the shell towards the minimum throat ( t ) radius ( i.e. , taking @xmath105 as the initial operation ) . formally , this means taking the limits in the following order @xmath107 . situation tb : this situation is achieved by , in the end the location of the shell approaches the throat ( t ) ( i.e. , taking @xmath105 as the last operation ) , after first turning the wormhole metric into a black hole ( b ) ( i.e. , taking @xmath103 as the initial operation ) . formally , this means taking the limits in the following order @xmath108 . note the case considered in @xcite for the metric ( [ 1 ] ) is a particular instance of the bt situation , since there @xmath109 always , and one only takes the @xmath103 limit . so the situation bt is the one that yields black foils , following the nomenclature of ds . one can calculate from equations ( [ n(r)])-([h(r ) ] ) that the components @xmath21 , @xmath22 and @xmath23 of the riemann tensor are always finite , and from equation ( [ 00 ] ) that both limits when applied to @xmath110 give zero , i.e. , @xmath111 . but for the quantities @xmath47 , @xmath112 and @xmath113 ( see equations ( [ k(r ) ] ) , ( [ 22 ] ) and ( [ sigmacapital ] ) ) the situation may be different depending on the order one takes the limits . now , as we have been seeing , in treating this problem there are many levels of distinction . first , we can specify two models of @xmath0-metrics with surgery which depend on the parameter @xmath0 and @xmath13 , namely the model considered in @xcite , where @xmath114 , and @xmath77 is non - extremal , or some appropriate generalization of it , which we will call model 1 , and the model given in @xcite , where @xmath115 and @xmath77 non - extremal , or some appropriate generalization of it , which we will call model 2 . second , within each of the two cases provided by model 1 and model 2 , we should study the situations bt and tb . furthermore , as we want to examine the regularity of the system under discussion , the relevant quantities which we are going to calculate are the spacetime curvature components , and the surface stresses which appear on the glued boundary , i.e. , the shell . we will also study the naked behavior of each case . so , within each situation we have to study the behavior of the scalars , and in addition the naked behavior . thus we have eight distinct cases to analyze . we consider these eight cases , each in turn . here we consider the one - parametric deformation , eqs . ( [ 1 ] ) and ( [ epsilonwormpot ] ) , such that for @xmath99 our metric represents the gluing of two non - extremal black holes . note that in this non - extremal case the function @xmath77 of the metric ( [ 1 ] ) does not need to contain the parameter @xmath0 , so we put @xmath116 . model 1 _ let the metric have the form ( [ 1 ] ) together with ( [ epsilonwormpot ] ) . for model 1 choose the metric potentials as @xmath117 where @xmath118 can be any function that satisfies @xmath101 and @xmath119 . for instance @xmath77 can be reissner - nordstrm , @xmath120 with @xmath121 and @xmath59 , with the case @xmath64 being the schwarzschild case , the one chosen in @xcite . let us now work out generically the general behavior of this model , i.e. , how the curvature and stress - tensor quantities behave , and also work out generically the naked behavior . then we apply these behaviors to the two situations bt and tb . in doing so , we will display the properties of the quantities @xmath47 , @xmath122 , which characterize regular or singular general behavior , and the properties of the quantities @xmath48 , @xmath113 which characterize non - naked or naked behavior . as for understanding the general behavior note that , from equations ( [ k(r ) ] ) and ( [ 22 ] ) , explicit calculations give @xmath123 and @xmath124 in relation to naked behavior , note that since from ( [ n(r ) ] ) one has @xmath125 , and from ( [ h(r ) ] ) one has @xmath126 , one finds @xmath127 . thus , from ( [ barn ] ) @xmath128 \ , . \label{newbarn}\ ] ] with this we can now study the situations bt and tb . 0.4 cm _ situation bt : _ as for the general behavior , one has that for a non - extremal system @xmath101 and @xmath119 . then , it follows from equation ( [ e ] ) that @xmath129 and so , @xmath130 correspondingly , the kretschmann scalar @xmath131 also diverges . it was shown in @xcite for the choice @xmath132 that for @xmath133 there exist geodesics which have no analogue for the schwarzschild black hole metric . the timelike particles which move along them oscillate between turning points which are situated at different sides of the throat . however , the problem is that in the limit @xmath103 these geodesics pass through a region of a strong gravitational field . this gives rise to tidal forces in the radial direction which are of order @xmath134 . if @xmath135is exponentially small @xcite , the tidal forces are exponentially large . in addition , from equation ( [ 22e ] ) , it follows that @xmath136 now let us analyze the naked behavior . here one has , @xmath137 , @xmath138 . so , @xmath139 , @xmath140 . thus , there is infinite contraction in the longitudinal direction and infinite transversal stretching . moreover , since @xmath101 and @xmath141 , we obtain immediately from ( [ sigmacapital2 ] ) and ( [ efv ] ) that in the situation bt one has @xmath142 0.4 cm _ situation tb : _ as for the general behavior , now one finds , @xmath143 . so , @xmath144 a result equal to that of a black hole . also , @xmath145 , so @xmath146 note that equation ( [ inf ] ) is in agreement with the behavior of surface stresses of a wormhole obtained by gluing two copies of the schwarzschild metric ( see equation 15.46 of @xcite ) . now let us analyze the naked behavior . one has , @xmath147 , so @xmath48 is finite and negative . thus , one obtains finite deformation in both directions . moreover , it also follows from ( [ sigmacapital ] ) that in the situation tb @xmath148 i.e. , @xmath113 diverges . thus , a free - falling observer encounters diverging surface energy density . the same conclusion applies to the flux @xmath149 . concluding here model 1 , we can say that there are two nonequivalent limits but each of them is bad " in that in the bt situation the kretschmann scalar diverges , whereas in the tb situation it is the surface stresses that diverge . 0.5 cm 7.4 cm _ _ model 2 _ _ 0.1 cm let the metric have the form ( [ 1 ] ) together with ( [ epsilonwormpot ] ) . for model 2 choose the metric potentials as @xmath150 with @xmath151 and @xmath0 being parameters . in @xcite the model with @xmath152 , was considered , in which case , when @xmath99 one has the schwarzschild metric , see also @xcite . here , @xmath153 can be any function that satisfies @xmath101 and @xmath154 , a typical example being the reissner - nordstrm @xmath77 potential . as for the general behavior , again after some calculations , we obtain @xmath155 and @xmath156 now let us analyze the naked behavior . also , again after some calculations , we obtain @xmath157 , so it follows from ( [ barn ] ) that @xmath158 \label{n2}\ ] ] with this we can now study the situations bt and tb . 0.4 cm _ situation bt : _ as for the general behavior , in this situation one finds , @xmath159 and @xmath160 now let us analyze the naked behavior . one has @xmath161 , and @xmath162 so one finds no longitudinal deformation , and an infinite transversal stretching . taking into account that @xmath101 , @xmath163\neq 0 $ ] , we obtain immediately from ( [ sigmacapital2 ] ) that in situation bt one finds @xmath164 . 0.4 cm _ situation tb : _ as for the general behavior in this situation , one finds , @xmath165 and @xmath166 now let us analyze the naked behavior . one finds , finite transversal stretching and finite longitudinal contraction ( @xmath167 ) or stretching ( @xmath168 ) . it follows from ( [ sigmacapital2 ] ) and ( [ efv ] ) that in the situation tb , @xmath169 i.e. , it diverges . thus , a free - falling observer encounters diverging surface energy density in the situation tb . the same conclusion applies to the flux @xmath170 . concluding here model 2 , we can say that in both situations , bt and tb , the curvature components remain finite but the limiting values of @xmath47 do not coincide . moreover , in both situations the surface stresses diverge . 0.5 cm _ overall conclusions for models 1 and 2 _ 0.1 cm as an overall conclusion for the situation tb in models 1 and 2 , we find the results agree for both models . this can be explained and generalized as follows . since , in the situation tb , the limit @xmath88 is taken first , the dependence of the metric on @xmath0 drops out in the final expressions for the curvature and surface stresses , so any model gives the same result . in addition , assuming that for @xmath99 one has @xmath171 , and taking into account that @xmath101 and @xmath172 , we obtain immediately from ( [ sigmacapital2 ] ) that in the situation bt , @xmath164 , and in the situation tb , @xmath113 diverges . this holds independently of the kind of the model used . thus , a free - falling observer encounters diverging surface energy density in the situation tb . the same conclusion applies to the flux @xmath173 . in the non - extremal cases just studied , it turned out that each of the limits under discussion is singular : either the kretschmann scalar or surface stresses on the throat ( or both ) diverge . thus , the limit is singular . in other words , a black hole mimicker made from a wormhole , and in particular a black hole foil , is not smooth . it is convenient to summarize the results in a table shown in figure [ table ] . 3.7 cm [ width=16.0 cm ] table.eps 0.5 cm although the singular or almost singular behavior of these black hole mimickers based on @xmath0-wormholes casts doubts on their real existence , it is worth our while to study a little more on the effects of such mimickers on infalling sources and their detection by far away observers . intuitively it is clear that there should be some observational effects if infalling sources are distorted by stronger than normal tidal fields . first , we point out that indeed the fact that tidal forces grow unbound when @xmath103 , can be used , in principle , to distinguish a black hole from an @xmath0-wormhole which mimics it . suppose a small mass falling freely into a massive body , such as an @xmath0-wormhole . consider , for example , the situation bt . one can compare two approaches . in the first approach , one neglects the size of the small mass and considers the geodesic along which such a point - like small mass moves @xcite . then , if the throat is very close to the would - be horizon and subsequent pulses are emitted near the throat , the intervals of time measured at infinity , grow unbound as @xmath174 in the limit @xmath175 . so an observer at infinity can not distinguish the fall of matter into a wormhole with vanishingly small @xmath11 from absorption of matter by a black hole , if the interval of observation time is less than @xmath176 , by construction a very long interval , see @xcite for details . in the second approach , however , one takes the finiteness of the size of the free - falling small mass into account . then , the overall picture changes since due to the growing tidal forces the small mass gets deformed . if the small mass is a luminescent source , such a change can in principle be detected by an observer at infinity . moreover , such changes happen much quicker than the typical times needed to penetrate the immediate vicinity of a horizon . the intervals of time during which the size of a small mass changes can be estimated from eq . ( [ kinf ] ) and the geodesic deviation equation . for a process occurring near the throat it yields for the proper time the value @xmath177 , so that @xmath178 , as @xmath103 . the corresponding interval of time @xmath179 at infinity is @xmath180 is finite . therefore , in case of stretching , an observer at infinity would see an extended distorted image of an infalling small mass . if the tidal forces are infinite , as they can be in some of the cases shown above for @xmath0-wormholes , then the stretching is correspondingly large . of course , it is also possible that infinite contraction in some direction converts an extended small mass to zero size in this direction , see @xcite . for the usual black holes , tidal forces are also present but such forces are much weaker than for @xmath0-wormholes , where near the throat they can be as large as one likes for sufficiently small @xmath0 . thus , the key point is to look not to single geodesics as in @xcite , but to the separation between geodesics of the same congruence . such a separation delivers , to an observer at infinity , meaningful information about the region of strong gravity where a black hole or a @xmath0-wormhole can be situated , and such information has very different properties , which depend on the massive body in question . second , we indicate what further physical changes could be expected by performing some interesting estimates of the effects . here , one can take advantage of known results ( see @xcite , section 32.6 ) . suppose then that a box of small proper mass @xmath181 and proper length @xmath30 is freelly falling towards an object of mass @xmath60 . to simplify we consider an uncharged object , @xmath95 . from @xcite one finds that the radial tidal force exerted on the box is equal to @xmath182 where @xmath47 has the meaning of a tidal radial acceleration . indeed , @xmath47 is given precisely in eq . ( [ k(r ) ] ) . for the schwarzschild metric one has @xmath183 , so that near the horizon of a schwarzschild black hole , where @xmath184 , one obtains the value @xmath185 . however , for model 1 , if we correspondingly choose the metric potential as @xmath186 in the situation bt , it follows from ( [ kinf ] ) that near the would - be horizon one has @xmath187 . let us suppose that a distant observer is able to recover from observational data the value of @xmath47 . if the observer thinks that the measured value of @xmath47 is due to a schwarzschild black hole , he should ascribe a mass @xmath188 to it . however , if the object turns out to be a black hole mimicker , then , for the same @xmath47 , the value of the actual mass will be much greater , given by @xmath189 . if , in addition , the observer , knowing the mass @xmath60 of the object , in this case @xmath190 , insists in explaining it in terms of the usual black hole metric , he will find that @xmath191 , and will certainly start asking about the hidden mass " . this example shows that hidden mass " in some situations may arise simply because the metrics of a black hole or of a black hole mimicker were not properly discerned . as far as gravastars are concerned , they contain , by construction , a thin layer of normal matter with positive density @xmath192 and positive radial pressure @xmath193 on the border of the tension matter with vacuum . in the model suggested in @xcite the stiff matter equation of state with @xmath194 was chosen . then , it follows immediately from the field equations and the conservation laws that , as the border approaches the gravitational radius , the gradient of the pressure becomes infinite . one can , in general , admit discontinuous radial pressure , giving rise to a surface pressure on the boundary between matter and vacuum . however , this surface pressure and other surface stresses also grow unbound in the horizon limit . surely , the more the border approaches the horizon , the better a black hole mimicker it becomes , but , at the same time , the closer the system approaches the singular state . this does not exclude in advance the astrophysical significance of gravastars as compact vacuum - like geometries but it shows that they can hardly pretend to be good black hole mimickers . in more detail , to see that the surface stresses go unbound , we introduce the quantity @xmath195 , so that we can rewrite eq . ( [ 22 ] ) as @xmath196 + \frac { 1 } { b(r_{0})}\left[\left ( \frac{db}{dl}\right ) _ { + } -\left ( \frac{db } { dl}\right ) _ { -}\right]$ ] , where we have taken into account that the continuity of the first fundamental form demands @xmath197 . in the outer region we have the schwarzschild metric , so @xmath198 . let the radius of the shell approach that of the would - be horizon , i.e. , @xmath105 , and @xmath199 . by definition of a gravastar , actually there is no horizon in the system , so the function @xmath200 can not cross @xmath4 at @xmath10 at all . in the inner region , either @xmath201 or @xmath202 . thus , since there is a @xmath203 @xmath204 term in @xmath205 , and the other terms remain finite and non - zero , we find that @xmath206 . this makes the gravastar unphysical in the near - horizon limit . it is worth noting that this conclusion and its derivation is very close to the statement that quasi - black holes can not be non - extremal , if only finite surface stresses are allowed ( see sec . iv of @xcite ) . the only difference is that in the whole inner region @xmath207 everywhere for quasi - black holes whereas for gravastars @xmath208 can be non - zero there . however , in the present context , only the vicinity of the would - be horizon is relevant , so the conclusions are similar . in section [ nonextremalepsilonsurg ] on wormholes on the basis of nonextremal @xmath0-metrics with surgery , we have discussed wormhole configurations which mimic non - extremal black holes . here we consider the one - parametric deformation ( [ 1 ] ) such that for @xmath99 our metric represents the gluing of two extremal reissner - nordstm black holes . note that in the non - extremal case , when using metric ( [ 1 ] ) we have chosen a function @xmath77 which itself does not contain the parameter @xmath2 . however , now such a simple construction can not be implemented . our goal is to trace the relationship between a wormhole metric with a generic @xmath0 parameter ( @xmath209 ) , and a black hole metric ( @xmath99 ) . if we start from equation ( [ 1 ] ) with @xmath210 , i.e. , @xmath211 , we encounter the immediate difficulty that spacetime is geodesically complete and represents an infinitely long horn , so one can not speak about a wormhole at all . therefore , we should deform the extremal reissner - nordstrm metric in a somewhat different way and include the parameter @xmath0 not only into the function @xmath212 but into @xmath77 as well . let us make the simplest choice for @xmath213 , namely , @xmath214 @xmath215 being such that @xmath216 , and @xmath217 . at some @xmath24 , we glue two copies of the spacetime , the `` @xmath218 '' branch to the `` @xmath219 '' branch . then , the behavior of @xmath47 and @xmath122 follows from eqs . ( [ e ] ) and ( [ 22e ] ) . note that if @xmath220 and @xmath221 we return to the deformed non - extremal reissner - nordstrm case and the results ( [ einf])-([0 ] ) are reproduced , @xmath222 being the schwarzschild case . 0.3 cm 7.4 cm _ model 1 _ 0.1 cm using ( [ efv ] ) , i.e. , @xmath223 , and ( [ vne ] ) , we can study model 1 in the extremal case . 0.4 cm _ situation bt : _ as for the general behavior one finds , @xmath224 and , @xmath225 let us analyze the naked behavior . in principle , the quantity @xmath226 may be finite or infinite depending on the model for @xmath227 . from eq . ( [ eqk ] ) , longitudinal contraction is finite if @xmath40 is finite , or infinite if @xmath40 is infinite . the value of @xmath48 which determines transverse deformation can be found from eq . ( [ newbarn ] ) . then , we obtain that in the limit under discussion @xmath228 where @xmath229 . then we have transverse stretching which is finite if @xmath230 is finite , or infinite if @xmath231 is infinite . the stress component @xmath232 is finite , indeed zero , in the limit . situation tb : _ as for the general behavior one finds , @xmath233 @xmath234 note that when @xmath235 in the situation bt , the quantities @xmath47 for two situations coincide . let us analyze the naked behavior . here there is finite longitudinal contraction , and no transverse deformation since , according to eq . ( [ newbarn ] ) , @xmath236 in the limit under consideration . thus , the only manifestation of naked behavior is connected with the surface stresses . according to ( [ sigmacapital2 ] ) , the quantity @xmath237 behaves as @xmath238 i.e. , it diverges in situation tb . 0.3 cm 7.4 cm _ model 2 _ 0.1 cm using eq . ( [ phimodel2 ] ) , i.e. , @xmath239 , and ( [ vne ] ) , we can study model 2 in the extremal case . _ situation bt : _ as for the general behavior one finds , @xmath159 @xmath240 let us analyze the naked behavior . there is no deformation in the radial direction . the behavior of @xmath48 which is responsible for transverse deformation , can be obtained from ( [ n2 ] ) and coincides with ( [ nb ] ) in the limit under discussion . thus , again transverse stretching is finite if @xmath231 is finite , or infinite if @xmath231 is infinite . note that it follows from the definitions of @xmath40 and @xmath231 that @xmath241 . according to the definition ( [ vne ] ) of @xmath242 , @xmath243 . therefore , if @xmath231 is finite , @xmath244 , then @xmath47 , @xmath122 , @xmath48 and @xmath113 are finite ( moreover , @xmath245 ) , so naked behavior is absent in this case . on the other hand , as @xmath246 , in case @xmath247 the quantity @xmath231 diverges and so does @xmath48 . this means transverse stretching is infinite . situation tb : _ as for the general behavior one finds , @xmath233 @xmath248 let us analyze the naked behavior . one finds , finite contraction in the longitudinal direction and no transverse deformation . @xmath238 i.e. , it diverges . 0.5 cm _ overall conclusions for models 1 and 2 _ 0.1 cm as an overall conclusion , assuming that for @xmath100 , one has @xmath249 and taking into account that @xmath250 , @xmath163\neq 0 $ ] , we obtain immediately from ( [ sigmacapital2 ] ) that in the situation bt , @xmath164 , and in the situation tb , @xmath113 diverges . this holds independently of the kind of the model , and is valid for the non - extremal spacetimes , as well as for the extremal spacetimes . thus , a free - falling observer encounters diverging surface energy density in the situation tb . the same conclusion applies to the flux @xmath251 . 0.1 cm _ light - like shells _ 0.1 cm one essential feature of the almost extremal configurations under discussion consists in that we deal with shells which are timelike but become light - like in the limit when they are approaching the would - be horizon . a natural question arises : what happens if we start out our analysis with two extremal black hole spacetimes , i.e. , putting @xmath252 and @xmath253 , and the shell in - between the two spacetimes lies on the event horizon and so is lightlike ? more precisely , we are interested whether the stresses on the shell remain finite or become infinite . one can expect from our previous results that they are finite but this is not so obvious in advance . we have seen already that taking different limiting procedures in the near - horizon limit is a rather delicate issue . moreover , the formalism for lightlike shells @xcite somewhat differs from that for timelike ones , so we perform our analysis anew . we restrict ourselves to the case of taking two extremal reissner - nordstrm black holes . the reason for this choice is that only extremal black holes are good candidates for gluing without severe singularities , i.e. , the gluing procedure maintains a finite kretschmann scalar throughout the spacetime as well as finite stresses on the glued surface . this case is also a majumdar - papapetrou case , and could be analyzed in the next section as well , but since it is also a limit of what has been done above , namely the null limit of a timelike wormhole throat with two extremal external vacuum spacetimes , we discuss it now . to match two extremal reissner - nordstrm spacetimes at the light - like surface @xmath253 we follow the general formalism for light - like shells @xcite . we write the metric in kruskal - type coordinates @xmath254 let the surface @xmath253 correspond , say , to @xmath255 . then , the effective energy density is @xmath256 , ( see , e.g. , equation ( 3.99 ) of @xcite ) . here @xmath257n^{\alpha } $ ] , where the indexes @xmath84 and @xmath83 refer to the different sides of the shell , and the null vector @xmath258 is such that @xmath259 , with @xmath260 being a null tangent vector . now , the only nonvanishing components of @xmath261 and @xmath262 , are by construction @xmath263 and @xmath264 . since there is no rest frame in the null case , the measured energy density @xmath265 depends on the chosen observer . to check that it is finite and non - zero , it is sufficient to check that @xmath266 $ ] is finite and non - zero . for this purpose , it is sufficient to exploit the result of @xcite where it is shown that @xmath267 on the horizon . in our case , as we deal with two black holes instead of a single one , the coordinate @xmath4 increases on both sides of the shell . thus , @xmath268 has different signs and takes the values @xmath269 on each side . so indeed , the difference is equal to @xmath270 and is finite and non - zero . in general , there are other contributions to the effective stress - energy tensor of the shell due to effective pressures and currents , but it is easy to show that in our case they are absent . it is worth noting that the gluing in our case differs from the gluing between different extremal reissner - nordstrm black holes considered in @xcite , in that we replaced the usual metric inside the horizon by their @xmath83 branch . it also differs from the wormhole construction used in @xcite where two spheres were cut out from a vacuum majumdar - papapetrou system , or more precisely , they were cut from a single spacetime containing two extremal reissner - nordstrm black holes . 0.3 cm _ classical electron models _ 0.1 cm as a by - product of this light - like shell construction , and a very interesting one , we have just found a configuration that represents a regular wormhole configuration which is also a black hole . more important perhaps , in addition , it can serve as a classical model for an elementary particle in that ( i ) the system is characterized by a minimum number of fixed parameters like mass and charge and ( ii ) it is free of a central singularity inside . we are aware that it is not entirely of electromagnetic nature because it has stresses on the horizon , a kind of poincar stresses . but , anyway , such a surface stress , can be considered as a mild singularity when compared to the usual central singularity . in this model a free - falling observer can penetrate to the inside but , of course , can not return back to the original asymptotic region , due to the existence of a horizon . so the wormhole is an untraversable one . thus , the body under discussion combines features of an untraversable wormhole and a regular black hole , and can be called a worm - black hole . since the proper distant to the extremal horizon is infinite , such a hybrid construction is similar to the null wormholes , or n - wormholes for short , see @xcite . note that it differs from configurations representing quasi - black holes . for example , in some quasi - black hole models , see @xcite , there is a region @xmath271 which becomes degenerate in the quasihorizon limit . this region is missing in our worm - black hole model , since the black hole metric beyond the horizon is replaced by a branch with an areal radius that grows away from the horizon . it also differs from the model considered in @xcite where the external reissner - nordstrm part was glued to the bertotti - robinson metric that leads to surface stresses that vanish in the horizon limit in the static frame but grow unbound in the free - falling one , and as a result , the inner region becomes impenetrable for a free - falling observer ( see @xcite for details ) . another candidate for the role of a black hole mimicker is a quasi - black hole . roughly speaking , it is an extremal object that appears when the system approaches the quasihorizon as nearly as one likes , along a family of quasi - static configuration . it was argued in @xcite that such a limit can correspond to an extremal quasihorizon only if we restrict ourselves to static configurations which are regular in the strong sense . the latter means that the kretschmann scalar should be finite everywhere in the system , and surface stresses at the quasihorizon should be finite as well . there are subtleties in the non - trivial relation between regular and singular features of quasi - black holes @xcite . for example , the whole region can look degenerate from the viewpoint of a distant observer and , nevertheless , the kretschmann scalar remains finite in that region . in more detail , consider the static spherically symmetric metric ( [ 1 ] ) , and let it represent a spacetime in which there is an inner matter configuration , attached to an asymptotic flat exterior region . the @xmath5 in ( [ 1 ] ) stand for the radius @xmath13 of the configurations and possibly some other parameters connected with the particular object one is analyzing . for instance , the parameter @xmath0 also enters in the analysis , and here has a slightly altered meaning . it means a small deviation from a quasi - black hole , rather than from a black hole solution as in section [ nonextremalepsilonsurg ] , see also below . suppose the spacetime in question has the following properties : ( a ) the function @xmath118 in ( [ 1 ] ) attains a minimum at some @xmath272 , such that @xmath273 , with @xmath274 , this minimum being achieved either from both sides of @xmath275 or from @xmath276 alone , ( b ) for such a small but nonzero @xmath0 the configuration is regular everywhere with a nonvanishing metric function @xmath277 , at most the metric contains only delta - function like shells , and ( c ) in the limit @xmath278 the metric coefficient @xmath279 for all @xmath280 . these three features define a quasi - black hole . in turn , these three features imply that , there are infinite redshift whole regions when @xmath281 , a free - falling observer finds in his own frame infinitely large tidal forces in the whole inner region , showing thus naked behavior , although the curvature scalars are finite . moreover it has some form of degeneracy since , although the spacetime curvature invariants remain perfectly regular everywhere , in the limit , outer and inner regions become mutually impenetrable and disjoint . for a free - falling external nearby observer it is as if a null singular horizon is being formed . for external far away observers the spacetime may be said to be naively indistinguishable from that of extremal black holes . however , if one makes experiments with infalling luminescent extended small masses , one might find differences , since as discussed previously @xcite , due to the naked behavior , quasi - black holes enlarge grossly the tidal forces on an infalling small mass when compared to the tiny effect of an extremal black hole on the same small mass . thus , as with the extremal @xmath0-metrics studied before , the naked behavior shows that quasi - black holes are not so good mimickers , as was previously thought , but they are still better than black foils @xcite where the singularity is more severe . a further important property is that quasi - black holes must be extremal . for a quasi - black hole the metric is well defined and everywhere regular . however , when @xmath99 , quasi - black hole spacetimes become degenerate , almost singular , see @xcite . the quasi - black hole is on the verge of forming an event horizon , but it never forms one , instead , a quasihorizon appears . in summary quasi - black holes have normal general behavior and singular naked behavior . quasi - black holes may appear from bonnor stars , i.e. , systems composed of extremal charged dust and vacuum , from self - gravitating higgs magnetic monopole systems , and from composite spacetimes even in the case of pure electrovacuum , in which these vacuum systems are composed of an exterior reissner - nordstrm part glued to an inner bertotti - robinson spacetime or of an exterior reissner - nordstrm part glued to an an inner minkowski spacetime , see @xcite for a full discussion and references . bonnor stars are majumdar - papapetrou matter systems with either a sharp or smooth boundary to an exterior vacuum . since bonnor stars are paradigmatic to understand the formation of quasi - black holes ( see @xcite ) , it is interesting to use those stars on the threshold of forming a quasi - black hole to understand whether wormholes on the basis of quasi - black holes which can be formed from bonnor stars , can mimic extremal black holes or not . this is an interesting variant , although with similarities , to wormholes on the basis of extremal @xmath0-metrics . we use bonnor stars , both in their compact version @xcite , as well as in their extended one @xcite . now , we start from the configuration which contains majumdar - papapetrou matter inside and vacuum outside , a compact bonnor star @xcite . let us have a compact object , a bonnor star , with extremal dust for @xmath282 , joined to an extremal reissner - nordstrm metric for @xmath283 . the potential @xmath284 can be written as @xmath285 with the mass density @xmath286 and the function @xmath287 being connected through @xmath288 . the function @xmath287 can be interpreted as the proper mass enclosed within a sphere of a radius @xmath4 . in addition , for extremal dust , one has @xmath289 , where @xmath290 is the electric charge within this sphere . at the boundary @xmath291 , one has @xmath292 , and @xmath293 , such that @xmath294 , @xmath60 and @xmath61 being the total mass and charge , respectively . to construct a wormhole , one can take the following procedure . cut the solution somewhere in the interior at some radius @xmath295 and discard the region @xmath296 . one obtains a majumdar - papapetrou matter region for @xmath297 , and a vacuum region for @xmath298 . for definiteness , let this spacetime be situated on the left . a symmetric right branch , also containing interior and exterior , is again cut at the radius @xmath299 , and glued to the symmetric left branch , producing thus a boundary shell at @xmath300 . then , it follows from the field equations that the left branch is given by @xmath301 \,,\quad r_{1}\leq r\leq r_{0}\ , , \notag \\ \exp ( \phi ) & = & 1-\frac{m}{r}\,,\quad r_{0}\leq r<\infty \ , , \label{phibon}\end{aligned}\ ] ] and that the proper radius @xmath30 and the coordinate radius @xmath4 are related by , @xmath302 as the bonnor star is here a compact object , the proper distance from @xmath300 to @xmath13 is finite , and tends to zero in the limit @xmath303 . therefore , in this limit , the matter between right and left boundaries becomes negligible and the construction corresponds to gluing two extremal reissner - nordstrm black holes in the situation tb of section [ nonextremalepsilonsurg ] . thus in the limit of our interest , @xmath304 , it is not surprising that the results coincide with ( [ 2bt1 ] ) and ( [ 2bt2 ] ) where @xmath305 . indeed , one finds @xmath306 and @xmath307 being finite as well . we have put @xmath294 , as is the case for these systems . there is yet another procedure to produce a wormhole . in the above considerations , we performed a symmetric construction , in the sense that the @xmath84 and @xmath83 branches differed by the sign of @xmath89 only . now , we start again from a compact bonnor star configuration which contains matter inside and vacuum outside . we want to preserve this feature , and so we have to make a non - symmetric deformation . thus , we change the procedure and consider the following construction . again , a bonnor star is made of majumdar - papapetrou matter for @xmath282 , which in turn is joined to an extremal reissner - nordstrm metric for @xmath283 . now , in the region @xmath282 , the left @xmath83 branch , choose the distribution with @xmath308 . the potential @xmath118 can be written again as in eq . ( [ vm ] ) . then , for the left branch , it follows from the field equations that @xmath309 \,,\quad 0\leq r\leq r_{0}\ , , \label{phibon2}\ ] ] and @xmath310 it is worth paying attention that because of the property @xmath311 , the matter distribution which was originally compact , turned after deformation into a non - compact one since at left infinity @xmath312 , where @xmath313 , @xmath314 . for the right @xmath84 branch we use the extremal reissner - nordstrm metric with the mass @xmath294 , which gives , @xmath315 and @xmath316 in the limit @xmath317 , @xmath318 one finds that @xmath319 and @xmath320 being finite as well . we have put @xmath294 , as is the case for these systems . thus , extremal quasi - black wormholes , made of majumdar - papapetrou matter are possible . their distinctive feature is the presence of finite non - zero surface stresses on the horizon . curvature components remain finite . it is worth noting that , although in this subsection we did not introduce the parameter @xmath0 explicitly , actually its role is played , say , by the difference @xmath321 in the sense that this quantity is responsible for the deviation of the spacetime from its limiting state ( a quasi - black hole or two quasi - black holes glued together ) . here we exploit the distribution of extremal charged dust given in @xcite . near the quasi - black hole limit , the first order corrected quasihorizon has radius @xmath275 given by @xmath322 \ , , \label{radiusquasi}\ ] ] where @xmath323 is the parameter that yields the deviation from the reissner - nordstrm solution , and @xmath324 is a quantity with units of electric charge , which is indeed the total charge @xmath61 when @xmath325 from the outset ( see @xcite for details ) . the quasi - black hole limit is such that @xmath326 , with @xmath327 . in a sense , the dimensionless parameter @xmath328 here corresponds to the @xmath0 parameter in the @xmath2-wormhole construction of a previous section . then the solution has asymptotics near the quasihorizon @xmath275 given by @xmath329 and @xmath330 consider again the @xmath84 and @xmath83 branches , symmetric relative to the first order corrected quasihorizon radius in the region @xmath331 , but with different signs of @xmath89 . each branch has the same dependence of the metric potential , eqs . ( [ as1])-([as2 ] ) , on @xmath4 . thus , for simplicity , we restrict ourselves to the analog of the situation bt considered previously , which translated to here means taking the limits as follows , @xmath332 . then , simple calculations show that @xmath333 and @xmath334 being finite as well . in ( [ lim ] ) we have put @xmath335 in the limit @xmath336 and @xmath294 also in this limit , where @xmath60 is the mass of the configuration . in the limit under consideration the metric in the region @xmath337 is given by the extremal reissner - nordstrm metric , whereas in the immediate vicinity of the quasihorizon the metric is described by the bertotti - robinson metric . therefore , as in the preceding subsection , our construction gives two extremal reissner - nordstrm black holes glued along the quasihorizon with different signs of @xmath89 on opposite sides . we have studied wormhole and other configurations as possible mimickers of black holes . we have separated the configurations into non - extremal and extremal . for wormholes , we have examined separately two limiting procedures in which the wormhole throat approaches the black hole horizon . in the first procedure , we fix the location of an observer ( exactly on the throat ) . then , we change the spacetime ( making a wormhole on the verge of being a black hole ) . this is the situation bt . in the second , we change spacetime ( making a wormhole on the verge of a black hole ) , then place the shell outside the throat and move it toward the throat ( which coincides with the horizon ) this is the situation tb . this procedure is carried out for non - extremal and extremal configurations separately . in the non - extremal case it turned out that each of the limits under discussion is singular : either the kretschmann scalar or surface stresses on the throat ( or both ) diverge . thus , the limit is singular . in other words , a mimicker of a non - extremal black hole , made from a non - extremal wormhole , including the black foil of @xcite , is not smooth . we have summarized the results for the non - extremal case in a table . for the extremal case both the kretschmann scalar and surface stresses remain finite . this pronounced distinction between properties of the limiting configurations in the non - extremal and extremal cases is one counterpart of the conclusion made in @xcite that quasi - black holes can be only extremal . however , one should not forget about some subtleties connected with the fact that singular behavior can , in general , manifest itself not only in the value of the kretschmann scalar . even if this scalar is finite , a naked behavior is possible or even inevitable as was shown in in the present paper and in @xcite ( see , e.g. , sec . v of @xcite for a discussion about other subtleties in which the singular features of quasi - black holes are revealed ) . there is also another candidate for the role of a non - extremal black hole mimicker , a gravastar @xcite . however , the corresponding surface stresses grow unbound when the radius approaches the gravitational radius , as we have seen . from an astrophysical viewpoint , the situation bt in the non - extremal case , i.e. , a black foil , is more interesting since it implies no necessity of making a shell by hand . it is the case considered in @xcite . one of the questions raised in @xcite is whether it is possible or not to distinguish between a true non - extremal black hole , a schwarzschild black hole say , and a wormhole . the main conclusion of @xcite is that it is impossible to distinguish for any finite time in the limit under discussion . this conclusion is reached on the basis of considering properties of bodies moving along separate fixed geodesics and emitting signals detected at infinity . however , if from single geodesics we shift our attention to a congruence of geodesics , it turns out that the strong gravity forces , on the near horizon region , leave their imprint on the form of a moving body and , thus , on the properties of signals which an observer at infinity is detecting . if the surface of the body is luminescent , an observer at infinity would see either a finite width instead of a point , a continuous detection instead of separate pulses , and so on . it is essential that in the case discussed in @xcite the corresponding proper time of deformation tends to zero when the curvature grows unbound , with the time at infinity being finite . thus , at least in principle , an observer can distinguish between a black hole and an almost singular wormhole . the singular nature of the limit in the non - extremal case makes also questionable the applicability of the membrane paradigm used in @xcite . the key point in this paradigm consists in boundary conditions according to which a free - falling observer sees a finite value of physical fields on the horizon ( see @xcite , sec . ii ) . however , in the problem under discussion , typically , this observer as well as the geometry itself become ill - defined . only in some cases ( see model 2 , situation tb ) the curvature components remain finite in the free - falling frame . but even in such situations the infinite surface stresses on the horizon surface makes the physical meaning of the membrane paradigm unclear since this paradigm relies heavily on the concept of a regular surface . objects based on nearly extremal wormholes , although of less interest astrophysically perhaps , have a much better behavior in the sense that both the geometry and surface stresses remain finite . moreover , typically , there is no naked behavior . in this case , the effect of strong curvature is much less pronounced than in the case of quasi - black holes where a naked behavior is typical @xcite . in this sense , a wormhole composed on the basis of two extremal black holes seems to be the best mimicker of an extremal black hole . as by - product , we have obtained a model of a regular black hole ! thus , if we try to arrange a ranking of black hole mimickers , both non - extremal and extremal , the list looks as follows from top to bottom : wormholes on the basis of extremal black holes or on the basis of quasi - black holes , quasi - black holes , wormholes on the basis of non - extremal black holes ( and within these the best are black foils ) , and gravastars . bearing in mind that in observational astrophysics it is difficult to find extremal configurations ( the would be best mimickers ) , whereas non - extremal configurations are really bad mimickers , the task of distinguishing black holes from their mimickers seems to be less difficult than one could think of it . in the present paper we have restricted ourselves to static spherically symmetric spacetimes . meanwhile , in a recent work @xcite the status of black hole mimickers is undermined in the rapidly rotating case as well since it is argued that they are unstable . we have also circumscribed our discussion to particular wormholes , gravastars and quasi - black holes @xcite-@xcite , since these objects are well adapted to our goal of examining the near - horizon properties and their connection with far away asymptotic properties . however , there are many other objects with properties that make them also potential black hole mimickers ( see , e.g. , @xcite ) and which are worthy of study within our formalism . o. z. thanks centro multidisciplinar de astrofsica centra for hospitality and a stimulating working atmosphere . this work was partially funded by fundao para a cincia e tecnologia ( fct ) - portugal , through project poci / fp/63943/2005 .

we present a systematic way of derivation of the algebraic curves of separation of variables for the classical kovalevskaya top and its generalizations , starting from the spectral curve of the corresponding lax representation found by reyman and semonov - tian - shansky . in particular , we show how the known kovalevskaya curve of separation can be obtained , by a simple one - step transformation , from the spectral curve . the algorithm works for the general constants of motion of the system and is based on w. barth s description of prym varieties via pencils of genus 3 curves . it also allows us to derive new curves of separation of variables in various generalizations of this system .

1606.08331

the famous kovalevskaya integrable case of heavy rigid body dynamics in @xmath0 has various nontrivial integrable generalizations . one of the most interesting ones is the motion of the kovalevskaya gyrostat in one or two constant force fields ( e.g. , gravitational and electric ) described in @xcite . the systems are hamiltonian on the cotangent bundle @xmath1 and , as was shown in @xcite , their generic invariant manifolds are 3-dimensional tori . in the present paper we consider only the cases of the presence of @xmath2-symmetries , which allow reductions to systems with two degrees of freedom with 2-dimensional generic invariant tori . ( the nature of the symmetries will be explained in section 2 . ) this occurs in two situations : \1 ) the two fields are orthogonal and the intensities are equal ; + 2 ) there is just one constant force field , the corresponding system is the kovalevskaya gyrostat , which , for zero value of the gyrostatic momentum , becomes the classical kovalevskaya top . in @xcite bobenko , reyman and semenov presented a lax representation for both cases and gave their algebraic geometric description . in particular , it was shown that , after factoring out a trivial involution , the corresponding spectral curve @xmath3 is non - hyperelliptic of genus 3 and is a 2-fold ramified covering of an elliptic curve @xmath4 . the reduced systems were linearized on 2-dimensional prym variety @xmath5 associated with the covering @xmath6 . in the case of the classical kovalevskaya top the authors of @xcite demonstrated that the generic complex invariant manifolds are open subsets of @xmath7 . this variety has polarization ( 1,2 ) , so it can not be the jacobian variety of a genus 2 curve which can be used for a separation of variables of the systems . on the other hand , in her celebrated paper @xcite , s. kovalevskaya presented a separation of variables on a genus 2 ( and , therefore , a hyperelliptic ) curve @xmath8 , whereas the complex flow of the system was linearized on the jacobian of @xmath8 . apparently , the papers @xcite were the first ones where this seeming contradiction had been explicitly explained : the subvariety @xmath7 can be viewed as 2-fold covering of the jacobians of 3 different , i.e. , birationally non - equivalent , genus 2 curves . in addition , there exist 3 other different genus 2 curves whose jacobians are 2-fold coverings of @xmath7 . in @xcite it was shown that , for the case of the kovalevskaya top , one of these hyperelliptic curves is birationally equivalent to the original curve of separation @xmath8 . ( these relations between @xmath7 and the jacobians are briefly described in section 3 . ) next , following a remarkable geometric construction of w. barth @xcite inspired by @xcite , the papers @xcite presented a family ( pencil ) @xmath9 , @xmath10 of ( generally non - hyperelliptic ) genus 3 curves covering elliptic curves , and the coverings give rise to one and the same prym variety _ dual _ to @xmath7 arising in the kovalevskaya problem . it was shown that all the above six genus 2 curves can explicitly be extracted from @xmath9 , either as regularizations of singular curves or as quotients of hyperelliptic curves of @xmath11 . curiously , the spectral genus 3 curve @xmath3 of @xcite did not appear in this analysis . the main observation of the present paper is that , in fact , the spectral curve @xmath3 naturally fits into the w. barth construction of pencils of genus 3 curves . using this property , we show that the kovalevskaya curve @xmath8 can be obtained from @xmath3 by a simple one - step transformation , even without knowing explicitly the family @xmath9 and avoiding complicated calculations . namely , @xmath8 appears as regularization of the curve _ dual _ to the spectral curve @xmath3 ( see section 3 . ) in the particular case of zero area integral , the curve @xmath3 itself becomes singular of genus 2 , then the above construction implies that @xmath12 must be isogeneous , whereas @xmath13 themselves are related via the richelot transformation . this property has previously been described in @xcite . the same main observation allows us to recover easily the curves of separation of variables in all known linearizations of the kovalevskaya gyrostat in one and two fields ( section 4 ) . we also derive some new separation curves for this problem ( without doing the linearization itself ) . in this section we reproduce some important known results on the dynamics and complex geometry of the kovalevskaya gyrostat . using , for convenience , the notation of @xcite , consider the motion of an axisymetric rigid body in @xmath14 with angular velocity vector @xmath15 about a fixed point in presence of constant force fields @xmath16 of different nature ( e.g. , gravitational and electric ) , which are not parallel . let @xmath17 be position vectors of the mass and , respectively , charge centers of the body . the equations of motion in the body frame have the form @xmath18 + [ c_1,g ] + [ c_2,h ] , \\ \frac{dg}{dt } = [ g,\omega ] , \quad \frac{dh}{dt } = [ h,\omega ] , \end{aligned}\ ] ] where @xmath19 is the total angular momentum and @xmath20 $ ] denotes the vector product . assume that the body carries a rotator with a fixed axis , then @xmath21 are related as @xmath22 , where @xmath23 is the inertia tensor of the body ( with the rotator ) and @xmath24 is related to the corresponding constant gyrostatic momentum . the phase space of the system is the cotangent bundle @xmath1 , and the components of @xmath25 stand for redundant coordinates on @xmath26 . equivalently , the phase space can be regarded as dual to the lie algebra @xmath27 , the semi - direct product of @xmath28 and @xmath29 with the corresponding lie poisson structure . then the system is hamiltonian with the hamilton function @xmath30 which is the total conserved energy . also , the intensities @xmath31 and the product @xmath32 are trivial integrals of the system . they are casimir functions of the lie poisson bracket of @xmath33 , and fixing generic values of the functions one obtains a 6-dimensional coadjoint orbit in the coalgebra . assuming now @xmath34 , @xmath35 , and @xmath36 ( the axes of symmetry of the body and of the rotator are parallel ) , we obtain the configuration of the _ kovalevskaya gyrostat _ with the corresponding hamiltonian @xmath37 for this case of equations , in @xcite bobenko , reyman , and semenov - tyan - shansky presented a @xmath38 matrix lax representation @xmath39 $ ] with a spectral parameter @xmath40 and @xmath41 ( we do not reproduce @xmath42 here ) with the spectral curve @xmath43 where @xmath44 are laurent polynomials . apart from the hamiltonian @xmath45 and the trivial integrals , the coefficients of @xmath46 involve two other independent integrals @xmath47 given by expressions ( 1.9 ) , ( 1.10 ) in @xcite , and all of them were shown to commute . as a result , the kovalevskaya gyrostat , is completely integrable , and its generic real invariant varieties are 3-dimensional tori . next , the complex flow of the system was shown to be linear on the jacobian variety of @xmath48 . the latter has an obvious involution @xmath49 , so it is a 2-fold covering of the quotient curve @xmath50 of maximal genus 4 . it follows that the jacobian variety of @xmath48 , @xmath51 contains @xmath52 as an abelian subvariety . as was shown in @xcite , the flow is confined to the subtorus @xmath53 . for this reason , it is sufficient to study only the properties of @xmath3 , which will be called the quotient spectral curve . [ [ the - cases - of - reduction - and - the - corresponding - spectral - curves . ] ] the cases of reduction and the corresponding spectral curves . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + as was first shown in @xcite , when the force fields are orthogonal of equal intensities , that is , @xmath54 , the system possesses an additional @xmath2-symmetry group consisting of simultaneous rotations about the symmetry axis of the gyrostat and the fixed in space vector @xmath55 $ ] . the group action is described by the hamiltonian @xmath56\rangle/|g|^2 $ ] , which is a conserved quantity . let @xmath57)^t \in so(3)$ ] be the rotation matrix of the gyrostat . following @xcite , the action of @xmath2 on @xmath58 is generated by conjugations @xmath59 as was shown in @xcite , in the quaternion realization of the group @xmath26 as the factor of @xmath60 by @xmath61 , the @xmath2-symmetry acts as the rotation in the 2-plane @xmath62 , then the quotient variety @xmath63 is the until sphere @xmath64 with @xmath65 . for any constant @xmath66 , the level variety @xmath67 is 5-dimensional , and the quotient space @xmath68 is isomorphic to a globally nontrivial fiber bundle @xmath69 . following @xcite , the system , admits a hamiltonian reduction onto @xmath70 endowed with the standard symplectic structure of @xmath71 plus a singular magnetic term . the reduced gyrostat in 2 fields possesses two independent commuting integrals , which are reductions of @xmath72 ( or @xmath73 , and is linearized on 2-dimensional invariant tori . in the sequel we will set @xmath74 , then the integrals @xmath75 read @xmath76 where the value of the momentum @xmath77 should be fixed . from section 5 of @xcite we extract the following expression for the quotient curve @xmath3 @xmath78 which , under the birational transformation @xmath79 , can be written as @xmath80 the curve has the involution @xmath81 , hence it is a 2-fold covering of the elliptic curve @xmath82 given by equation @xmath83 the covering is ramified over 4 points on @xmath82 , therefore , by the riemann hurwitz formula , for generic values of the constants of motion , genus(@xmath84 . as we will see in section 4 , @xmath85 is hyperelliptic ( i.e. , can be transformed to a hyperelliptic form ) , this quite special property will simplify our analysis considerably . in the other case of @xmath2-symmetry , the kovalevskaya gyrostat moves only in one constant force field ( one can set @xmath86 ) , and for @xmath87 the system becomes the classical kovalevskaya top . the @xmath2-action is the rotation about the fixed vector @xmath88 and is generated by the hamiltonian @xmath89 , i.e. , the area integral . upon fixing its value , the system admits a hamiltonian reduction onto the cotangent bundle @xmath90 of @xmath91 with two independent commuting integrals and 2-dimensional generic invariant tori . equivalently , the reduced kovalevskaya gyrostat can be regarded as a hamiltonian system on the coalgebra @xmath92 with the two integrals @xmath93 and the casimir functions @xmath94 . in the considered case the quotient spectral curve @xmath3 takes the following concrete form @xmath95 where now @xmath96 , and @xmath97 is obtained from by setting @xmath86 . under the same birational transformation @xmath79 the curve @xmath98 can be written in the form @xmath99 the factor of @xmath98 by the involution @xmath81 is the elliptic curve @xmath100 and again , generally ( also for @xmath101 ) , genus@xmath102 . note that , in contrast to @xmath85 , the curve @xmath98 is not hyperelliptic , even for @xmath101 . in both considered cases the involution @xmath103 on @xmath3 extends to its jacobian . thus the latter contains two abelian subvarieties : the elliptic curve @xmath104 or @xmath82 itself and the 2-dimensional prym variety , @xmath7 , which is anti - symmetric with respect to the extended involution , whereas @xmath4 is invariant . equivalently , following @xcite , @xmath105 . using the approach of vector baker akhiezer functions , the authors of @xcite linearized the complex flow of the reduced gyrostat in one field on the subvariety @xmath106 and proved that , for @xmath87 and @xmath107 , the generic 2-dimensional complex invariant manifold @xmath108 of the system , the common level surface of the integrals @xmath109 , is an open subset of the prym subvariety . it is known ( see @xcite and section 3 below ) that this variety has polarization ( 1,2 ) , so it can not be the jacobian of a genus 2 curve which can be used for a separation of variables of the system . one can compare this result with the famous kovalevskaya reduction of the problem to hyperelliptic quadratures @xmath110 where the pairs @xmath111 are coordinates of two points on the genus 2 kovalevskaya s separation curve : @xmath112 \left ( ( s - h)^2-i_2/4 \right ) .\ ] ] the integral form of the quadratures defines the abel jacobi map @xmath113 , which suggests that the complex invariant tori @xmath108 of the system and @xmath114 must be algebraically related . this relation was first described in @xcite : let @xmath115 be a compactification of @xmath108 converting it to an abelian torus , then @xmath115 is an 8-fold unramified covering on @xmath114 . equivalently , if we replace @xmath114 by its 16-fold covering , obtained by doubling all the 4 period vectors ( both abelian tori are conformally equivalent ) , then @xmath114 can be regarded as a 2-fold unramified covering of @xmath116 . note that the detailed algebraic geometric analysis of @xcite did not involve neither the lax pair , nor the quotient spectral curve @xmath98 in , but a pencil of genus 3 curves covering elliptic curves and leading to the same prym variety , which was identified with the manifold @xmath116 . [ [ remark-1 . ] ] remark 1 . + + + + + + + + + as was noticed in @xcite , for the case of zero area integral ( @xmath117 ) the quotient spectral curve @xmath98 of the kovalevskaya top ( the gyrostatic component @xmath118 is zero ) is singular , and its regularization is a genus 2 curve @xmath119 with hyperelliptic form the curve @xmath98 is singular of genus 2 also for @xmath120 . ] @xmath121 which can be obtained from by the birational change of coordinates @xmath122 in section 7.6 of @xcite it was also claimed that in this case @xmath123 coincides with @xmath124 and that the complex invariant manifold @xmath125 is an open subset of a 2-fold unramified covering of @xmath124 . in fact , a careful analysis based on the procedure described in @xcite shows that @xmath123 can be recovered as an abelian subvariety of the 3-dimensional generalized jacobian of the singular curve @xmath98 , which is an extension of @xmath124 by @xmath126 . then @xmath123 appears as a smooth 2-fold covering of @xmath124 . as a result , @xmath125 is an open subset of @xmath123 in both cases : @xmath107 and @xmath127 . moreover , the proof of this statement equally holds for @xmath128 . thus we have [ prym - tori ] for general constants of motion , including @xmath127 , the compactified 2-dimensional complex invariant manifold @xmath116 of the reduced kovalevskaya gyrostat in one field is @xmath123 . next , when performing a linearization of the complex flow on @xmath124 in the case @xmath127 , bobenko , reyman , and semenov @xcite observed that @xmath119 is not birationally equivalent to the kovalevskaya curve with @xmath87 . a decade later leprvost and markushevich @xcite proved that the jacobians of both curves are isogeneous : @xmath114 is a 4-fold unramified covering of @xmath124 , obtained by duplicating two of the 4 period vectors of @xmath124 ( namely , by duplicating the riemann period matrix of @xmath124 ) . the curves @xmath129 themselves are related algebraically via the richelot transformation , whose description can be found in @xcite . below in section 4 we will see that the richelot relation between @xmath129 becomes quite natural in the context of pencils of genus 3 curves and dual prym varieties . this global construction described in the next section will also allow us to recover all the genus 2 curves that previously appeared in the linearization of the kovalevskaya gyrostat in one and two force fields , as well as to construct new genus 2 curves of separation of variables . apart from the mentioned generalizations , there are also various deformations of the kovalevskaya gyrostat obtained by modifications of the poisson bracket , see e.g. , @xcite . we do not consider here the corresponding curves of separations . we conclude the section with the following analog of theorem [ prym - tori ] for the kovalevskaya gyrostat in two fields . [ prym - tori2 ] for general constants of motion , the complex 2-dimensional invariant manifold @xmath125 of the reduced kovalevskaya gyrostat in two fields is an open subset of @xmath130 . _ let @xmath131 be the complex 3-dimensional invariant manifold of the gyrostat in two fields obtained by fixing the independent constants of motion @xmath132 and , therefore , the quotient spectral curve @xmath85 in . obviously , @xmath133 is isomorphic to the complex isospectral manifold @xmath134 , the set of all the matrix laurent polynomials @xmath135 of the form having the same spectral curve @xmath46 and its quotient @xmath85 . consider _ the eigenvector map _ @xmath136 defined in the following standard way : a matrix @xmath137 induces the eigenvector bundle @xmath138 : for any point @xmath139 @xmath140 we assume that the eigenvector @xmath141 is normalized : @xmath142 , for a certain @xmath143 . this defines the divisor @xmath144 of poles of @xmath145 on @xmath46 . for any choice of normalization , such divisors form an equivalence class @xmath146 . for a base point @xmath147 , the class defines a point @xmath148 , where @xmath149 . then one has @xmath150 . in lemma 4.5 of @xcite it was shown that under the time evolution of @xmath135 the image @xmath151 is restricted to a translation of @xmath152 let @xmath153 be the maximal subgroup of @xmath154 which acts freely on @xmath134 by conjugations and preserves the form of @xmath135 . for any @xmath155 , the @xmath156-images of @xmath135 and @xmath157 coincide . then the eigenvector map pushes down to @xmath158 and , as was shown in e.g. , @xcite , such a map is _ injective_. now note that @xmath153 must be a stabilizer of @xmath159 in , as well as of the constant @xmath160 block @xmath161 of @xmath162 . it also must preserve the symmetry of @xmath163 . then , nesesarily , any element of @xmath153 has the block form @xmath164 since both varieties @xmath165 have dimension 2 and @xmath166 is injective , the factor @xmath167 is an open subset of @xmath130 . we now show that the action of @xmath168 is generated by the action of @xmath169 on @xmath58 induced by with @xmath170 . in view of the expressions for @xmath171 in terms of euler angles and their time derivatives , implies the transformation @xmath172 with @xmath173 the latter can be described as the conjugation @xmath174 as a result , the action of @xmath168 on @xmath134 coincides with that of @xmath169 on @xmath133 . then @xmath175 , which is the complex invariant manifold of the reduced gyrostat in 2 fields , is isomorphic to @xmath167 , which , in turn , is isomorphic to an open subset of @xmath130 . @xmath176 we start this section with the description of the general form of genus 3 curves ( over the field @xmath178 ) covering elliptic curves and the related prym varieties . [ gen_cover ] let @xmath4 be a generic elliptic curve in @xmath179 whose affine part in @xmath180 is given by equation @xmath181 \1 ) any 2-fold covering @xmath182 ramified at 4 arbitrary chosen finite points + @xmath183 on @xmath4 can be written in form @xmath184 where @xmath185 is a polynomial of degree at most 3 and @xmath186 are constants such that @xmath187 for a certain linear function @xmath188 ( it is a constant if degree of @xmath189 and @xmath190 ) . \2 ) for any divisor @xmath191 on @xmath4 , the set @xmath192 of 2-fold covers of @xmath4 ramified exactly at @xmath193 consists of @xmath194 birationally non - equivalent covering curves @xmath3 . the idea of the proof of item ( 1 ) of the theorem is due to a. levin @xcite . for reasons of brevity we do not give it here , just mention that the condition says that on the curve @xmath4 the meromorphic function @xmath195 has 4 simple zeros , one double zero ( given by the zero of @xmath188 ) and a pole of order 6 . multiple zeros or poles of @xmath195 of even order give singularities of @xmath3 , which , after the regularization , do not contribute to branching of @xmath6 . as a result this covering has precisely 4 ramification points . note that the quotient spectral curves @xmath196 in , have the structure . the involution @xmath197 on @xmath3 extends to @xmath52 , which then contains two abelian subvarieties : the elliptic curve @xmath4 itself and the 2-dimensional prym variety denoted as @xmath198 , the latter being anti - symmetric with respect to @xmath103 , whereas @xmath4 is invariant . the complex subtori @xmath199 intersect at 4 points , which are the half - periods of @xmath4 . as an abelian variety , @xmath7 is a complex torus @xmath200 , with @xmath201 being a lattice generated by 4 period vectors . in appropriate coordinates @xmath202 in @xmath203 , its period matrix takes the form @xmath204 the right half of @xmath205 is the riemann matrix @xmath206 with i m @xmath207 , and the diagonal of the left half indicates the polarization of @xmath7 , namely ( 1,2 ) . according to @xcite , any 2-dimensional abelian variety @xmath208 with polarization ( 1,2 ) can be realized as the prym variety of a covering @xmath182 described in theorem [ gen_cover ] . moreover , for a generic @xmath208 there is a one - parametric family of 2-fold coverings @xmath209 , in which both curves @xmath3 and @xmath4 depend on the parameter @xmath40 ( see also below ) . it is also natural to consider the ( 1,2)-polarized variety @xmath210 _ dual _ to @xmath7 , its period matrix can be written as @xmath211 this matrix is obtained by dividing by 2 the 1st and 3rd period vectors of @xmath212 and rescaling @xmath213 to get a riemann matrix . namely , @xmath214 note that , modulo the conformal equivalence , the dual to @xmath215 is again @xmath216 , and both varieties can be regarded as 4-fold coverings of each other . [ [ prym - varieties - and - isogeneous - jacobians . ] ] prym varieties and isogeneous jacobians . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + as was mentioned in section 6 of @xcite , given an arbitrary ( 1,2)-polarized 2-dimensional abelian variety @xmath217 having the period matrix @xmath205 as in , there are several ways to obtain a principally polarized 2-dimensional abelian variety ( and therefore , the jacobian of a genus 2 hyperelliptic curve ) isogeneous to @xmath217 . an obvious way is to divide the second period vector of @xmath205 by 2 , to get the period matrix @xmath218 , @xmath219 . on the other hand , one can double the first period vector of @xmath205 to obtain the matrix @xmath220 . the latter defines a torus with polarization ( 2,2 ) , which is conformally equivalent to a principally polarized variety with the period matrix @xmath221 . there are others , less obvious , transformations which produce jacobian varieties symplectically non - equivalent to the above ones ( their explicit description is given in @xcite , @xcite ) . there is , however , a very limited number of them . [ main_th ] for a generic covering @xmath182 described in theorem [ gen_cover ] there are exactly 6 birationally non - equivalent genus 2 curves @xmath222 and @xmath223 such that @xmath7 is a 2-fold unramified covering of @xmath224 , and @xmath225 are 2-fold unramified coverings of @xmath7 . + next , @xmath224 are also 2 fold coverings of the dual variety @xmath210 , whereas the latter is a 2-fold covering of tori conformally equivalent to @xmath226 , @xmath227 , @xmath228 as depicted in the following diagram , where the arrows denote the corresponding 2:1 coverings and double arrows indicate 4-fold coverings : @xmath229 below we identify conformally equivalent abelian varieties obtained one from another by duplication of all of the 4 periods . under this convention , @xmath230 can be regarded as an 8-fold covering of any of @xmath226 , @xmath227 , @xmath228 . [ [ the - richelot - correspondences . ] ] the richelot correspondences . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + as was shown in @xcite , the period matrices of some of the jacobians below and above @xmath216 are obtained from one another by doubling their riemann matrix @xmath231 . then , upon a proper ordering of the jacobians , the above diagram can be accompanied with the following one @xmath232 where arrows denote the duplication @xmath233 ) . explicit relations between the riemann matrices of the jacobians are given in @xcite . the corresponding curves @xmath234 are related algebraically via the _ richelot transformation _ ( explicit formulas can be found in @xcite ) . one should stress that such a transformation is only a correspondence : for example , @xmath235 is obtained by duplication of symplectically non - equivalent riemann matrices of @xmath236 and @xmath237 . [ [ the - case - of - two - pairs - of - elliptic - curves . ] ] the case of two pairs of elliptic curves . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + as was noted in @xcite , theorem [ main_th ] does not completely hold in special cases when @xmath7 contains abelian subtori . in particular , this happens when one of the associated curves in say @xmath238 , is a 2-fold covering of two elliptic curves @xmath239 with normalized periods @xmath240 respectively . then , according to the weierstrass poincar theory of reduction ( see e.g. , @xcite ) , @xmath241 contains the curves @xmath239 , although @xmath241 is not the direct product @xmath242 , but there is 4-fold isogeny @xmath243 . then , according to @xcite , theorem [ main_th ] must be replaced by the following [ ell_case ] let @xmath7 be a 2-fold unramified covering of the direct product @xmath242 . then there are two non - equivalent genus 2 curves @xmath244 , both being 2-fold ramified coverings of @xmath239 , such that @xmath245 are 2-fold unramified coverings of @xmath7 . next , @xmath7 is a 2-fold unramified coverings of jacobians of other non - equivalent genus 2 curves @xmath246 , both being 2-fold ramified coverings of elliptic curves @xmath247 with normalized periods @xmath248 respectively . finally , the direct product @xmath249 is a 2-fold unramified covering of @xmath7 . the relations between the genus 2 curves , their jacobians , and the elliptic curves are described by the following diagrams ( on the first one the arrows denote 4-fold coverings such that their inversions describe the richelot transformations ) . @xmath250 here the products @xmath249 , @xmath242 replace @xmath251 and @xmath252 in diagrams , for the general case . we stress that , as above , all the abelian varieties are considered modulo conformal equivalence . in this context , in view of their periods , the curves @xmath253 are 2-fold unramified coverings of each other ( the same for @xmath254 ) , and their equations are obtained from one another by the classical landen transformation ( see also proposition [ landen ] for details and references ) . then @xmath249 is a 4-fold covering of @xmath242 and vice versa . we will use proposition [ ell_case ] in subsection 4.2 in the analysis of a particular case of linearization of the kovalevskaya top on two elliptic curves . in the sequel we will call the 2-dimensional variety @xmath7 _ generic _ , if it does not contain elliptic curves . [ [ dual - pencils - of - genus - le-3-curves . ] ] dual pencils of genus @xmath255 curves . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + following @xcite ( see also @xcite ) , a generic torus @xmath7 carries a line bundle @xmath256 defining polarization ( 1,2 ) and @xmath257 gives a one - parametric family @xmath258 of algebraic curves @xmath259 . geometrically , each curve @xmath260 is the intersection @xmath261 , where @xmath262 is an appropriate translation of the theta - divisor @xmath263 , the zero locus of the theta - function of @xmath3 . more precisely , following @xcite , all @xmath264 are translations of @xmath263 by vectors of @xmath265 . a general curve of @xmath260 is smooth of genus 3 and non - hyperelliptic . the family @xmath266 is , in fact , a pencil of curves is a pencil only in @xmath7 , the family of equivalent curves in @xmath179 does not form a pencil . ] whose base locus consists of 4 points @xmath267 . next , the extended involution @xmath268 acts on @xmath5 as a reflection having @xmath267 and the 4 half - periods of @xmath4 as fixed points . as was shown in @xcite , for a generic smooth curve @xmath260 , @xmath269 , and @xmath270 is an elliptic curve @xmath271 . the latter is a member of a family @xmath272 of generally elliptic curves , which , along with @xmath260 , forms a family of 2-fold coverings @xmath273 , all of them ramified at @xmath267 . remarkably , all these coverings define the same prym variety , denoted as @xmath274 , which is dual to @xmath7 . then , as one may expect , the whole construction can be reflected : there is a pencil of ( generally smooth and non - hyperelliptic ) genus 3 curves @xmath275 having 4 base points @xmath276 , and a family of 2-fold coverings @xmath209 , all ramified at @xmath277 , and a general curve of @xmath278 is elliptic . all these coverings lead to the same prym variety @xmath7 dual to @xmath274 . in this connection it is natural to refer to the pencils @xmath279 as _ dual _ pencils . a pencil @xmath280 will be called generic if the corresponding variety @xmath7 is generic . in the case the dual pencil @xmath266 and the variety @xmath274 are generic as well . this beautiful purely algebraic geometric construction of barth has been inspired by the work of haine @xcite on the complex geometry of the 4-dimensional integrable top . yet , it would not be complete without important observations made by horozov and van moerbeke in @xcite for the pencil @xmath280 related to the tori of the kovalevskaya top , although they hold for generic pencils as well . we summarize these results of @xcite in form of the following theorem . [ fibers ] a generic pencil @xmath280 of curves of genus @xmath255 in @xmath210 contains 6 hyperelliptic genus 3 curves @xmath281 , which are 2-fold unramified coverings of genus 2 curves . among the latter there are only 3 non - equivalent curves , and they are birationally equivalent to @xmath223 in diagram . next , @xmath280 also contains 12 singular fibers @xmath282 , which upon desingularization lead to curves of genus 2 . the set of these 12 curves consists of 3 distinct groups of four equivalent curves , defining @xmath222 in . respectively , the dual pencil @xmath266 in @xmath7 contains 6 hyperelliptic genus 3 curves , which are 2-fold coverings of curves equivalent to @xmath222 , and 12 singular curves whose regularizations are birationally equivalent to @xmath223 . thus , once a family of curves in @xmath179 , equivalent to the curves @xmath283 , is known , one can calculate explicitly equations of all the genus 2 curves @xmath284 in theorem [ main_th ] . for the prym variety arising as the complex invariant torus of the classical kovalevskaya top , such a family of curves was found in @xcite , although without any relation with the quotient spectral curve @xmath3 in . apparently , for a generic covering @xmath182 described in theorem [ gen_cover ] there is no general recipe of reconstruction of the families @xmath279 in @xmath179 . [ [ remark-2 . ] ] remark 2 . + + + + + + + + + in some special cases of covering @xmath182 ( see subsection 4.2 ) the dual pencil @xmath285 may contain a reducible curve @xmath286 being a union of two elliptic curves , say @xmath247 , and @xmath287 is the direct product @xmath249 . in the same case @xmath283 may contain a hyperelliptic genus 3 curve @xmath288 , which is a 2-fold unramified covering of a genus 2 curve @xmath289 , which , in turn , is 2-fold ramified covering of @xmath247 . there is a natural conjecture that this case corresponds to the situation described in proposition [ ell_case ] , when the jacobians of some of related genus 2 curves are replaced by direct products of elliptic curves , however we did not find a rigorous proof of this . theorem [ fibers ] can be accompanied with following well - known property ( see , e.g. , @xcite ) : [ 3->2 ] any hyperelliptic genus 3 curve which is also a 2-fold covering of an elliptic curve , can be written in form @xmath290 it has two obvious commuting involutions @xmath291 the first one having 4 fixed points , the second one no such points . then @xmath292 is the elliptic curve @xmath293 , whereas @xmath294 is the genus 2 curve + @xmath295 . @xmath296 is a 2-fold unramified covering of @xmath297 . note that the last statement is consistent with those of theorems [ fibers ] , [ main_th ] . [ [ dual - genus-3-curves . ] ] dual genus 3 curves . + + + + + + + + + + + + + + + + + + + + the last necessary ingredient is a simple scheme which , for a given genus 3 curve @xmath298 , allows to obtain a curve @xmath299 . let , as above , @xmath3 be a covering of the elliptic curve @xmath4 as described in theorem [ gen_cover ] . consider the tower of curves @xmath300 where @xmath301 denote the corresponding coverings with branch points @xmath302 and , respectively , @xmath303 . now let @xmath304 be another curve covering an elliptic curve @xmath305 with the corresponding tower of 2-fold coverings @xmath306 where @xmath307 are ramified at @xmath308 , respectively , @xmath309 . we now require @xmath310 that is , @xmath311-coordinates of the branch loci of @xmath182 ( of @xmath312 ) coincide with @xmath311-coordinates of the branch loci of @xmath313 ( of @xmath314 ) . for a given generic covering @xmath182 , the above condition itself does not determine @xmath304 uniquely : item 2 ) of theorem [ gen_cover ] says that there are four non - equivalent curves @xmath304 with the same branch points of @xmath314 . we then choose the curve @xmath304 in the form @xmath315 ^ 2 - \psi(x ) = 0 , \notag\end{gathered}\ ] ] with the polynomials @xmath316 specified in . observe that @xmath304 is a 2-fold covering of the elliptic curve @xmath317 , and the covering is ramified at 4 points with @xmath318 , which are the branch points of @xmath312 . thus the conditions are satisfied . a similar transformation applied to @xmath304 yields back the curve isomorphic to @xmath3 in . the following theorem is a reformulation of the result obtained in section 3 of @xcite . [ bigonal ] for the curves @xmath319 described by , , the corresponding varieties @xmath320 are dual . as a result , @xmath321 and , therefore , @xmath322 . it is natural to call the curves @xmath323 _ dual _ , or following @xcite , _ bigonally related _ , as the above construction is reminiscent ( but not equivalent ) to donagi s tetragonal construction involving unramified coverings of curves and @xmath304 are not birationally equivalent . ] . we now apply the above construction of dual pencils to the curves arising in the reductions of the kovalevskaya gyrostat . recall that , according to theorems [ prym - tori ] , [ prym - tori2 ] , generic complex invariant manifolds of this system are open subsets of @xmath7 , where the quotient spectral curve @xmath3 and the underlying elliptic curve @xmath4 have the properties of theorem [ gen_cover ] . if the system is linearized on the jacobian of a genus 2 curve @xmath289 ( a curve of separation of variables ) , then @xmath324 is isogeneous to @xmath7 . and , if the isogeny is of degree 2 , then , according to theorem [ main_th ] , @xmath289 must be equivalent to one of the genus 2 curves @xmath325 in diagram . we intend to calculate some of these curves by using theorem [ fibers ] without finding the family @xmath283 explicitly . as was already noticed , for a non - zero value of the area integral @xmath326 , the version @xmath98 of @xmath3 in is smooth and non - hyperelliptic ( even for @xmath101 ) . observe however that , by the formula , the dual to @xmath98 reads @xmath327 for @xmath128 , @xmath328 is again smooth , genus 3 , and non - hyperelliptic , but for @xmath329 ( the case of the classical kovalevskaya top ) the polynomial @xmath330 , defining the underlying elliptic curve @xmath331 , has a double root @xmath332 , and @xmath328 becomes singular of geometric genus 2 . by the birational transformation @xmath333 it is send to the hyperelliptic form @xmath334 now one can easily observe that , under the substitution @xmath335 , @xmath336 , the above equation takes the form of the original kovalevskaya curve @xmath8 in . as a result , we conclude : _ modulo birational transformations , the kovalevskaya curve of separation of variables is dual to the quotient genus 3 spectral curve of the lax pair given in @xcite . _ then , according to theorems [ fibers ] , [ main_th ] , @xmath8 is equivalent to one of the genus 2 curves @xmath337 in diagram , and @xmath114 is a 2-fold covering of @xmath7 , the compactified complex invariant manifold @xmath116 of the kovalevskaya top . equivalently , @xmath116 is an 8-fold covering of the torus conformally equivalent to @xmath114 , as was previously shown in @xcite without using the spectral curve @xmath98 . [ [ the - case - gamma0-i_10 . ] ] the case @xmath338 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + the above configuration also holds for zero value of the area integral : the kovalevskaya curve becomes @xmath339 it is still smooth of genus 2 , and its jacobian is a 2-fold covering of @xmath108 . on the other hand , as was already noticed in @xcite , in this case the quotient spectral curve @xmath98 is itself singular of genus 2 having the hyperelliptic form @xmath119 in . then , theorem [ fibers ] implies that @xmath119 must be equivalent to one of the curves @xmath340 in , and , by theorems [ prym - tori ] , [ main_th ] , @xmath341 is a 2-fold covering of @xmath124 . thus @xmath342 and @xmath124 are isogeneous , one is a 4-fold covering of another , so it is natural to expect that @xmath343 are connected via the richelot transformation . and the authors of @xcite showed that this indeed holds by presenting this transformation explicitly . in this connection it is also worth mentioning another genus 3 curve @xmath344 appearing in @xcite as the spectral curve of a @xmath160 lax pair for the kovalevskaya goryachev chaplygin top , under the condition @xmath117 . as was shown in @xcite , @xmath344 can be transformed to the following hyperelliptic form in variables @xmath345 @xmath346 this curve has the structure in proposition [ 3->2 ] , hence it is 2-fold covering of an elliptic curve @xmath4 and of the genus 2 curve @xmath347 , the latter being precisely the curve @xmath119 in . then @xmath124 is a 2-fold covering of @xmath348 and the following diagram of coverings holds ( see also @xcite ) @xmath349 [ [ the - case - gammane-0-i_10 . ] ] the case @xmath350 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + as in the previous case , the curve @xmath98 in is singular of genus 2 whose hyperelliptic form is a slight generalization of , namely @xmath351 here we set @xmath352 , @xmath353 . then , following theorem [ fibers ] , @xmath354 can be identified with one of the curves @xmath340 in diagram , and the invariant manifold @xmath115 is a 2-fold covering of @xmath355 . hence @xmath354 is a curve of separation of variables for the kovalevskaya gyrostat on the level @xmath117 . in the same case , the dual curve remains to be smooth of genus 3 and non - hyperelliptic , so one can not immediately apply theorem [ fibers ] to find equations of one of the curves in diagram . however , given a curve @xmath340 , to calculate one of the curves @xmath337 one can make use of the richelot relations , that is , to apply the richelot transformation to the known curve @xmath354 . omitting intermediate calculations , we present the result in the form @xmath356 where @xmath357 is the simplest root of the cubic polynomial in the equation of @xmath354 , @xmath358 and @xmath359 is a solution of the quadratic equation @xmath360 it is a simple exercise to show that @xmath361 as @xmath362 ( and , therefore , @xmath363 ) , then the equation of @xmath364 reduces to @xmath365 , which transforms to the kovalevskaya curve @xmath366 for @xmath117 by the birational change @xmath367 , @xmath368 . thus , due to theorem [ main_th ] , we conclude that the jacobian of @xmath364 is a 2-fold unramified covering of @xmath369 , or , equivalently , the latter is an 8-fold covering of the complex torus conformally equivalent to @xmath370 , hence the curve is a separation curve for the gyrostat on the level @xmath117 . this can be summarized in the diagram @xmath371 the structure of @xmath364 suggests that in the construction of a separation of variables for the kovalevskaya gyrostat leading to a generalization of the curve @xmath8 for @xmath120 , one can not avoid solving algebraic equations of degree @xmath372 . there are numerous publications devoted to linearization of the kovalevskaya gyrostat on the jacobians of genus 2 curves , both for @xmath117 and @xmath373 , see @xcite , however we did not find there any explicit expression for a separation genus 2 curve . for various deformations of the gyrostat , the lax pairs and separation curves have been presented in @xcite . [ [ the - general - case - gammane-0-i_1-ne-0 . ] ] the general case @xmath374 . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + as was already mentioned , in this case neither the original quotient spectral curve @xmath98 not its dual @xmath328 are singular or hyperelliptic , hence one can not apply theorem [ fibers ] to obtain the corresponding genus 2 curves @xmath375 in diagram without knowing explicitly the pencil of genus 3 curves @xmath283 . here , as above , we consider only the case @xmath376 , the corresponding quotient spectral curve @xmath85 given in . this situation is , in a sense , simpler than the previous one : as maple command weierstrassform shows , for generic constants of motion , @xmath85 is hyperelliptic of genus 3 and is equivalent to @xmath377 where , as above , @xmath352 , @xmath353 . the birational transformation between and is described by @xmath378 thus @xmath379 has the structure of , and , by proposition [ 3->2 ] , it is 2-fold covering of the elliptic curve @xmath380 ( which is birationally equivalent to @xmath82 in , as expected ) and of the genus 2 curve @xmath381 . according to theorems [ fibers ] , [ main_th ] , @xmath289 is equivalent to one of the curves @xmath337 in diagram , and the following holds . [ gg ] the complex invariant torus @xmath382 of the gyrostat in two fields is isogeneous to the jacobian of @xmath383 , @xmath384 being specified in . more precisely , @xmath324 is a 2-fold covering of @xmath130 or , equivalently , @xmath116 is an 8-fold covering of the torus conformally equivalent to @xmath324 . note that under the shift @xmath385 , the equation of @xmath289 simplifies to @xmath386 where we set @xmath387 . it follows that the curve @xmath289 can be a curve of separation of variables for the reduced kovalevskaya gyrostat in two fields with general constants of motion . to our knowledge , an explicit linearization of this system is still an open problem . there are however several papers presenting linearizations for particular cases of motion . in section 3 of @xcite , harlamov and yehia gave an explicit separation of variables for the case @xmath388 , where , as above , @xmath389 \rangle$ ] is the hamiltonian of the @xmath2-symmetry . following @xcite , in this case the reduced lagrangian of the problem ( the routh function ) does not contain linear terms in velocities . then the corresponding hamilton function on @xmath90 has a stckel form , which ensures a separation of variables . the corresponding quadratures involve a genus 2 curve , which , for @xmath74 , _ in our notation _ , reads @xmath390 here one should mention that previously , in @xcite , the kovalevskaya gyrostat for @xmath388 was shown to be isomorphic with goryachev s particular case of integrability of the kirchhoff equations ( @xcite ) , and the latter system had been reduced to quadratures , by different methods , in @xcite . now , observe that , in view of , the condition @xmath388 implies @xmath391 . substituting this into and replacing @xmath392 by @xmath393 , we obtain the following equation of the curve @xmath289 @xmath394 since @xmath324 is isogeneous to the invariant torus @xmath116 , it can also be a curve of separation of variables for @xmath391 . an algorithm given in @xcite allows to compare the absolute invariants of the curves , in a numerical example , and the result indicates that the curves are not birationally equivalent . on the other hand , the structure of is especially convenient to apply the richelot transformation to it , which yields the following curve @xmath395 under the trivial substitutions @xmath396 , the latter takes the form . as a result , we have recovered the separation curve of @xcite for the case @xmath388 from the quotient spectral curve @xmath85 . due to the richelot relation between the curves @xmath397 and according to diagram , @xmath324 is a 4-fold covering @xmath398 . then theorem [ main_th ] implies the following tower of unramified coverings @xmath399 more curves of separation in particular cases of motion can be obtained by considering the curve dual to @xmath85 in , which reads @xmath400 and it is 2-fold covering of the generally elliptic curve @xmath401 for @xmath120 and general constants of motion , @xmath402 is smooth of genus 3 and , in contrast to @xmath85 , non - hyperelliptic . it becomes singular of geometric genus 2 in two obvious cases : + 1 ) @xmath87 ( @xmath403 is singular ) ; 2 ) @xmath404 ( @xmath403 is of genus 0 ) . in the first , i.e. , gyroscopic - free , case , in view of , there is the relation @xmath405 , and the curve @xmath402 admits the following hyperelliptic form @xmath406 \ , , \ ] ] the corresponding birational transformation is @xmath407 thus , following theorem [ fibers ] , for @xmath87 , @xmath408 is a 2-fold covering of the invariant torus @xmath130 . [ [ remark-3 . ] ] remark 3 . + + + + + + + + + one can then compare @xmath409 with the curve @xmath289 in proposition [ gg ] for @xmath87 , which takes the form @xmath410 both @xmath411 and @xmath408 are 2-fold coverings of @xmath130 , however the curves themselves are not birationally equivalent : in a numerical example their igusa absolute invariants are distinct . hence , @xmath412 represent two different curves @xmath337 in diagram , and both curves can serve as separation curves for the case @xmath87 . observe that the second special case @xmath413 is different from @xmath414 considered in @xcite and above . now the curve @xmath402 in can be transformed to the following hyperelliptic form @xmath415\ ] ] with the transformation relations @xmath416 it follows that @xmath417 is a 2-fold covering of the invariant torus @xmath130 , and @xmath418 is a possible curve of separation for @xmath413 . apparently , an explicit separation of variables in this case is still an open problem . [ [ the - case - gamma0-f_20-linearization - on - two - elliptic - curves . ] ] the case @xmath419 : linearization on two elliptic curves . + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + as one can observe , in the limit @xmath419 , which implies the relation @xmath391 , the equation of the above curve @xmath418 becomes @xmath420 , so it is singular of genus 0 . the same holds for the curves @xmath421 in for @xmath87 and @xmath409 in for @xmath422 . in this limit the original dual curve @xmath402 in takes the algebraic form + @xmath423 , which admits factorization @xmath424 it follows that @xmath425 is isomorphic to the direct product of two elliptic curves @xmath426 this fact can be compared with the observation of @xcite that in the case @xmath422 the two field kovalevskaya top has a certain analogy with chaplygin s particular integrable case of the problem of motion of a body in a fluid @xcite . the latter case reduces to quadratures which consist of two different elliptic integrals . and in section 5 of @xcite a similar reduction was made for the kovalevskaya top : the corresponding quadratures involve the elliptic curves given by equations @xmath427 one can check however that these curves are not equivalent to @xmath428 in . on the other hand , observe that under the same conditions @xmath419 the alternative genus 2 curve @xmath429 in remains smooth and takes the form @xmath430 then theorem [ fibers ] implies that @xmath431 is a 2-fold unramified covering of the complex invariant manifold @xmath115 . upon setting again @xmath432 the curve reads @xmath433 one immediately sees that @xmath434 . this case corresponds to the simplest possible reduction of curves and abelian functions studied by jacobi and described in many books , see , e.g. , @xcite . [ jacobi ] under the condition @xmath435 , the curve of the form is a 2-fold ramified covering of two generally different elliptic curves @xmath436 the jacobian of @xmath437 contains @xmath239 as abelian subvarieties , which intersect at 4 points . @xmath431 is not the direct product @xmath438 , but isogeneous to it : @xmath439 . in view of , @xmath440 , hence in our case @xmath441 , and , under the change @xmath442 , @xmath443 , the curves @xmath444 take the form @xmath445 the latter coincide with the separation curves obtained in @xcite for the case @xmath419 . the appearance of the direct product of curves @xmath428 , the pair @xmath444 , and of the double coverings @xmath446 implies that the considered case is described by proposition [ ell_case ] and diagrams : + @xmath431 and the product @xmath447 are ( different ) 2-fold unramified coverings of the complex invariant torus @xmath115 , which , itself , is a 2-fold unramified covering of the direct product @xmath448 . then , following proposition [ ell_case ] , the elliptic curves @xmath444 and @xmath428 must be isogeneous . and , indeed , @xmath449 ( respectively , @xmath450 ) is obtained from @xmath451(respectively , @xmath452 ) by duplicating one of the two period vectors . the equations of the corresponding curves are obtained from each other by a second order transformation ( also called _ landen s transformation _ , see , e.g. , @xcite ) , which can be described as follows . [ landen ] the equation of any double unramified covering of an elliptic curve @xmath453 is given by regularization @xmath454 of the curve @xmath455 @xmath456 . for example , @xmath457 the covering @xmath458 is obtained by doubling the period @xmath459 ( respectively , @xmath460 ) in the parallelogram of periods of @xmath4 . following this procedure , we make the second order substitution @xmath461 in the equation of @xmath451 obtaining @xmath462 .\ ] ] regularizing this curve by omitting full squares , we get the following equation of the 2-fold covering of @xmath451 @xmath463 .\ ] ] then the birational substitution @xmath464 , @xmath465 takes the above equation precisely to the curve @xmath449 in . the same composition of changes takes @xmath452 to @xmath450 . as a result , the product @xmath447 can be regarded as a 4-fold covering of @xmath466 , and vice versa ( if we identify conformally equivalent abelian tori ) . and , as stated in proposition [ jacobi ] , @xmath466 is also a 4-fold covering of @xmath431 . all the observations lead to the following diagram of coverings of abelian varieties in the case @xmath419 . according to proposition [ ell_case ] , apart from the curves @xmath467 , and @xmath437 there exist three other genus 2 curves : a curve @xmath468 , which is also a 2-fold covering of @xmath239 , and two non - equivalent curves @xmath246 , being such coverings of @xmath247 ( we do not present their equations here ) . all of them can be used as curves of separation in the considered special case , as their jacobians are isogeneous to the complex invariant torus @xmath115 . we conclude by summarizing all the curves of this subsection , as well as the relations between them , in the following diagram where the arrows denote reductions to particular cases of motion or coverings , or the richelot transformation , and the symbol @xmath469 indicates non - equivalence . from the spectral curve of the lax pair of the kovalevskaya gyrostat in one and two fields presented in @xcite we managed to recover all known curves of separation of variables , including the original kovalevskaya curve , and constructed new ones . naturally , the approach we used does not allow to make an explicit reduction of the systems to quadratures . however , the obtained separation curves can be used in analysis of bifurcations of complex invariant manifolds of the systems . next , the approach itself does not permit to reconstruct explicitly the whole pencil of genus 3 curves @xmath283 , which might be used to derive alternative genus 2 curves , whose jacobians are degree 2 isogeneous to @xmath7 , and which are depicted in diagram . returning to the case of general configuration of the kovalevskaya gyrostat in two fields , when the conditions @xmath376 do not hold , it is natural to conjecture that generic complexified invariant tori of the system are isomorphic to the 3-dimensional prym variety @xmath470 associated with the quotient genus 4 curve @xmath3 in covering an elliptic curve . then an interesting open problem is to describe principally polarized abelian tori isogeneous to @xmath470 as jacobians of genus 3 curves . the latter will serve as curves of separation of variables for this integrable system . the authors are grateful to a. borisov and p. ryabov for discussions and clarifying remarks . the contribution of y.f was partially supported by the spanish mineco - feder grants mtm2012 - 31714 , mtm2012 - 37070 . he also gratefully acknowledges the hospitality and a financial support of unam , mexico city , where a part of this work has been completed . lgn acknowledges support for his research from the program unam - dgapa - papiit - ia103815 . sokolov , v. v. ; tsyganov , a. v. lax pairs for deformed kovalevskaya and goryachev - chaplygin tops . ( russian ) _ teoret . mat . * 131 * ( 2002 ) , no . 1 , 118125 ; translation in _ theoret . and math . * 131 * ( 2002 ) , no . 1 , 543549 tsiganov , a. v. on the kowalevski goryachev chaplygin gyrostat . _ j. phys . a _ * 35 * ( 2002 ) , no . 22 , l309l318 tsiganov , a. v. on the generalized chaplygin system . _ zap . nauchn . sem . otdel . mat . inst . steklov . ( pomi ) _ * 374 * ( 2010 ) , 250267 ( russian ) ; translation in _ j. math . _ * 168 * ( 2010 ) , no . 6 , 901911

we consider hnon - like families of diffeomorphisms of and study the boundary of the region of parameter values for which the nonwandering set is uniformly hyperbolic . assuming sufficient dissipativity , we show that the loss of hyperbolicity is caused by a first homoclinic or heteroclinic tangency and that uniform hyperbolicity estimates hold _ uniformly in the parameter _ up to this bifurcation parameter and even , to some extent , at the bifurcation parameter .

math0502235

our aim in this paper is to study the _ boundary of hyperbolicity _ of certain families of two dimensional maps . we say that a family of @xmath2 plane diffeomorphisms is called a _ hnon - like family _ if it can be written in the form @xmath3 where @xmath4 , @xmath5 and @xmath6 is a @xmath0 `` perturbation '' of the standard _ hnon family _ @xcite satisfying @xmath8 in this paper we consider @xmath9 fixed sufficiently small and investigate the dynamics as the parameter @xmath10 is varied . for simplicity we shall therefore omit @xmath11 and @xmath12 from the notation and denote a hnon - like family by @xmath13 . for future reference we remark that the inverse of @xmath14 is given by an equation of a similar form : @xmath15 where @xmath16 as @xmath17 . we shall suppose without loss of generality that @xmath18 hnon and hnon - like families have been extensively studied over the last almost 30 years . one of the earliest rigorous results on the subject is @xcite in which it was shown that the non - wandering set @xmath19 is uniformly hyperbolic for all @xmath20 and all sufficiently large @xmath10 ( depending on @xmath21 ) . on the other hand , for small @xmath22 and @xmath23 there exists positive probability of `` strange attractors '' which contain tangencies between stable and unstable leaves . this was first proved in @xcite for the hnon family and later generalized in @xcite to hnon - like families , see also @xcite . these attractors can not be uniformly hyperbolic due to the presence of tangencies but turn out to satisfy weaker _ nonuniform _ hyperbolicity conditions @xcite . more recently bedford and smillie have described the transition between these two regimes for hnon families by identifying and describing some of the properties of the _ boundary of uniform hyperbolicity _ @xcite . in particular they show that for small @xmath24 , the nonwandering set is uniformly hyperbolic up until the first parameter @xmath10 at which a tangency occurs between certain stable and unstable manifolds . combining this with the statements contained in @xcite their results also imply uniform bounds on the lyapunov exponents of all invariant probability measures at the bifurcation parameter @xcite . their methods rely crucially on previous work @xcite which in turn is based on the polynomial nature of the hnon family , a feature which allows them to consider its complexification and to apply original and highly sophisticated arguments of holomorphic dynamics . in this paper we develop a new and completely different strategy to the problem , based purely on geometric `` real '' arguments , which have the advantage of applying to general @xmath0 _ hnon - like _ families . we also obtain the analogous _ uniformity _ results by showing that the hyperbolicity expansion and contraction rates are uniform right up to the point of tangency and that even _ at _ the point of tangency some strong version of nonuniform hyperbolicity continues to hold : all lyapunov exponents of all invariant measures are uniformly bounded away from 0 . for all @xmath25 and @xmath26 sufficiently small we have the following property . for every hnon - like family @xmath27 of plane diffeomorphisms there exists a unique @xmath28 such that 1 . for all @xmath29 the nonwandering set @xmath30 is uniformly hyperbolic ; 2 . for @xmath31 the nonwandering set @xmath32 contains an orbit of tangency but is `` almost uniformly hyperbolic '' in the sense that all lyapunov exponents of all invariant probability measures supported on @xmath33 are uniformly bounded away from 0 . moreover , the bounds on the expansion and contraction rates for all @xmath34 are independent of @xmath10 and of the family . we remark that this is not the only existing definition of hnon - like in the literature . one standard approach is to consider `` singular '' perturbations of the limiting one - dimensional map corresponding to the case @xmath35 : @xmath36 this formulation however has some slight technical issues . for example , one can not assume that @xmath37 is small on all of @xmath38 since that would violate the requirement that @xmath39 be a global diffeomorphism of @xmath1 . this can be dealt with by restricting our attention to some compact region , say @xmath40 \times [ -2,2 ] $ ] , and supposing only that @xmath41 in this region . our arguments apply in this case also and yield a more local result on the hyperbolicity of the nonwandering set restricted to @xmath40 \times [ -2,2 ] $ ] . we recall that a point @xmath42 belongs to the _ nonwandering _ set @xmath43 of @xmath44 if it has the property that for every neighbourhood @xmath45 of @xmath42 there exists some @xmath46 such that @xmath47 . the nonwandering set is always invariant and closed ( and thus if bounded , also compact ) . we say that a compact invariant set @xmath33 is _ uniformly hyperbolic _ ( with respect to @xmath44 ) if there exists constants @xmath48 and a _ continuous _ decomposition @xmath49 of the tangent bundle such that for every @xmath50 , every non - zero vector @xmath51 and @xmath52 and every @xmath46 we have @xmath53 by standard hyperbolic theory , the stable and unstable subspaces @xmath54 are tangent everywhere to the stable and unstable manifolds . in particular uniform hyperbolicity is incompatible with the presence of any tangencies in @xmath33 between any stable and any unstable invariant manifolds associated to points of @xmath55 . a weaker notion of hyperbolicity can be formulated in terms of invariant measures . for simplicity we restrict our discussion to the two - dimensional setting , as relevant to the situation we consider in this paper . let @xmath56 be an @xmath44-invariant ergodic probability measure with support in some compact invariant set @xmath33 . by oseledec s ergodic theorem @xcite there exist constants @xmath57 and a measurable decomposition @xmath49 such that for @xmath56-almost every @xmath58 and every non - zero vector @xmath51 and @xmath52 we have @xmath59 the constants @xmath60 and @xmath61 are called the _ lyapunov exponents _ associated to the measure @xmath56 . we say that @xmath56 is _ hyperbolic _ @xcite if @xmath62 clearly implies for any @xmath56 . the converse however is false in general : the measurable decomposition may not extend to a continuous one on all of @xmath33 and the exponential expansion and contraction in implies only a limited version of in which the constants @xmath63 are measurable functions of @xmath64 and not bounded away from 0 . this definition of hyperbolicity in terms of lyapunov exponents is sometimes called _ nonuniform hyperbolicity _ and is consistent in principle with the existence of tangencies between stable and unstable manifolds . in general there may be many ergodic invariant probability measures supported in @xmath33 of which some may be hyperbolic and some not . even if they are all hyperbolic the corresponding lyapunov exponents may not be uniformly bounded way from @xmath65 . the situation in which all lyapunov exponents of all ergodic invariant measures are uniformly bounded away from zero is , in some sense , as `` uniformly hyperbolic '' as one can get while admitting the existence of tangencies . this situation can indeed occur , for example in the present context of hnon - like maps . a first example of a set satisfying this property was given in @xcite . after completing the proof of the theorem 1 we realized that _ much simpler _ versions of our arguments yield an analogous , new and non - trivial , result in the context of one - dimensional maps . we explain and give a precise formulation of this result . we consider first the quadratic family @xmath66 we choose this particular parametrization for convenience and consistency with our two dimensional results , but any choice of smooth family of unimodal or even multimodal maps with negative schwarzian derivative would work in exactly the same way . it is well known that for @xmath67 the nonwandering set @xmath30 is uniformly expanding although we emphasize here that this depends crucially on the negative schwarzian derivative property . the negative schwarzian property is not robust with respect to @xmath0 perturbations and standard methods do not therefore yield this statement for such perturbations . there exists a constant @xmath68 such that if a family @xmath69 of @xmath70 one - dimensional maps satisfies @xmath71 then there exists a unique parameter value @xmath28 such that 1 . for all @xmath72 the non - wandering set @xmath33 is uniformly hyperbolic ; 2 . for @xmath31 the lyapunov exponents of all ergodic invariant probability measures are all positive and uniformly bounded away from 0 . moreover the rates of expansion and the bound on the lyapunov exponents are uniform , independent of the family and of the parameter . the proof of this result is exactly the same as that of theorem 1 but hugely simpler as all more geometrical arguments concerning curvature etc become essentially trivial . we emphasize that the uniform expansivity of @xmath30 for a particular parameter value @xmath67 is of course robust under sufficiently small perturbations of @xmath39 , by standard hyperbolic theory . however this approach _ requires the size of the perturbation to depend on the parameter _ @xmath10 and in particular to shrink to zero as @xmath10 tends to @xmath73 . the crucial point of our approach , both in this one - dimensional setting , as in the two - dimensional setting is that the size of the perturbation _ does not _ depend on the parameter . we have divided our argument into three main sections . in section [ proofnonwan ] we analyze the geometric structure of stable and unstable manifolds of the two fixed points and define the parameter @xmath28 as the first parameter for which a tangency occurs between some compact parts of these manifolds . we also identify a region @xmath74 which we show contains the non - wandering set . in section [ sectionhyp ] we define a `` critical neighbourhood '' @xmath75 outside of which our maps are uniformly hyperbolic by simple perturbation arguments . however @xmath76 does contain points of @xmath33 and thus we can not ignore this region . to control the hyperbolicity in @xmath75 we introduce the notions of hyperbolic coordinates and critical points which form the key technical core of our approach . finally , in section [ sectionhypest ] we apply these techniques to prove the required hyperbolicity properties . in this section we define the parameter @xmath77 as in the statement of our main theorem , and show that for @xmath78 the nonwandering set is contained in the closure if the unstable manifold of a hyperbolic fixed point restricted to a certain compact region of @xmath79 . we define the bifurcation parameter @xmath28 below as the first parameter for which there is a tangency between certain compact parts of the stable and unstable manifolds of the fixed points . this does not immediately imply that it is a first parameter of tangency though this will follow from our proof of the fact that the nonwandering set is uniformly hyperbolic for all @xmath29 . for the endomorphisms @xmath81 with @xmath82 , there are two fixed points , @xmath83 both hyperbolic . for the special parameter value @xmath84 , to simplify the notation below , we write @xmath85 since @xmath86 and @xmath87 are repelling in the horizontal direction , their stable sets are simply their preimages : @xmath88 in particular these sets contain the following curves @xmath89 and @xmath90 and @xmath91.,scaledwidth=30.0% ] the first preimage of @xmath86 is a parabola with a minimum at @xmath92 and intersecting the @xmath64-axis at @xmath93 , and having slope equal to @xmath94 at the point @xmath95 and the second is a parabola with a minimum at @xmath96 . similarly we can compute @xmath97 which is a parabola with a minimum at @xmath98 , intersecting the @xmath64-axis at @xmath99 and having slope equal to @xmath73 at the point @xmath100 , and @xmath101 which is a parabola with a minimum at @xmath102 . the unstable manifolds going @xmath103 and @xmath104 can be defined and computed in a similar way and are easily seen to be horizontal . consider first the _ hnon family _ @xmath105 for @xmath106 , @xmath107 is a diffeomorphism . the hyperbolicity of the fixed points implies that there exists a neighbourhood of the set @xmath108 corresponding to pairs of parameters for which there is an _ analytic continuation _ @xmath109 as hyperbolic fixed points of @xmath107 . considering @xmath110 small , we also have that the analytic continuations @xmath111 and @xmath112 are also well defined and hyperbolic . for simplicity we shall often just refer to these two points as @xmath113 leaving implicit their dependence on @xmath44 . explicit formulas for @xmath109 can be easily derived from the equation @xmath114 but these would not be particular useful . instead we just observe that the fixed points must lie on the line @xmath115 and so in particular this means that for @xmath116 and @xmath117 the vertical coordinates of @xmath118 and @xmath119 are negative and positive respectively , and the converse for @xmath120 . clearly the same holds for @xmath121 and @xmath122 if @xmath12 is sufficiently small . moreover , the determinant of @xmath123 is given by @xmath124 in particular the determinant is constant and negative if @xmath21 is positive and positive if @xmath21 is negative . we thus refer to the case @xmath125 as the _ orientation - reversing _ case , and the case @xmath126 as the _ orientation - preserving _ case . ) and the orientation - preserving ( @xmath127 ) case ( dotted curves indicate negative eigenvalues),title="fig:",scaledwidth=45.0% ] ) and the orientation - preserving ( @xmath127 ) case ( dotted curves indicate negative eigenvalues),title="fig:",scaledwidth=45.0% ] recall that the determinant of a matrix is the product of the eigenvalues , and thus in particular , the sign of the determinant has implications for the sign of the eigenvalues which , as we shall see , in turn has implications for the geometry of the stable and unstable manifolds of the fixed points . for @xmath35 the fixed points @xmath128 and @xmath129 have derivatives @xmath130 and @xmath131 respectively , and thus , for @xmath22 and @xmath12 small , the expanding eigenvalues of @xmath132 and @xmath133 are @xmath134 and @xmath135 respectively . this implies that for @xmath136 , the orientation - reversing case , the contracting eigenvalues of @xmath137 and @xmath132 must be @xmath138 and @xmath139 respectively , while for @xmath127 , the orientation preserving case , they must be @xmath139 and @xmath140 respectively . the two situations are illustrated in figure [ fixedpoints ] with dashed lines showing the invariant manifolds corresponding to negative eigenvalues . by classical hyperbolic theory , compact parts of the stable manifolds depend continuously on the map ( see e.g. @xcite ) . therefore , for small @xmath21 and small @xmath12 the analytic continuations @xmath113 of the fixed points @xmath129 and @xmath128 have stable and unstable manifolds which are close to those computed above for the limiting case . elementary calculations show that the actual geometrical relations between these continuations depend on whether we consider the orientation reversing ( @xmath125 ) or the orientation preserving ( @xmath127 ) case , and are as illustrated in figure [ intersection ] . , width=340 ] we let @xmath141 denote the compact parts of the stable and unstable manifold as shown in figure [ intersection ] and notice that , in particular , since for @xmath35 and @xmath142 the unstable manifold of @xmath143 and @xmath144 extend to the whole of the line , for each @xmath67 and @xmath125 sufficiently small we have that @xmath145 crosses @xmath146 four times , and for each @xmath67 and @xmath127 sufficiently small we have that @xmath147 crosses @xmath148 four times , and also we can ensure that the compact parts defined above and in the figure intersect transversally . again this continues to hold also for a hnon - like family for sufficiently small @xmath149 . we are now ready to define the parameter @xmath28 . we fix @xmath22 . + for an _ orientation - reversing _ ( @xmath125 ) hnon - like family @xmath39 , we let @xmath150 for an _ orientation - preserving _ @xmath127 hnon - like family @xmath39 , we let @xmath151 we also define a parameter @xmath152 as the @xmath153 of parameters @xmath10 for which @xmath145 crosses @xmath146 four times ( @xmath125 ) or @xmath154 crosses @xmath155 four times ( @xmath127 ) . clearly this is a weaker condition and thus @xmath156 . notice that @xmath28 and @xmath157 converge to @xmath158 as @xmath21 and @xmath159 tend to 0 . in this section we carry out a detailed geometrical study aimed at showing that the nonwandering set is contained in a relatively restricted region . to prove hyperbolicity we will then just have to focus our efforts in this region . for the moment we restrict ourselves to the orientation - reversing case . at the end we shall remark how the orientation - preserving case follows by identical arguments with a few minor changes of notation . first of all we let @xmath160 denote the closed topological disc bounded by compact pieces of the @xmath161 and @xmath162 as shown in figure [ regiond ] . , width=151 ] the main result of this section is the following [ nonwan ] for all @xmath163 we have @xmath164 \times ( -4b , 4b)\}.\ ] ] in this paper we are interested in parameters @xmath165 , but it is worth observing that if follows from proposition [ nonwan ] that for all @xmath166 , and so in particular for a certain range of parameter values which may contain multiple tangencies the recurrent dynamics is captured to some extent by the dynamics on @xmath161 . this includes in particular all complex dynamical phenomena associated to the unfolding of the tangency at the parameter @xmath28 ( indeed , this includes the range of parameters considered by benedicks - carleson in @xcite . we split the proof of proposition [ nonwan ] in several lemmas . once again we deal first with the case @xmath125 . at the end of the proof we indicate the necessary modifications in order to deal with the case @xmath167 . we first define a relatively `` large '' region @xmath168 and show that @xmath169 and then show in separate arguments that @xmath170 and @xmath171 , and finally refine our estimate to obtain the statement in the proposition . let @xmath172 we also define the following 6 ( overlapping ) regions ( see figure [ regionv ] ) : @xmath173 to @xmath174,width=340 ] then @xmath175 we prove the following two statements . [ 1.1 ] @xmath176 . we show that the orbit of every point @xmath177 , @xmath178 is unbounded in either backward or forward time . this implies in particular that no such point is nonwandering . for @xmath179 , let @xmath180 . we shall use repeatedly the fact that @xmath116 and @xmath181 . for @xmath182 we have @xmath183 and therefore @xmath184 and @xmath185 , and @xmath186 . repeating the calculation we have @xmath187 and so @xmath188 . for @xmath189 we have @xmath190 and @xmath191 . thus @xmath192 and so @xmath188 . similarly , for @xmath193 we have @xmath194 and @xmath191 . thus @xmath192 and we argue as above . for @xmath195 we consider backward iterations of @xmath196 . note that @xmath197 . then @xmath198 and @xmath199 . thus @xmath200 and @xmath201 . therefore @xmath202 and so @xmath203 we have @xmath204 . thus @xmath205 and @xmath206 . so we have that @xmath200 , and we argue as above . for @xmath207 , we have @xmath208 and @xmath209 . therefore , @xmath210 and again , we argue as above . @xmath211 . the arguments used above have implications for the locations of the stable and unstable manifolds of the fixed points . indeed the stable manifolds of the fixed points can not intersect @xmath212 since all points in this region tend to infinity in forward time , whereas , by definition , points in the stable manifolds tend to the fixed points under forward iteration . similarly the unstable manifolds of the fixed points can not intersect @xmath213 since all points in this region tend to infinity in backwards time . by definition @xmath74 is bounded by arcs of stable and unstable manifolds of the fixed point as in the figure and therefore @xmath214 . , width=340 ] @xmath215 to show that @xmath216 we refine the strategy used in the proof of the previous lemma , and show that the orbits of all points outside @xmath217 are unbounded in either backward or forward time . since we have already shown that @xmath218 , we need to consider only points in the region @xmath219 . we write @xmath220 where the regions @xmath221 are defined as follows . consider the points @xmath222 and @xmath223 of intersection of @xmath224 and @xmath225 and @xmath226 , @xmath227 of intersection of @xmath225 and @xmath228 as in the figure [ regionsr ] . we let @xmath229 denote the closed region bounded by the arcs of manifold @xmath230 , @xmath231 and @xmath232 , and the segment @xmath233 . we let @xmath234 denote the region bounded by the arc of manifolds @xmath235 , @xmath236 and @xmath230 , and the segment @xmath237 . we let @xmath238 denote the region bounded by the arcs of @xmath239 and @xmath225 between the points @xmath240 and @xmath241 , as in figure [ regionsr ] . we let @xmath242 . we also define @xmath243 as the region satisfying @xmath244 at the left side of the arc of @xmath225 between the points @xmath245 and @xmath246 , of intersection of that manifold with the lines @xmath247 and @xmath248 , as in the figure [ regionsrnew ] ( b ) . since @xmath250 is small , we have that all the points @xmath251 satisfy @xmath252 . notice that , for the unperturbed hnon map @xmath253 , we have that any piece of curve @xmath254 with slope less than @xmath255 contained in the region where @xmath256 is mapped to another curve with slope less than @xmath255 . indeed , letting @xmath257 denote a tangent vector to @xmath254 with @xmath258 , we have @xmath259 whose slope is @xmath260 , provided @xmath250 is small and @xmath261 is close to 2 . for future reference , notice that , if @xmath262 , we also have that the norm of @xmath263 is uniformly expanded . so , since @xmath264 is close to @xmath107 in the @xmath265 topology , we can assume that @xmath264 also has this property in @xmath229 . now call @xmath266 the successive images of the segment @xmath233 intersected to @xmath229 . since they can not intersect each other , and @xmath233 has a point of the stable manifold of @xmath267 , the curves @xmath266 determine a system of `` fundamental domains '' in @xmath229 : they cross @xmath229 from one stable boundary to the other , and they converge to the arc of unstable manifold @xmath231 . call @xmath268 the region of @xmath229 between @xmath269 and @xmath270 , @xmath271 , and notice that @xmath272 . we also have that @xmath273 falls outside @xmath274 . that implies that @xmath275 does not intersect @xmath276 , and any point which has an iterate in @xmath277 is not in @xmath278 . we show that @xmath281 . indeed , the unstable eigenvalue of @xmath132 is positive and therefore @xmath282 must remain on the same side of @xmath283 as @xmath284 . also , since @xmath285 $ ] we have that @xmath282 does not intersect any of @xmath286 . to @xmath287,width=340 ] we now show that @xmath289 . again , we use the fact that @xmath285 $ ] . then , since one of the components of the boundary of @xmath290 is an arc of stable manifold of @xmath291 containing the fixed point @xmath291 , and the unstable eigenvalue of @xmath291 is positive , we conclude that the image of @xmath292 is contained @xmath293 . we can assume , if @xmath250 is small , that all the points @xmath294 in @xmath295 satisfy @xmath296 ( notice that , for @xmath297 , we have @xmath298 . take @xmath299 a point in @xmath300 , and connect @xmath299 to the boundary of @xmath295 through a horizontal line inside @xmath301 , determining a point @xmath302 . again , by the proximity of @xmath303 and @xmath107 , and the fact that vectors with slope smaller than @xmath304 in @xmath305 are sent by @xmath306 in vectors with slope smaller than @xmath304 , and uniformly expanded , we have that the horizontal distance between @xmath307 and @xmath308 is uniformly expanded . applying @xmath303 repeatedly , as long as the image is inside @xmath309 , we have that the horizontal distance between the successive images of @xmath299 and @xmath225 increases exponentially . then the forward images of @xmath299 leave @xmath274 for some positive time . notice that @xmath311 since all the other regions in @xmath274 outside @xmath312 are mapped forward to the region @xmath295 , and so do not contain points of the backward image of @xmath313 . moreover , the unstable boundary of @xmath313 belongs to @xmath239 and approaches @xmath267 as we apply @xmath314 , and the stable boundary can not cross @xmath225 , then @xmath315 does not intersect @xmath312 . since @xmath316 , the points in there that are not in @xmath239 leave @xmath274 for backward iterations . for @xmath167,width=340 ] @xmath317 . notice first of all that by the @xmath318-lemma we have @xmath319 . now suppose by contradiction that there exists @xmath320 with @xmath321 . then there exists @xmath322 and an @xmath322 neighbourhood @xmath323 of @xmath42 with @xmath324 . since @xmath33 is @xmath44-invariant we have @xmath325 for all @xmath326 and therefore @xmath327 for all notice that the boundary @xmath329 ( where @xmath330 denotes the piece of @xmath331 between @xmath332 and @xmath333 , as in figure [ regionsr ] . it is enough therefore to show that , for large @xmath334 , the @xmath335 is @xmath322-dense in @xmath336 as this will imply that @xmath337 , contradicting the assumptions . this follows easily from the fact that @xmath44 is ( strongly ) area contracting , and thus the area of @xmath338 tends to zero as @xmath339 . in particular we must have that @xmath340 for all @xmath341 sufficiently large . moreover , the length of the part of the boundary which belongs to @xmath283 also tends to zero . thus most of the boundary belongs to @xmath161 and thus we must have @xmath342 for all @xmath343 sufficiently large . combining the results of the lemmas stated above we have that @xmath344 . the statement in the proposition now follows immediately by observing that @xmath345 implies @xmath346 and that @xmath347\times [ -4b , 4b ] $ ] directly from the definition of @xmath44 if @xmath12 is sufficiently small . finally , in the case @xmath348 , we consider the stable and unstable manifolds of @xmath291 crossing as in figure [ negativeb ] ( the rectangle @xmath274 is exactly the same ) , determining the region @xmath349 in this case . the proof is entirely analogous considering the points @xmath350 , etc . , corresponding to the points @xmath351 etc . the key idea of our whole strategy is the notion of _ dynamically defined critical point _ which relies on the fundamental notion of _ hyperbolic coordinates_. in this section we introduce these notions and develop the main technical ideas which we will use . in section [ prelim ] we clarify the relations between various constants used in the argument and introduce some preliminary geometric constructions . in section [ hypcoord ] we discuss the definition and basic theory of hyperbolic coordinates . in section [ curvature ] we introduce the idea of _ admissible curves _ and prove certain estimates concerning the images of admissible curves . finally , in section [ hypcrit ] we introduce the notion of dynamically defined critical point and prove that such critical points always exist in images of certain admissible curves . we now explain the required relations between the different constants used in the proof , and the order in which these constants are chosen . all constants are positive . first of all we fix two constants @xmath352 these will be introduced in sections [ fixptnhbd ] and [ curvature ] below . even though we specify the actual numerical value of these constants we shall continue to use the constants in the argument because they have some specific geometric meaning and it is useful to keep track of their occurrence throughout the paper . we then fix a constant @xmath353 large enough so that @xmath354 in section [ critnhbd ] we fix a constant @xmath322 which will then remain unchanged . finally , at some finite number of places in the argument , we will require @xmath10 to be sufficiently close to 2 and @xmath24 and @xmath12 to be sufficiently small . we remark that we can suppose that @xmath10 is close to 2 without compromising the fact that hyperbolicity holds for all larger values of @xmath10 . indeed , once we fix a neighbourhood of @xmath73 in the @xmath10 parameter space , we can always guarantee uniform hyperbolicity for values of @xmath355 outside this neighbourhood by taking @xmath24 and @xmath12 sufficiently small , ( depending on this neighbourhood ) . recall first of all that the map @xmath356 has two fixed points @xmath128 and @xmath357 with @xmath358 . for @xmath359 we let @xmath360 be the open ball of radius @xmath361 centred at @xmath129 and @xmath362 be the component of @xmath363 not intersecting @xmath364 , see figure [ q ] . the expanding eigenvalue at the point @xmath129 is equal to 4 and so we can suppose that @xmath365 are all small enough so that @xmath366 for all @xmath367 . then , for @xmath368 , let @xmath369 notice that @xmath370 is just the component of @xmath371 containing @xmath372 . and @xmath373,width=340 ] since @xmath374 is a neighbourhood of @xmath133 for every @xmath343 , @xmath375 , where @xmath376 denotes the connected component containing @xmath133 of @xmath377 , has two components : we let @xmath378 notice that a piece of @xmath283 forms the boundary between @xmath379 and @xmath380 . we mention for future reference a simple estimate which we shall use below . [ distance1 ] @xmath381 for all @xmath382 @xmath383 implies , by definition , @xmath384 . for points @xmath42 close to @xmath385 this also means @xmath386 since such points come very close to the fixed point @xmath137 and escape the @xmath361 neighbourhood of @xmath133 along the direction of @xmath387 . thus , using the fact that the norm of the derivative @xmath388 in @xmath74 is uniformly bounded above by 5 we obtain the result . for @xmath389 we define a _ critical neighbourhood _ @xmath390 notice that we can take @xmath322 sufficiently small so that @xmath391 and @xmath392 from now on we consider @xmath322 fixed . we also let @xmath393 for @xmath10 sufficiently close to @xmath73 and @xmath24 and @xmath12 sufficiently small we have uniform hyperbolicity outside @xmath394 . we state this fact more formally in the following [ lemmaunifhyp ] for every @xmath395 and @xmath396 sufficiently small , there exists a constant @xmath397 such that for all @xmath398 and points @xmath42 with @xmath399 , and vector @xmath400 with slope @xmath401 we have @xmath402 @xmath403 if , _ moreover _ , @xmath404 then we have @xmath405 this is a standard result ( see for example @xcite or @xcite ) and so we omit the details . we just mention that it follows from the fact that the limiting one - dimensional map @xmath406 satisfies uniform expansivity estimates outside an arbitrary critical neighbourhood @xmath75 ( with constant @xmath407 arbitrarily close to @xmath408 but constant @xmath409 depending on @xmath322 and arbitrarily small for @xmath410 small ) , see e.g. @xcite . considering this one - dimensional map as embedded in the space of two - dimensional maps and using the fact that uniform hyperbolicity is an open condition we obtain the statement in the lemma for @xmath411 sufficiently small . the notion of hyperbolic coordinates is inspired by some constructions in @xcite , developed in @xcite and formalized in @xcite as an alternative framework with which to approach the classical theory of invariant manifolds . here we review the basic definitions and theory to the extent to which they will be required for our purposes . we recall the notion of hyperbolic coordinates and give the basic definitions and properties in the general context of @xmath0 diffeomorphisms of a riemannian surface @xmath412 . for @xmath413 we let @xmath414 denote the maximum expansion and the maximum contraction respectively of @xmath415 . then we think of the quantity @xmath416 as the _ hyperbolicity _ of @xmath415 . notice that @xmath417 always . the condition @xmath418 implies that the linear map @xmath419 maps the unit circle @xmath420 to an ellipse @xmath421 with well defined major and minor axes . the unit vectors @xmath422 which are mapped to the minor and major axis respectively of the ellipse , and are thus the _ most contracted _ and _ most expanded _ vectors respectively , are given analytically as solutions to the differential equation @xmath423 which can be solved to give the explicit formula @xmath424 } { ( { \ensuremath{\partial_{x}f_{1}}}^{k})^2+({\ensuremath{\partial_{x}f_{2}}}^{k})^2 - ( { \ensuremath{\partial_{y}f_{1}}}^{k})^2 -({\ensuremath{\partial_{y}f_{2}}}^{k})^2}.\ ] ] here @xmath425 are the two coordinate functions of @xmath44 . notice that @xmath426 and @xmath427 are always _ orthogonal _ and _ do not _ in general correspond to the stable and unstable eigenspaces of @xmath419 . we define the _ hyperbolic coordinates of order _ @xmath428 at the point @xmath42 as the orthogonal coordinates @xmath429 given at @xmath42 by the most contracted and most expanded directions for @xmath430 . if @xmath44 is @xmath0 and @xmath431 then hyperbolic coordinates are defined in some neighbourhood of @xmath42 and define two orthogonal @xmath432 vector fields . in particular they are locally integrable and thus give rise to two orthogonal foliations . we let @xmath433 denote the _ stable foliation of order _ @xmath428 formed by the integral curves of the vector field @xmath434 and @xmath435 denote the _ unstable foliation of order _ @xmath428 formed by the integral curves of the vector field @xmath436 . a crucial property of hyperbolic coordinates and finite order stable and unstable foliations is that , under very mild assumptions , they _ converge _ in quite a strong sense as @xmath437 . we formulate a version of this property here in our specific context . [ c2close ] for every @xmath438 , hyperbolic coordinates @xmath439 and stable and unstable foliations @xmath433 and @xmath435 are defined in @xmath440 moreover 1 . the angle between each stable direction @xmath426 and the slope of @xmath441 . the curvature of each stable leaf , are both @xmath442 . also , the @xmath0 distance between leafs of @xmath433 and leaves @xmath443 is @xmath444 . analogous convergence results are formulated and proved in great generality in @xcite under weak ( subexponential ) growth of the derivative . here we shall need only some very particular cases of these estimates and therefore we first describe the specific setting in which they will be applied here . the main ingredient for the proof is that fact that by our choice of @xmath361 and assuming that @xmath445 , @xmath446 and @xmath12 are small enough we have that @xmath447 is small for all @xmath448 and thus in particular @xmath449 it follows immediately that hyperbolic coordinates , and their associated foliations , of order @xmath428 are well defined in @xmath450 . points in @xmath451 are then re - injected into @xmath452 and these hyperbolicity estimates can no longer be guaranteed , a priori , for all time . points in @xmath453 however are outside @xmath74 and therefore , by the arguments of section [ proofnonwan ] , eventually escape towards infinity . in particular the required hyperbolicity conditions can be guaranteed to hold for all positive iterates . this implies that hyperbolic coordinates of order @xmath428 are well defined in @xmath454 as in the statement of the proposition . the statements about the direction of the stable directions , the curvature of the leaves and the @xmath0 distance between stable leaves of different orders , all follow directly from @xcite*main theorem . these calculations are purely technical and do not add to our geometrical understanding of this situation , we therefore omit the details and refer the reader to that paper . recall that the curvature @xmath455 of a parametrized curve @xmath456 is given by @xmath457 the equivalence between the two formulas is given by the formula @xmath458 . for @xmath459 , we say that a @xmath0 curve @xmath460 is _ admissible _ if @xmath461 and @xmath462 for all @xmath463 . we remark that both the curvature and the slope of tangent vectors of a curve are independent of the parametrization , and thus so is the definition of admissibility . we shall want to compare the curvature at a point of a curve and at the corresponding point of its image . so , we suppose @xmath464 is a parametrized @xmath0 curve and @xmath465 . for simplicity we shall often omit the parameter @xmath463 and simply write @xmath388 to denote the derivative at the point @xmath464 . [ smallcurv ] let @xmath466 be a sequence of @xmath0 curves with @xmath467 . suppose that for some @xmath463 , @xmath343 is a `` hyperbolic time '' in the sense that @xmath468 for all @xmath469 . then for @xmath470 sufficiently small , @xmath471 implies @xmath472 [ admisstoadmiss ] if @xmath473 is admissible , then @xmath474 is also admissible this follows from proposition [ smallcurv ] and the hyperbolicity outside @xmath75 . condition implies that the slope of each tangent vector to @xmath475 is @xmath476 and condition together with lemma [ smallcurv ] gives the curvature @xmath476 . to prove proposition [ smallcurv ] we first prove a general curvature estimate . we fix some bounded neighbourhood @xmath477 of @xmath168 and , as above , we suppose @xmath466 is a sequence of @xmath0 ( not necessarily admissible ) curves with @xmath467 , all contained in @xmath478 . [ curv ] there exists @xmath479 independent of @xmath480 such that for all @xmath481 we have @xmath482 we use the formula @xmath483 for the curvature . we have @xmath484 and @xmath485 therefore @xmath486 is given by @xmath487 where @xmath488 and @xmath489 we shall estimate the two terms of separately . these will yield the two terms in the statement of the proposition . for the second term we have @xmath490 is the area of the parallelogram defined by the two vectors @xmath491 and @xmath492 , and @xmath493 is the are of the parallelogram defined by the two vectors @xmath494 and @xmath495 which of course just the image of the first parallelogram under @xmath388 . the second equality just follows immediately from the definition of @xmath496 . so it just remains to show that the value of @xmath497 is bounded above by some multiple of @xmath11 and @xmath12 . indeed , writing @xmath498 we have , by the `` row - linearity '' of the determinant , @xmath499 using @xmath500 , @xmath501 , @xmath502 , @xmath503 and @xmath504 this gives @xmath505 where in the last step we have used the fact that @xmath10 is close to @xmath506 . substituting this above gives the required bound for the second term of . to bound the first term we write @xmath507 then the norm of this expression is bounded above by @xmath508 all the terms contain a factor @xmath494 ; each of the terms @xmath509 , see , contains a bounded constant multiplied by the factor @xmath12 ; the term @xmath510 , see , is of the order of @xmath511 but here it is multiplied by either @xmath512 or @xmath513 each one of which contains a term which is bounded by @xmath12 . therefore , there exists a constant @xmath514 such that @xmath515 applying lemma [ curv ] recursively we get @xmath516 using the expansivity assumption and @xmath517 small , this gives @xmath518 the next proposition makes precise the notion of a _ critical point of order _ @xmath519 . we recall that @xmath254 is a @xmath0 _ admissible curve _ if all its tangent vectors have slope @xmath476 and it has curvature @xmath520 . we say that @xmath254 is a _ long admissible curve _ if it is an admissible curve which crosses the entire length of @xmath75 . [ tang ] let @xmath521 be a long admissible curve . then there exists a unique point @xmath522 such that @xmath523 has a ( quadratic ) tangency at @xmath524 with the stable foliation @xmath525 , for any @xmath526 . moreover there exists a constant @xmath527 , independent of @xmath517 , such that @xmath528 . in particular , the sequence @xmath529 is cauchy . we call @xmath530 and @xmath531 respectively a _ critical point _ and _ critical value _ of order @xmath428 , associated to the long admissible curve @xmath254 . we remark that critical values @xmath531 of finite order are not guaranteed to be outside @xmath74 , however we shall show below that their limit points as @xmath532 , i.e. the `` real '' critical points always fall strictly outside @xmath74 for @xmath533 . given a parametrized curve @xmath534 and its image @xmath535 we denote by @xmath536 the curvature of @xmath537 at the point @xmath538 and by @xmath539 the curvature of @xmath540 at the point @xmath541 . [ poscurv ] let @xmath538 be an admissible curve and let @xmath542 . suppose that for some @xmath543 we have @xmath544 and @xmath545 . then @xmath546 . lemma [ poscurv ] essentially says that if the tangent direction of the image of an admissible curve at a certain point is roughly vertical ( or at least contained in the `` vertical '' cone between the positive and the negative diagonals ) then the curvature at this point is strictly bounded away from 0 . this does not apply to admissible curves outside @xmath75 since we have shown above ( corollary [ admisstoadmiss ] ) that images of such curves are still admissible and therefore their tangent directions are roughly horizontal . we will instead apply it below to the images of admissible curves inside @xmath75 as a way of pinpointing the location of _ folds_. first recall that the curvature @xmath547 is independent of the choice of parametrization and also the condition @xmath545 is independent of the parametrization since @xmath548 is just the slope of the tangent vector . therefore we choose the parametrization @xmath549 for simplicity we also omit the subscript @xmath550 from the coordinate functions of @xmath540 and just write @xmath551 . from the definition of @xmath44 we have @xmath552^{2 } , d^{2}\varphi_{2}(\gamma_{0}(t))[\dot\gamma_{0}(t)]^{2 } ) \end{aligned}\ ] ] choosing @xmath12 sufficiently small , for example so that @xmath553 this implies @xmath554 we can now compute the curvature @xmath547 . first notice that the condition @xmath545 implies in particular @xmath555 dividing numerator and denominator by @xmath556 , using the condition @xmath545 and we get @xmath557 finally , from the formulas for @xmath558 and @xmath559 and the fact that @xmath560 by the admissibility of @xmath537 , we get @xmath561 as long as @xmath12 is sufficiently small . the existence of a tangency between @xmath475 and the stable foliation @xmath433 follows by the simple geometric observation that the image of a long admissible curve necessarily `` changes direction '' between one endpoint and the other . thus , by a simple intermediate value argument it follows that there is some point of tangency . now , proposition [ c2close ] says that the leaves of the stable foliations @xmath562 are close to the piece of stable manifold @xmath563 and thus have slope close to @xmath73 , and that their curvature is small . in particular the point of tangency must occur at some point at which the tangent direction to @xmath475 is close to 2 and therefore proposition [ tang ] shows that at this point of tangency @xmath475 has strictly positive curvature . this implies that this tangency is quadratic as well as unique . this is the final and main section of the paper . we apply the notion of hyperbolic coordinates and dynamical defined critical points to prove theorem 1 . in section [ shadowing ] we combine the hyperbolic coordinates and the curvature estimates to show that all components of the unstable manifold @xmath161 in @xmath75 are almost horizontal curves with small curvature . in particular they all have well - defined critical points . in section [ hypret ] we take advantage of the structure of critical points on such components to show that points in the critical region @xmath564 recover hyperbolicity after some bounded number of iterations depending only on the parameter @xmath10 . in section [ unifhyp ] we then extend these estimates to uniform expansion estimates on all of @xmath161 with a hyperbolicity constant @xmath565 depending only on the parameter . in section [ unifhypom ] we then show how to extend this hyperbolicity to the closure of @xmath161 and thus to the whole nonwandering set @xmath33 . finally , in section [ lyap ] , we consider the bifurcation parameter value @xmath31 and show that all lyapunov exponents are uniformly bounded away from 0 . [ recovering ] for all @xmath78 all components of @xmath567 are long admissible curves . moreover , for all @xmath568 and any vector @xmath569 tangent to @xmath161 at @xmath42 and @xmath438 such that @xmath570 we have @xmath571 if @xmath572 is a long admissible curve we consider the tangent vectors @xmath573 and their images @xmath574 . by proposition [ tang ] , @xmath575 is tangent to the stable direction @xmath426 at the point @xmath531 . for this and other nearby points on @xmath254 we can write the tangent vector @xmath576 where @xmath577 is the orthogonal basis given by the most expanded and most contracted direction for @xmath419 and @xmath578 and @xmath579 are the components of @xmath580 in this basis . notice that the basis itself depends on the point . proposition [ c2close ] implies that the basis varies _ very slowly _ with the base point , and proposition [ tang ] implies that the tangent vector @xmath575 is varying at _ positive speed _ with respect to this basis . we omit the calculations which are relatively standard , see for example @xcite . specifically this implies that the component @xmath581 of the tangent vector @xmath575 at some point @xmath582 is proportional to the distance between @xmath42 and the critical point of order @xmath428 , @xmath530 . in our setting , the constants actually give @xmath583 we can now prove the following [ binding ] suppose @xmath584 is an admissible curve , @xmath585 , @xmath586 and @xmath530 is the critical point of order @xmath428 in @xmath254 . then for a vector @xmath587 tangent to @xmath254 at @xmath42 and all @xmath588 we have @xmath589 in particular @xmath590 the first equality follows immediately from and . to prove the second we need to find a bound for @xmath591 in terms of @xmath428 . using the quadratic nature of @xmath537 and the proximity to the one - dimensional map @xmath592 with @xmath116 , we obtain @xmath593 to estimate @xmath594 we use the observation that the `` real '' critical value @xmath595 on @xmath537 , i.e. the point of tangency between @xmath537 and the limiting stable foliation @xmath596 lies necessarily either on @xmath597 ( this is only a possibility if @xmath31 ) or to the right of @xmath283 in @xmath598 . we write this as @xmath599 . combining this with lemma [ distance1 ] and the rate of convergence of critical points of finite order @xmath600 as mentioned in proposition [ tang ] and taking @xmath21 sufficiently small , we get @xmath601 substituting this into and using the fact that we can assume @xmath602 as well as the definition of @xmath353 in and of @xmath318 in , we have @xmath603 returning to the proof of the proposition , to obtain the statement about admissibility , notice first of all that combining lemma [ binding ] with lemma [ smallcurv ] we immediately obtain the statement that if @xmath604 is admissible and @xmath428 is the first time that @xmath605 , then @xmath606 is admissible . now , by choosing @xmath24 and @xmath12 small we can guarantee that @xmath607 is a long admissible curve . moreover , every piece of @xmath608 is the image of some curve in @xmath609 and is therefore admissible . proposition [ recovering ] gives a pointwise recovery time for the hyperbolicity of points in the critical region , based on their position . the following proposition gives a key uniformity estimate in the phase space for each parameter @xmath29 . [ n ] for all @xmath29 there exists a constant @xmath611 such that for @xmath612 , and @xmath569 a tangent vector to @xmath161 at @xmath613 , there exists @xmath614 such that @xmath615 is almost horizontal and @xmath616 we remark that the constant @xmath617 is _ not _ uniformly bounded in @xmath618 and in particular _ does not _ apply to @xmath31 . however it gives us a _ uniformity statement _ in @xmath42 which will implies , as we shall see below , uniform hyperbolicity for each given parameter value @xmath29 . for the proof we need to extend the definition of admissibility naturally to curves which are only differentiable of class @xmath619 ( lipschitz continuous derivative ) . [ unique critical ] for every @xmath72 , every @xmath628 lies on a @xmath619 admissible curve @xmath254 which is the limit of @xmath0 admissible curves in @xmath161 and @xmath254 contains a unique critical point @xmath629 with @xmath630 . we show first of all that every point @xmath631 lies on a @xmath619 admissible curve which is the limit of @xmath0 admissible curves in @xmath161 . let @xmath631 and let @xmath632 be a sequence in with @xmath633 . by proposition [ recovering ] each @xmath634 belongs to a long admissible curve @xmath635 . we can write these as functions @xmath636 with @xmath637 $ ] and suppose that converge pointwise to @xmath638 . since @xmath639 is compact and @xmath640 are bounded and equicontinuous sequences we have that @xmath254 is @xmath432 and @xmath641 in the @xmath432 topology . to see that @xmath642 is lipschitz , let @xmath643 and observe that each @xmath644 is a lipschitz function with uniformly bounded lipschitz constant @xmath645 . then we have @xmath646 and hence @xmath646 . we now show that any such curve @xmath254 contains a unique critical point . we show first that it must contain at most one , and then argue that it must contain at least one . let @xmath647 be the angle between the vectors @xmath648 and @xmath649 . since the image of each admissible curve is quadratic with respect to @xmath596 we have that @xmath647 has a strictly non - zero derivative having at most one root corresponding to a point of tangency between @xmath650 and @xmath596 . since @xmath651 in the @xmath652 topology , we have that @xmath653 also has strictly non - zero derivative having at most one root also corresponding to a point of tangency between @xmath475 and @xmath596 . to see that such a point exists , observe that if @xmath29 then the graph of @xmath654 crosses the boundary of @xmath655 twice and @xmath656 is outside @xmath74 where the foliation @xmath596 is well defined and the extreme points of @xmath656 both lie on a piece of @xmath283 which is a leaf of the foliation @xmath657 . this implies that there exists a point outside the interior of @xmath74 where @xmath475 is tangent to @xmath657 . lemma [ unique critical ] allows us to define a canonical set @xmath658 of _ critical points _ as the union of all critical points @xmath629 for every @xmath619 which are @xmath432 limits of long admissible curves of @xmath659 . in the next lemma we show that this set is bounded away from the set of nonwandering points . notice first of all that @xmath663 and thus in particular is bounded . let @xmath664 be a sequence converging to some point @xmath665 . we need to show that @xmath666 . since each @xmath667 is the limit of long admissible curves , we can consider sequences @xmath668 for each @xmath428 . using lemma [ unique critical ] and the fact that @xmath669 converges pointwise to @xmath254 , we conclude that this convergence is in fact @xmath432 . since @xmath670 we have that @xmath671 and this implies that @xmath665 is a critical point as required . we have therefore shown that the critical set @xmath672 is compact . since @xmath33 is also compact , it is sufficient to show that @xmath673 to imply that they are at some positive distance apart . disjointness follows from the observation that the image of a critical point is always outside @xmath74 , while @xmath33 is an invariant set contained in @xmath674 . by lemma [ crit ] and the uniform approximation of the critical set @xmath675 by the finite order critical sets @xmath676 , there exists @xmath662 sufficiently large so that the following two conditions hold ( using also @xmath677 : @xmath678 now consider @xmath679 and let @xmath46 be such that @xmath680 . recall @xmath681 and therefore such an @xmath343 is always well defined except for those points which map exactly to the the curve @xmath682 which forms the boundary between @xmath683 and @xmath684 . for these points we let @xmath685 . then we let @xmath686 if @xmath687 the statement follows from proposition [ recovering ] . otherwise our choice of @xmath662 in gives @xmath688 the first inequality follows from lemma [ binding ] , the second one follows from @xmath689 , the third one follows from the first part of , and the last one follows from the second part of . the following proposition is is essentially a corollary of proposition [ n ] . however we state it separately as it gives an explicit construction of the constant @xmath565 of hyperbolicity for each @xmath29 . before stating the result we define this constant . let @xmath694 and @xmath695 denote the maximum possible contraction and the maximum possible expansion exhibited by any vector @xmath696 for any point @xmath697 in at most @xmath662 iterations . letting @xmath409 denote the constant of hyperbolicity as in on page , we then let @xmath698 let @xmath699 and let @xmath587 tangent to @xmath161 at @xmath42 . since we do not assume anything about the location of @xmath42 the vector @xmath587 may or may not be almost horizontal . we distinguish these two possibilities . let @xmath702 be the sequence of returns of the iterates of @xmath42 to @xmath75 ( with @xmath703 if @xmath704 and @xmath705 otherwise ) . then for each @xmath706 we have an integer @xmath707 given by proposition [ recovering ] . then we can write @xmath708 where @xmath709 is the number of iterates during which the point remains outside @xmath75 . from proposition [ recovering ] properties and , the images of the vector at these iterates remains horizontal and we have @xmath710 for all @xmath711 , in particular for @xmath712 . if @xmath713 , applying to the remaining iterates gives @xmath714 as required . if @xmath715 we have expansion for the first @xmath716 iterates which gives @xmath717 . there follow @xmath718 iterates ( since @xmath719 ) during which we can bound the contraction coarsely by the @xmath662th power of the maximum contraction in the region @xmath74 which gives @xmath720 we show first of all that some preimage of @xmath42 lies in @xmath724 . indeed , @xmath725 implies that @xmath726 as @xmath727 and therefore that @xmath728 is almost horizontal for sufficiently large @xmath343 since the local unstable manifold of @xmath132 is admissible . by the invariance of the unstable conefield outside @xmath75 images of @xmath729 remain almost horizontal unless some return to @xmath724 takes place . now let @xmath730 be the smallest integer such that @xmath731 . then @xmath732 is almost horizontal since every component of @xmath733 in @xmath75 is almost horizontal . from proposition [ n ] it follows that @xmath734 is almost horizontal and therefore it follows necessarily that @xmath735 . otherwise @xmath736 will be almost horizontal . returning to the proof of the proposition , we can now argue as in the previous case to obtain exponential growth starting from time @xmath737 : @xmath738 where @xmath739 moreover @xmath740 substituting this back into completes the proof . we have obtained uniform expansion estimates for vectors tangent to @xmath161 . in this section we show that these estimates can be extended to @xmath33 . this part of the argument uses very little of the specific hnon - like form of the map and therefore we state it in a more abstract and general context . [ hypprop2 ] let @xmath742 be a @xmath432 diffeomorphism and @xmath33 a compact invariant set with @xmath743 on @xmath33 . suppose that there exists some invariant submanifold @xmath744 dense in @xmath33 and such that there exist constants @xmath745 such that @xmath746 for all @xmath747 and @xmath569 tangent to @xmath744 . then @xmath33 is uniformly hyperbolic with hyperbolic constants @xmath748 and @xmath318 . we shall show that @xmath33 is uniformly hyperbolic by constructing an invariant hyperbolic splitting @xmath750 at every point of @xmath33 and then showing that this splitting is continuous . we carry out this construction in several steps . the starting point is the observation that @xmath751 is already given by the tangent direction to @xmath744 for all points @xmath752 . [ limit ] for any @xmath753 and any sequence @xmath754 with @xmath755 , the sequence @xmath756 converges to a ( unique ) limit direction @xmath757 . each vector @xmath758 satisfies @xmath759 for all @xmath46 . suppose @xmath753 and let @xmath754 be a sequence with @xmath760 . consider the sequence of corresponding directions @xmath756 . by compactness ( of the space @xmath761 of possible directions ) there must exist some subsequence @xmath762 such that the corresponding directions @xmath763 converge to some direction which we call @xmath757 . notice that a priori this direction is not unique since it depends on the choice of subsequence . we shall show first that the forward expansion and backward contraction estimates hold and then show that this actually implies uniqueness . let @xmath764 and @xmath765 be a sequence with @xmath766 . then , for each @xmath767 we have , by the continuity of @xmath768 , @xmath769 by assumption we know that @xmath770 and therefore @xmath771 for any @xmath389 . therefore @xmath772 and , since this holds for every @xmath343 , we have the required statement as far as the expansion in forward time is concerned . to prove contraction in backward time it is sufficient to prove it for points on @xmath744 and then apply exactly the same approximation argument . for @xmath773 this follows immediately from the uniform expansivity assumption in forward time . indeed , letting @xmath774 , the expansivity assumption gives @xmath775 which immediately implies @xmath776 . it remains to show uniqueness of @xmath757 for each @xmath777 . suppose by contradiction that we could find two sequences @xmath778 and @xmath779 with corresponding directions @xmath780 and @xmath781 converging to two different directions @xmath751 and @xmath782 . let @xmath783 and @xmath784 . then @xmath785 must be linearly independent and thus every other vector @xmath786 can be written as a linear combination @xmath787 for some @xmath788 . by linearity and the backward contraction estimates obtained above this implies that @xmath789 as @xmath727 . since @xmath587 was arbitrary this implies that all vectors are shrinking to zero in backward time . but this is impossible since we have assumed that @xmath790 and thus @xmath791 at every point @xmath792 there exists a unique tangent space splitting @xmath793 which is invariant by the derivative @xmath388 and which satisfies the standard uniform hyperbolicity expansion and contraction estimates . lemma [ limit ] gives the expanding direction @xmath751 of the splitting with the required hyperbolic expansion estimates in forward time . the invariance for points in @xmath744 is automatic ( since tangent directions to @xmath744 are mapped to tangent directions to @xmath744 ) , and the invariance for general points follows immediately from the definition of @xmath794 , the invariance of @xmath780 for @xmath754 , and the continuity of @xmath388 . the stable direction @xmath795 is given immediately by as the limit of the sequence @xmath796 of vectors most contracted by @xmath797 , as discussed in section [ shadowing ] . this also automatically gives the uniqueness and invariance . to complete the proof of the proposition , we just need to show that the given tangent space splitting is continuous this follows by standard arguments from the uniqueness proved in corollary [ unique ] . indeed , for any @xmath753 and any sequence @xmath798 with @xmath799 , every limit point of the corresponding sequence of splittings @xmath800 must also be a splitting @xmath801 . by approximation arguments identical to those used above it follows that this splitting must also satisfy the uniform hyperbolic contraction and expansion estimates . therefore . by uniqueness , it must coincide with the existing splitting @xmath793 . this completes the proof that @xmath33 is uniformly hyperbolic . finally it remains to consider the dynamics of @xmath803 . recall that @xmath28 is defined on page as the first parameter for which a tangency occurs between the compact parts of @xmath283 and @xmath161 , see figure [ tangency ] for the pictures in the two cases @xmath125 and @xmath167 . we need to show that , for @xmath31 , all lyapunov exponents are uniformly bounded away from 0 . we show that for each point @xmath804 _ not in the orbit of tangency _ @xmath805 ( it is not necessary to consider the orbit of tangency since this is a countable set without recurrence and can therefore not support any invariant probability measure ) there exists a constant @xmath806 , a vector @xmath807 , and a sequence @xmath808 with @xmath809 such that , for all @xmath810 , @xmath811 this is obviously true if the orbit of @xmath42 never enters @xmath75 in forward time or it enters @xmath75 only a finite number of times . indeed suppose that there exists some @xmath428 such that @xmath812 for all @xmath813 . then let @xmath587 be a vector which is mapped to the horizontal vector @xmath814 after @xmath428 iterations . then by we have @xmath815 for all @xmath46 . this implies that there exists a constant @xmath816 such that @xmath817 for all @xmath46 . otherwise there exists an infinite sequence @xmath818 such that @xmath819 and @xmath820 . by lemma [ unique critical ] , @xmath821 lies on either a @xmath0 long admissible curve or a @xmath619 long admissible curve which is the @xmath432 limit of @xmath0 long admissible curves in @xmath161 . since @xmath42 has an infinite number of returns to @xmath394 , this implies in particular that @xmath822 and so @xmath823 and so there exists @xmath824 such that @xmath825 . therefore _ exactly _ the same arguments as in lemmas [ binding ] and [ limit ] show that for a vector @xmath826 tangent to such an admissible curve @xmath254 at @xmath827 we have @xmath828 notice that since the @xmath432 limits of @xmath0 admissible curves are unique , as proved above , we have @xmath829 . then , by and we have @xmath830 then we can define @xmath831 and we have @xmath832 . where the constant @xmath806 is required simply to compensate for the possible lack of expansion for the first @xmath833 iterates . in particular it can be chosen by considering the maximum possible contraction along the orbit of @xmath42 for the first @xmath833 iterations @xmath834 we have shown therefore that for each @xmath753 @xmath835 this clearly implies the same bound for the limit wherever it exists . in particular any point which is typical for some ergodic invariant probability measure and for which therefore such a limit does exist , will have a positive lyapunov exponent @xmath836 . by dissipativity this immediately implies also that the other lyapunov exponent is negative and uniformly bounded away from 0 both in the dynamical and in the parameter space .

we prove a uniformly computable version of de finetti s theorem on exchangeable sequences of real random variables . as a consequence , exchangeable stochastic processes in probabilistic functional programming languages can be automatically rewritten as procedures that do not modify non - local state . along the way , we prove that a distribution on the unit interval is computable if and only if its moments are uniformly computable . de finetti s theorem , exchangeability , computable probability theory , probabilistic programming languages , mutation 03d78 , 60g09 , 68q10 , 03f60 , 68n18

0912.1072

the classical de finetti theorem states that an exchangeable sequence of real random variables is a mixture of independent and identically distributed ( i.i.d . ) sequences of random variables . moreover , there is an ( almost surely unique ) measure - valued random variable , called the _ directing random measure _ , conditioned on which the random sequence is i.i.d . the distribution of the directing random measure is called the _ de finetti measure _ or the _ mixing measure_. this paper examines the _ computable _ probability theory of exchangeable sequences of real - valued random variables . we prove a uniformly computable version of de finetti s theorem , which implies that _ computable _ exchangeable sequences of real random variables have _ computable _ de finetti measures . the classical _ proofs _ do not readily effectivize ; instead , we show how to directly compute the de finetti measure ( as characterized by the classical theorem ) in terms of a computable representation of the distribution of the exchangeable sequence . along the way , we prove that a distribution on @xmath0^\omega$ ] is computable if and only if its moments are uniformly computable , which may be of independent interest . a key step in the proof is to describe the de finetti measure in terms of the moments of a set of random variables derived from the exchangeable sequence . when the directing random measure is ( almost surely ) continuous , we can show that these moments are computable , which suffices to complete the proof of the main theorem in this case . in the general case , we give a proof inspired by a randomized algorithm which succeeds with probability one . these results are formulated in the turing - machine - based bit - model for computation over the reals ( for a general survey , see braverman and cook @xcite ) . this computational model has been explored both via the type-2 theory of effectivity ( tte ) framework for computable analysis , and via effective domain - theoretic representations of measures . computable analysis has its origins in in the study of recursive real functions , and can be seen as a way to provide `` automated numerical analysis '' ( for a tutorial , see brattka , hertling , and weihrauch @xcite ) . effective domain theory has its origins in the semantics of programming languages , where it continues to have many applications ( for a survey , see edalat @xcite ) . here we use methods from these approaches to transfer a representational result from probability theory to a setting where it can directly transform statistical objects as represented on a computer . the computable probability measures in the bit - model coincide with those distributions from which we can generate exact samples to arbitrary precision on a computer . as such , our results have direct implications for programming languages which manipulate probability measures on real numbers via exact interfaces . in many areas of statistics and computer science , especially machine learning , one is often concerned with distributions on data structures that are higher - order or are defined using recursion . probabilistic functional programming languages provide a convenient setting for describing and manipulating such distributions . exchangeable sequences play a fundamental role in both statistical models and their implementation on computers . description of an exchangeable process , in which one uses previous samples or sufficient statistics to sample the next element in the sequence , a direct implementation in these languages would need to use non - local communication ( to record new samples or update sufficient statistics ) . this is often implemented by modifying the program s internal state directly ( i.e. , using _ mutation _ ) , or via some indirect method such as a state monad . the classical de finetti theorem implies that ( for such sequences over the reals ) there is an alternative description in which samples are conditionally independent ( and so could be implemented without non - local communication ) , thereby allowing parallel implementations . but the classical result does not imply that there is a _ program _ which computes the sequence according to this description . even when there is such a program , the classical theorem does not provide a method for finding it . the computable de finetti theorem states that such a program _ does _ exist . moreover , the proof itself provides the _ method _ for constructing the desired program . in section [ funcpure ] we describe how an implementation of the computable de finetti theorem uniformly transforms procedures which induce exchangeable stochastic processes into equivalent procedures which do not modify non - local state . this transformation is of interest beyond its implications for programming language semantics . in statistics and machine learning , it is often desirable to know the representation of an exchangeable stochastic process in terms of its de finetti measure ( for several examples , see section [ partialexch ] ) . many such processes in machine learning have very complicated ( though computable ) distributions , and it is not always feasible to find the de finetti representation by hand . the computable de finetti theorem provides a method for automatically obtaining such representations . we assume familiarity with the standard measure - theoretic formulation of probability theory ( see , e.g. , billingsley @xcite or kallenberg @xcite ) . fix a basic probability space @xmath1 and let @xmath2 denote the borel sets of @xmath3 . note that we will use @xmath4 to denote the set of nonnegative integers ( as in logic ) , rather than an element of the basic probability space @xmath5 ( as in probability theory ) . by a _ random measure _ we mean a random element in the space of borel measures on @xmath3 , i.e. , a kernel from @xmath6 to @xmath7 . an event @xmath8 is said to occur _ almost surely _ ( a.s . ) if @xmath9 . we denote the indicator function of a set @xmath10 by @xmath11 . let @xmath12 be a sequence of real random variables . we say that @xmath13 is _ exchangeable _ if , for every finite set @xmath14 of distinct indices , @xmath15 is equal in distribution to @xmath16 . let @xmath12 be an exchangeable sequence of real - valued random variables . there is a random probability measure @xmath17 on @xmath3 such that @xmath18 is conditionally i.i.d . with respect to @xmath17 . that is , @xmath19 = \nu^\infty \quad a.s . \label{ciid}\ ] ] moreover , @xmath17 is a.s . unique and given by @xmath20 where @xmath10 ranges over @xmath2 . the random measure @xmath17 is called the _ directing random measure_. its distribution ( a measure on probability measures ) , which we denote by @xmath21 , is called the _ de finetti measure _ or the _ mixing measure_. as in kallenberg ( * ? ? ? * chap . 1 , eq . 3 ) , we may take expectations on both sides of to arrive at a characterization @xmath22 of an exchangeable sequence as a mixture of i.i.d . sequences . a bayesian perspective suggests the following interpretation : exchangeable sequences arise from independent observations from some latent random measure . posterior analysis follows from placing a prior distribution on @xmath17 . for further discussion of the implications of de finetti s theorem for the foundations of statistical inference , see dawid @xcite and lauritzen @xcite . in 1931 , de finetti @xcite proved the classical result for binary exchangeable sequences , in which case the de finetti measure is simply a mixture of bernoulli distributions ; the exchangeable sequence is equivalent to repeatedly flipping a coin whose weight is drawn from some distribution on @xmath0 $ ] . in the 1950s , hewitt and savage @xcite and ryll - nardzewski @xcite extended the result to arbitrary real - valued exchangeable sequences . we will refer to this more general version as the _ de finetti theorem_. hewitt and savage @xcite provide a history of the early developments , and a discussion of some subsequent extensions can be found in kingman @xcite , diaconis and freedman @xcite , and aldous @xcite . a recent book by kallenberg @xcite provides a comprehensive view of the area of probability theory that has grown out of de finetti s theorem , stressing the role of invariance under symmetries . consider an exchangeable sequence of @xmath0$]-valued random variables . in this case , the de finetti measure is a distribution on the ( borel ) measures on @xmath0 $ ] . for example , if the de finetti measure is a dirac measure on the uniform distribution on @xmath0 $ ] ( i.e. , the distribution of a random measure which is almost surely the uniform distribution ) , then the induced exchangeable sequence consists of independent , uniformly distributed random variables on @xmath0 $ ] . as another example , let @xmath23 be a random variable , uniformly distributed on @xmath0 $ ] , and let @xmath24 . then the de finetti measure is the uniform distribution on dirac measures on @xmath0 $ ] , and the corresponding exchangeable sequence is @xmath25 , i.e. , a constant sequence , marginally uniformly distributed . as a further example , we consider the stochastic process @xmath18 whose finite marginals are given by @xmath26 where @xmath27 , and where @xmath28 is the gamma function and @xmath29 are positive real numbers . ( one can verify that these marginals satisfy kolmogorov s extension theorem ( * ? ? ? * theorem 6.16 ) , and so there is a stochastic process @xmath18 with these finite marginals . ) clearly this process is exchangeable , as @xmath30 and @xmath31 are invariant to order . this process can also be described by a sequential scheme known as plya s urn ( * ? ? ? each @xmath32 is sampled in turn according to the conditional distribution @xmath33 this is often described as a urn which starts with @xmath34 red balls and @xmath35 black balls , where at each stage , a ball is drawn uniformly at random , and is returned to the urn along with an additional ball of the same color . by de finetti s theorem , there exists a random coin weight @xmath36 with respect to which this process is conditionally independent and @xmath37 for each @xmath38 . in fact , @xmath39 = \prod_{i\le n } { \mathbb{p } } [ x_i = x_i { \mid}\theta\ , ] = \theta^{s_n } ( 1- \theta)^{(n- s_n)}.\ ] ] furthermore , one can show that here @xmath36 is @xmath40-distributed , and so the process given by the marginals is called the beta - bernoulli process . here the de finetti measure is the distribution of the bernoulli measure with random coin weight @xmath36 . in each of these examples , the de finetti measure is a _ computable measure_. ( in section [ representation ] , we make this and related notions precise . for an explicit computable representation of the beta - bernoulli process , see in section [ funcpure ] . ) a natural question to ask is whether computable exchangeable sequences always arise from computable de finetti measures . in fact , computable de finetti measures give rise to computable distributions on exchangeable sequences ( see proposition [ mucomp ] ) . our main result is the converse : every computable distribution on real - valued exchangeable sequences arises from a computable de finetti measure . [ cdeftheorem ] let @xmath41 be the distribution of a real - valued exchangeable sequence @xmath13 , and let @xmath21 be the distribution of its directing random measure @xmath17 . then @xmath21 is uniformly computable in @xmath41 , and @xmath41 is uniformly computable in @xmath21 . in particular , @xmath41 is computable if and only if @xmath21 is computable . the directing random measure is classically given a.s . by the explicit limiting expression . without a computable handle on the rate of convergence , the limit is not directly computable , and so we can not use this limit directly to compute the de finetti measure . however , we are able to reconstruct the de finetti measure using the moments of a set of derived random variables . recall that @xmath2 denotes the borel sets of @xmath3 . let @xmath42 denote the set of open intervals , and let @xmath43 denote the set of open intervals with rational endpoints . then @xmath44 . for @xmath45 and @xmath46 , we write @xmath47 to denote the @xmath38th coordinate of @xmath48 . we will denote by @xmath49 the algebra generated by @xmath50 ( i.e. , finite unions of open rectangles in @xmath51 ) , and denote by @xmath52 the algebra generated by @xmath53 . note that @xmath54 . let @xmath55 be an exchangeable sequence of real random variables , with distribution @xmath41 and directing random measure @xmath17 . for every @xmath56 , we define a @xmath0$]-valued random variable @xmath57 . a classical result in probability theory ( * ? ? ? * 1.17 ) implies that a borel measure on @xmath3 is uniquely characterized by the mass it places on the open intervals with rational endpoints . therefore , the distribution of the stochastic process @xmath58 determines the de finetti measure @xmath21 ( the distribution of @xmath17 ) . let @xmath59 . the _ mixed moments _ of the variables @xmath60 , are the set of all expectations @xmath61 , for @xmath62 and @xmath63 . we can now restate the consequence of de finetti s theorem described in eq . , in terms of the finite - dimensional marginals of the exchangeable sequence @xmath13 and the mixed moments of @xmath64 . [ link1 ] @xmath65 for @xmath66 and @xmath67 . as we will show in lemma [ compdiststandard ] , when @xmath41 is computable , we can enumerate all rational lower bounds on quantities of the form @xmath68 where @xmath69 . we can also enumerate rational upper bounds on , provided that @xmath13 places no mass on the boundary of any @xmath70 . in particular , if @xmath17 is a.s . continuous ( i.e. , with probability one , @xmath71 for every @xmath72 ) , then we can use @xmath41 to compute the mixed moments of @xmath73 . in section [ momentproblem ] , we show how to computably recover a distribution from its moments . this suffices to recover the de finetti measure when @xmath17 is a.s . continuous , as we show in section [ easycase ] . in the general case , fixed point masses in @xmath17 prevent us from computing the mixed moments . here we use a proof inspired by a randomized algorithm which almost surely avoids the point masses and recovers the de finetti measure . for the complete proof , see section [ mainproof ] . before beginning the proof of the computable de finetti theorem , we first define computable probability measures on various spaces . these definitions follow from more general tte notions , though we will sometimes derive simpler equivalent representations for the concrete spaces we need ( such as the real numbers , borel measures on reals , and borel measures on borel measures on reals ) . for details , see the original papers , as noted . we assume familiarity with the standard notions of computability theory and computably enumerable ( c.e . ) sets ( see , e.g. , rogers @xcite or soare @xcite ) . recall that @xmath74 is called a c.e . real when the set of all rationals less than @xmath75 is a c.e . similarly , @xmath75 is a co - c.e . real when the set of all rationals greater than @xmath75 is c.e . a real @xmath75 is a computable real when it is both a c.e . and co - c.e . real . to represent more general spaces , we work in terms of an effectively presented topology . suppose that @xmath76 is a second - countable @xmath77 topological space with subbasis @xmath78 . for every point in @xmath79 , define the set @xmath80 . because @xmath76 is @xmath77 , we have @xmath81 when @xmath82 , i.e. , @xmath83 is the unique point that is contained in those subbasis sets that are elements of @xmath84 and no others . since every point in a @xmath77 space is determined by the subbasis in this way , it is convenient to define representations on topological spaces under the assumption that the space is @xmath77 . in the specific cases below , we often have much more structure , which we use to simplify the representations . we now develop these definitions more formally . [ ctzero ] let @xmath76 be a second - countable @xmath77 topological space with a countable subbasis @xmath78 . let @xmath85 be an enumeration of @xmath78 . we say that @xmath76 is a _ computable topological space ( with respect to @xmath86 ) _ when the set @xmath87 is c.e . in terms of the @xmath86-indices for @xmath88 , i.e. , @xmath89 is a c.e . subset of @xmath4 , where @xmath90 is the standard pairing function . computable topological spaces , as defined here , are instances of the computable @xmath77 spaces defined in grubba , schrder , and weihrauch @xcite , when the subbasis is replaced by the basis it generates ( and the enumeration is extended in a canonical way ) . it is often possible to pick a subbasis @xmath78 ( and enumeration @xmath86 ) for which the elemental `` observations '' that one can computably observe are those of the form @xmath91 , where @xmath92 . then the set @xmath93 is computably enumerable ( with respect to @xmath86 ) when the point @xmath83 is such that it is eventually noticed to be in each basic open set containing it ; we will call such a point @xmath83 _ computable_. this is one motivation for the definition of computable point in a @xmath77 space below . note that in a @xmath94 space , two computable points are computably distinguishable , but in a @xmath77 space , computable points will be , in general , distinguishable only in a computably enumerable fashion . however , this is essentially the best that is possible , if the open sets are the those we can `` observe '' . ( for more details on this approach to considering datatypes as topological spaces , in which basic open sets correspond to `` observations '' , see battenfeld , schrder , and simpson @xcite . ) note that the choice of topology and subbasis are essential ; for example , we can recover both computable reals and c.e . reals as instances of `` computable point '' for appropriate computable topological spaces , as we describe in section [ repreals ] . let @xmath95 be a computable topological space with respect to an enumeration @xmath86 . we say that a point @xmath96 is _ computable ( with respect to @xmath86 ) _ when the set @xmath97 is c.e . ( in terms of the @xmath86-indices for @xmath92 ) . we call the set the _ representation _ of @xmath83 . suppose that @xmath98 and @xmath10 are computable objects ( either points , as defined here , or computable functions or measures , as defined below ) . we will say that @xmath98 is _ uniformly computable in _ @xmath10 , or that @xmath98 is _ uniformly computable relative to _ @xmath10 , when there is a single program ( even as @xmath10 varies ) that computes the representation of @xmath98 using the representation of @xmath10 as an oracle ( and similarly for computable enumerability ) . these representations , in turn , can be made into explicit subsets of @xmath4 using the relevant enumerations , as in . we will use both the standard topology and right order topology on the real line @xmath3 . the reals under the standard topology are a computable topological space using the basis @xmath43 with respect to the canonical enumeration . the reals under the _ right order topology _ are a computable topological space using the basis @xmath99 under the canonical enumeration . note that the computable points of @xmath100 are precisely the computable reals , and the computable points of @xmath101 are precisely the c.e . reals . we now consider computable representations for continuous functions on the reals . let @xmath78 and @xmath102 each be either of @xmath43 or @xmath103 , under their canonical enumeration @xmath86 and @xmath104 . let @xmath105 denote the closure of the set @xmath98 . fix @xmath66 . we say that a continuous function @xmath106 is _ computable _ when @xmath107 is a c.e . set ( in terms of the @xmath108-indices for @xmath109 and @xmath104-indices for @xmath110 , where @xmath108 is the canonical enumeration determined by @xmath86 ) . we call the set the _ representation _ of @xmath111 . this definition is computably equivalent to the canonical construction of computable functions between two represented spaces ( * ? ? ? * ch . 6 ) . note that when @xmath112 , this recovers the standard definition of a computable real function . when @xmath113 and @xmath114 , this recovers the standard definition of a lower - semicomputable real function @xcite . the following representations for probability measures on computable topological spaces are devised from more general tte representations in schrder @xcite and bosserhoff @xcite , and agree with weihrauch @xcite in the case of the unit interval . in particular , the representation for @xmath115 below is admissible with respect to the weak topology , hence computably equivalent ( see weihrauch ( * ? ? ? 3 ) ) to the canonical representation for borel measures given in schrder @xcite . schrder @xcite has also shown the equivalence of this representation for probability measures ( as a computable space under the weak topology ) with _ probabilistic processes_. a probabilistic process ( see schrder and simpson @xcite ) formalizes the notion of a program which uses randomness to sample points in terms of their representations of the form . let @xmath115 denote the set of borel probability measures on a second - countable @xmath77 topological space @xmath76 ( i.e. , the probability measures on the @xmath116-algebra generated by @xmath78 ) . provided that the subbasis @xmath78 is closed under finite intersection , such measures are determined by the measure they assign to elements of @xmath78 . note that @xmath115 is a second - countable @xmath77 space . let @xmath76 be a computable topological space . we will describe a subbasis for @xmath115 which makes it a computable topological space . let @xmath117 denote the algebra generated by the subbasis @xmath78 of @xmath76 ( i.e. , the closure of @xmath78 under finite union and complementation ) . then , the class of sets @xmath118 where @xmath119 and @xmath120 , is a subbasis for the weak topology on @xmath115 . an effective enumeration of this subbasis can be constructed in a canonical fashion from the enumeration of @xmath78 and the rationals , making @xmath115 a computable topological space . [ cdist ] let @xmath76 be a computable topological space with respect to @xmath86 , and let @xmath117 be the algebra generated by the subbasis @xmath78 , under its canonical enumeration @xmath108 induced by @xmath86 . a borel probability measure @xmath121 is _ computable ( with respect to @xmath86 ) _ when @xmath122 is a c.e . real , uniformly in the @xmath108-index of @xmath123 . we call the set @xmath124 the _ representation _ of @xmath125 . in particular , this implies that the measure of a c.e . open set ( i.e. , the c.e . union of basic open sets ) is a c.e . real ( uniformly in the enumeration of the terms in the union ) , and that the measure of a co - c.e . closed set ( i.e. , the complement of a c.e . open set ) is a co - c.e . real ( similarly uniformly ) ; see , e.g. , @xcite for details . note that on a discrete space , where singletons are both c.e . open and co - c.e . closed , the measure of each singleton is a computable real . but for a general space , it is too strong to require that even basic open sets have computable measure ( as it is more than is needed to ensure that exact samples be computably described to arbitrary accuracy ) . we will be interested in computable measures @xmath126 where @xmath76 is either @xmath127 , @xmath0^k$ ] , or @xmath128 , topologized as described below . consider @xmath127 under the product topology , where @xmath3 is endowed with its standard topology . note that @xmath127 is a computable topological space with respect to the canonical enumeration on the basis given by the cylinders @xmath129 for @xmath130 . using definition [ cdist ] , we can characterize the class of computable distributions on real sequences . let @xmath131 be a sequence of real - valued random variables ( e.g. , the exchangeable sequence @xmath13 , or the derived random variables @xmath132 under the canonical enumeration of @xmath43 ) . thus , the joint distribution @xmath125 of @xmath133 is computable when @xmath134 is a c.e . real , uniformly in @xmath135 . the following simpler representation was shown to be equivalent by mller ( * ? ? ? [ compdiststandard ] + let @xmath131 be a sequence of real - valued random variables with joint distribution @xmath125 . then @xmath125 is computable if and only if @xmath136 is a c.e . real , uniformly in @xmath62 and @xmath137 . therefore knowing the measure of the sets in @xmath138 is sufficient . note that is precisely the form of the first expression in corollary [ link1 ] . note also that one obtains a characterization of the computability of a finite - dimensional vector by embedding it in an initial segment of a sequence . distributions on c.e . reals play an important role when representing measures on measures , because , as definition [ cdist ] indicates , measures are themselves represented by collections of c.e . reals lower - bounding the measure of an element of an algebra . the set @xmath139 is a basis for the product of the right order topology on @xmath51 that makes @xmath140 is a computable topological space ( under the canonical enumeration of @xmath139 ) . [ rodist ] for @xmath141 , let @xmath142 be a random vector in @xmath51 , and let @xmath143 . the joint distribution of @xmath144 is computable under the right order topology when @xmath145 is a c.e . real , uniformly in @xmath146 . note that the joint distribution of a random vector in @xmath127 is computable under the right order topology when all its restrictions to @xmath51 are computable under the right order topology , uniformly in @xmath147 . note also that if a distribution on @xmath51 is computable under the standard topology , then it is clearly computable under the right order topology . the above representation is used in the next section as well as in proposition [ mucomp ] , where we must compute an integral with respect to a topology that is weaker than the standard topology . the de finetti measure @xmath21 is the distribution of the directing random measure @xmath17 , an @xmath128-valued random variable . recall the definition @xmath148 , for @xmath149 . then @xmath21 is computable ( with respect to an enumeration @xmath86 of the weak topology ) when @xmath150 is a c.e . real , uniformly in @xmath151 and @xmath143 . as an immediate consequence of and corollary [ rodist ] , we obtain the following representation for the de finetti measure . [ muiff ] the de finetti measure @xmath21 is computable if and only if the joint distribution of @xmath152 is computable under the right order topology . the following lemma is a restatement of an integration result by schrder ( * ? ? ? 3.6 ) , which itself generalizes integration results on standard topologies of finite - dimensional euclidean spaces by mller @xcite and the unit interval by weihrauch @xcite . define @xmath153 { \,:\,}a \in { { \mathcal{i}}_{\ensuremath{\mathbf{q}}}}\},\ ] ] which is a basis for the standard topology on @xmath0 $ ] , and define @xmath154 { \,:\,}a \in { \mathcal{r}}_{{<}}\},\ ] ] which is a basis for the right order topology on @xmath0 $ ] . [ ceintegration ] let @xmath66 and let @xmath78 be either @xmath43 or @xmath103 . let @xmath155 , { \mathbb{i}}_{{<}})\ ] ] be a computable function and let @xmath21 be a computable distribution on @xmath156 . then @xmath157 is a c.e . real , uniformly in @xmath111 and @xmath21 . the following result of mller @xcite is an immediate corollary . [ compintegration ] let @xmath158,{\mathbb{i}})\ ] ] be a computable function and let @xmath21 be a computable distribution on @xmath159 . then @xmath160 is a computable real , uniformly in @xmath161 and @xmath21 . one often has access to the moments of a distribution , and wishes to recover the underlying distribution . let @xmath162 be a random vector in @xmath0^\omega$ ] with distribution @xmath125 . classically , the distribution of @xmath133 is uniquely determined by the mixed moments of @xmath133 . we show that the distribution is in fact _ computable _ from the mixed moments . one classical way to pass from the moments of @xmath133 to its distribution is via the lvy inversion formula , which maps the characteristic function @xmath163 , given by @xmath164 to the distribution of @xmath133 . however , even in the finite - dimensional case , the inversion formula involves a limit for which we have no direct handle on the rate of convergence , and so the distribution it defines is not obviously computable . instead , we use computable versions of urysohn s lemma and the weierstrass approximation theorem to compute a representation of a distribution from its moments . to show that @xmath125 is computable , it suffices to show that @xmath165^\omega ) = { \mathbb{e}}\bigl({\mathbf 1}_{\sigma}(x_1,\dotsc , x_k)\bigr)$ ] is a c.e . real , uniformly in @xmath166 . we begin by showing how to build sequences of polynomials that converge pointwise from below to indicator functions of the form @xmath167 for @xmath168 . [ effindicator ] there is a computable array @xmath169 of rational polynomials where , for @xmath170 and @xmath135 , the polynomial @xmath171 is in @xmath147 variables and , for each @xmath172 ^k$ ] , we have @xmath173 fix @xmath170 and @xmath135 . let @xmath174 denote the set of @xmath175 for which the @xmath176-ball centered at @xmath177 is contained in @xmath116 . we will construct ( using a procedure that is uniform in @xmath30 and @xmath116 ) a rational polynomial @xmath171 which satisfies @xmath178 for all @xmath172 ^k$ ] , and which satisfies @xmath179 for all @xmath180 and for all @xmath172^k \setminus \sigma$ ] . then the values @xmath181 all equal @xmath182 on @xmath172^k \setminus \sigma$ ] , and also equal @xmath182 for @xmath177 on an increasing sequence of subsets of @xmath116 whose union is @xmath116 . for @xmath183 we may take @xmath184 . thus assume @xmath185 . using a computable version of urysohn s lemma ( see weihrauch @xcite ) , we can find ( uniformly in @xmath30 and @xmath116 ) a computable real function @xmath186 which equals @xmath187 for all @xmath180 , equals @xmath188 on @xmath172^k \setminus \sigma$ ] , and is in between these two values on @xmath189 . then by the effective weierstrass approximation theorem ( see pour - el and richards @xcite ) , we can find ( uniformly in @xmath30 and @xmath116 ) a polynomial @xmath171 with rational coefficients which uniformly approximates @xmath186 to within @xmath190 on @xmath0^k$ ] . this polynomial @xmath171 has the desired properties . using these polynomials , we can compute the distribution from the moments . the other direction follows from computable integration results . [ computablemoments ] let @xmath162 be a random vector in @xmath0^\omega$ ] with distribution @xmath125 . then @xmath125 is uniformly computable in the mixed moments of @xmath191 , and vice versa . in particular , @xmath125 is computable if and only if the mixed moments of @xmath191 are uniformly computable . any monic monomial in @xmath192 , considered as a real function , computably maps @xmath0^k$ ] into @xmath0 $ ] ( under the standard topology ) . furthermore , @xmath125 restricts to a computable measure on @xmath193 . therefore , by corollary [ compintegration ] , the mixed moments are uniformly computable in @xmath125 and the index of the monomial . to prove that @xmath125 is computable in the mixed moments , we give a proof which relativizes to representation of the mixed moments , and so we assume that the mixed moments of @xmath191 are uniformly computable , without loss of generality . let @xmath194 and @xmath195 . to establish the computability of @xmath125 , it suffices to show that @xmath196^\omega ) = { \mathbb{e}}\bigl ( { \mathbf 1}_{\sigma \times [ 0,1]^\omega } ( \vec x ) \bigr ) = { \mathbb{e}}\bigl ( { \mathbf 1}_{\sigma } ( x_1,\dotsc , x_k ) \bigr).\ ] ] is a c.e . real , uniformly in @xmath116 . by lemma [ effindicator ] , we are given a uniformly computable sequence of polynomials @xmath197 which converge pointwise from below to the indicator @xmath198 . by the dominated convergence theorem , @xmath199 the expectation @xmath200 is a @xmath201-linear combination of moments , hence a computable real . thus their supremum is a c.e . in the remainder of the paper , let @xmath13 be a real - valued exchangeable sequence with distribution @xmath41 , let @xmath17 denote its directing random measure , and let @xmath21 denote the de finetti measure . classically , the joint distribution of @xmath13 is uniquely determined by the de finetti measure ( see equation [ mixture ] ) . we now show that the joint distribution of @xmath13 is in fact _ computable _ from the de finetti measure . [ mucomp ] the distribution @xmath41 is uniformly computable in @xmath21 . we give a proof that relativizes to @xmath21 ; without loss of generality , assume that @xmath21 is computable . in order to show that @xmath41 , the distribution of @xmath13 , is computable , we must show that @xmath202 is a c.e . real , uniformly in @xmath195 . fix @xmath195 . note that , by corollary [ link1 ] , @xmath203 let @xmath125 be the joint distribution of @xmath204 and let @xmath205^k \to [ 0,1]$ ] be defined by @xmath206 to complete the proof , we now show that @xmath207 is a c.e . note that the computability of @xmath21 implies that @xmath125 is computable under the right order topology . furthermore , @xmath111 is order - preserving , and so is a continuous ( and obviously computable ) function from @xmath208^k,{\mathbb{i}}_{{<}}^k)$ ] to @xmath208,{\mathbb{i}}_{{<}})$ ] . therefore , by lemma [ ceintegration ] , we have that @xmath209 is a c.e . we will first prove the main theorem under the additional hypothesis that the directing random measure is almost surely continuous . we then sketch a randomized argument which succeeds with probability one . finally , we present the proof of the main result , which can be seen as a derandomization . for @xmath66 and @xmath210 , we say that @xmath211 is an _ @xmath13-continuity set _ when @xmath212 a.s . for @xmath213 ( informally , @xmath13 places no mass on the boundary of @xmath211 ) . [ cemon1 ] relative to @xmath41 , the mixed moments of @xmath152 are uniformly c.e . reals and the mixed moments of @xmath214 are uniformly co - c.e . reals ; in particular , if @xmath215 is an @xmath13-continuity set , then the mixed moment @xmath216 is a computable real . without loss of generality , we may assume that @xmath41 is computable . let @xmath45 and @xmath151 . by corollary [ link1 ] , @xmath217 which is a c.e . real because @xmath41 is computable . the set @xmath218 is a co - c.e . closed set in @xmath51 because we can computably enumerate all @xmath219 contained in the complement of @xmath116 . therefore , @xmath220 is the measure of a co - c.e . closed set , hence a co - c.e . real . when @xmath116 is an @xmath13-continuity set , @xmath221 and so the expectation is a computable real . [ accdeftheorem ] assume that @xmath17 is continuous with probability one . then @xmath21 is uniformly computable in @xmath41 . we give a proof that relativizes to @xmath41 ; assume without loss of generality that @xmath41 is computable . let @xmath222 and consider @xmath151 . the almost sure continuity of @xmath17 implies that @xmath223 a.s . , i.e. , @xmath116 is an @xmath13-continuity set . therefore , by lemma [ cemon1 ] , the moment @xmath224 is a computable real . the computable moment theorem ( theorem [ computablemoments ] ) then implies that joint distribution of the variables @xmath225 is computable under the standard topology , and so @xmath21 is computable under the weaker right order topology . by corollary [ muiff ] , this implies that @xmath21 is computable . in general , the joint distribution of @xmath226 is not computable under the standard topology because the directing random measure @xmath17 may , with nonzero probability , have a fixed point mass on a rational . in this case , the mixed moments of @xmath152 are c.e . , but not co - c.e . , reals ( relative to @xmath41 ) . in this case , the computable moment theorem ( theorem [ computablemoments ] ) is inapplicable . for arbitrary directing random measures , we give a proof of the computable de finetti theorem which works regardless of the location of point masses . consider the following sketch of a `` randomized algorithm '' : we independently sample a countably infinite set of real numbers @xmath227 from a computable , absolutely continuous distribution which has support everywhere on the real line ( e.g. , a gaussian or cauchy ) . let @xmath228 denote the algebra generated by open intervals with endpoints in @xmath229 . note that , with probability one ( over the draw ) , @xmath229 will be dense in @xmath3 and @xmath230 almost surely , and so @xmath13 will place no mass on the boundary of any @xmath231 ; suppose we have chosen such an @xmath229 . now we can proceed analogously to the case where @xmath17 is almost surely continuous , relative to an oracle @xmath98 which encodes @xmath229 . we begin by proving an extension of lemma [ cemon1 ] that shows that the mixed moments of variables in terms of the new basis are @xmath98-computable , relative to @xmath41 . we call @xmath232 a _ refinement _ of @xmath233 , and write @xmath234 , when @xmath235 for all @xmath236 . [ cemonreal ] let @xmath229 be sampled as above . relative to @xmath41 , the mixed moments of @xmath237 are uniformly @xmath98-c.e . reals and the mixed moments of @xmath238 are uniformly @xmath98-co - c.e . reals . in particular , if @xmath239 is an @xmath13-continuity set , then the mixed moment @xmath240 is an @xmath98-computable real relative to @xmath41 . fix @xmath241 . all computability claims are relative to @xmath41 . we may @xmath98-compute a sequence @xmath242 such that for each @xmath243 , @xmath244 for each @xmath30 , @xmath245 is a c.e . real , and so their supremum is an @xmath98-c.e . real . if @xmath246 satisfy @xmath247 then @xmath248 ( a.s . ) . multiplication is continuous , and so the dominated convergence theorem gives us @xmath249 which we have already noted is an @xmath98-c.e . real . the co - c.e . result follows similarly from a sequence of nested unions of rational intervals whose intersection is @xmath250 . the final claim is immediate as in lemma [ cemon1 ] . the proof of the computable moment theorem ( theorem [ computablemoments ] ) relativizes , and so the joint distribution of @xmath251 is @xmath98-computable in @xmath41 . this joint distribution also classically determines the de finetti measure . moreover , we can @xmath98-compute ( relative to @xmath41 ) the desired representation with respect to the original basis . let @xmath229 be sampled as above . the de finetti measure @xmath21 is uniformly @xmath98-computable in @xmath41 . we show that the joint distribution of @xmath252 is @xmath98-computable under the right order topology . let @xmath141 , let @xmath143 , and let @xmath253 . note that @xmath254 is an @xmath98-c.e . open set with respect to the basis @xmath255 , and so we can @xmath98-computably enumerate a sequence @xmath256 such that @xmath257 . without loss of generality , we may assume that @xmath258 . note that @xmath259 ( a.s . ) for all @xmath260 and @xmath261 . by the continuity of @xmath262 , @xmath263 the right hand side is the supremum of a uniformly @xmath98-computable set of @xmath98-computable reals , relative to @xmath41 , hence an @xmath98-c.e . real relative to @xmath41 . if we consider @xmath98 to be a random oracle , then the representation of @xmath21 that was computed ( relative to @xmath41 ) using @xmath98 is also a random variable . however , while @xmath98 is almost surely noncomputable , the random representation of @xmath21 is almost surely constant . recall that the distribution of the oracle @xmath98 is computable because we started with a computable distribution on @xmath229 . hence we could compute @xmath21 by composing the distribution of @xmath229 with the ( almost surely constant ) representation of @xmath21 . a proof along these lines could be made precise using a simulation argument to calculate the pushforward measure of the random representation of @xmath21 . the basis elements assigned positive probability ( and therefore probability one ) are precisely those which hold of @xmath21 . instead , in section [ mainproof ] , we complete the proof by explicitly computing the representation of @xmath21 in terms of our rational basis . this construction can be seen as a `` derandomization '' of the above algorithm . alternatively , the above sketch could be interpreted as a degenerate _ probabilistic process _ ( see schrder and simpson @xcite ) which returns a representation of the de finetti measure with probability one . schrder @xcite shows that representations in terms of probabilistic processes are computably reducible to representations of computable distributions . the structure of the derandomized argument occurs in other proofs in computable analysis and probability theory . weihrauch ( * ? ? ? 3.6 ) proves a computable integration result via an argument that could likewise be seen as a derandomization of an algorithm which densely subdivides the unit interval at random locations to find continuity sets . bosserhoff ( * ? ? ? * lem . 2.15 ) uses a similar argument to compute a basis for a computable metric space , for which every basis element is a continuity set ; this suggests an alternative approach to completing our proof . mller ( * ? ? ? 3.7 ) uses a similar construction to find open hypercubes such that for any @xmath264 , the probability on their boundaries is less than @xmath265 . these arguments also resemble the classic proof of the portmanteau theorem ( * ? ? ? * thm . 4.25 ) , in which an uncountable family of sets with disjoint boundaries is defined , almost all of which are continuity sets . fix @xmath141 and let @xmath266 . define @xmath267^k \to [ 0,1]\ ] ] to be the indicator function for the set @xmath268 \times \dotsm \times ( c_{ik},1].\ ] ] for @xmath269 , we denote by @xmath270 the polynomial @xmath171 ( as defined in lemma [ effindicator ] ) , where @xmath271 here , we arbitrarily chose @xmath272 so that the sequence of polynomials @xmath273 converges pointwise from below to @xmath274 on @xmath0^k$ ] . let @xmath275 and @xmath276 . we can write @xmath277 where @xmath278 and @xmath279 are polynomials with positive coefficients . define the @xmath280-variable polynomial @xmath281 we denote @xmath282 by @xmath283 , and similarly with @xmath270 . let @xmath269 , let @xmath284 , let @xmath285 , and let @xmath286 . then @xmath287 is a c.e . real relative to @xmath41 , uniformly in @xmath30 , @xmath116 and @xmath146 . [ polypoly ] by lemma [ cemon1 ] , relative to @xmath41 , each monomial of @xmath288 has a c.e . real expectation , and each monomial of @xmath289 has a co - c.e . real expectation , and so by the linearity of expectation , @xmath287 is a c.e . we are now ready to prove the main theorem . the distribution @xmath41 is + uniformly computable in @xmath21 by proposition [ mucomp ] . we now give a proof of the other direction , showing that the joint distribution of @xmath226 is computable , relative to @xmath41 , under the right order topology . let @xmath284 , let @xmath290 and let @xmath143 . for @xmath291 , let @xmath292 denote the @xmath147-tuple @xmath293 and similarly for @xmath294 . take @xmath295 to be defined as above in . it suffices to show that @xmath296 is a c.e . real relative to @xmath41 , uniformly in @xmath297 and @xmath298 . we do this by a series of reductions , which results in a supremum over quantities of the form @xmath299 for @xmath285 . by proposition [ polypoly ] , these quantities are c.e . reals relative to @xmath41 . note that is an open set in the right order topology on @xmath0^k$ ] and so @xmath274 is an order preserving function . in particular , if @xmath300 satisfy @xmath247 , then @xmath248 ( a.s . ) , and so @xmath301 for all @xmath232 such that @xmath302 . therefore , by the density of the reals and the dominated convergence theorem , we have that @xmath303 where @xmath211 ranges over @xmath304 . recall that the polynomials @xmath305 converge pointwise from below to @xmath274 in @xmath0^k$ ] . therefore , by the dominated convergence theorem , @xmath306 multiplication is continuous , and so the dominated convergence theorem gives us @xmath307 where @xmath116 and @xmath254 range over @xmath308 . therefore , by the linearity of expectation , @xmath309 if @xmath211 is an @xmath13-continuity set , then @xmath310 a.s . for @xmath236 , and so @xmath311 because @xmath17 has at most countably many point masses , the @xmath312 which are @xmath13-continuity sets are dense in @xmath53 . therefore , @xmath313 note that @xmath314 hence @xmath315 again by dominated convergence we have @xmath316 combining , , , , , and , we have @xmath317 finally , by proposition [ polypoly ] , @xmath318 is a set of uniformly c.e . reals , relative to @xmath41 . the computable de finetti theorem has implications for the semantics of probabilistic functional programming languages , and in particular , gives conditions under which it is possible to remove uses of mutation ( i.e. , code which modifies a program s internal state ) . furthermore , an implementation of the computable de finetti theorem itself performs this code transformation automatically . for context , we provide some background on probabilistic functional programming languages . we then describe the code transformation performed by the computable de finetti theorem , using the example of the plya urn and beta - bernoulli process discussed earlier . finally , we discuss partial exchangeability and its role in recent machine learning applications . functional programming languages with probabilistic choice operators have recently been proposed as universal languages for statistical modeling ( e.g. , ibal @xcite , @xmath319@xcite , church @xcite , and hansei @xcite ) . within domain theory , idealized functional languages that can manipulate exact real numbers , such as escard s realpcf+ @xcite and plotkin s pcf++ @xcite , have also been extended by probabilistic choice operators ( e.g. , by escard @xcite and saheb - djahromi @xcite ) . the semantics of probabilistic programs have been studied extensively in theoretical computer science in the context of randomized algorithms , probabilistic model checking , and other areas . however , the application of probabilistic programs to universal statistical modeling has a somewhat different character from much of the other work on probabilistic programming languages . in bayesian analysis , the goal is to use observed data to understand hidden variables in a probabilistic model . this type of inductive reasoning , from evidence to hypothesis , can be thought of as inferring the hidden states of a program that generates the observed output . one speaks of the _ conditional execution _ of probabilistic programs , in which they are `` run backwards '' to sample from the conditional probability distribution given the observed data . one important difference is the algorithms used for conditional inference . goodman et al . @xcite describe the language church , which extends scheme , and which implements approximate conditional execution via markov chain monte carlo ( which can be thought of as a random walk over computational histories ) . park , pfenning , and thrun @xcite describe the language @xmath319 , which extends ocaml , and which implements approximate conditional execution by monte carlo importance sampling . ramsey and pfeffer @xcite describe an implementation of the probability monad in haskell , and develop `` measure terms '' suitable for implementing the sum - product algorithm , giving a more efficient method for calculating expectations . these languages also stress the flexibility of representations . in statistics and especially nonparametric statistics , there is an emphasis on higher - order distributions ( e.g. , distributions on distributions , or distributions on trees ) , and so it is crucial to work in a language which can express these types . functional languages with randomness are therefore a natural choice . the de finetti theorem gives two different representations for exchangeable sequences , each of which has its own advantages with respect to space and time complexity . in section [ partialexch ] we provide several examples of the representational changes provided by the de finetti transformation . some of these representational issues are also examined in roy et al . mansinghka @xcite has also described some uses of exchangeable sequences , and situations where one can exploit conditional independence for improved parallel execution . we now describe the code transformation performed by the computable de finetti theorem . for concreteness , we will explore this connection using church , a probabilistic functional programming language . church extends scheme ( a dialect of lisp ) with a binary - valued _ _ procedure , which denotes the bernoulli distribution . in church , an expression denotes the distribution induced by evaluation . for example , denotes @xmath324 . ( for more examples , see @xcite . ) in scheme , one can modify the state of a non - local variable using mutation via the set ! procedure . in other functional programming languages , non - local state may be implemented via other methods . ( for example , in haskell , one could use the state monad . ) if an expression modifies its environment ( using mutation or otherwise ) , it might not denote a fixed distribution . for example , a procedure may keep a counter variable and return an increasing sequence of integers on repeated calls . consider the beta - bernoulli process and the plya urn scheme written in church . ( for their mathematical characterization , see section [ polyaex ] . ) while these two processes look different , they induce the same distribution on sequences . we two define two procedures , sample - beta - coin and sample - plya - coin , such that the application of either procedure returns a procedure of no arguments whose application returns ( random ) binary values . fix @xmath325 . consider the following definitions of sample - beta - coin and sample - plya - coin ( and recall that the ( @xmath321 ( ) @xmath326 ) special form creates a procedure of no arguments ) : evaluating ( my - beta - coin ) returns a 1 with probability weight and a 0 otherwise , where the shared weight parameter is itself drawn from a beta@xmath327 distribution on @xmath0 $ ] . note that the sequence of values obtained by evaluating ( my - beta - coin ) is exchangeable but not i.i.d . ( e.g. , an initial sequence of ten 1 s leads one to predict that the next draw is more likely to be 1 than 0 ) . however , conditioned on the weight ( a random variable within the opaque procedure my - beta - coin ) the sequence is i.i.d . a second random coin constructed by will have a different weight ( a.s . ) and will generate a sequence that is independent of that generated by my - beta - coin . the sequence induced by repeated applications of my - beta - coin is exchangeable because applications of _ flip _ return independent samples . the code in _ ( ii ) _ implements the plya urn scheme with @xmath328 red balls and @xmath329 black balls ( see ( * ? ? ? * chap . 11.4 ) ) , and so the sequence of return values from my - plya - coin is exchangeable . ( one can see that the sequence is exchangeable by noting that its joint distribution depends only on the number of red and black balls and not on their order . ) because the sequence induced by repeated applications of my - plya - coin is exchangeable , de finetti s theorem implies that its distribution is equivalent to that induced by i.i.d . draws from _ some _ random measure ( in particular , the directing random measure ) . in the case of the plya urn scheme , the directing random measure is a random bernoulli whose weight parameter has a beta@xmath327 distribution . therefore ( sample - beta - coin ) denotes the de finetti measure of the sequence of repeated applications of my - beta - coin ( and also of your - beta - coin , incidentally ) . the distributions of the sequences induced by repeated applications of my - beta - coin and my - plya - coin are identical . however , there is an important difference between these two implementations . while ( sample - beta - coin ) denotes the de finetti measure , ( sample - plya - coin ) does not , because my - plya - coin does not denote a fixed distribution : the internal state of the procedure my - plya - coin changes after each iteration , as the sufficient statistics are updated ( using the mutation operator set ! ) . therefore , each element of the sequence of repeated applications of my - plya - coin is generated from a different distribution . even though the sequence of applications of my - plya - coin has the same distribution as those given by repeated applications of my - beta - coin , the procedure my - plya - coin denotes a probability kernel which depends on the state . in contrast , my - beta - coin does not modify itself via mutation ; after it is randomly initialized , the value of weight does not change during the execution of the program . therefore , in each run of the program , my - beta - coin denotes a fixed distribution a particular bernoulli . the procedure my - beta - coin is precisely the directing random measure of the beta - bernoulli process produced by repeated applications of my - beta - coin . an implementation of the computable de finetti theorem ( theorem [ cdeftheorem ] ) transforms _ ( ii ) _ into a procedure which does not use mutation and whose application is equivalent in distribution to the evaluation of ( sample - beta - coin ) . in fact , assuming we know that the code generates a binary sequence ( and therefore that the directing random measure will be a random bernoulli ) , we can transform the de finetti measure into a procedure that does not use mutation and whose application is equivalent in distribution to the evaluation of ( _ beta _ a b ) . in the general case , given a program that generates an exchangeable sequence of reals , an implementation of the computable de finetti proof can generate a mutation - free procedure generated - code such that _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ .... ( { \bf\ttfamily{define } } ( sample - directing - random - measure ) ( { \bf\ttfamily{let } } ( ( shared - randomness ( { \sl\ttfamily{uniform } } 0 1 ) ) ) ( $ \lambda$ ( ) ( generated - code shared - randomness ) ) ) ) .... _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ in addition to their simpler semantics , mutation - free procedures are often desirable for practical reasons . for example , having sampled the directing random measure , an exchangeable sequence of random variables can be efficiently sampled in parallel without the overhead necessary to communicate sufficient statistics . the example above involved binary sequences , but the computable de finetti theorem can be used to transform implementations of real exchangeable sequences . consider the following exchangeable sequence whose combinatorial structure is known as the chinese restaurant process ( see aldous @xcite ) . let @xmath330 be a computable real and let @xmath331 be a computable distribution on @xmath3 . for @xmath243 , each @xmath332 is sampled in turn according to the conditional distribution @xmath333 = \frac{n}{n+\alpha } \sum_{i=1}^n \delta_{x_i } \,+\ , \frac{\alpha}{n+\alpha } h.\ ] ] the sequence @xmath334 is exchangeable and the directing random measure is a dirichlet process with parameter @xmath335 . given such a program , we can automatically recover the underlying dirichlet process prior , samples from which produce random measures whose discrete structure was characterized by sethuraman s `` stick - breaking construction '' @xcite . note that the random measure is not produced in the same manner as sethuraman s construction and certainly is not of closed form . but the resulting mathematical objects have the same structure and distribution . sequences of random objects other than reals can sometimes be given de finetti - type representations . for example , the indian buffet process , defined by griffiths and ghahramani @xcite , is the combinatorial process underlying a _ set - valued _ exchangeable sequence that can be written in a way analogous to the plya urn in _ ( ii)_. just as the chinese restaurant process gives rise to the dirichlet process , the indian buffet process gives rise to a non - homogeneous beta - bernoulli process . unlike in the chinese restaurant process example , which was a sequence of random reals , the computable de finetti theorem is not applicable to exchangeable sequences of random sets . the directing random measure was identified as a random bernoulli process , whose base measure is a beta process ( see thibaux and jordan @xcite ) . a non - homogenous bernoulli process is a sequence of independent ( but not necessarily identically distributed ) bernoulli random variables @xmath336 , and the _ bernoulli sequence _ is the random set @xmath337 . for the beta - bernoulli process , the bernoulli sequence is almost surely finite . a `` stick - breaking construction '' of the beta - bernoulli process given by teh , grr , and ghahramani @xcite is analogous to the code in _ ( i ) _ , but gives only a @xmath338-index for the bernoulli sequence , rather than its canonical index ( see soare ( * ? ? ? * ii.2 ) ) . this observation was first noted by roy et al . the computability of sampling independent bernoulli sequences of a beta - bernoulli process remains open . however , an indian buffet process on a _ discrete _ base measure can be classically transformed into an exchangeable sequence of _ integer _ indices ( representing the finite subsets of the discrete support ) . if we are given such a representation , the computable de finetti theorem implies the existence of a computable de finetti measure that gives canonical indices for the sets . combinatorial structures other than sequences have been given de finetti - type representational theorems . for example , an array of random variables is called _ separately _ ( or _ jointly _ ) exchangeable when its distribution is invariant under ( simultaneous ) permutations of the rows and columns and their higher - dimensional analogues . nearly fifty years after de finetti s result , aldous @xcite and hoover @xcite showed that the entries of an infinite array satisfying either separate or joint exchangeability are conditionally i.i.d . these results have been connected with the theory of graph limits by diaconis and janson @xcite and austin @xcite by considering the adjacency matrix of an exchangeable random graph . as in the case of the indian buffet process , structured probabilistic models in machine learning can often be represented in multiple ways , each of which has different advantages ( e.g. , efficient inference procedures , representational simplicity , compositionality , etc . ) . the infinite relational model of kemp et al . @xcite can be viewed as a partially exchangeable array , while the hierarchical stochastic block model built with the mondrian process of roy and teh @xcite is described using the aldous - hoover representation , making the conditional independence explicit . if the computable de finetti theorem could be extended to partially exchangeable settings , it would provide analogous uniform transformations on a wider range of data structures . + the research of roy has been supported in part by an nsf graduate research fellowship . some of the results in this paper were presented at the _ computability in europe _ conference in heidelberg , germany , july 1924 , 2009 , and an extended abstract @xcite was published in the proceedings . the authors would like to thank nate ackerman , vikash mansinghka , hartley rogers , and the anonymous referees of the extended abstract for helpful comments .

while estimation of the marginal ( total ) causal effect of a point exposure on an outcome is arguably the most common objective of experimental and observational studies in the health and social sciences , in recent years , investigators have also become increasingly interested in mediation analysis . specifically , upon evaluating the total effect of the exposure , investigators routinely wish to make inferences about the direct or indirect pathways of the effect of the exposure , through a mediator variable or not , that occurs subsequently to the exposure and prior to the outcome . although powerful semiparametric methodologies have been developed to analyze observational studies that produce double robust and highly efficient estimates of the marginal total causal effect , similar methods for mediation analysis are currently lacking . thus , this paper develops a general semiparametric framework for obtaining inferences about so - called marginal natural direct and indirect causal effects , while appropriately accounting for a large number of pre - exposure confounding factors for the exposure and the mediator variables . our analytic framework is particularly appealing , because it gives new insights on issues of efficiency and robustness in the context of mediation analysis . in particular , we propose new multiply robust locally efficient estimators of the marginal natural indirect and direct causal effects , and develop a novel double robust sensitivity analysis framework for the assumption of ignorability of the mediator variable . .

1210.4654

the evaluation of the total causal effect of a given point exposure , treatment or intervention on an outcome of interest is arguably the most common objective of experimental and observational studies in the fields of epidemiology , biostatistics and in the social sciences . however , in recent years , investigators in these various fields have become increasingly interested in making inferences about the direct or indirect pathways of the exposure effect , through a mediator variable or not , that occurs subsequently to the exposure and prior to the outcome . recently , the counterfactual language of causal inference has proven particularly useful for formalizing mediation analysis . indeed , causal inference offers a formal mathematical framework for defining varieties of direct and indirect effects , and for establishing necessary and sufficient identifying conditions of these effects . a notable contribution of causal inference to the literature on mediation analysis is the key distinction drawn between so - called controlled direct effects versus natural direct effects . in words , the controlled direct effect refers to the exposure effect that arises upon intervening to set the mediator to a fixed level that may differ from its actual observed value [ @xcite , @xcite , @xcite ] . in contrast , the natural ( also known as pure ) direct effect captures the effect of the exposure when one intervenes to set the mediator to the ( random ) level it would have been in the absence of exposure [ @xcite , @xcite ] . as noted by @xcite , controlled direct and indirect effects are particularly relevant for policy making , whereas natural direct and indirect effects are more useful for understanding the underlying mechanism by which the exposure operates . in fact , natural direct and indirect effects combine to produce the exposure total effect . to formally define natural direct and indirect effects first requires defining counterfactuals . we assume that for each level of a binary exposure @xmath0 , and of a mediator variable @xmath1 , there exist a counterfactual variable @xmath2 corresponding to the outcome @xmath3 had possibly contrary to fact the exposure and mediator variables taken the value @xmath4 . similarly , for @xmath5 , we assume there exists a counterfactual variable @xmath6 corresponding to the mediator variable had possibly contrary to fact the exposure variable taken the value @xmath7 . the current paper concerns the decomposition of the total effect of @xmath0 on @xmath3 , in terms of natural direct and natural indirect effects , which , expressed on the mean difference scale , is given by@xmath8\\[-8pt ] \qquad&=&\overbrace{\mathbb{e } ( y_{e=1,m_{e=1}}-y_{e=1,m_{e=0 } } ) } ^{\mathrm{natural\ indirect\ effect}}+ \overbrace{\mathbb{e } ( y_{e=1,m_{e=0}}-y_{e=0,m_{e=0 } } ) } % ^{\mathrm{natural\ direct\ effect } } , \nonumber\hspace*{-25pt}\end{aligned}\ ] ] where @xmath9 stands for expectation . in an effort to account for confounding bias when estimating causal effects , such as the average total effect ( [ total_effect ] ) from nonexperimental data , investigators routinely collect and adjust for in data analysis , a large number of confounding factors . because of the curse of dimensionality , nonparametric methods of estimation are typically not practical in such settings , and one usually resorts to one of two dimension - reduction strategies ; either one relies on a model for the outcome given exposure and counfounders , or alternately one relies on a model for the exposure , that is , the propensity score . recently , powerful semiparametric methods have been developed to analyze observational studies that produce so - called double robust and highly efficient estimates of the exposure total causal effect [ robins ( @xcite ) , @xcite , @xcite , @xcite ] and similar methods have also been developed to estimate controlled direct effects [ @xcite ] . an important advantage of a double robust method is that it carefully combines both of the aforementioned dimension reduction strategies for confounding adjustment , to produce an estimator of the causal effect that remains consistent and asymptotically normal , provided at least one of the two strategies is correct , without necessarily knowing which strategy is indeed correct [ @xcite ] . unfortunately , similar methods for making semiparametric inferences about marginal natural direct and indirect effects are currently lacking . thus , this paper develops a general semiparametric framework for obtaining inferences about marginal natural direct and indirect effects on the mean of an outcome , while appropriately accounting for a large number of confounding factors for the exposure and the mediator variables . our semiparametric framework is particularly appealing , as it gives new insight on issues of efficiency and robustness in the context of mediation analysis . specifically , in section [ sec2 ] , we adopt the sequential ignorability assumption of imai , keele and tingley ( @xcite ) under which , in conjunction with the standard consistency and positivity assumptions , we derive the efficient influence function and thus obtain the semiparametric efficiency bound for the natural direct and natural indirect marginal mean causal effects , in the nonparametric model @xmath10 in which the observed data likelihood is left unrestricted . we further show that in order to conduct mediation inferences in @xmath10 , one must estimate at least a subset of the following quantities : the conditional expectation of the outcome given the mediator , exposure and confounding factors ; the density of the mediator given the exposure and the confounders ; the density of the exposure given the confounders . ideally , to minimize the possibility of modeling bias , one may wish to estimate each of these quantities nonparametrically ; however , as previously argued , when as we assume throughout , we wish to account for numerous confounders , such nonparametric estimates will likely perform poorly in finite samples . thus , in section [ sec2.3 ] we develop an alternative multiply robust strategy . to do so , we propose to model ( i ) , ( ii ) and ( iii ) parametrically ( or semiparametrically ) , but rather than obtaining mediation inferences that rely on the correct specification of a specific subset of these models , instead we carefully combine these three models to produce estimators of the marginal mean direct and indirect effects that remain consistent and asymptotically normal ( can ) in a union model , where at least one but not necessarily all of the following conditions hold : the parametric or semi - parametric models for the conditional expectation of the outcome ( i ) and for the conditional density of the mediator ( ii ) are correctly specified ; the parametric or semiparametric models for the conditional expectation of the outcome ( i ) and for the conditional density of the exposure ( iii ) are correctly specified ; the parametric or semiparametric models for the conditional densities of the exposure and the mediator ( ii ) and ( iii ) are correctly specified . accordingly , we define submodels @xmath11 , @xmath12 and @xmath13 of @xmath10 corresponding to models ( a ) , ( b ) and ( c ) respectively . thus , the proposed approach is triply robust as it produces valid inferences about natural direct and indirect effects in the union model @xmath14 . furthermore , as we later show in section [ sec2.3 ] , the proposed estimators are also locally semiparametric efficient in the sense that they achieve the respective efficiency bounds for estimating the natural direct and indirect effects in @xmath15 , at the intersection submodel @xmath16 . section [ sec3 ] summarizes a simulation study illustrating the finite sample performance of the various estimators described in section [ sec2 ] , and section [ sec4 ] gives a real data application of these methods . section [ sec5 ] describes a strategy to improve the stability of the proposed multiply robust estimator which directly depends on inverse exposure and mediator density weights , when such weights are highly variable , and section [ sec6 ] demonstrates the favorable performance of two modified multiply robust estimators in the context of such highly variable weights . in section [ sec7 ] , we compare the proposed methodology to the prevailing estimators in the literature . based on this comparison , we conclude that the new approach should generally be preferred because an inference under the proposed method is guaranteed to remain valid under many more data generating laws than an inference based on each of the other existing approaches . in particular , as we argue below the approach of @xcite is not entirely satisfactory because , despite producing a can estimator of the marginal direct effect under the union model @xmath17 ( and therefore an estimator that is double robust ) , their estimator requires a correct model for the density of the mediator . thus , unlike the direct effect estimator developed in this paper , the van der laan estimator fails to be consistent under the submodel @xmath18 . nonetheless , the estimator of van der laan is in fact locally efficient in model @xmath17 , provided the model for the mediator s conditional density is either known , or can be efficiently estimated . this property is confirmed in a supplementary online appendix [ @xcite ] , where we also provide a general map that relates the efficient influence function for model @xmath19 to the corresponding efficient influence function for model @xmath20 , assuming an arbitrary parametric or semiparametric model for the mediator conditional density is correctly specified . in section [ sec8 ] , we describe a novel double robust sensitivity analysis framework to assess the impact on inferences about the natural direct effect , of a departure from the ignorability assumption of the mediator variable . we conclude with a brief discussion . suppose i.i.d . data on @xmath21 is collected for @xmath22 subjects . recall that @xmath3 is an outcome of interest , @xmath0 is a binary exposure variable , @xmath1 is a mediator variable with support @xmath23 , known to occur subsequently to @xmath0 and prior to @xmath3 and @xmath24 is a vector of pre - exposure variables with support @xmath25 that confound the association between @xmath26 and @xmath3 . the overarching goal of this paper is to provide some theory of inference about the fundamental functional of mediation analysis which judea pearl calls `` the mediation causal formula '' [ pearl ( @xcite ) ] and which , expressed on the mean scale , is @xmath27\\[-8pt ] & & \hspace*{19pt}{}\times f_{m|e , x } ( m|e=0,x = x ) f_{x}(x)\,d\mu(m , x),\nonumber\end{aligned}\ ] ] @xmath28 and @xmath29 are respectively the conditional density of the mediator @xmath1 given @xmath30 and the density of @xmath24 , and @xmath31 is a dominating measure for the distribution of @xmath32 . hereafter , to keep with standard statistical parlance , we shall simply refer to @xmath33 as the `` mediation functional '' or `` m - functional '' since it is formally a functional on the nonparametric statistical model @xmath34 of all regular laws @xmath35 of the observed data @xmath36 that satisfy the positivity assumption given below ; that is , @xmath37 , with @xmath38 the real line . the functional @xmath33 is of keen interest here because it arises in the estimation of natural direct and indirect effects as we describe next . to do so , we make the consistency assumption . _ consistency : _ @xmath39 in addition , we adopt the sequential ignorability assumption of imai , keele and tingley ( @xcite ) which states that for @xmath40 . _ _ sequential ignorability:__@xmath41 where @xmath42 states that @xmath43 is independent of @xmath44 given @xmath45 ; paired with the following : _ positivity : _ @xmath46 then , under the consistency , sequential ignorability and positivity assumptions , imai , keele and tingley ( @xcite ) showed that @xmath47 so that @xmath48 and @xmath49 , @xmath50 , are identified from the observed data , and so is the mean natural direct effect @xmath51 and the mean natural indirect effect @xmath52 . for binary @xmath3 , one might alternatively consider the natural direct effect on the risk ratio scale @xmath53 or on the odds ratio scale @xmath54 and similarly defined natural indirect effects on the risk ratio and odds ratio scales . it is instructive to contrast the expression ( [ main_functional ] ) for @xmath55 with the expression ( [ ate functional ] ) for @xmath56 corresponding to @xmath57 , and to note that the two expressions bare a striking resemblance except the density of the mediator in the first expression conditions on the unexposed ( with @xmath58 ) , whereas in the second expression , the mediator density is conditional on the exposed ( with ) . as we demonstrate below , this subtle difference has remarkable implications for inference . @xcite was the first to derive the m - functional @xmath59 under a different set of assumptions . others have since contributed alternative sets of identifying assumptions . in this paper , we have chosen to work under the sequential ignorability assumption of @xcite , @xcite , but note that alternative related assumptions exist in the literature [ @xcite , @xcite , @xcite , hafeman and vanderweele ( @xcite ) ] ; however , we note that @xcite disagree with the label `` sequential ignorability '' because its terminology has previously carried a different interpretation in the literature . nonetheless , the assumption entails two ignorability - like assumptions that are made sequentially . first , given the observed pre - exposure confounders , the exposure assignment is assumed to be ignorable , that is , statistically independent of potential outcomes and potential mediators . the second part of the assumption states that the mediator is ignorable given the observed exposure and pre - exposure confounders . specifically , the second part of the sequential ignorability assumption is conditional on the observed value of the ignorable treatment and the observed pretreatment confounders . we note that the second part of the sequential ignorability assumption is particularly strong and must be made with care . this is partly because it is always possible that there might be unobserved variables that confound the relationship between the outcome and the mediator variables , even upon conditioning on the observed exposure and covariates . furthermore , the confounders @xmath24 must all be pre - exposure variables ; that is , they must precede @xmath0 . in fact , @xcite proved that without additional assumptions , one can not identify natural direct and indirect effects if there are confounding variables that are affected by the exposure , even if such variables are observed by the investigator [ also see tchetgen tchetgen and vanderweele ( @xcite ) ] . this implies that , similarly to the ignorability of the exposure in observational studies , ignorability of the mediator can not be established with certainty , even after collecting as many pre - exposure confounders as possible . furthermore , as @xcite point out , whereas the first part of the sequential ignorability assumption could , in principle , be enforced in a randomized study , by randomizing @xmath0 within levels of @xmath60 the second part of the sequential ignorability assumption can not similarly be enforced experimentally , even by randomization . and thus , for this latter assumption to hold , one must entirely rely on expert knowledge about the mechanism under study . for this reason , it will be crucial in practice to supplement mediation analyses with a sensitivity analysis that accurately quantifies the degree to which results are robust to a potential violation of the sequential ignorability assumption . later in the paper , we develop a variety of sensitivity analysis techniques that allow the analyst to quantify the degree to which his or her mediation analysis results are robust to a potential violation of the sequential ignorability assumption . in this section , we derive the efficient influence function for the m - functional @xmath33 in @xmath10 . this result is then combined with the efficient influence function for the functional @xmath61 [ @xcite , @xcite ] to obtain the efficient influence function for the natural direct and indirect effects on the mean difference scale . thus , in the following , we shall use the efficient influence function @xmath62 of @xmath63 which is well known to be @xmath64 where for @xmath65 , we define @xmath66 so that @xmath67 , @xmath68 . the following theorem is proved in the . [ teo1 ] under the consistency , sequential ignorability and positivity assumptions , the efficient influence function of the m - functional @xmath33 in model @xmath10 is given by @xmath69 and the efficient influence function of the natural direct and indirect effects on the mean difference scale in model @xmath70 are respectively given by @xmath71 and @xmath72 thus , the semiparametric efficiency bound for estimating the natural direct and the natural indirect effects in @xmath73 are respectively given by @xmath74 and @xmath75 . although not presented here , theorem [ teo1 ] is easily extended to obtain the efficient influence functions and the respective semiparametric efficiency bounds for the direct and indirect effects on the risk ratio and the odds ratio scales by a straightforward application of the delta method . an important implication of the theorem is that all regular and asymptotically linear ( ral ) estimators of @xmath33 , @xmath76 and @xmath77 in model @xmath10 share the common influence functions @xmath78 and @xmath79 , respectively . specifically , any ral estimator @xmath80 of the m - functional @xmath33 in model @xmath10 , shares a common asymptotic expansion , @xmath81 where @xmath82 = n^{-1}\sum_{i } [ \cdot ] _ { i}$ ] . to illustrate this property of nonparametric ral estimators , and as a motivation to multiply robust estimation when nonparametric methods are not appropriate , we provide a detailed study of three nonparametric strategies for estimating the m - functional in a simple yet instructive setting in which @xmath24 and @xmath1 are both discrete with finite support . _ strategy _ 1 : the first strategy entails obtaining the maximum likelihood estimator upon evaluating the m - functional under the empirical law of the observed data , @xmath83 where @xmath84 and @xmath85 are the empirical probability mass functions , and @xmath86 is the expectation of @xmath3 under @xmath84 . _ strategy _ 2 : the second strategy is based on the following alternative representation of the m - functional : @xmath87 thus , our second estimator takes the form @xmath88 with @xmath89 the empirical estimate of the probability mass function @xmath90 . _ strategy _ 3 : the last strategy is based on a third representation of the m - functional @xmath91 thus , our third estimator takes the form @xmath92 at first glance the three estimators @xmath93 , @xmath94 and @xmath95 might appear to be distinct ; however , we observe that provided the empirical distribution function @xmath96 satisfies the positivity assumption , and thus @xmath97 , then actually @xmath98 since the three representations agree on the nonparametric model @xmath99 . therefore we may conclude that these three estimators are in fact asymptotically efficient in @xmath10 with common influence function @xmath100 . furthermore , from this observation , one further concludes that ( asymptotic ) inferences obtained using one of the three representations are identical to inferences using either of the other two representations . at this juncture , we note that the above equivalence no longer applies when as we have previously argued will likely occur in practice , @xmath32 contains 3 or more continuous variables and/or @xmath24 is too high dimensional for models to be saturated or nonparametric , and thus parametric ( or semiparametric ) models are specified for dimension reduction . specifically , for such settings , we observe that three distinct modeling strategies are available . under the first strategy , the estimator @xmath101 is obtained @xmath95 using parametric model estimates @xmath102 and @xmath103 instead of their nonparametric counterparts ; similarly under the second strategy , the estimator @xmath104 is obtained similarly to @xmath94 using estimates of parametric models @xmath105 and @xmath106 and finally , under the third strategy , @xmath107 is obtained similarly to @xmath93 using @xmath106 and @xmath108 . then , it follows that @xmath109 is can under the submodel @xmath11 , but is generally inconsistent if either @xmath110 or @xmath111 fails to be consistent . similarly , @xmath104 and @xmath112 are , respectively , can under the submodels @xmath113 and @xmath114 , but each estimator generally fails to be consistent outside of the corresponding submodel . in the next section , we propose an approach that produces a triply robust estimator by combining the above three strategies so that only one of models @xmath115 and @xmath114 needs to be valid for consistency of the estimator . the proposed triply robust estimator @xmath116 solves@xmath117 where @xmath118 is equal to @xmath119 evaluated at @xmath120 , @xmath103 , @xmath121 ; that is,@xmath122\\[-8pt ] & & \hspace*{14pt } { } + \frac{i(e=0)}{\widehat{f}_{e|x}^{\,\mathrm{par}}(0|x ) } \ { \widehat{% \mathbb{e}}^{\mathrm{par } } ( y|x , m , e=1 ) \nonumber\\ & & \hspace*{32pt } { } -\widehat{\eta}^{\,\mathrm{par } } ( 1,0,x ) \ } + \widehat{\eta}^{\,\mathrm{par } } ( 1,0,x ) \biggr ] , \nonumber\end{aligned}\ ] ] is can in model @xmath123 , where @xmath124 in the next theorem , the estimator in the above display is combined with a doubly robust estimator @xmath125 of @xmath61 [ see @xcite or @xcite ] , to obtain multiply robust estimators of natural direct and indirect effects , where @xmath126 .\ ] ] to state the result , we set @xmath127 , where @xmath128 is a known link function , and @xmath129 is a user specified function of ( @xmath130 so that @xmath131 entails a working regression model for @xmath132 , and @xmath133 solves the estimating equation @xmath134 = % \mathbb{p}_{n } \bigl [ h(x , m , e ) \bigl ( y - g^{-1 } ( \widehat{\beta}% _ { y}^{t}h(x , m , e ) ) \bigr ) \bigr ] .\ ] ] similarly , we set @xmath135 for @xmath136 , a parametric model for the density of @xmath137 $ ] with @xmath138 , solving @xmath139 = % \mathbb{p}_{n } \biggl [ \frac{\partial}{\partial\beta_{m}}\log f_{m|e , x}^{\,\mathrm{par } } ( m|e , x;\widehat{\beta}_{m } ) \biggr ] , \ ] ] and we set @xmath140 for @xmath141 , a parametric model for the density of @xmath142 $ ] with @xmath143 solving @xmath144 = % \mathbb{p}_{n } \biggl [ \frac{\partial}{\partial\beta_{e}}\log f_{e|x}^{\,\mathrm{par } } ( e|x;\widehat{\beta}_{e } ) \biggr ] .\ ] ] [ teo2 ] suppose that the assumptions of theorem [ teo1 ] hold , and that the regularity conditions stated in the hold and that @xmath145 and @xmath146 are variation independent . mediation functional : then , @xmath147 is ral under model @xmath148 with influence function @xmath149 and thus converges in distribution to a @xmath150 , where @xmath151 with @xmath152 and @xmath153 , and with @xmath154 denoting the probability limit of the estimator @xmath155 . natural direct effect : similarly , @xmath156 is ral under model @xmath19 with influence function @xmath157 defined as @xmath158 with @xmath159 replacing @xmath160 , and asymptotic variance @xmath161 defined accordingly . natural indirect effect : similarly , @xmath162 is ral under model @xmath163 with influence function @xmath164 defined as @xmath165 with @xmath166 replacing @xmath167 , and asymptotic variance @xmath168 defined accordingly . @xmath116 , @xmath169 and @xmath170 are semiparametric locally efficient in the sense that they are ral under model @xmath171 and respectively achieve the semiparametric efficiency bound for @xmath33 , @xmath172 , and @xmath173 under model @xmath19 at the intersection submodel @xmath174 , with respective efficient influence functions : @xmath175 and @xmath176 . empirical versions of @xmath177 and @xmath168 are easily obtained , and the corresponding wald - type confidence intervals can be used to make formal inferences about natural direct and indirect effects . it is also straightforward to extend the approach to the risk ratio and odds ratio scales for binary @xmath3 . by a theorem due to @xcite , part ( iv ) of the theorem implies that when all models are correct , @xmath178 , @xmath179 and @xmath180 are semiparametric efficient in model @xmath181 at the intersection submodel @xmath182 . in this section , we report a simulation study which illustrates the finite sample performance of the various estimators described in previous sections . we generated 1000 samples of size @xmath183 from the following model : 1 . @xmath184\sim \mathit{bernoulli}(0.3 + 0.4x_{1})$ ] ; 2 . @xmath185\sim -0.024 - 0.4x_{1}+0.4x_{2}+n(0,1)$ ] ; 3 . @xmath186\sim \mathit{bernoulli } ( [ 1+\exp\{- ( 0.4+x_{1}-x_{2}+0.1x_{3}-1.5x_{1}x_{3 } ) \}]^{-1 } ) $ ] ; 4 . @xmath187\sim \mathit{bernoulli}([1+\exp\{-(0.5-x_{1}+0.5x_{2}$ ] 5 . @xmath188^{-1});$ ] 6 . @xmath189\sim1 + 0.2x_{1}+0.3x_{2}+1.4x_{3}$ ] 7 . @xmath190 . we then evaluated the performance of the following four estimators of the natural direct effect @xmath191 , @xmath192 , @xmath193 and @xmath194 . note that the doubly robust estimator @xmath195 was used throughout to estimate @xmath196 . to assess the impact of modeling error , we evaluated these estimators in four separate scenarios . in the first scenario , all models were correctly specified , whereas the remaining three scenarios respectively mis - specified only one of model e , model m and model y. in order to mis - specify model e and model m , we respectively left out the @xmath197 interaction when fitting each model , and we assumed an incorrect log log link function . the incorrect model for @xmath198 simply assumed no @xmath199 interaction . @lld2.3d2.3d2.4c@ & & & & & @xmath200 + all correct & bias & 0.002 & 0.008 & 0.002 & 0.005 + @xmath201 & @xmath202 s.e . & 0.005 & 0.007 & 0.006 & 0.006 + @xmath3 wrong & bias & -0.500 & -0.500 & 0.0001 & 0.004 + @xmath201 & @xmath202 s.e . & 0.005 & 0.006 & 0.006 & 0.006 + @xmath1 wrong & bias & 0.038 & 0.008 & -0.054 & 0.003 + & @xmath202 s.e . & 0.005 & 0.007 & 0.006 & 0.006 + @xmath0 wrong & bias & 0.003 & 0.027 & 0.059 & 0.004 + @xmath201 & @xmath202 s.e . & 0.005 & 0.005 & 0.005 & 0.005 + @lld2.3d2.3d2.3c@ & & & & & @xmath200 + all correct & bias & 0.001 & 0.009 & 0.001 & 0.001 + @xmath201 & @xmath202 s.e . & 0.004 & 0.005 & 0.004 & 0.004 + @xmath3 wrong & bias & -0.484 & -0.484 & 0.003 & 0.003 + @xmath201 & @xmath202 s.e . & 0.004 & 0.004 & 0.004 & 0.004 + @xmath1 wrong & bias & 0.136 & -0.008 & 0.056 & 0.01 + & @xmath202 s.e . & 0.004 & 0.05 & 0.004 & 0.01 + @xmath0 wrong & bias & 0.001 & -0.024 & -0.054 & 0.001 + @xmath201 & @xmath202 s.e . & 0.004 & 0.004 & 0.004 & 0.004 + tables [ tab1 ] and [ tab2 ] summarize the simulation results which largely agree with the theory developed in the previous sections . mainly , all proposed estimators performed well at both moderate and large sample sizes in the absence of modeling error . furthermore , under the partially mis - specified model in which model.y was incorrect , both estimators , @xmath203 and @xmath204 , showed significant bias irrespective of sample size , while @xmath205 and @xmath206 both performed well . similarly when model m was incorrect , the estimators @xmath207 and @xmath208 resulted in large bias , when compared to the relatively small bias of @xmath203 and @xmath179 . finally , mis - specifying model e lead to estimators @xmath209 and @xmath210 that were significantly more biased than the estimators @xmath211 and @xmath212 . interestingly , the efficiency loss of the multiply robust estimator remained relatively small when compared to the consistent nonrobust estimator under the various scenarios , suggesting that , at least in this simulation study , the benefits of robustness appear to outweigh the loss of efficiency . in this section , we illustrate the methods in a real world application from the psychology literature on mediation . we re - analyze data from the job search intervention study ( jobs ii ) also analyzed by @xcite . jobs ii is a randomized field experiment that investigates the efficacy of a job training intervention on unemployed workers . the program is designed not only to increase reemployment among the unemployed but also to enhance the mental health of the job seekers . in the study , 1801 unemployed workers received a pre - screening questionnaire and were then randomly assigned to treatment and control groups . the treatment group with @xmath213 participated in job skills workshops in which participants learned job search skills and coping strategies for dealing with setbacks in the job search process . the control group with @xmath58 received a booklet describing job search tips . an analysis considers a continuous outcome measure @xmath198of depressive symptoms based on the hopkins symptom checklist [ @xcite ] . in the jobs ii data , a continuous measure of job search self - efficacy represented the hypothesized mediating variable @xmath1 . the data also included baseline covariates @xmath24 measured before administering the treatment including : pretreatment level of depression , education , income , race , marital status , age , sex , previous occupation , and the level of economic hardship . note that by randomization , the density of @xmath142 $ ] was known by design not to depend on covariates , and therefore its estimation is not prone to modeling error . the continuous outcome and mediator variables were modeled using linear regression models with gaussian error , with main effects for @xmath214 included in the outcome regression and main effects for @xmath30 included in the mediator regression . table [ tab3 ] summarizes results obtained using @xmath93 , @xmath94 , @xmath215 and @xmath116 together with @xmath216 , @xmath50 , to estimate the direct and indirect effects of the treatment . @lld2.4d2.4d2.4d2.4@ & & & & & + direct effect & estimate & -0.0310 & -0.0310 & 0.0280 & -0.0409 + @xmath201 & s.e . & 0.0124 & 0.0620 & 0.0465 & 0.021 7 + indirect effect & estimate & -0.0160 & -0.0160 & -0.0750 & -0.0070 + & s.e . & 0.0372 & 0.0620 & 0.0434 & 0.021 7 + point estimates of both natural direct and indirect effects closely agreed under models @xmath217 and @xmath218 , and also agreed with the results of @xcite . we should note that inferences under our choice of @xmath217 are actually robust to the normality assumption and , as in @xcite , only require that the mean structure of @xmath219 $ ] and @xmath220 $ ] is correct . in contrast , inferences under model @xmath221 require a correct model for the mediator density . this distinction may partly explain the apparent disagreement in the estimated direct effect under @xmath221 when compared to the other methods , also suggesting that the gaussian error model for @xmath1 is not entirely appropriate . the multiply robust estimate of the natural direct effect is consistent with estimates obtained under models @xmath217 and @xmath218 , and is statistically significant , suggesting that the intervention may have beneficial direct effects on participants mental health ; while the multiply robust approach suggests a much smaller indirect effect than all other estimators although none achieved statistical significance . the triply robust estimator @xmath116 which involves inverse probability weights for the exposure and mediator variables , clearly relies on the positivity assumption , for good finite sample performance . but as recently shown by @xcite in the context of missing outcome data , a practical violation of positivity in data analysis can severely compromise inferences based on such methodology ; although their analysis did not directly concern the m - functional @xmath222 . thus , it is crucial to critically examine , as we do below in a simulation study , the extent to which the various estimators discussed in this paper are susceptible to a practical violation of the positivity assumption , and to consider possible approaches to improve the finite sample performance of these estimators in the context of highly variable empirical weights . methodology to enhance the finite sample behavior of @xmath223 is well studied in the literature and is not considered here ; see , for example , @xcite , @xcite and tan ( @xcite ) . we first describe an approach to enhance the finite sample performance of @xmath224 , particularly in the presence of highly variable empirical weights . to focus the exposition , we only consider the case of a continuous @xmath3 and a binary @xmath1 , but in principle , the approach could be generalized to a more general setting . the proposed enhancement involves two modifications . the first modification adapts to the mediation context , an approach developed for the missing data context ( and for the estimation of total effects ) in @xcite . the basic guiding principle of the approach is to carefully modify the estimation of the outcome and mediator models in order to ensure that the triply robust estimator given by equation ( [ triply ] ) has the simple m - functional representation @xmath225 where @xmath226 is carefully estimated to ensure multiple robustness . the reason for favoring an estimator with the above representation is that it is expected to be more robust to practical positivity violation because it does not directly depend on inverse probability weights . however , as we show next , to ensure multiple robustness , estimation of @xmath227 involves inverse probability weights , and therefore , @xmath228 indirectly depends on such weights . our strategy involves a second step to minimize the potential impact of this indirect dependence on weights . in the following , we assume , to simplify the exposition , that a simple linear model is used : @xmath229 \beta_{y}.\ ] ] then , similar to @xcite , one can verify that the above m - functional representation of a triply robust estimator is obtained by estimating @xmath230 with @xmath231 obtained via weighted logistic regression in the unexposed - only , with weight @xmath232 ; and by estimating @xmath233 using weighted ols of @xmath3 on @xmath32 in the exposed - only , with weight @xmath234 provided that both working models include an intercept . the second enhancement to minimize undue influence of variable weights on the m - functional estimator , entails using @xmath235 in the previous step instead of @xmath236 , where @xmath237 with @xmath238\bigr).\ ] ] this second modification ensures a certain boundedness property of inverse propensity score - weighting . specifically , for any bounded function @xmath239 of @xmath3 and @xmath1 ; consider for a moment the goal of estimating the counterfactual mean @xmath240 ; then it is well known that even though @xmath241 is bounded , the simple inverse - probability weighting estimator @xmath242 could easily be unbounded , particularly if positivity is practically violated . in contrast , as we show next , the estimator @xmath243 is generally bounded . to see why , note that @xmath244}\bigl(1-\mathbb{p}_{n}(e)\bigr ) \biggr\}\\ & & { } + \mathbb{p}% _ { n } \ { r \}\end{aligned}\ ] ] which is bounded since the second term is bounded , and the first term is a convex combination of bounded variables , and therefore is also bounded . furthermore , @xmath245 $ ] converges in probability to @xmath246 provided that @xmath236 converges to @xmath90 , ensuring that the expression in the above display is consistent for @xmath247 . the nonparametric bootstrap is most convenient for inference using @xmath235 . in the next section , we study , in the context of highly variable weights , the behavior of our previous estimators of @xmath33 , together with that of the enhanced estimators @xmath248 , @xmath249 , where @xmath250 is constructed as described above using @xmath251 , and @xmath252 uses @xmath235 . we adapted to the mediation setting , the missing data simulation scenarios in @xcite which were specifically designed so that , when misspecified , working models are nonetheless nearly correct , but yield highly variable inverse probability weights with practical positivity violation in the context of estimation . we generated 1000 samples of size @xmath253 from the following model : 1 . @xmath254 2 . @xmath255 ; @xmath256 3 . and @xmath257 , so that @xmath258 may be expressed in terms of @xmath24 . @xmath186\sim \mathit{bernoulli } ( [ 1+\exp\ { ( z_{1}-0.5z_{2}+0.25z_{3}$ ] 5 . @xmath259^{-1 } ) $ ] ; 6 . @xmath187\sim \mathit{bernoulli}([1+\exp\{-(0.5-z_{1}+0.5z_{2}$ ] 7 . @xmath260^{-1})$ ] ; 8 . @xmath189\sim210 + 27.4z_{1}+13.7z_{3}+13.7z_{3}$ ] 9 . @xmath261 . correctly specified working models were thus achieved when an additive linear regression of @xmath3 on @xmath258 , a logistic regression of @xmath1 with linear predictor additive in @xmath258 and @xmath0 and a logistic regression of @xmath0 with linear predictor additive in the @xmath258 , respectively . incorrect specification involved fitting these models with @xmath24 replacing @xmath258 , which produces higly variable weights . for instance , an estimated propensity score as small as @xmath262 occurred in the simulation study reflecting an effective violation of positivity ; similarly , a mediator predicted probability as small as @xmath263 also occured in the simulation study . @lld2.3d3.3d4.3d3.4d2.4d2.8@ & & & & & & & + all correct & bias & 0.001 & -0.207 & 0.498 & 0.003 & -0.08 & -0.079 + & @xmath202 s.e . & 2 . 615 1 & 2 . 615 5 & 2 . 615 3 + @xmath3 wrong & bias & -9 . 221 & 0.498 & -0.147 & -0.502 & -0.202 + & @xmath202 s.e . & 3 . 141 + @xmath1 wrong & bias & -0.033 & -0.207 & -9 . 497 & 0.001 & 0.046 & 0.046 + & @xmath202 s.e . & 2 . 614 + @xmath0 wrong & bias & -0.001 & 0.132 & 210 . 450 & 0.066 & -0.089 & -0.087 + & @xmath202 s.e . & 2.614 & 4.373 & 2336.92 & 4.891 & 2.619 & 2.615 + @xmath264 wrong & bias & -9.869 & -13.535 & 210.454 & -33.090 & -1.4609 & -2.487 + & @xmath202 s.e . & 3.322 & 5.256 & 2336.92 & 375.334 & 5.187 & 4.245 + @xmath265 wrong & bias & -9.355 & -10.220 & -9.496 & -4.346 & -3.579 & -3.579 + & @xmath202 s.e . & 3.224 & 10.539 & 15.376 & 3.912 & 3.480 & 3.441 + @xmath266 wrong & bias & -0.032 & 0.132 & 205.060 & 0.088 & -0.001 & -3.7710 ^ -5 + & @xmath202 s.e . & 2.614 & 4.373 & 2289.788 & 4.763 & 2.623 & 2.618 + @xmath267 wrong & bias & -9.355 & -13.535 & 205.060 & -37.757 & -4.223 & -5.253 + & @xmath202 s.e . & 3.224 & 5.356 & 2289.78 & 379.122 & 5.835 & 4.828 + @lld2.8d3.3d2.6d5.6d2.4d2.4@ & & & & & & & + all correct & bias & 0.0324 & 0.004 & -0.106 & 0.034 & -0.047 & -0.047 + @xmath201 & @xmath202 s.e . & 1.136 & 3.06 & 6.490 & 1.136 & 1.137 & 1.137 + @xmath3 wrong & bias & -10.256 & -10.305 & -0.106 & 0.063 & -0.147 & -0.148 + @xmath201 & @xmath202 s.e . & 1.675 & 4.005 & 6.490 & 1.769 & 1.419 & 1.407 + @xmath1 wrong & bias & & 0.004 & -9.706 & 0.033 & 0.076 & 0.076 + & @xmath202 s.e . & 1.136 & 3.060 & 5.395 & 1.137 & 1.137 & 1.135 + @xmath0 wrong & bias & 0.032 & 0.135 & 2.410 ^ 6 & 1908.76 & -0.038 & -0.030 + @xmath201 & @xmath202 s.e . & 1.136 & 1.794 & 4.310 ^ 7 & 53911.63 & 1.400 & 1.242 + @xmath264 wrong & bias & -10.256 & -14.011 & 2.410 ^ 6 & -1.110 ^ 6 & 6.201 & 1.024 + @xmath201 & @xmath202 s.e . & 1.675 & 2.386 & 4.310 ^ 7 & 2.110 ^ 7 & 9.406 & 5.097 + @xmath265 wrong & bias & -9.705 & -10.305 & -9.706 & -4.216 & -3.555 & -3.557 + & @xmath202 s.e . & 1.626 & 4.004 & 5.395 & 1.667 & 1.527 & 1.510 + @xmath266 wrong & bias & 5.710 ^ -4 & 0.135 & 2.510 ^ 6 & 2034.83 & 0.0539 & 0.0599 + @xmath201 & @xmath202 s.e . & 1.136 & 1.794 & 4.610 ^ 7 & 56090.10 & 1.429 & 1.272 + @xmath267 wrong & bias & -9.075 & -14.011 & 2.510 ^ 6 & -1.210 ^ 6 & 4.659 & -0.755 + @xmath201 & @xmath202 s.e . & 1.626 & 2.386 & 4.610 ^ 7 & 2.210 ^ 7 & 10.121 & 5.910 + tables [ tab4 ] and [ tab5 ] summarize simulation results for @xmath268 , @xmath269 , @xmath93 , @xmath270 , @xmath271 and @xmath272 . when all three working models are correct , all estimators perform well in terms of bias , but there are clear differences between the estimators in terms of efficiency . in fact , @xmath273 , @xmath116 , @xmath274 and @xmath275 have comparable efficiency for @xmath253 , but @xmath94 , @xmath93 is far more variable . moreover , under mis - specification of a single model , @xmath116 , @xmath276 and @xmath277 remain nearly unbiased , and for the most part substantially more efficient than the corresponding consistent estimator in @xmath278 , @xmath94 , @xmath279 . when at least two models are mis - specified , the multiply robust estimators @xmath116 , @xmath280 and @xmath275 generally outperform the other estimators , although @xmath270 occasionally succumbs to the unstable weights resulting in disastrous mean squared error ; see table [ tab5 ] when model m and model e are both incorrect . in contrast , @xmath275 generally improves on @xmath271 which generally outperforms @xmath116 and for the most part @xmath281 and @xmath275 appear to eliminate any possible deleterious impact of highly variable weights . in this section , we briefly compare the proposed approach to some existing estimators in the literature . perhaps the most common approach for estimating direct and indirect effects when @xmath3 is continuous uses a system of linear structural equations ; whereby , a linear structural equation for the outcome , given the exposure , the mediator and the confounders , is combined with a linear structural equation for the mediator , given the exposure and confounders , to produce an estimator of natural direct and indirect effects . the classical approach of @xcite is a particular instance of this approach . in recent work , mainly motivated by pearl s mediation functional , several authors [ imai , keele and tingley ( @xcite ) , imai , keele and yamamoto ( @xcite ) , pearl ( @xcite ) , @xcite , @xcite ] have demonstrated how the simple linear structural equation approach generalizes to accommodate both , the presence of an interaction between exposure and mediator variables , and a nonlinear link function , either in the regression model for the outcome , or in the regression model for the mediator , or both . in fact , when the effect of confounders is also modeled in such structural equations , inferences based on the latter can be viewed as special instances of inferences obtained under a particular specification of model @xmath11 for the outcome and the mediator densities . and thus , as previously shown in the simulations , an estimator obtained under a system of structural equations will generally fail to produce a consistent estimator of natural direct and indirect effects when model @xmath11 is incorrect , whereas , by using the proposed multiply robust estimator , valid inferences can be recovered under the union model @xmath282 , even if @xmath283 fails . a notable improvement on the system of structural equations approach is the double robust estimator of a natural direct effect due to @xcite . their estimator solves the estimating equation constructed using an empirical version of @xmath284 given in the online appendix . they show their estimator remains can in the larger submodel @xmath17 and therefore , they can recover valid inferences even when the outcome model is incorrect , provided both the exposure and mediator models are correct . unfortunately , the van der laan estimator is still not entirely satisfactory because unlike the proposed multiply robust estimator , it requires that the model for the mediator density is correct . nonetheless , if the mediator model is correct , the authors establish that their estimator achieves the efficiency bound for model @xmath17 at the intersection submodel @xmath285 where all models are correct ; and thus it is locally semiparametric efficient in @xmath286 . interestingly , as we report in the online supplement , the semiparametric efficiency bounds for models @xmath287 and @xmath288 are distinct , because the density of the mediator variable is not ancillary for inferences about the m - functional . thus , any restriction placed on the mediator s conditional density can , when correct , produce improvements in efficiency . this is in stark contrast with the role played by the density of the exposure variable , which as in the estimation of the marginal causal effect , remains ancillary for inferences about the m - functional and thus the efficiency bound for the latter is unaltered by any additional information on the former [ @xcite ] . in the online appendix , we provide a general functional map that relates the efficient influence function for the larger model @xmath288 to the efficient influence for the smaller model @xmath287 where the model for the mediator is either parametric or semiparametric . our map is instructive because it makes explicit using simple geometric arguments , the information that is gained from increasing restrictions on the law of the mediator . in the online appendix , we illustrate the map by recovering the efficient influence function of van der laan and petersen in the case of a singleton model ( i.e. , a known conditional density ) for the mediator and in the case of a parametric model for the mediator . we describe a semiparametric sensitivity analysis framework to assess the extent to which a violation of the ignorability assumption for the mediator might alter inferences about natural direct and indirect effects . although only results for the natural direct effect are given here , the extension for the indirect effect is easily deduced from the presentation . let @xmath289 -\mathbb{e}% [ y_{1,m}|e = e , m\neq m , x = x ] , \ ] ] then @xmath290 that is , a violation of the ignorability assumption for the mediator variable , generally implies that @xmath291 thus , we proceed as in @xcite , and propose to recover inferences by assuming the selection bias function @xmath292 is known , which encodes the magnitude and direction of the unmeasured confounding for the mediator . in the following , the support of @xmath1 , @xmath23 is assumed to be finite . to motivate the proposed approach , suppose for the moment that @xmath293 is known ; then under the assumption that the exposure is ignorable given @xmath24 , we show in the that @xmath294\\ & & \qquad = \mathbb{e } [ y_{1,m}|e=0,m = m , x = x ] \\ & & \qquad=\mathbb{e } [ y|e=1,m = m , x = x ] -t ( 1,m , x ) \bigl ( 1-f_{m|e , x } ( m|e=1,x = x ) \bigr ) \\ & & \quad\qquad{}+t ( 0,m , x ) \bigl ( 1-f_{m|e , x } ( m|e=0,x = x ) \bigr ) , \end{aligned}\ ] ] and therefore the m - functional is identified by@xmath295 -\mbox { } t ( 1,m , x ) \bigl ( 1-f_{m|e , x } ( m|e=1,x ) \bigr ) \nonumber\\ & & \hspace*{137pt } { } + t ( 0,m , x ) \bigl ( 1-f_{m|e , x } ( m|e=0,x ) \bigr ) \bigr\}\\ & & \qquad{}\times f_{m|e , x } ( m|e=0,x ) , \nonumber\end{aligned}\ ] ] which is equivalently represented as @xmath296 . \nonumber\end{aligned}\ ] ] below , these two equivalent representations , ( [ rep1 ] ) and ( [ rep2 ] ) , are carefully combined to obtain a double robust estimator of the m - functional , assuming @xmath297 is known . a sensitivity analysis is then obtained by repeating this process and reporting inferences for each choice of @xmath298 in a finite set of user - specified functions @xmath299 @xmath300 indexed by a finite dimensional parameter @xmath301 with @xmath302 corresponding to the unmeasured confounding assumption , that is , @xmath303 . throughout , the model @xmath304 for the probability mass function of @xmath1 is assumed to be correct . thus , to implement the sensitivity analysis , we develop a semiparametric estimator of the natural direct effect in the union model @xmath305 , assuming @xmath297 = @xmath306 for a fixed @xmath307 . the proposed doubly robust estimator of the natural direct effect is then given by @xmath308 where @xmath195 is as previously described , and @xmath309,\end{aligned}\ ] ] with@xmath310 & & \hspace*{-5pt}\quad=\sum_{m\in \mathcal{s } } \bigl\ { \widehat{\mathbb{e}}^{\mathrm{par } } ( y|x , m = m , e=1 ) + t_{\lambda^{\ast } } ( 0,m , x ) \bigl ( 1-\widehat{f}% _ { m|e , x}^{\,\mathrm{par } } ( m|e=0,x ) \bigr ) \\[-2pt ] & & \hspace*{-5pt}\hspace*{165pt } { } -t_{\lambda^{\ast } } ( 1,m , x ) \bigl ( 1-\widehat{f}% _ { m|e , x}^{\,\mathrm{par } } ( m|e=1,x ) \bigr ) \bigr\}\\[-2pt ] & & \hspace*{-5pt}\hspace*{17pt}\qquad{}\times \widehat{f}% _ { m|e , x}^{\,\mathrm{par } } ( m|e=0,x ) .\end{aligned}\ ] ] our sensitivity analysis then entails reporting the set @xmath311 ( and the associated confidence intervals ) , which summarizes how sensitive inferences are to a deviation from the ignorability assumption @xmath312 . a theoretical justification for the approach is given by the following formal result , which is proved in the supplemental appendix . [ teo4 ] suppose @xmath313 ; then under the consistency , positivity assumptions and the ignorability assumption for the exposure,@xmath314 is a can estimator of the natural direct effect in @xmath17 . the influence function of @xmath315 is provided in the , and can be used to construct a corresponding confidence interval . it is important to note that the sensitivity analysis technique presented here differs in crucial ways from previous techniques developed by @xcite , @xcite and @xcite . first , the methodology of @xcite postulates the existence of an unmeasured confounder @xmath316 ( possibly vector valued ) which , when included in @xmath24 , recovers the sequential ignorability assumption . the sensitivity analysis then requires specification of a sensitivity parameter encoding the effect of the unmeasured confounder on the outcome within levels of @xmath317 , and another parameter for the effect of the exposure on the density of the unmeasured confounder given @xmath318 . this is a daunting task which renders the approach generally impractical , except perhaps in the simple setting where it is reasonable to postulate a single binary confounder is unobserved , and one is willing to make further simplifying assumptions about the required sensitivity parameters [ @xcite ] . in comparison , the proposed approach circumvents this difficulty by concisely encoding a violation of the ignorability assumption for the mediator through the selection bias function @xmath319 . thus the approach makes no reference and thus is agnostic about the existence , dimension and nature of unmeasured confounders @xmath316 . furthermore , in our proposal , the ignorability violation can arise due to an unmeasured confounder of the mediator - outcome relationship that is also an effect of the exposure variable , a setting not handled by the technique of @xcite . the method of @xcite which is restricted to binary data , shares some of the limitations given above . finally , in contrast with our proposed double robust approach , a coherent implementation of the sensitivity analysis techniques of @xcite , @xcite and @xcite rely on correct specification of all posited models . we refer the reader to @xcite for further discussion of @xcite and . the main contribution of the current paper is a theoretically rigorous yet practically relevant semiparametric framework for making inferences about natural direct and indirect causal effects in the presence of a large number of confounding factors . semiparametric efficiency bounds are given for the nonparametric model , and multiply robust locally efficient estimators are developed that can be used when nonparametric estimation is not possible . although the paper focuses on a binary exposure , we note that the extension to a polytomous exposure is trivial . in future work , we shall extend our results for marginal effects by considering conditional natural direct and indirect effects , given a subset of pre - exposure variables [ tchetgen tchetgen and shpitser ( @xcite ) ] . these models are particularly important in making inferences about so - called moderated mediation effects , a topic of growing interest , particularly in the field of psychology [ @xcite ] . in related work , we have recently extended our results to a survival analysis setting [ @xcite ] . a major limitation of the current paper is that it assumes that the mediator is measured without error , an assumption that may be unrealistic in practice and , if incorrect , may result in biased inferences about mediated effects . we note that much of the recent literature on causal mediation analysis makes a similar assumption . in future work , it will be important to build on the results derived in the current paper to appropriately account for a mis - measured mediator [ tchetgen tchetgen and lin ( @xcite ) ] . proof of theorem [ teo1 ] let @xmath320 denote a one - dimensional regular parametric submodel of @xmath10 , with @xmath321 , and let @xmath322 the efficient influence function @xmath323 is the unique random variable to satisfy the following equation : @xmath324 for @xmath316 the score of @xmath325 at @xmath326 , and @xmath327 denoting differentiation w.r.t . @xmath328 at @xmath326 . we observe that @xmath329 considering the first term , it is straightforward to verify that @xmath330.\end{aligned}\ ] ] similarly , one can easily verify that @xmath331,\end{aligned}\ ] ] and finally , one can also verify that @xmath332.\end{aligned}\ ] ] thus we obtain @xmath333 given @xmath62 , the results for the direct and indirect effect follow from the fact that the influence function of a difference of two functionals equals the difference of the respective influence functions . because the model is nonparametric , there is a unique influence function for each functional , and it is efficient in the model , leading to the efficiency bound results . proof of theorem [ teo2 ] we begin by showing that @xmath334\\[-8pt ] & & \qquad = 0 \nonumber\end{aligned}\ ] ] under model @xmath19 . first note that @xmath335 under model @xmath11 . equality ( [ appzero ] ) now follows because @xmath336 and @xmath337 = \eta ( 1,0,x ) $ ] : @xmath338 } ^{\mathrm{=0 } } } \biggr ] \\ & & \quad\qquad { } + \mathbb{e } [ \eta ( 1,0,x;\beta_{y},\beta_{m } ) ] -\theta_{0 } \\ & & \qquad = 0.\end{aligned}\ ] ] second , @xmath339 under model @xmath12 . equality ( appzero ) now follows because @xmath340 and @xmath341 : @xmath342 \\ & & \quad\qquad{}+\mathbb{e } \biggl [ \frac{i(e=0)}{f_{e|x}^{\,\mathrm{par}}(1|x;\beta_{e})}\\ & & \hspace*{56pt}{}\times\mathbb{e } [ \ { \mathbb{e}^{\mathrm{par } } ( y|x , m , e=1;\beta_{y } ) -\eta ( 1,0,x;\beta_{y},\beta_{m}^{\ast } ) \ } |e=0,x ] \biggr ] \\ & & \quad\qquad{}+\mathbb{e } [ \eta ( 1,0,x;\beta_{y},\beta_{m}^{\ast } ) ] -\theta_{0 } \\ & & \qquad=\mathbb{e } \bigl [ \mathbb{e } [ \ { \mathbb{e}^{\mathrm{par } } ( y|x , m , e=1;\beta_{y } ) \ } |e=0,x ] \bigr ] -\theta_{0}=0.\end{aligned}\ ] ] third , equality ( [ appzero ] ) holds under model @xmath114 because @xmath343 \\ & & \quad\qquad{}+\mathbb{e } \biggl [ \frac{i(e=0)}{f_{e|x}^{\,\mathrm{par}}(1|x;\beta_{e})}\\ & & \hspace*{57pt}{}\times\mathbb{e } [ \ { \mathbb{e}^{\mathrm{par } } ( y|x , m , e=1;\beta_{y}^{\ast } ) -\eta ( 1,0,x;\beta_{y}^{\ast},\beta_{m } ) \ } |e=0,x ] \biggr ] \\ & & \quad\qquad{}+\mathbb{e } [ \eta ( 1,0,x;\beta_{y}^{\ast},\beta_{m } ) ] -\theta_{0 } \\ & & \qquad=\mathbb{e } \bigl [ \mathbb{e } [ \ { \mathbb{e } ( y|x , m , e=1 ) \ } |e=0,x ] \bigr]\\ & & \quad\qquad { } -\mathbb{e } \bigl [ \mathbb{e } [ \mathbb{e}^{\mathrm{par } } ( y|x , m , e=1;\beta_{y}^{\ast } ) |e=0,x ] \bigr ] \\ & & \quad\qquad{}+\mathbb{e } \bigl [ \mathbb{e } [ \mathbb{e}^{\mathrm{par } } ( y|x , m , e=1;\beta _ { y}^{\ast } ) |e=0,x ] \bigr ] -\mathbb{e } [ \eta ( 1,0,x;\beta_{y}^{\ast},\beta_{m } ) ] \\ & & \quad\qquad{}+\mathbb{e } [ \eta ( 1,0,x;\beta_{y}^{\ast},\beta_{m } ) ] -\theta_{0 } \\ & & \qquad=\mathbb{e } \bigl [ \mathbb{e } [ \ { \mathbb{e } ( y|x , m , e=1 ) \ } |e=0,x ] \bigr ] -\theta_{0}.\end{aligned}\ ] ] assuming that the regularity conditions of theorem 1a in @xcite hold for @xmath344 , the expression for @xmath165 follows by standard taylor expansion arguments , and it now follows that @xmath345 the asymptotic distribution of @xmath346 under model @xmath19 follows from the previous equation by slutsky s theorem and the central limit theorem . we note that @xmath125 is can in the union model @xmath19 since it is can in the larger model where either the density for the exposure is correct , or the density of the mediator and the outcome regression are both correct and thus @xmath347 . this gives the multiply robust result for direct and indirect effects . the asymptotic distribution of direct and indirect effect estimates then follows from similar arguments as above . at the intersection submodel @xmath348 hence @xmath349 the semiparametric efficiency claim then follows for @xmath116 , and a similar argument gives the result for direct and indirect effects . proofs of theorems 3 and [ teo4 ] the proofs are given in the online appendix . the authors would like to acknowledge andrea rotnitzky who provided invaluable comments that improved the presentation of the results given in section [ sec7 ] . the authors also thank james robins and tyler vanderweele for useful comments that significantly improved the presentation of this article .

we present a new numerical scheme for solving the advection equation and its application to vlasov simulations . the scheme treats not only point values of a profile but also its zeroth to second order piecewise moments as dependent variables , for better conservation of the information entropy . we have developed one- and two - dimensional schemes and show that they provide quite accurate solutions within reasonable usage of computational resources compared to other existing schemes . the two - dimensional scheme can accurately solve the solid body rotation problem of a gaussian profile for more than hundred rotation periods with little numerical diffusion . this is crucially important for vlasov simulations of magnetized plasmas . applications of the one- and two - dimensional schemes to electrostatic and electromagnetic vlasov simulations are presented with some benchmark tests . advection equation , conservative form , multi - moment , information entropy , vlasov simulations , magnetized plasmas

1105.3061

the kinematics of collisionless plasmas has been studied in a wide variety of fields , such as in laboratory plasma physics , space physics , and astrophysics . evolution of collisionless plasmas and self - consistent electromagnetic fields is fully described by the vlasov - maxwell ( or vlasov - poisson ) equations . thanks to recent development in computational technology , self - consistent numerical simulations of collisionless plasmas have been successfully performed from the first - principle vlasov - maxwell system of equations . there are two numerical methods to solve the vlasov equation . the most popular one is the particle - in - cell ( pic ) method @xcite , which approximates the plasma by a finite number of macro - particles . their trajectories calculated from the equation of motion are continuous in space , whereas electromagnetic fields are calculated on grid points in space . the pic method has been used for a wide variety of plasma phenomena , because it gives satisfying results even with a relatively small number of particles . however , the pic method inherently has the large statistical noise due to an approximation of the distribution function by a finite number of particles . this noise only decreases in @xmath0 when the number of particles @xmath1 is increased , making it difficult to study such as particle acceleration and thermal transport processes , in which a small number of high energy particles play an important role . to overcome this problem , an alternative method free from the statistical noise has been used , in which the vlasov equation is directly discretized on grid points in phase space . the so - called vlasov simulation involves solving the advection equation in multidimensions ( up to six ) . however , it has been widely known that a numerical solution of the advection equation suffers from spurious oscillations and numerical diffusion . a highly accurate scheme is required to preserve characteristics of the vlasov equation ( i.e. , the liouville theorem ) as much as possible . in contrast to the pic simulation , no standard scheme for the vlasov simulation has been established thus far . @xcite proposed a splitting scheme , in which the electrostatic vlasov equation is split into two advection equations in one - dimensional configuration and velocity spaces , and then are alternately advanced . both equations reduce to a simple form of the linear advection equation . moreover , their splitting method is equivalent to the second - order symplectic integration so that the conservation of energy is very well . following them , many authors have proposed high order advection schemes and applied them to the vlasov simulation . one of standard time integration methods for the advection equation is the semi - lagrangian method , which advances a physical variable by interpolating its profile between grid points and then following the characteristics backward in time . for example , @xcite and @xcite employed this method with the cubic b - spline interpolation . @xcite developed a semi - lagrangian scheme in a conservative form with an upwind - biased lagrange polynomial interpolation , called the positive and flux conservative ( pfc ) scheme . the pfc preserves mass and positivity . proposed a non - oscillatory type of the pfc scheme . the above and many popular schemes consider the time integration of a single physical variable . on the other hand , the concept of `` multi - moment '' that treats multiple dependent variables has been proposed . @xcite and @xcite developed a multi - moment semi - lagrangian scheme , called the constrained interpolation profile ( cip ) scheme . the cip scheme employs the cubic hermite interpolation . difference of the cip scheme from the others is that it treats not only a profile but also its first derivative in space as dependent variables , and their governing equations are solved as coupled equations . the cip scheme provides comparable or better solutions even with a relatively smaller number of grid points compared to the others , although it requires a higher memory cost to store multiple dependent variables . due to its high capability , the cip scheme has been applied not only to the vlasov simulation @xcite but also to magnetohydrodynamic simulations @xcite . conservative schemes are preferable for the vlasov simulation , because the integration of the distribution function in phase space is equal to the number of particles . @xcite and @xcite also proposed a conservative form of the cip , called the cip - csl2 ( cip - conservative semi - lagrangian scheme with a second - order polynomial ) . the cip - csl2 scheme treats point values of a profile and its cell - integrated values as dependent variables . the cell - integrated value is advanced in a conservative form to guarantee the conservation of mass . therefore , the cip - csl2 scheme will be more suitable than the cip scheme for the vlasov simulation . various cip - csl type schemes have been proposed thus far ( e.g. , * ? ? ? a non - oscillatory type of the cip - csl scheme was proposed by @xcite , and is successfully extended to fluid dynamic equations @xcite . @xcite developed a conservative eulerian scheme , which is applied to the electrostatic vlasov - poisson simulation @xcite . although there has been a number of numerical schemes , they have been applied mainly to the electrostatic vlasov - poisson simulation . there is an increasing interest for the application of electromagnetic vlasov simulations to magnetized plasmas @xcite . however , the vlasov simulation of magnetized plasmas is more difficult than the electrostatic one , because no suitable scheme for solving the gyro motion ( solid body rotation in velocity space ) has been established thus far . @xcite proposed a rigorous time - splitting method to solve the solid body rotation problem using the one - dimensional pfc scheme , called the backsubstitution . even with this method , however , we find considerable numerical diffusion in a long time calculation that causes non - physical plasma heating during the gyration . for this reason , it is still quite limited to deal with plasma phenomena such as heating and acceleration on the basis of the electromagnetic vlasov simulation . in this paper , we propose a new numerical scheme for the advection equation , specifically designed to solve the vlasov equation in magnetized plasmas . we first argue in section [ sec : concept - mma - scheme ] that it is important to preserve high order moments of the distribution function to reduce numerical diffusion , on the basis of the concept of the conservation of the information entropy . we develop an advection scheme , in which not only point values of a profile but also its zeroth to second order piecewise moments are treated as dependent variables . details of the scheme are described in sections [ sec : mma - one - dimension ] and [ sec : mma - two - dimension ] for one and two dimensions , respectively . benchmark tests of the scheme and its application to electrostatic and electromagnetic vlasov simulations are presented in sections [ sec : numer - tests - mma1d ] and [ sec : numer - tests - mma2d ] . finally , we summarize the paper in section [ sec : conclusion ] . we consider that the conservation of the information entropy is essential to develop a dissipationless scheme for the advection equation . when a profile @xmath2 follows the advection equation , @xmath3 , its entropy function , @xmath4 , also follows the equation , @xmath5 where @xmath6 is the self - information function of @xmath7 , and @xmath2 is regarded as the probability function that rapidly decreases toward @xmath8 . the conventional information entropy function is @xmath9 . @xcite defined more general function of the form , @xmath10 . the shape of a profile will be preserved during the advection with constant velocity when its information entropy is conserved in a numerical simulation , @xmath11 where @xmath12 is the position in physical space . however , it is difficult to develop an advection scheme that exactly guarantees the entropy conservation , because @xmath6 is nonlinear in general . then , we approximate it by expanding in the taylor series with respect to @xmath7 , @xmath13dx.\nonumber\end{aligned}\ ] ] this equation tells that the entropy can be described as a linear combination of the zeroth to @xmath14-th order moments , @xmath15 . we then consider that better conservation of the entropy may be achieved by preserving information of high order moments . for example , the entropy of a gaussian profile , @xmath16/\sqrt{2 \pi}\sigma_x$ ] , is described by up to the second order moment , @xmath17.\nonumber\end{aligned}\ ] ] since the gaussian profile is the most fundamental distribution in velocity space that maximizes the entropy in thermal equilibrium , preserving the entropy of this profile is crucially important for vlasov simulations . we speculate from this discussion that it is essential to treat , at least , up to the second order moment as dependent variables , to keep the characteristics of the vlasov equation . a `` multi - moment advection ( mma ) '' scheme is designed to treat the zeroth to second order moments as well as the profile itself as dependent variables , and advance them on the basis of their governing equations . we consider the time evolution of a one - dimensional profile @xmath18 and its zeroth to second order moments defined as @xmath19 the conservative advection equation of @xmath20 and governing equations of the moments are written as @xmath21 where @xmath22 is the velocity . here , the conservative advection equation of @xmath20 is cast into the advective form . equations ( [ eq:2 ] ) and ( [ eq:3 ] ) are obtained by multiplying equation ( [ eq:1 ] ) by @xmath23 and then integrating over space . note that the cip - csl2 scheme considers equations ( [ eq:1 ] ) and ( [ eq:2 ] ) ( @xmath24 only ) . hereafter , we assume the constant velocity case , @xmath25 , because we are concerned with the vlasov equation . to solve a set of these equations , the one - dimensional mma scheme ( mma1d ) treats four dependent variables ; the point value of the profile @xmath26 , and the piecewise moments , @xmath27 and constructs a piecewise interpolation for @xmath20 in a cell with a fourth order polynomial , @xmath28 which gives an interpolation function for @xmath29 as @xmath30 where @xmath31 to determine the coefficients @xmath32 , we use the variables at the upwind position as constraints , @xmath33 where @xmath34 are the position of the upwind grid and cell , @xmath35 stands for the sign of @xmath36 , and @xmath37 is the distance of the upwind departure position relative to @xmath38 . consequently , the coefficients are explicitly determined as @xmath39 , \label{eq:18}\\ c_{3;i } & = & \frac{10}{\delta x^2 } \left [ 2f_{i } + f_{iup } - \frac{2 { { \rm sgn}}\left(\zeta\right)}{\delta x^3 } \times \right . \nonumber \\ & & \left . \left\ { \left(10x_{iup}^2 + 25x_{i}x_{iup } + 7x_{i}^2 \right)m^0_{icell}\right.\right.\nonumber \\ & & \left.\left . - \left(45x_{iup}+39x_{i}\right ) m^1_{icell } + 84 m^2_{icell } \right\}\right ] , \label{eq:19}\\ c_{4;i } & = & \frac{-5}{\delta x^3 } \left [ 4f_{i } + 3f_{iup } - \frac{3 { { \rm sgn}}\left(\zeta\right)}{\delta x^3 } \times \right . \nonumber \\ & & \left . \left\ { \left(15x_{iup}^2 + 42x_{i}x_{iup } + 13x_{i}^2 \right)m^0_{icell}\right.\right.\nonumber \\ & & \left.\left . - 4\left(18x_{iup}+17x_{i}\right ) m^1_{icell } + 140 m^2_{icell } \right\}\right ] , \label{eq:20}\\ c_{5;i } & = & \frac{7}{\delta x^4 } \left [ f_{i } + f_{iup } - \frac{12 { { \rm sgn}}\left(\zeta\right)}{\delta x^3 } \times \right . \nonumber \\ & & \left . \left\ { \left(x_{iup}^2 + 3x_{i}x_{iup } + x_{i}^2 \right)m^0_{icell}\right.\right.\nonumber\\ & & \left.\left . - 5\left(x_{iup}+x_{i}\right ) m^1_{icell } + 10 m^2_{icell } \right\}\right ] . \label{eq:21}\end{aligned}\ ] ] where @xmath40 . then , let us consider the time integration of the variables . the mma1d employs the semi - lagrangian method . equations ( [ eq:1])-([eq:3 ] ) are split into two phases , advection and non - advection phases . we first calculate the advection phase as @xmath41 \nonumber \\ & = & { } ^nm^m_{i+1/2 } + \left\{{}^ng^m_{i+1}(x_{i+1}+\zeta ) - { } ^ng^m_{i}(x_{i}+\zeta ) \right\},\ ; \left(m=1,2\right),\label{eq:24}\end{aligned}\ ] ] where the left - superscript indicates the number of time steps and the asterisk means that the variables are at the intermediate step . here , we consider the case of the cfl number @xmath42 for simplicity . we next advance the non - advection phase . since @xmath22 is constant , the right - hand side of equation ( [ eq:3 ] ) is equal to @xmath43 . therefore , the non - advection phase is calculated as @xmath44 it is noteworthy that in the advection problem with constant velocity , the mma scheme exactly guarantees the conservation of the zeroth to second order central moments defined as @xmath45 the conservation of the zeroth order central moment ( i.e. , mass ) is evidently guaranteed by equation ( [ eq:23 ] ) . using equations ( [ eq:24 ] ) through ( [ eq:28 ] ) , the conservation of the first order central moment is obtained as @xmath46 and that of the second order central moment ( i.e. , variance ) is also as @xmath47\nonumber \\ & = & \sum_{i } { } ^{n}m^2_{i+1/2 } = \frac{1}{2 } { \int_{-\infty}^{\infty } } x^2 \cdot { } ^{n}f dx.\end{aligned}\ ] ] we consider the time evolution of a two - dimensional profile @xmath48 and its zeroth to second order moments in the @xmath7 and @xmath49 directions defined as @xmath50 the conservative advection equation of @xmath20 and governing equations of the moments are written as @xmath51 where @xmath22 and @xmath52 are the velocity component in the @xmath7 and @xmath49 directions . here , the conservative advection equation of @xmath20 is cast into the advective form . equations ( [ eq:29])-([eq:31 ] ) are obtained by multiplying equation ( [ eq:12 ] ) by @xmath23 or @xmath53 and then integrating over space . hereafter , we assume @xmath54 , but not necessarily @xmath55 , and use vector forms @xmath56 , @xmath57 , and @xmath58 . to solve a set of these equations , the two - dimensional mma scheme ( mma2d ) treats six dependent variables ; the point value of the profile @xmath59 , and the piecewise moments , @xmath60 and constructs a piecewise interpolation for @xmath20 in a cell with a polynomial , @xmath61 which gives an interpolation function for @xmath62 as @xmath63 where @xmath64 , @xmath65 , and @xmath66 has been already defined in equation ( [ eq:10 ] ) . note that the order of the interpolation function of @xmath20 is same as that used in the two - dimensional cip - csl2 scheme @xcite . to determine the coefficients @xmath67 , we use the variables at the upwind position as constraints , @xmath68 where @xmath69 are the position of the upwind grid and cell in the @xmath7 and @xmath49 directions , and @xmath70 is the distance of the upwind departure position relative to @xmath71 , determined with a second order accuracy , @xmath72 ( note again that we have assumed @xmath54 , but not necessarily @xmath55 . ) consequently , the coefficients are explicitly determined , which are listed in [ sec : coefficients - mma2d ] . then , let us consider the time integration of the variables . for the vlasov equation in magnetized plasmas , we consider two problems of the advection with constant velocity and the solid body rotation . when the velocity is constant in space , it is easy to implement the semi - lagrangian method to the two - dimensional scheme , because the volume element of the fluid does not change its shape . the advection phase of equations ( [ eq:12])-([eq:31 ] ) is calculated as @xmath73,\label{eq:39}\\ { } ^{*}{\mbox{\boldmath $ m$}}_{icell , jcell}^{m } & = & { { \rm sgn}}\left(\zeta\right){{\rm sgn}}\left(\eta\right)\times \nonumber \\ & & \left[{\int_{y_j+\eta}^{y_{jup}}}\!\!\!{\int_{x_i+\zeta}^{x_{iup}}}{\mbox{\boldmath $ x$}}^m\cdot{}^{n}f_{i , j}(x , y)dxdy \right.\nonumber \\ & & \left . + { \int_{y_j+\eta}^{y_{jup}}}\!\!\!{\int_{x_{iup}}^{x_{iup}+\zeta}}{\mbox{\boldmath $ x$}}^m\cdot{}^{n}f_{iup , j}(x , y)dxdy\right.\nonumber \\ & & \left . + { \int_{y_{jup}}^{y_{jup}+\eta}}\!\!\!{\int_{x_{i}+\zeta}^{x_{iup}}}{\mbox{\boldmath $ x$}}^m\cdot{}^{n}f_{i , jup}(x , y)dxdy\right . \nonumber \\ & & \left . + { \int_{y_{jup}}^{y_{jup}+\eta}}\!\!\!{\int_{x_{iup}}^{x_{iup}+\zeta}}{\mbox{\boldmath $ x$}}^m\cdot{}^{n}f_{iup , jup}(x , y)dxdy\right],\;\;\;\left(m=1,2\right).\label{eq:41}\end{aligned}\ ] ] here , we consider the case of the cfl number @xmath74 . using equation ( [ eq:36 ] ) , integrations in the right - hand side of equations ( [ eq:39 ] ) and ( [ eq:41 ] ) are exactly calculated as , e.g. , @xmath75{\mbox{\boldmath $ x$}}^m \cdot{}^{n}f_{i , j}(x , y ) dxdy\nonumber \\ & = & \frac{{}^{n}{\mbox{\boldmath $ m$}}_{icell , jcell}^m}{{{\rm sgn}}\left(\zeta\right){{\rm sgn}}\left(\eta\right)}-{}^{n}{\mbox{\boldmath $ g$}}_{i , j}^m(x_{iup},y_{j}+\eta ) -{}^{n}{\mbox{\boldmath $ g$}}_{i , j}^m(x_i+\zeta , y_{jup})+{}^{n}{\mbox{\boldmath $ g$}}_{i , j}^m(x_i+\zeta , y_j+\eta).\nonumber\end{aligned}\ ] ] we next advance the non - advection phase . since @xmath76 is constant , the right - hand sides of equations ( [ eq:30 ] ) and ( [ eq:31 ] ) are equal to @xmath77 and @xmath78 , respectively . therefore , the non - advection phase is calculated as @xmath79 similar to the one - dimensional case , in the two - dimensional advection problem with constant velocity , the mma scheme exactly guarantees the conservation of the zeroth to second order central moments in the @xmath7 and @xmath49 directions . when the velocity varies in space , it is quite complicated to implement the semi - lagrangian method , because the exact integration of the interpolation function over a distorted volume element is required . in such a general case ( including the solid body rotation problem ) , it is rather convenient to use the finite volume method for the time integration of the moments . @xcite implemented a convenient time integration method to the cip - csl2 scheme , in which cell - integrated values are advanced by the finite volume method with the runge - kutta time integration , whereas point values are advanced by the semi - lagrangian method . the mma2d employs their strategy . when the velocity is given as @xmath80 , equations ( [ eq:29])-([eq:31 ] ) are written in the following finite volume formulation , @xmath81 evidently , equation ( [ eq:67 ] ) guarantees the conservation of mass . line - integrated variables of @xmath20 ( up to third order ) appearing in the right - hand side of equations ( [ eq:67])-([eq:69 ] ) are constructed from the interpolation function as @xmath82 where we expand the definition of @xmath66 for the third order moment , @xmath83 the non - advection terms ( third terms on the right - hand side ) of equations ( [ eq:68 ] ) and ( [ eq:69 ] ) are equal to @xmath84 and @xmath85 for @xmath86 . for @xmath87 , the cross moment appears , which is also constructed as @xmath88 for stable calculation , the time integration of equations ( [ eq:67])-([eq:69 ] ) is implemented with the third - order tvd runge - kutta method @xcite . to construct the variables ( equations ( [ eq:62 ] ) , ( [ eq:63 ] ) , and ( [ eq:71 ] ) ) at each runge - kutta stage , we should calculate the coefficients of the interpolation function at each stage , which depend on @xmath20 as well as @xmath89 . therefore , intermediate values of @xmath20 at each stage are necessary . to obtain these , we first solve the equations of the characteristics with the runge - kutta method , @xmath90 to get intermediate values of the upwind departure position at each stage , @xmath91 , where the left - superscript @xmath92 denotes the runge - kutta stage . the solution is exactly obtained as @xmath93 where @xmath94 is the runge - kutta coefficient , @xmath95 for the third - order tvd method . thus , the intermediate values of @xmath20 are calculated with the semi - lagrangian method , @xmath96 consequently , we can calculate the coefficients at each stage , and then advance the moments with the third - order tvd runge - kutta method , which is implemented as follows , @xmath97 where @xmath98 stands for the right - hand side of equations ( [ eq:67])-([eq:69 ] ) . finally , we calculate the point value as @xmath99 unlike the advection problem , the first and second order central moments are not necessarily conserved quantities in the solid body rotation problem . on the other hand , the sum of the square of the first order moments ( i.e. , the radius of rotation ) , @xmath100 and the sum of the second order moments , @xmath101 should be conserved . equations ( [ eq:68 ] ) and ( [ eq:69 ] ) @xmath102 guarantee the conservation of equation ( [ eq:61 ] ) . taking the sum of equations ( [ eq:68 ] ) and ( [ eq:69 ] ) @xmath103 over @xmath104 and then implementing the runge - kutta calculation , we obtain @xmath105 { } ^{n } m^{1}_{x } - \omega \delta t \left[1-\frac{\left(\omega \delta t\right)^2}{6}\right ] { } ^{n } m^{1}_{y } \right\ } , \nonumber \\ \sum_{i , j } { } ^{n+1 } m^{1}_{y } & = & \sum_{i , j}\left\ { \left[1-\frac{\left(\omega \delta t\right)^2}{2}\right ] { } ^{n } m^{1}_{y } + \omega \delta t \left[1-\frac{\left(\omega \delta t\right)^2}{6}\right ] { } ^{n } m^{1}_{x } \right\}. \nonumber \end{aligned}\ ] ] this is a third - order accurate solution of the centroid position of a profile . equation ( [ eq:59 ] ) is numerically calculated as @xmath106 \left [ \left ( \sum_{i , j } { } ^{n } m^{1}_{x } \right)^2 + \left ( \sum_{i , j } { } ^{n } m^{1}_{y } \right)^2 \right].\end{aligned}\ ] ] therefore , the mma scheme will preserve the orbit of rotation with high accuracy , although it does not guarantee the exact conservation . we perform some numerical tests with the mma1d , and compare them to the cip - csl2 and pfc schemes . the accuracy of the cip - csl2 scheme is almost same as that of the cip scheme @xcite . comparisons of the cip and pfc schemes with others are found in , e.g. , @xcite , and . note that it is important for the mma scheme to accurately give the initial condition of the piecewise moments . in the following simulations , we determine the initial condition of the moments by a gauss quadrature method . we test the one - dimensional advection problem with constant velocity , @xmath107 of a gaussian profile , @xmath108,\end{aligned}\ ] ] with @xmath109 and @xmath110 . since the numbers of dependent variables are different among the three schemes ( four for the mma , two for the cip - csl2 , and one for the pfc ) , we use different grid sizes ( 0.01 for the mma , 0.005 for the cip - csl2 , and 0.0025 for the pfc ) so that the total memory usage is equal . the cfl number is 0.2 . figure [ fig:1 ] shows the results at @xmath111 solved by the mma , cip - csl2 , and pfc schemes . while the cip - csl2 and pfc schemes show numerical diffusion , the mma scheme provides quite an accurate solution . from simulation runs with different grid sizes , we evaluate the order of accuracy of the schemes . figure [ fig:2 ] shows the local error at the center of the profile , @xmath112 , as a function of the grid size . the slope of a line corresponds to the order of accuracy in space . while the cip - csl2 ( diamonds ) and pfc ( crosses ) schemes show the third order accuracy ( dashed line ) , the mma scheme ( triangles ) achieves the seventh order accuracy ( dot - dashed line ) . in figure [ fig:2 ] , the simulation results with the same memory usage are denoted as symbols with the same colors . the error of the mma scheme is much smaller than the others with the same memory usage . figure [ fig : sqr ] shows the numerical test of the square wave advection after 4,000 time steps . the grid size and cfl number are 0.005 and 0.2 . compared to the cip - csl2 and pfc schemes , the mma scheme accurately keeps the discontinuity , although it shows small numerical oscillations . we apply the mma1d to electrostatic vlasov - poisson simulations . the one - dimensional electrostatic vlasov - poisson system of equations is written as follows , @xmath113 where @xmath114 is the electron phase space distribution function , @xmath52 is the electron velocity , @xmath7 is the position in the configuration space , @xmath115 is the charge , @xmath116 is the electron rest mass , @xmath117 is the density of ambient immobile ions , and @xmath118 and @xmath119 are the electrostatic field and potential , respectively . to solve these equations , we treat eight dependent variables ; point values of the distribution function , piecewise moments in the velocity space , and their cell - integrated values in the configuration space , @xmath120 where the subscripts @xmath121 and @xmath122 denote the grid position in the velocity and configuration spaces , and @xmath123 . we then split the vlasov equation ( [ eq:75 ] ) into two phases , the advections in the velocity and configuration spaces , which are solved by the mma and cip - csl2 schemes as follows : \(1 ) the advection equation in the velocity space and its integration over the configuration space are @xmath124 where we approximate @xmath125 . these equations are solved by the mma scheme for @xmath126 and @xmath127 , respectively . \(2 ) the advection equations in the configuration space , @xmath128 are solved by the cip - csl2 scheme for @xmath129 . multiplying these equations by @xmath130 and then integrating over velocity gives the governing equations of the moments , @xmath131 since the ( @xmath116 + 1)-th order moment is required for the advance of the @xmath116-th order one , the system is not closed ( the closure problem ) . in this study , we approximate the equations as @xmath132 these advection equations are solved by the cip - csl2 scheme for @xmath133 , @xmath134 , and @xmath135 . the poisson equation ( [ eq:76 ] ) is solved by the fourier transform . the electric charge density on the right - hand side can be immediately calculated by taking the sum of @xmath136 or @xmath137 over the velocity space . since the mma scheme guarantees the conservation of @xmath136 and @xmath138 ( see equation ( [ eq:23 ] ) ) , and the cip - csl2 scheme guarantees the conservation of @xmath139 and @xmath140 , our simulation exactly conserves @xmath138 . therefore , we use @xmath138 for the charge density calculation . physical variables in the system are normalized as follows ; velocity by the thermal velocity @xmath141 , time by the inverse electron plasma frequency @xmath142 , and position by the debye length @xmath143 . the boundary conditions are periodic in the configuration space , and open in the velocity space where constant incoming fluxes are assumed while outgoing fluxes are perfectly lost . we employ the splitting method of @xcite for the time advance of the system . we first test the linear landau damping . the initial condition of the electron distribution function is given as follows , @xmath144.\label{eq:80}\end{aligned}\ ] ] the simulation domain is @xmath145 $ ] in the velocity space with 32 grid points , and @xmath146 $ ] in the configuration space with 32 grid points . the time step is 0.01 . figure [ fig:4 ] is the time profile of the electric field ( normalized by the initial value ) , showing the exponential decrease . the simulation ( solid line ) agrees well with the linear theory ( dashed line ) . the recurrence effect occurs at @xmath147 , which is comparable to the theoretical value @xmath148 obtained from the free - streaming equation @xcite . we next test the two stream instability . the initial condition of the electron distribution function is given as follows , @xmath149.\label{eq:81}\end{aligned}\ ] ] the simulation domain is @xmath150 $ ] in the velocity space with 32 , 64 , or 128 grid points , and @xmath146 $ ] in the configuration space with 64 grid points . the time step is 0.01 . for comparison , we also perform the simulation in which the advections in the velocity space as well as the configuration space are solved by the cip - csl2 scheme . figure [ fig:5 ] shows the electron distribution function at @xmath151 and @xmath152 simulated with the mma and cip - csl2 schemes . both simulations show the well - known electron hole structure in the phase space . figure [ fig:1dcut ] shows the velocity space distribution taken at @xmath153 and @xmath152 . the mma scheme ( blue line ) describes finer structures than the cip - csl2 scheme with the same grid points ( dashed line ) , and shows a similar result to the cip - csl2 scheme with the double number of grid points ( solid line ) . in figure [ fig:6 ] , we check the conservation of the total energy and the entropy , @xmath154 with the different schemes ( mma or cip - csl2 ) and the different numbers of grid points ( 32 , 64 , or 128 ) in the velocity space . the mma scheme provides the energy and entropy conservations well , even with the smallest number of grid points ( 32 , green lines ) . it is noteworthy that the mma scheme with 32 grid points shows a better result than the cip - csl2 scheme with 128 grid points ( solid lines ) , which uses the memory twice as much as the mma . figure [ fig : twst_err ] shows the relative error of the total energy @xmath155 at @xmath156 obtained from the both schemes , as a function of the number of grid points in the velocity space . with the sufficiently large number of grid points , the both schemes show the same overall accuracy , because the same method is applied to solve the advection in the configuration space . we perform some numerical tests with the mma2d , and compare them to the cip - csl2 and pfc schemes . we also apply the mma2d to electromagnetic vlasov - maxwell simulations and compare them to the pic method . we test long time simulations of the two - dimensional solid body rotation problem , @xmath157 of a gaussian profile , @xmath158.\label{eq:77}\end{aligned}\ ] ] we compare the simulations with the mma to the two - dimensional cip - csl2 @xcite and backsubstitution @xcite schemes . since the numbers of dependent variables are different among these three schemes ( six for the mma , four for the cip - csl2 , and one for the backsubstitution ) , we use the different numbers of grid points ( @xmath159 for the mma , @xmath160 for the cip - csl2 , and @xmath161 for the backsubstitution ) so that the total memory usage is equal . the simulation domain is [ -0.5,0.5 ] in both the @xmath7 and @xmath49 directions . the open boundary condition is employed where constant incoming fluxes are assumed while outgoing fluxes are perfectly lost . the time steps are 0.004@xmath162 for the mma and cip - csl2 schemes , and 0.002@xmath162 for the backsubstitution scheme so that the cfl number is close among the three simulations . figure [ fig:7 ] shows the results for a symmetric gaussian profile with @xmath163 after 50 and 300 rotations . the cip - csl2 scheme shows the most serious numerical diffusion . the backsubstitution scheme is better than the cip - csl2 , but also shows considerable diffusion after about a hundred of rotations . on the other hand , the mma scheme provides quite an accurate solution with little diffusion even after hundreds of rotations . figure [ fig:8 ] shows the results for an asymmetric gaussian profile with @xmath164 . the mma scheme provides a better solution than the others . the gaussian centroid @xmath165 and the standard deviation @xmath166 calculated by the mma scheme are kept correct even after hundreds of rotations , although a small error is seen on the tail of the profile ( the outermost contour ) . it is also found from this simulation that the phase error of the rotation is very small . we next test the solid body rotation with the advection problem , @xmath167 which describes the rotation around @xmath168 . for magnetized plasmas , this corresponds to the electric field @xmath169 drift motion . to solve this equation , we split it into two phases , the rotation phase ( equation ( [ eq:79 ] ) ) , and the advection phase , @xmath170 figure [ fig:9 ] shows the results for a symmetric gaussian profile with @xmath171 . in the problem of the rotation with advection , the backsubstitution scheme shows the most serious diffusion after several tens of rotations . again , the mma scheme provides a better solution with keeping @xmath165 and @xmath166 constant . figure [ fig : a ] shows the temporal variation of @xmath166 obtained by fitting the profile with the gaussian function . this result indicates that the mma scheme can accurately solve the @xmath172 drift motion for a long time with little numerical heating . from these simulation runs with different grid sizes , we evaluate the order of accuracy of the schemes . figure [ fig : err2d ] shows the error @xmath173 after 50 rotations , as a function of the grid size ( @xmath1 is the number of grid points ) . all schemes show the third order accuracy in space ( dashed line ) . in figure [ fig : err2d ] , the simulation results with the same memory usage are denoted as symbols with the same colors . the error of the mma scheme ( triangles ) is @xmath174 times smaller than the others with the same memory usage . the errors of the schemes increase with time . we note that the order of the interpolation function of the mma scheme ( equation ( [ eq:35 ] ) ) is exactly same as that of the cip - csl2 scheme . nevertheless , significant differences are found between the two schemes , suggesting that selection of constraints for the interpolation function is important to improve the accuracy . in this respect , we consider that the mma scheme has the reasonable selection on the basis of the concept of the entropy conservation ( section [ sec : concept - mma - scheme ] ) , as it is designed specifically for vlasov simulations . we apply the mma2d to electromagnetic vlasov - maxwell simulations . the one - dimensional electromagnetic vlasov - maxwell system of equations is written as follows , @xmath175 where @xmath176 and @xmath177 are the electric and magnetic fields , @xmath178 is the current density , @xmath179 is the speed of light , @xmath180 is the charge , and the subscript @xmath181 denotes particle species ( @xmath182 for protons and @xmath115 for electrons ) . we assume the two dimensionality in the velocity space , @xmath183 , @xmath184 , and @xmath185 . to solve these equations , we treat twelve dependent variables for both protons and electrons ; point values of the distribution function , piecewise moments in the velocity space , and their cell - integrated values in the configuration space , @xmath186 where the subscripts @xmath121 , @xmath122 , and @xmath92 denote the grid position in the @xmath187 , @xmath188 , and @xmath7 directions , @xmath189 , @xmath190 , @xmath191 , @xmath192 , and @xmath123 . we then split the vlasov equation ( [ eq:84 ] ) into three phases , the advection in the configuration space , the acceleration by electric fields , and the gyro motion @xcite . the acceleration and the gyro motion are solved by the mma scheme ( sections [ sec : mma2d - line - advect ] and [ sec : mma2d - solid - rotation-1 ] ) for @xmath193 and @xmath194 . the advection in the configuration space is solved by the cip - csl2 scheme for @xmath129 , @xmath195 , @xmath196 , and @xmath197 . those are carried out in the same manner as the electrostatic vlasov simulation that avoids the closure problem ( see section [ sec : es - vlas - simul ] ) . we solve the maxwell equation with the implicit scheme @xcite . the current density can be immediately calculated from the sum of @xmath198 or @xmath199 over the velocity space . we use the latter for the current density calculation , because they are conserved in the configuration space ( but not in the velocity space ) . physical variables in the system are normalized as follows ; velocity by the speed of light , time by the inverse electron plasma frequency @xmath200 , electromagnetic fields by an ambient magnetic field , and position by the debye length @xmath201 or the electron inertia length @xmath202 depending on problems . the boundary conditions are periodic in the configuration space and open in the velocity space . the time advance of the system is implemented as follows ; ( 1 ) advection in the configuration space with a half time step @xmath203 , ( 2 ) advance of the electromagnetic fields with a half time step , ( 3 ) acceleration by electric fields with a half time step , ( 4 ) gyro motion with a full time step @xmath204 , ( 5 ) acceleration by electric fields with a half time step , ( 6 ) advection in the configuration space with a half time step , ( 7 ) current density calculation , ( 8) advance of the electromagnetic fields with a half time step . this algorithm is a second order leap - flog time integration , also employed in pic simulations . we first test the wave propagation perpendicular to magnetic fields . the initial condition of the particle distribution function is given as follows , @xmath205,\;\;\;\left(s = p , e\right),\label{eq:96}\end{aligned}\ ] ] where @xmath206 and @xmath207 are the thermal velocity and the density distribution of @xmath181-species particles . we set a uniform density for both protons and electrons , and then add a small ( 1% ) uniform random noise only for the electron density . the initial magnetic field distribution is uniform , @xmath208 is zero , and @xmath209 is determined from the poisson equation ( [ eq:76 ] ) . simulation parameters are @xmath210 , @xmath211 , @xmath212 ( corresponding to a ratio of the proton to electron temperature being unity ) , and a ratio of the electron gyro to plasma frequency @xmath213 . the simulation domain in the velocity space is @xmath214 $ ] with 50 grid points in both the @xmath187 and @xmath188 directions . the position is normalized by @xmath201 . the grid size in the configuration space is equal to @xmath201 , and the spatial length is @xmath215 . the time step is @xmath216 . figure [ fig:10](a ) shows the fourier spectrum of @xmath209 obtained from the electromagnetic vlasov simulation integrated until @xmath217 . for comparison , we also perform the electromagnetic pic simulation with the same parameters , which is shown in figure [ fig:10](b ) . the number of particles is 5,000 in each cell so that the total memory usage is comparable between the two simulations . we can clearly identify the electron cyclotron ( bernstein ) modes . we also identify the high - frequency x - mode and z - mode at very low wavenumbers and frequencies close to the r- and l - mode cutoff ( @xmath218 and @xmath219 ) , respectively . as expected , the vlasov simulation provides a result with less noise , because it is free from the statistical noise . at very low wavenumber and frequency in figure [ fig:10](a ) , the contribution of protons is slightly seen . then , we continue the same simulation for a long time until @xmath220 . the ion bernstein modes and the lower - hybrid waves can be seen in figure [ fig:11 ] . even at this moment ( electrons gyrate more than a hundred times ) , the total energy is conserved very well within an error of @xmath221 , due to the fact that the mma scheme can solve the solid body rotation problem with little numerical diffusion . we next test the one - dimensional harris current sheet configuration , in which a dense plasma is confined at the center of the anti - parallel magnetic field configuration . the initial conditions of the particle distribution function and the magnetic field are given as a double harris current sheet , for utilizing the periodic condition in the configuration space , @xmath222\nonumber \\ & + & \frac{n_s^{+}(x)}{\pi v_{s;th}^2 } \exp\left[-\frac{v_x^2+\left(v_y - u_{s}^{+}\right)^2}{v_{s;th}^2}\right],\;\;\;\left(s = p , e\right),\label{eq:98}\\ n_{s}^{\pm}(x ) & = & n_{0 } { \rm sech}^2 \left(\frac{x \mp l/4}{\lambda}\right),\label{eq:99}\\ u_{s}^{\pm } & = & \pm { { \rm sgn}}\left(q_s\right ) \frac{v_{s;th}^2}{\omega_{gs } \lambda},\label{eq:100}\\ b_{z}(x ) & = & b_0 \left[\tanh \left(\frac{x+l/4}{\lambda}\right)-\tanh \left(\frac{x - l/4}{\lambda}\right)-1\right],\label{eq:101}\end{aligned}\ ] ] where @xmath223 is the diamagnetic drift velocity , @xmath224 is the gyro frequency outside the sheet , @xmath225 is the half thickness of the current sheet , and @xmath226 is the spatial length . the one - dimensional harris current sheet is a steady - state solution of the vlasov - maxwell system , in which the plasma and magnetic pressures balance with each other . therefore , it is a suitable benchmark test for electromagnetic vlasov simulations : an accurate scheme is required for solving the advection and rotation in the presence of spatially - inhomogeneous plasma and magnetic field distributions and thereby keeping the equilibrium . simulation parameters are @xmath227 , @xmath228 , @xmath229 , @xmath230 ( corresponding to a temperature ratio of unity ) , @xmath231 ( proton inertia length ) , and @xmath232 . the thermal velocity is determined so as to satisfy the pressure balance . the position is normalized by @xmath233 . the simulation domain is @xmath234 $ ] in the velocity space with 32 grid points in both the @xmath187 and @xmath188 directions , and @xmath235 $ ] in the configuration space with 1024 @xmath236 or 256 @xmath237 grid points . the time step is @xmath238 . figure [ fig:13](a ) shows the magnetic field and density @xmath239 distributions at @xmath240 ( corresponding to 100 gyrations of electrons outside the sheet ) , obtained from the simulation with @xmath241 . the simulation keeps the equilibrium of the current sheet with small numerical dissipation . the amplitude of numerically - produced electric fields is smaller than @xmath242 . we check the pressure balance along the @xmath7 direction between the magnetic field and plasma , @xmath243 where @xmath244 and @xmath245 are the @xmath7-diagonal component of the pressure and temperature tensors . in the simulation with the mma scheme , diagonal components of the pressure tensor can be easily estimated from dependent variables , @xmath246,\nonumber\end{aligned}\ ] ] where @xmath247 is the bulk velocity in the @xmath7 direction , also estimated from them , @xmath248 figure [ fig:13](b ) shows the pressure distribution . the pressure balance is preserved very well within an error of 0.1 % . the plasma pressure is kept correct within an error of 2 % . for comparison , we also perform the simulation in which the cip - csl2 scheme is applied to the advections in the velocity space as well as the configuration space . the number of grid points in the velocity space is @xmath249 for this simulation so that the total memory usage is comparable to the simulation with the mma scheme . the result is shown in figure [ fig : harris_csl ] . compared to the mma scheme , the cip - csl2 scheme causes the dissipation of the current sheet ( increase of the sheet thickness ) , through the numerical heating of the plasma . the proton and electron pressures increase by 10 % and 40 % around the edge of the sheet , respectively . the total pressure increases by 2 % . figure [ fig:14 ] shows the result from the same simulation with the mma scheme , but with a larger grid size of @xmath250 . the simulation again keeps the equilibrium of the current sheet . the order of a numerical error is comparable to the result with @xmath241 . the result confirms that our electromagnetic vlasov simulation code is stable even with the grid size larger than the debye length . this is an advantage over an explicit pic simulation , as suggested by , e.g. , @xcite . we have presented a new numerical scheme for solving the advection equation and the vlasov equation . the present scheme solves not only point values of a profile but also its zeroth to second order piecewise moments as dependent variables , for better conservation of the information entropy . we have developed one- and two - dimensional schemes , and have shown their high capabilities . the present scheme provides quite accurate solutions even with smaller numbers of grid points , although it requires a higher memory cost than other existing schemes . we have shown that , however , the total memory usage of the present scheme is smaller than the others for the same accuracy of solutions applications of the one - dimensional scheme to the electrostatic vlasov simulations ( linear landau damping and two stream instability ) , and the two - dimensional scheme to the electromagnetic vlasov simulations ( perpendicular wave propagation and harris current sheet equilibrium ) have been presented . the two - dimensional scheme allows us to solve the gyro motion and the @xmath251 drift motion for a long time with little numerical heating . since the present scheme treats the zeroth to second order moments and advances them on the basis of their governing equations , the particle momentum and energy as well as mass are conserved very well . this is important for studying plasma phenomena such as convection , heating , and acceleration . although the present scheme correctly solves the moments up to the second order , the entropy is numerically increased in the two stream instability ( section [ sec : two - stre - inst ] ) , by the dissipation of fine structures in velocity space . this is understood that the perturbation of lower order moments decreases and higher order moments contribute to the entropy when the so - called filamentation phenomena proceed ( e.g. , * ? ? ? since the filamentation is inevitable as long as discretizing velocity space , it is essentially impossible to exactly conserve the entropy with finite information . the present scheme is designed specifically to solve the advection and rotation in velocity space in the vlasov equation . in our vlasov simulation code , the cip - csl2 scheme has been employed to solve the advection in configuration space . we note that , however , another scheme can be applied to the advection in configuration space and be combined with the present scheme . several advection schemes proposed for vlasov simulations are designed to preserve positivity and non - oscillatory property , in order to suppress a non - physical growth of plasma waves caused by numerically - produced positive gradient in velocity space . @xcite argued that the positivity - preserving and non - oscillatory are important properties for reliable vlasov simulations . on the other hand , it may cause a non - physical global evolution driven by plasma pressure gradient that is increased by the numerical diffusion in velocity space , even when the plasma should be in an equilibrium state . although the present scheme is not positivity - preserving or non - oscillatory , we have shown that the obtained results are better than the others . we thus consider that preserving high order moments is another important property for vlasov simulations . magnetized plasma phenomena that allow to assume the two dimensionality in velocity space are limited , e.g. , strictly perpendicular shocks @xcite and the two - dimensional kelvin - helmholtz instability @xcite . even considering the one dimension in configuration space , there are many phenomena that should treat the full three - dimensional velocity space . we are now developing a three - dimensional scheme and its application to the full electromagnetic vlasov simulation , which will enable us to study a wide variety of collisionless plasma phenomena . we would like to thank t. umeda , t. miyoshi , t. sugiyama , and k. kusano for insightful comments on our manuscript , and anonymous referees for carefully reviewing the manuscript . t. m. is supported by a grand - in - aid for young scientists , ( b ) # 21740135 . @xmath252,\label{eq:46}\\ c_{13;i , j } & = & \frac{1}{\delta x^2 } \left[f_{i , j}+f_{iup , j}-\frac{6{{\rm sgn}}\left(\zeta_{i , j}\right){{\rm sgn}}\left(\eta_{i , j}\right)}{\delta x \delta y^3 } g(y_{j},y_{jup},m_{y;icell , jcell}^m)\right],\label{eq:47}\\ c_{21;i , j } & = & \frac{-1}{\delta y } \left[2f_{i , j}+f_{i , jup}-\frac{9{{\rm sgn}}\left(\zeta_{i , j}\right){{\rm sgn}}\left(\eta_{i , j}\right)}{\delta x^3 \delta y } g(x_{i},x_{iup},m_{x;icell , jcell}^m)\right],\label{eq:48}\\ c_{22;i , j } & = & \frac{1}{\delta x \delta y}\left[4f_{i , j}+2\left(f_{iup , j}+f_{i , jup}\right)+f_{iup , jup } - \frac{9{{\rm sgn}}\left(\zeta_{i , j}\right ) { { \rm sgn}}\left(\eta_{i , j}\right)}{\delta x \delta y}\times \right.\nonumber \\ & & \left . \left\{\left(30\left\{\frac{x_i x_{iup}}{\delta x^2}+\frac{y_j y_{jup}}{\delta y^2}\right\}+2\left\{\frac{x_i}{\delta x}+\frac{y_j}{\delta y}\right\}+13\right)m^0_{icell , jcell } \right.\right.\nonumber \\ & & \left.\left . -4\left(\left\{8x_{iup}+7x_{i}\right\}\frac{m^1_{x;icell , jcell}}{\delta x^2 } + \left\{8y_{jup}+7y_{j}\right\}\frac{m^1_{y;icell , jcell}}{\delta y^2}\right ) \right.\right.\nonumber \\ & & \left . \left.+ 60\left(\frac{m^2_{x;icell , jcell}}{\delta x^2 } + \frac{m^2_{y;icell , jcell}}{\delta y^2}\right)\right\}\right],\label{eq:49}\\ c_{23;i , j } & = & \frac{-1}{\delta x^2 \delta y}\left[2\left(f_{i , j}+f_{iup , j}\right)+f_{i , jup}+f_{iup , jup } - \frac{12{{\rm sgn}}\left(\zeta_{i , j}\right ) { { \rm sgn}}\left(\eta_{i , j}\right)}{\delta x \delta y}\times \right.\nonumber \\ & & \left . \left\{\left(\frac{15 x_i x_{iup}}{\delta x^2}+\frac{6y_{jup}^2 + 4y_{j}y_{jup}+5 y_j^2}{\delta y^2}\right)m^0_{icell , jcell } \right.\right.\nonumber \\ & & \left.\left . -\left(15\left\{x_{iup}+x_{i}\right\}\frac{m^1_{x;icell , jcell}}{\delta x^2 } + 2\left\{8y_{jup}+7y_{j}\right\}\frac{m^1_{y;icell , jcell}}{\delta y^2}\right)\right.\right.\nonumber \\ & & \left.\left . + 30\left(\frac{m^2_{x;icell , jcell}}{\delta x^2 } + \frac{m^2_{y;icell , jcell}}{\delta y^2}\right)\right\}\right],\label{eq:50}\\ c_{31;i , j } & = & \frac{1}{\delta y^2 } \left[f_{i , j}+f_{i , jup}-\frac{6{{\rm sgn}}\left(\zeta_{i , j}\right){{\rm sgn}}\left(\eta_{i , j}\right)}{\delta x^3 \delta y } g(x_{i},x_{iup},m_{x;icell , jcell}^m)\right],\label{eq:51}\\ c_{32;i , j } & = & \frac{-1}{\delta x \delta y^2}\left[2\left(f_{i , j}+f_{i , jup}\right)+f_{iup , j}+f_{iup , jup } - \frac{12{{\rm sgn}}\left(\zeta_{i , j}\right ) { { \rm sgn}}\left(\eta_{i , j}\right)}{\delta x \delta y}\times \right.\nonumber \\ & & \left . \left\{\left(\frac{6x_{iup}^2 + 4x_{i}x_{iup}+5 x_i^2}{\delta x^2}+\frac{15 y_j y_{jup}}{\delta y^2}\right)m^0_{icell , jcell}\right.\right.\nonumber \\ & & \left.\left . -\left(2\left\{8x_{iup}+7x_{i}\right\}\frac{m^1_{x;icell , jcell}}{\delta x^2 } + 15\left\{y_{jup}+y_{j}\right\}\frac{m^1_{y;icell , jcell}}{\delta y^2}\right)\right.\right.\nonumber \\ & & \left.\left . + 30\left(\frac{m^2_{x;icell , jcell}}{\delta x^2 } + \frac{m^2_{y;icell , jcell}}{\delta y^2}\right)\right\}\right],\label{eq:52}\\ c_{33;i , j } & = & \frac{1}{\delta x^2 \delta y^2}\left[f_{i , j}+f_{iup , j}+f_{i , jup}+f_{iup , jup } - \frac{4{{\rm sgn}}\left(\zeta_{i , j}\right ) { { \rm sgn}}\left(\eta_{i , j}\right)}{\delta x \delta y}\times \right.\nonumber \\ & & \left . \left\{\left(30\left\{\frac{x_i x_{iup}}{\delta x^2}+\frac{y_j y_{jup}}{\delta y^2}\right\}+11\right)m^0_{icell , jcell } \right.\right.\nonumber \\ & & \left.\left . -30\left(\left\{x_{iup}+x_{i}\right\}\frac{m^1_{x;icell , jcell}}{\delta x^2 } + \left\{y_{jup}+y_{j}\right\}\frac{m^1_{y;icell , jcell}}{\delta y^2}\right)\right.\right.\nonumber \\ & & \left.\left . + 60\left(\frac{m^2_{x;icell , jcell}}{\delta x^2 } + \frac{m^2_{y;icell , jcell}}{\delta y^2}\right)\right\}\right],\label{eq:53}\end{aligned}\ ] ] where @xmath253 , @xmath254 , and , @xmath255 , t. , hoshino , m. , jan . 2009 . electron shock surfing acceleration in multidimensions : two - dimensional particle - in - cell simulation of collisionless perpendicular shock . astrophysical journal 690 , 244251 . , k. , kishimoto , y. , saito , d. , li , j. , utsumi , t. , dec . 2009 . a numerical method for solving the vlasov - poisson equation based on the conservative ido scheme . journal of computational physics 228 , 89198943 . , y. , aoki , t. , takizawa , k. , feb . conservative form of interpolated differential operator scheme for compressible and incompressible fluid dynamics . journal of computational physics 227 , 22632285 . , t. , miwa , j. , matsumoto , y. , nakamura , t. k. m. , togano , k. , fukazawa , k. , shinohara , i. , may 2010 . full electromagnetic vlasov code simulation of the kelvin - helmholtz instability . physics of plasmas 17 ( 5 ) , 052311+ . , t. , togano , k. , ogino , t. , mar . two - dimensional full - electromagnetic vlasov code with conservative scheme and its application to magnetic reconnection . computer physics communications 180 , 365374 . , f. , apr . 2004 . unified formulation for compressible and incompressible flows by using multi - integrated moments i : one - dimensional inviscid compressible flow . journal of computational physics 195 , 629654 . , f. , akoh , r. , , ii , s. , mar . 2006 . unified formulation for compressible and incompressible flows by using multi - integrated moments ii : multi - dimensional version for compressible and incompressible flows . journal of computational physics 213 , 3156 . , t. , ishikawa , t. , wang , p. y. , aoki , t. , kadota , y. , ikeda , f. , sep . 1991 . a universal solver for hyperbolic equations by cubic - polynomial interpolation ii . two- and three - dimensional solvers . computer physics communications 66 , 233242 . and ( c , d ) @xmath152 in the two stream instability simulation , with ( a , c ) the mma and ( b , d ) the cip - csl2 schemes . the numbers of grid points are 64 and 128 in the @xmath7 and @xmath52 directions . the time , position , and velocity are normalized by the inverse electron plasma frequency , debye length , and thermal velocity , respectively . ] and @xmath152 in the two stream instability simulation . a blue line is obtained from the mma scheme with 64 grid points in the velocity space . black solid and dashed lines are from the cip - csl2 scheme with 128 and 64 points , respectively . the time , position , and velocity are normalized by the inverse electron plasma frequency , debye length , and thermal velocity , respectively . ] , as a function of the number of grid points in the velocity space . triangles and diamonds are the results obtained from the mma and cip - csl2 schemes . symbols with the same colors are obtained from the simulations with the same memory usage . ] . horizontal and vertical axes are the wavenumber and frequency normalized by the inverse electron gyro radius and the electron gyro frequency . color contour shows the fourier component of @xmath209 ( normalized by its maximum value ) to the power of 0.15 ( for illustration ) . from top to bottom , dashed lines represent the r - mode cutoff , upper hybrid , l - mode cutoff , and lower hybrid frequencies . dot - dashed lines represent the dispersion relation of the light mode in vacuum . ] . horizontal and vertical axes are the wavenumber and frequency normalized by the inverse proton gyro radius and the proton gyro frequency . color contour shows the fourier component of @xmath209 ( normalized by its maximum value ) to the power of 0.15 ( for illustration ) . a dot - dashed line represents the dispersion relation of the alfvn wave . ]

the impending javalambre physics of the accelerating universe astrophysical survey ( j - pas ) will be the first wide - field survey of 8500 deg to reach the ` stage iv ' category . because of the redshift resolution afforded by 54 narrow - band filters , j - pas is particularly suitable for cluster detection in the range z . the photometric redshift dispersion is estimated to be only with few outliers 4% for galaxies brighter than ab , because of the sensitivity of narrow band imaging to absorption and emission lines . here we evaluate the cluster selection function for j - pas using n - body+semi - analytical realistic mock catalogues . we optimally detect clusters from this simulation with the bayesian cluster finder , and we assess the completeness and purity of cluster detection against the mock data . the minimum halo mass threshold we find for detections of galaxy clusters and groups with both% completeness and purity is up to . we also model the optical observable ,-halo mass relation , finding a non - evolution with redshift and main scatter of down to a factor two lower in mass than other planned broad - band stage iv surveys , at least . for the planck mass limit , j - pas will arrive up to with a . therefore j - pas will provide the largest sample of clusters and groups up to with a mass calibration accuracy comparable to x - ray data . [ firstpage ] cosmology : large - scale structure of universe - cosmology : observations - surveys - cosmology : dark matter - cosmology : miscellaneous - galaxies : clusters : general

1601.00656

we are living exciting times in cosmology . roughly 15 years after the discovery of the inconsistency of a @xmath15 universe with the magnitude - redshift observed relation for the type ia supernova @xcite , many cosmological probes have pointed towards the universe passing through a phase of accelerated expansion : the cosmic microwave background anisotropies ( e.g. @xcite ) , the baryonic acoustic oscillations @xcite , the clustering of galaxies ( e.g. @xcite ) and the growth of massive galaxy clusters ( e.g. @xcite ) , among others . one possible explanation for this acceleration can be postulated by introducing a new energy component in the form of a dark energy with negative pressure ( for a review , see @xcite ) . consequently , the dark energy task force ( detf , @xcite ) has been created , urging the cosmology community to invest its effort in understanding the origin and nature of the dark energy . with this goal in mind , a number of surveys have been planned for the upcoming years intending to constrain the values of the dark energy by a factor of @xmath1610 times better than at present ( the so - called stage iv surveys , @xcite ) . some of these surveys , cited in chronological order of predicted start off , are : the javalambre physics of the accelerating universe astrophysical survey ( j - pas @xcite ) , the dark energy spectroscopic instrument ( desi , @xcite ) survey , the large synoptic sky telescope ( lsst , @xcite ) survey , the euclid @xcite survey and the wide - field infrared survey telescope ( wfirst ) , among many others . for an excellent review of some of these surveys , we refer the reader to @xcite . these surveys will follow complementary observational strategies , allowing constraints covering different regions of the cosmological parameter space for testing competing cosmological models . @xcite recently explored the properties of the photometric redshift capabilities we may expect for the euclid and lsst surveys , showing their different behaviours . in this work , we explore the unique photometric capability of the impending j - pas multiple ( @xmath6 50 ) narrow - band survey now being commissioned . this survey samples the optical spectrum with 54 narrow - bands of @xmath17 , providing photometric redshift accuracies close to what we would expect for a low resolution spectroscopic survey , where emission and absorption features will be detected photometrically . this kind of data , in the frontier between spectroscopic and photometric surveys , has never been explored before and it will allow us to reliably map the large - scale structure in 3d down to fainter magnitudes and larger areas than previous spectroscopic samples . to realise the full scientific potential of these data , new algorithms and techniques are being developed and tested and cosmological constraints will be forecasted @xcite . in particular , we focus on the expected performance of j - pas for galaxy clusters and groups related to cosmology . clusters , by virtue of their extreme masses are of great importance for the purpose of setting cosmology constraints and the study of the large - scale structure ( e.g. @xcite or see for a review @xcite ) . modeling accurately the cluster selection function and the uncertainties in the observable- theoretical mass relation is of major importance for extracting cosmological information from j - pas . the amplitude of the cluster power - spectrum ( e.g @xcite and references herein ) is expected to be rapidly evolving over the redshift range accessible to j - pas ( [email protected] ) , hence becoming very sensitive to the growth rate of structure ( see review by @xcite ) . presently , only relatively weak constraints are derived from redshift space distortions ( e.g. sdss , wiggle - z , boss , etc ) and even less derived from massive clusters owing to the current difficulties of completing even modest sized surveys with sufficient redshift information . furthermore , current wide angle surveys sensitive to clusters through weak lensing ( wl ) , x - ray measurements or the sunyaev - zeldovich ( sz ) effect are still dealing with alleviating the tension between the different scaling of the sz , wl and x - ray observable mass and the theoretical cluster mass in order to make the connection to the cosmological predictions ( e.g. @xcite ) . numerous techniques have been developed to detect galaxy clusters using x - ray data , the sz effect , wl or optical / ir data ( see @xcite and references herein ) , being their selection function carefully modeled . for the optical and ir techniques , there is a large diversity of selection functions depending on the technique or the survey considered ( for a review see @xcite ) . for instance , only few surveys with large number of medium or narrow - bands ( i.e. good photometric redshift resolution ) had their cluster samples fully exploited ( cosmos , @xcite ; alhambra , @xcite ) . therefore , the selection function of clusters in narrow - band surveys such as j - pas are still in the process of being explored . in this paper we provide a comprehensive estimate of the cluster selection function for j - pas , accounting for the expected photometric limits and redshift accuracy of our multi - narrow - bands . it must be stressed that j - pas provides near optimal efficiency for separating cluster members from foreground and background galaxies because of its photometric redshift precision . this accuracy of the photometric redshifts is matched to the typical velocity dispersion of massive clusters , and therefore we can detect clusters above the noise to much lower masses and to higher redshifts than the wide - field surveys using conventional filters . this cluster selection function will be useful not only for providing cosmological forecasts from cluster counts but also for performing extended studies on galaxy evolution and large - scale structure in clusters the structure of the paper is as follows . in section 2 , we describe j - pas , the survey used in this paper , giving an overview of its main characteristics . 3 describes the simulation used in this work , the original mock catalogue and the posterior modification with ` photreal ` intended to mimic the photometry and photometric redshifts realistically . in section 4 , we describe the photometric redshift properties of the j - pas data and compare with other next - generation surveys such as the lsst and euclid . section 5 presents the results of detecting galaxy clusters in these mocks . it first provides an explanation of the bayesian cluster finder , the cluster detector used in this work . then , it shows the results regarding the cluster selection function expected from the j - pas - mock catalogue and it finally models the cluster observable - halo mass relation and its evolution with redshift . finally , we draw conclusions of the work in section 6 . the cosmology adopted throughout this paper is @xmath18 @xmath19 , @xmath20 , @xmath21 , @xmath22 , @xmath23 , corresponding to the cosmology assumed in the millennium simulation used in this work for consistency . all the magnitudes in this work are provided in the ab system @xcite . the javalambre - physics of the accelerating universe astrophysical survey ( j - pas , @xcite ) is the first stage iv survey , starting in 2016 . the observations will be taken from the javalambre survey telescope ( jst / t250 ) , a new fully dedicated 2.5 m telescope located at the observatorio astrofsico de javalambre in teruel ( spain ) , using jpcam , a panoramic camera with a mosaic of 14 large - format ccds amounting to 1200 mpix , that provides an effective field of view of @xmath244.7deg@xmath1 ( see @xcite ) . with the main purpose of constraining the dark energy parameters with at least 10 times higher precision than present surveys , j - pas will image @xmath0 8500 @xmath25 of the northern sky with 54 narrow - band filters plus 2 medium - band and 3 broad - band @xmath26-like filters in the whole optical range . each narrow - band filter will have a width of @xmath27 and will be spaced by 100 . the filter transmission curves of the 54 narrow - band overlapping filters plus the two medium - band filters for j - pas are displayed in fig . [ fig : transmissioncurves ] ( see also @xcite ) . for comparison , we also plot the five broad - band filters of the sloan digital sky survey ( sdss ) . as we can see , the optical wavelength range for a low - redshift object will be sampled with more than 50 data points allowing , not only to recover a good estimation of the photometric redshift , but also to infer intrinsic properties of the galaxies . the expected depth of the survey ( 5@xmath28 detection magnitudes ) for all the different bands are provided in tables 3 - 5 in @xcite from realistic simulations using the characteristics of the telescope , camera and site . in addition , we have created a synthetic @xmath29-band as a combination of the narrow - band filters of the survey , by following a similar procedure to that described in @xcite for the alhambra survey . this has been made in order to use the same pass - band to detect galaxy clusters as some other work in the literature ( e.g. @xcite ) . due to the large coverage of the visible spectrum , the expected photometric redshift accuracy will be @xmath30 for more than @xmath31 galaxies down to the flux limit of the survey @xcite . this photometric redshift resolution makes this survey comparable to a low resolution integral field unity ( ifu ) of the northern sky . the excellent photometric redshift precision that j - pas will achieve , makes this survey ideal for characterizing the overall galaxy population in terms of colours , morphology or chemical composition and therefore , for determining the cluster galaxy membership . in this paper , we use a mock catalogue generated by using the same procedure as in @xcite . indeed , we use the 500 @xmath25 wide mock cone catalogue by @xcite designed to mimic euclid and , we transform it into a j - pas mock catalogue by using ` photreal ` . this technique , described in @xcite , obtains a new photometry and photometric error set for a particular survey to reproduce the observational properties of the galaxies with fidelity . after that , photometric redshifts have been derived by using bpz2.0 ( @xcite , bentez et al . in prep ) . in this section , we give a brief description of the mock catalogue construction . we use a mock catalogue constructed from the millennium dark matter simulation @xcite . the dark matter haloes have been populated with galaxies created through the semi - analytic galaxy formation model galform @xcite . the light - cone is built from different simulation s snapshots up to @xmath32 , allowing for interpolation between snapshots in order to properly model the evolution of structures along the line of sight . for a detailed explanation , we refer the reader to @xcite . the solid angle of the cone used is 500 @xmath25 , which is @xmath24 16 times smaller than the actual surveyed area of j - pas . while this fact does not affect the recovery of the photometric redshift accuracy ; it might cause the absence of some rare , very massive clusters . however , these clusters are always well - identified through different techniques and for the purpose of characterizing the selection function , the results will remain virtually unchanged . the mock catalogue contains a large quantity of physical parameters related with the dark matter haloes ( namely : dark matter mass , dark matter i d , center of halo , galaxies belonging to each halo ) and the galaxies ( galaxy i d , ra , dec , redshift , mass of cold gas , quiescent sfr in disk , stellar mass of galaxy , among others ) . the catalogue also includes spectroscopic redshifts , defined as the cosmological redshift with peculiar velocities added . finally , the photometry of the galaxies in the five @xmath26 sdss broad - bands together with some other different bands mimicking euclid and other surveys are also included . unfortunately , no information on the original spectrum was kept due to disk space issues . while this fact makes a direct comparison impossible , we can obtain an estimation of their spectral type by fitting the photometry to a library of templates , as detailed in 3.2 . while the advantages of these semi - analytic mock catalogues are crearly recognized , well - known issues have been widely reported in the literature related to unrealistic galaxy colours @xcite . many of these issues consequently lead to an overestimation of the mean dispersion of the photometric redshifts @xcite and flawed stellar mass estimations @xcite . some of these disagreements for the mock catalogues have been reported in @xcite and @xcite . these issues , together with the lack of photometric errors in the original mock catalogue motivated us to create a new mock catalogue that reproduces the properties of the observed galaxies as well as their photometric redshift precision in all the bands of j - pas . in order to do this , we apply ` photreal ` ( @xcite , bentez et al . in prep . ) . this procedure , already applied in several papers @xcite , ensures the accurate reproduction of the magnitudes , colours and photometric redshifts of the galaxies . in this section , we provide a brief summary of the method . for further details , we refer the reader to @xcite and bentez et al . in prep . ` photreal ` first obtains an estimate of the spectral type of the original catalogue by matching the original rest - frame mock photometry to a well - calibrated library of galaxy templates . this library includes eight different empirical templates representing a complete representation of the colors of any galaxy populations sampled by j - pas ( bentez , private communication ) . indeed , this library represents the observed properties of the alhambra and cosmos surveys with an outlier rate of @xmath24 1 - 2% @xcite . once we have a realistic representation of the spectral type distribution of the original mock catalogue , we generate photometry in the different j - pas bands by using their filter response and with their expected depths @xcite . we include an empirically calibrated systematic error of about 7% . this error remains constant with magnitude and seems to be intrinsic to the galaxy colors in multi - band photometry . furthermore , photometric and instrumental errors are added to these magnitudes . the photometric errors are estimated as in @xcite from the telescope response and are normalized to the j - pas depths in each band . thereafter , we run ` bpz2.0 ` on the new photometry to obtain photometric redshifts and redshift probability distribution functions , @xmath33 . one might claim that the use of the same library could be introducing an optimistic behavior of the performance of the photometric redshifts . however , we checked that this was not the case in a previous work @xcite . first of all , we ensured that the photometry generated with ` photreal ` was very similar to the one observed in the literature , which if any , it should make the derived photometric redshifts more realistic . secondly , we checked that the photometric redshifts derived from this photometry perfectly matched with the photometric dispersion and bias measurement in real data . for example , we measured the original photometric redshift dispersion of the mock catalogue generated for the advanced large , homogeneous area medium band redshift astronomical survey ( alhambra , @xcite ) data , finding it to be three times higher than the one expected from real data @xcite . after applying ` photreal ` , the photometric redshift dispersion exactly matched the data expectations @xcite . note that the choice of the library was motivated from an excellent calibration of the template library in representing @xmath698% of the known galaxies . moreover , possible interpolations between their templates are allowed when obtaining photometric redshifts . in this section , we describe the performance of the photometric redshifts estimated for j - pas obtained from the mock catalog described in section 3 . following @xcite , we define the photometric redshift dispersion as : @xmath34 where the difference @xmath35 is defined as the photometric redshift bias ( see @xcite and references herein ) and @xmath36 and @xmath37 refer to the bayesian photometric and the spectroscopic redshifts respectively . the overall photometric redshift dispersion obtained for the global j - pas mock sample is @xmath38 , equivalent to @xmath39 km s@xmath40 . [ fig : zszb ] shows the normalized density plot of the spectroscopic redshift versus photometric redshift for all the sample . it is noticeable the tightness of the relation as expected from the excellent spectrum coverage of j - pas . in addition , figs [ fig : deltazmag ] and [ fig : sigmazz ] show the density plot of the relation of the photometric redshift bias ( upper panel ) and dispersion ( bottom panel ) as a function of magnitude and redshift , respectively . we complement this information with the first two columns of tables [ tab : photozmag ] and [ tab : photozz ] , showing the photometric redshift precision and photometric redshift bias as a function of magnitude and redshift , respectively . the photometric redshift bias keeps well below 0.0015 up to @xmath41 mag and @xmath42 , at least . the photometric redshift dispersion remains below 0.003 up to @xmath43 mag and up to @xmath14 . this performance is markedly better than similar next - generation surveys in the same range of redshift and magnitude @xcite , as we discuss in 4.2 , putting in evidence the very good behavior of a pseudo - spectra - like survey as j - pas . we can argue that the utilized mock catalogues might be too simplistic and therefore unrealistic since they do not include other sources of errors except the photometric errors . however , as seen in previous analysis based on real data such as the alhambra survey @xcite , other sources of errors such as those coming , for instance , from the chosen library of templates , has a small impact on the results . therefore , we are confident that the present results in this work represent a realistic expectation of the capacity of the j - pas survey . versus spectroscopic redshift @xmath37 for the overall j - pas sample , colour - coded by normalized density . the photometric redshift dispersion is quoted . the j - pas photometric redshift will perform with a similar resolution to this obtained for low - resolution spectra up to redshift 1.2 . ] we have also characterized the rate of outliers expected in the survey . to do this , we have chosen to use two different definitions as in @xcite . first , we call a galaxy an outlier if it satisfies the following condition : @xmath44 hence , the rate of outliers , @xmath45 , is defined as : @xmath46 where @xmath47 refers to the total number of galaxies . we decided to use this definition since it has been the referred quantity in a large number of papers in the literature for historical reasons ( e.g. @xcite ) . however , as noted in @xcite , this definition becomes kind of arbitrary , particularly for surveys with a large - number of narrow - medium bands . hence , we also use the definition of an outlier galaxy , as that accomplishing the following condition : @xmath48 as already introduced in @xcite . as before , the rate of outliers , @xmath49 , is defined as @xmath50 the solid lines in figs . [ fig : outmag ] and [ fig : outz ] display the rate of outliers ( @xmath45 top panel , @xmath49 bottom panel ) expected for j - pas as a function of magnitude and redshift respectively . also , these values are collected in the third and fourth columns in tables [ tab : photozmag ] and [ tab : photozz ] , respectively . the rate of outliers becomes almost negligible ( @xmath511% down to i@xmath24 22.5 and @xmath523% down to [email protected] ) and it increases up to @xmath514% and up to @xmath5213.7% down to i@xmath2423 . this rate is slightly higher ( @xmath515% and @xmath5212.5% ) up to [email protected] , as expected from the inclusion of fainter low - redshift galaxies in this sample . these values are significantly ( @xmath245 - 20 times ) smaller than the values that other next - generation surveys will achieve @xcite , in agreement with what it is expected for surveys with a large number of narrow - band filter @xcite . a more detailed comparison is performed in 4.2 . ( top panel ) and @xmath49 ( bottom panel ) as a function of magnitude for three different j - pas samples : the overall sample ( solid line ) , the 50% best quality photometric redshift sample ( dotted line ) and the 25% best quality photometric redshift sample ( dashed line ) . the outliers rate is very small ( @xmath511% and @xmath526.5% ) down to @xmath53 , increasing to higher rates at fainter magnitudes for the overall sample , and for all the 50% and 25% best quality j - pas samples . ] ( top panel ) and @xmath49 ( bottom panel ) as a function of redshift for three different j - pas samples : the overall sample ( solid line ) , the 50% best quality photometric redshift sample ( dotted line ) and the 25% best quality photometric redshift sample ( dashed line ) . the @xmath45 outliers rate keeps below 5% for the overall sample ( @xmath54 ) down to @xmath55 and almost negligible ( @xmath45 and @xmath521% ) for the 50% and 25% best quality j - pas samples . ] complementarily , @xcite introduced an indicator of the quality of the photometric redshift of a survey with a parameter called _ odds_. this parameter is defined as the integral of the full redshift probability @xmath33 centered in the maximum peak of the probability within a given interval : @xmath56 and it becomes a very useful quantity to select the best quality photo - z samples . in fig . [ fig : zszbodd ] , we show the recovery of photometric redshifts for the j - pas survey for the 50% best quality _ odds _ sample , which have a mean dispersion of 0.0018 . we also display the rate of outliers for the 50% and 25% best _ odds _ samples as a function of magnitude and redshift in figs [ fig : outmag ] and [ fig : outz ] respectively . likewise , in tables [ tab : photozmag ] and [ tab : photozz ] , we list the photometric redshift precision , the photometric redshift bias and the rate of outliers , @xmath45 and @xmath49 , as a function of magnitude and redshift , respectively , for different _ odds _ cuts . versus spectroscopic redshift @xmath37 for the j - pas sample resulting after selecting the 50% best quality photometric redshifts in the sample , selected through the _ odds _ parameter ( see the text for an explanation ) , colour - coded by normalized density . the photometric redshift dispersion is quoted . this dispersion is comparable to that obtained for a low - resolution spectroscopic survey . ] while the global photometric redshift dispersion , @xmath57 remains below 0.003 down to @xmath58 , performing a selection of the 50% highest _ odds _ allows the mean photometric redshift dispersion to decrease down to 0.0015 or to even smaller values down to @xmath59 . in the same way , the overall outliers rate , @xmath45 ( @xmath49 ) , remains below 1% ( 6.3% ) for magnitude @xmath60 mag but it increases to @xmath64% ( @xmath614% ) at deeper magnitudes than this . an _ odds _ cut leaving 50% of the sample , automatically reduces the outliers rate to @xmath511% and @xmath522% down to @xmath61 and to @xmath510.25% and @xmath520.5% down to @xmath59 if we consider the 25% best quality _ odds _ galaxies in the sample . @xmath62 [ tab : photozmag ] similarly , the overall photometric redshift resolution remains below @xmath63 up to redshift 0.8 , increasing at higher redshifts . however , by selecting the 50% best quality _ odds _ sample , these values remain @xmath20.0015 up to redshift 1.0 . more strikingly , while the @xmath45 outlier rates range between 3% and 11% and @xmath49 between 8% and 23% for the whole redshift range , selecting the best 50% of the sample can decrease these assessments to less than 1% and 2% for @xmath45 and @xmath49 respectively . @xmath64 [ tab : photozz ] we compare our results with those shown in @xcite for the lsst and euclid surveys . the authors considered two euclid surveys consisting of the three infrared @xmath65 euclid bands and two different optical counterparts . the so - called euclid - pessimistic includes the five @xmath66 des optical bands as the optical counterpart and the euclid - optimistic includes the previously mentioned five optical des bands and the 6 @xmath67 lsst deep optical bands . for further details on these surveys , we refer the reader to @xcite . in order to be consistent in our comparison and to use the same definition of outlier , we only use @xmath45 in this subsection and refer to it as @xmath68 . in fig . [ fig : comparisonphotoz ] , we show the mean photometric redshift bias , the mean photometric redshift dispersion and the mean outlier rate ( top , middle and bottom panel , respectively ) as a function of the i - band magnitude ( for the j - pas and lsst surveys ) , h - band magnitude ( for the euclid surveys ) and redshift ( left , centre and right panel , respectively ) . we find that the photometric redshift bias as a function of the i - band magnitude obtained for j - pas is @xmath6100 times smaller than those found for the lsst . the values found for the euclid - optimistic as a function of the h - band magnitude are @xmath65 times smaller than those found for the euclid - pessimistic . comparing the photometric redshift bias as a function of redshift , the values found for j - pas are a factor 1 - 20 smaller than those found for the lsst and euclid - pessimistic and in the same range of magnitude as those found for the euclid - optimistic survey . the mean photometric redshift dispersion for the j - pas survey is more than a factor of 20 smaller than the lsst and euclid - pessimistic survey , and more than a factor of 10 smaller for the euclid - optimistic survey for similar magnitudes ranges . these differences remain similar , although slightly smaller , when we consider it as a function of redshift , up to redshift @xmath21 . at higher redshifts , the photometric redshift dispersion increases almost exponentially for the j - pas survey . finally , the rate of outliers for the j - pas survey as a function of magnitude is more than 15 times smaller than for the lsst and euclid - pessimistic and similar to the euclid - optimistic survey down to @xmath53 . however , the j - pas outliers rate becomes comparable to the lsst values ( @xmath24 3 - 5% ) as a function of redshift and higher than the ones expected for the euclid - optimistic survey . the results discussed in this section illustrate that the different observational strategies of the next - generations surveys provide different photometric redshift performances . for instance , deep ir surveys such as euclid will allow to reach high redshift regimes and very deep optical surveys , such as the lsst , will sample a wide range of the luminosity function , reaching very deep magnitudes . however , as shown in @xcite , although somewhat counterintuitive , medium and narrow band filter systems produce much more robust photometric redshifts , and there much larger photometric redshift depth , than broad - band systems reaching higher s / n . indeed , only the surveys that use the combination of multiple bands ( e.g j - pas or euclid - optimistic ) will reach very small bias and outliers rates . in addition , if these bands are narrow and cover the whole optical spectrum as j - pas will do , the photometric redshift dispersion will be reduced to very low - levels down to the depth limit of the survey . the bayesian cluster finder , ( bcf , @xcite ) is an optical / ir galaxy cluster detector , which was developed with the purpose of detecting galaxy clusters independently of the presence or absence of a red sequence or a central dominant brightest cluster galaxy ( bcg ) but using this information if it is present . in other words , the algorithm uses the presence of a red sequence or a dominant bcg if present , but it can still detect galaxy clusters or groups if this information is absent . in this section , we summarise the main details of the method and we refer the reader to the original work for further details . the bcf first calculates the probability at a given redshift that there is a cluster with a determined density and luminosity profile centered on each galaxy , including different priors related to the colour - magnitude relation of the cluster or the bcg magnitude - redshift relation . then , we perform a search in a predefined number of redshift slices , where the minimum threshold comes from the minimum redshift we can resolve ( usually determined from the geometry of the survey for small area surveys or 0 , otherwise ) and the maximum redshift is obtained from the wavelength coverage and the depth of the survey . the bin width is fixed according to the expected photometric resolution of the survey . for instance , we fix the photometric redshift resolution to 0.01 for the j - pas survey . effects of stars and masking of edges of the frames have been incorporated . then , clusters are selected as the density peaks of those probability maps and the center is located at the peak of the probability . finally , if we find two or more detections with separations less than 0.5 mpc in projected space and up to two bins in redshift space , we merge them into a single one . the bcf has been applied to a number of optical surveys in the literature : a wide ( 141 @xmath25 ) survey , the cfhtls - archive research survey ( cars , @xcite ; @xcite ) ; a very deep ( [email protected] mag depth ) survey , the deep lens survey ( dls , @xcite ; @xcite ) ; and a 20 medium - band survey , the alhambra survey ( @xcite ) . we remark different performances as a function of the different properties of the data . particularly , we see that multiple narrow - band surveys such as the alhambra survey are better at resolving the galactic population and therefore , at increasing the purity rates and setting a lower mass limit threshold for detecting clusters and groups . in this work , we apply the bcf to our j - pas mock catalogue . in order to assess the performance of the bcf on the j - pas mock data , we match the original mock sample to the recovered sample following the same friends - of - friends ( fof , @xcite ) algorithm described in @xcite . this procedure searches for each detection found in the recovered sample candidates or friends in the original mock catalogue whose centers are placed within a comoving distance of 3 mpc including the photometric redshift errors . then , a search of fof is done until no more candidates are found . afterwards , the candidate with the closest photometric redshift to the original detection is selected . finally , if this detection is found within a distance of 1 mpc , we consider this to be a match . we compute our observable richness , the total stellar mass , @xmath9 , defined as the sum of stellar mass of all the galaxies belonging to the clusters brighter than the magnitude limit within a certain radius @xcite . this observable has been chosen instead of the @xmath69 used in other work @xcite or @xmath70 used usually in red sequence based methods as it has been proved one the optical measurables with smaller intrinsic scatter with the halo mass ( e.g @xcite ) . in future work , we will explore the possibility of using the photometric redshift functions , @xmath33 , to obtain proxies for the kinematical mass of the cluster . in order to choose the optimal aperture to compute the total stellar mass for the data presented in this work , we chose six different apertures ranging from 0.5 to 1.5 mpc in steps of 0.25 mpc . we fit the @xmath9- @xmath71 relation following the procedure that it is described in section 5.3 and we computed the main scatter , @xmath72 , between these two variables . in fig . [ fig : radiusmstellar ] , we show the main measured scatter obtained for the different radius . the radius for which the minimum scatter is achieved is @xmath73 . therefore , we adopt this radius for computing the @xmath9 . , between the total stellar mass @xmath9 and the mass halo , @xmath71 , as a function of different considered radius to compute @xmath9 . the minimum scatter is obtained for a @xmath74 mpc for this data . ] we have also investigated which is photometric redshift performance of the bcf cluster finder for the j - pas clusters . in fig . [ fig : photozbcf ] , we show the density map of the cluster photometric redshift estimation versus the cluster spectroscopic redshift for the sample of clusters selected with j - pas . as we see the sequence is really tight , being the mean dispersion @xmath75 , comparable to the individual accuracy of photo - z s of the bright red galaxies . this result is comparable to the photometric redshift accuracy found in @xcite between optical and x - ray center up to @xmath76 with a negligible percentage of outliers ( @xmath25% with a normalized redshift difference @xmath77 ) and it is also significantly better than obtained by other works ( e.g. @xcite ) . this outcome illustrates the nice performance of the bcf on recovering the main redshift of the main structures detected with j - pas . furthermore , we have computed the completeness and purity of the results as a function of redshift and richness . we define completeness as the rate of clusters detected with the bcf out of the total simulated sample , and the purity as the rate of clusters simulated that were detected with the bcf out of the total detected sample . in fig . [ fig : drate ] , we show the completeness and purity rates as a function of redshift for different values of total stellar mass , @xmath9 . their equivalent halo mass @xmath71 bins have been obtained through the calibration described in 5.3 and are also quoted in the bottom panel . the results show that we will be able to detect galaxy clusters with completeness and purity rates @xmath78% for clusters and groups down to @xmath79 up to redshift 0.7 . at higher redshifts , the completeness rates decay , so that we can detect clusters down to @xmath80 up to redshift 0.85 with completeness rates higher than @xmath78% . if instead , we relaxed the completeness rates to be @xmath81% , the bcf would be able to detect clusters and groups down to @xmath82 up to redshift 0.7 ; down to @xmath79 up to redshift 0.8 and down to @xmath80 up to redshift 0.9 . in this work , we choose to work with the most conservative threshold of both completeness and purity rates @xmath680% . we then define the clusters selection function as the predicted minimum halo mass threshold for which we can detect galaxy clusters and groups with a completeness and purity rates @xmath83 as a function of redshift . for j - pas , the expected selection function is constant with @xmath84 up to z = 0.7 and progressively increases at higher redshifts ( see fig . [ fig : drate ] ) . ) and their equivalent total stellar mass ( @xmath9 ) bins . the plotted lines have been smoothed by linear ( top panel ) and fourth order polynomial ( bottom panel ) interpolation . while purity remains almost constant as a function of redshift , being lower for lower masses , we find a decreasing trend in the completeness rate with both redshift and mass . according to these rates , we expect to find reliably ( @xmath680% completeness and purity ) galaxy clusters and groups with total masses down to @xmath85 up to [email protected] and @xmath670% completeness rates down to @xmath86 up to the same redshift . ] modified fof algorithms have also been explored to detect galaxy clusters in j - pas - like narrow - band surveys @xcite , obtaining lower completeness and purity rates . this is expected since the performance of the fof algorithms decays for non - spectroscopic surveys due to their sensibility to the linking length . the results with the bcf suffer less contamination and therefore achieve higher completeness and purity rates for the same mass threshold . this result shows the benefit of using cluster finders that use other information either than the spatial to detect galaxy clusters . it is important to note that the extremely good quality of the photometric redshifts in the j - pas survey make these results comparable to what we would expect for a low - resolution spectroscopic survey . indeed , cluster and group searches in spectroscopic surveys such as the vimos vlt deep survey ( vvds , @xcite ) , where a search using voronoi - delaunay tessellation techniques was performed @xcite or the deep2 survey where a group catalogue was obtained based on the voronoi - delaunay technique @xcite and another one based on the fof algorithm @xcite found that they could obtain high purity and completeness rates for a similar threshold and with cluster velocity dispersion of @xmath87 , equivalent to @xmath88 @xcite . in order to illustrate the enormous benefits of using narrow - band surveys in terms of producing cluster catalogs , we show in fig . [ fig : selectionfnextg ] a comparison between the cluster selection function of different next - generation surveys using different observational techniques : x - ray ( erosita , @xcite ) , sunyaev - zeldovich ( actpol , @xcite and sptpol , @xcite ) and optical surveys ( des , @xcite ; lsst , @xcite and j - pas ) . all these functions , with the exception of the j - pas , have been extracted with ` dexter ` from @xcite . from this figure , we can clearly see that the selection functions of the optical surveys , while having very similar shapes , also show a large offset with respect to the j - pas . the knee of the curve is starting at z@xmath24 0.7 for the j - pas survey , whereas for des it happens at @xmath89 and for the lsst at @xmath90 . this behavior is related with the depth of the different surveys ( @xmath53 for j - pas , @xmath91 for des and @xmath92 for the lsst ) . the x - ray erosita selection function shows an increasing mass threshold as a function of redshift , obtaining similar mass - groups at low redshift as the j - pas . on the contrary , the cluster selection functions obtained from the sz cluster samples show a decreasing lower mass threshold as a function of redshift . . ] the impact of the previously shown j - pas selection function can be seen in fig . [ fig : nclustersnextg ] , where we plot the _ total number of clusters _ as a function of redshift that each survey will observe . as in fig . [ fig : selectionfnextg ] , the x - ray and sz curves have been taken from @xcite . according to this figure , the number of bound structures detected by j - pas will be comparable to those found by lsst and erosita at least , up to redshift @xmath240.7 and ten times superior to those found by des . while the latter surveys will sample with more number statistics the high - end of the mass function , j - pas will sample the mass function within a wider range of masses . complementarily , the des and lsst surveys will image a substantial part of the southern sky , whereas j - pas will provide an optical counterpart of many of the clusters / groups in common with erosita in the northern hemisphere up to z=0.7 and some of the most massive higher redshift clusters between @xmath93 . this will create an important synergy between the different next - generation surveys that will become very useful for a number of purposes , such as , for instance cosmological purposes . 0.7 , at least . ] the optical observable - dark matter halo mass is a crucial relation for cosmological purposes since it allows us to translate an optical measurement into a physical cluster mass ( e.g. @xcite and references herein ) . while several efforts have been invested in probing that optical cluster mass tracers can achieve accuracies similar to sz or x - ray tracers , so far it only has been probed up to moderate redshift @xcite or massive clusters @xcite . in this section , we empirically calibrate the total stellar mass observable - theoretical dark matter halo mass relation , @xmath94 , from the j - pas simulations . the fact that the observable used in this work , the total stellar mass , @xmath9 , is defined down to the flux limit where the survey is complete prevents us from introducing any bias up to the redshift limit where the survey is complete ( @xmath95 for j - pas ) . inspired by different works ( e.g. @xcite ) , we have model the @xmath94 relation with a log - log relation as follows : @xmath96 where @xmath97 refers to the decimal logarithmic , @xmath98 is the redshift of the cluster and @xmath99 are free parameters . we choose @xmath100 as a reasonable value that represents the expected cluster population . we have fit our data restricted to @xmath101 and @xmath102 to this model by using an iterative non - linear least - squares minimization method based on the levenberg - marquardt algorithm @xcite . we performed a monte carlo simulation sampling 8000 different initial values to compute the fit . the best fitting parameters for the model , together with their 68% confidence level are listed in table [ tab : fitmm ] . note that the results of this fit have been used to obtain the completeness and purity curves for different observable @xmath9 in 5.2 . @xmath103 [ tab : fitmm ] the @xmath94 relation appears not to evolve with redshift , in agreement with other works @xcite . in fig . [ fig : mhalolamda ] , we show the density plot of the relation between the total stellar mass parameter and the dark matter halo mass for different redshift bins . the solid line shows the fit for a particular redshift bin up to @xmath104 0.75 . the last redshift bin , 0.75@xmath105 is only shown to illustrate our our inability to measure correctly @xmath9 at this redshift range . while it becomes difficult to compare the values of @xmath106 and @xmath107 with different works due to the dependence of the definition of @xmath9 on the considered survey @xcite , or other observable used @xcite , we can compare the scatter of the relation . the main scatter found for j - pas is slightly smaller than this measured with the alhambra survey @xcite . this scatter also becomes comparable to this found by the sample of 52 local ( @xmath108 ) clusters and groups ranging a similar mass range @xcite . other works have found slightly smaller scatter for a sample of clusters ranging a similar range of redshift but five times more massive that in this work . for instance , @xcite find @xmath6 0.065 dex for very massive @xmath109 clusters and @xmath6 0.087 dex for @xmath110 clusters . the results in this work are encouraging since they show that we can extend previous findings to larger redshift ranges using an optical richness estimator . in the first three redshift bins . each panel shows the scatter measured for each different redshift bin . ] in parallel , we have estimated @xmath111 , the scatter in the dark matter halo mass at a fixed value of the @xmath9 . many authors have noticed the importance of measuring accurately this quantity in order to compute an observational cluster mass function from cluster counts @xcite . following a similar approach as in @xcite , we have performed 10000 monte carlo samplings of the possible halo mass values obtained directly from the simulation ( see fig . [ fig : mhalolamda ] ) to obtain a mean value and scatter for each fixed @xmath9 . in fig . [ fig : smassmhalo ] , we show this calibration for different redshift bins together with the main @xmath111 , obtained for each redshift bin . the mean @xmath111 value obtained is @xmath112 down to the mass limit of @xmath113 and @xmath114 down to the mass limit of @xmath115 . this value is somewhat smaller than the values found for the alhambra survey ( @xmath116 ) down to the same limit and comparable to values found by other authors for @xmath245 times higher mass limits . for instance , @xcite found @xmath114 at @xmath117 ( equivalent to @xmath118 ) in their calibration between @xmath119 and @xmath70 , where @xmath120 is obtained from x - ray and wl proxies and @xmath70 is the number of red galaxies lying within @xmath121 , the radius where the critical density is 200 times the mean density of the universe . similarly , @xcite find a scatter of @xmath122 for a sample of 53 local clusters between @xmath123 estimated from caustic or velocity dispersion measurements and @xmath70 . other authors have recently improved significantly this value for massive clusters . for instance , @xcite find @xmath124 at @xmath125 ( corresponding to @xmath126 ) between the sz mass estimate and the redmapper @xcite richness estimator , @xmath127 . comparing with simulations , similar values have also been found . for instance , @xcite find @xmath128 scatter from n - body simulations for the same m and @xmath70 variables down to @xmath113 and @xmath129 scatter from n - body simulations for the same variables down to @xmath115 . likewise , @xcite found slightly smaller scatter values ( @xmath130 ) from the millennium - xxl simulation between the cluster virial mass @xmath123 and the optical richness @xmath131 down to @xmath132 for a limited redshift sample ( [email protected] ) . while the simulations might result too simplistic to describe the real data , @xcite already showed that existing multiple narrow - band surveys , such as alhambra , are able to decrease substantially the scatter between different optical observables and the dark matter halo mass . hence , we present results that firmly support the fact that narrow - band surveys not only allow to detect clusters and groups down to smaller mass limits than broad - band surveys but they also enable to calibrate the observable - halo mass relation with an accuracy comparable to that obtained from broad - band survey for more massive clusters . the imminent start of the survey will confirm this point in the nearby future . 80% and @xmath670% respectively for each redshift bin . the average scatter @xmath111 , measured as the standard deviation between the two variables is also shown for each redshift bin . the last redshift bin is shown to illustrate the inability to fit the relation at [email protected] . ] in this work , we have first characterized the photometric redshift properties of the j - pas survey in terms of photometric redshift bias , photometric redshift dispersion and rate of catastrophic outliers using an n - body and semi - analytical simulations @xcite and a posterior modification with ` photreal ` ( @xcite , bentez et al . in prep ) . we have seen that the mean photometric redshift precision of j - pas is @xmath133 down to @xmath134 23.0 , is in agreement with what was expected from previous simulations @xcite . furthermore , the photometric redshift bias is fully consistent with zero down to the same magnitude limit and up to moderate redshift ( z@xmath24 0.7 ) without performing any preselection of the survey . the rate of outliers , @xmath45 , is always lower than 4% down to @xmath134 22.5 and at least within @xmath135 . in addition , we have compared the photometric redshift predictions for j - pas with similar predictions obtained for the lsst and euclid , using the same techniques to transform the same original mock catalogues @xcite . in this comparison , we conclude that the photometric redshift performance of j - pas will be outstanding in comparison with other next - generation surveys up to [email protected] at least . the photometric redshift dispersion becomes more than 20 times smaller than euclid+des or the lsst and more than 10 times smaller than euclid+des+lsst surveys together . complementarily , we have also explored the performance of the bayesian cluster finder ( bcf ) applied to our narrow - band next - generation j - pas survey . we have demonstrated with realistic simulations that we will be able to recover groups and masses down to @xmath136 up to redshifts 0.7 with completeness @xmath670% and purity higher than 80% and higher masses at @xmath137 . restricting completeness to be @xmath680% makes the minimum mass to be detected to be @xmath138 up to redshift 0.7 . we have compared these selection functions with other selection functions coming from different surveys in different wavelengths and we have concluded that j - pas will reach at least a factor of 2 lower mass threshold than other similar next - generation and present surveys such as the des and lsst , done with bigger telescopes ( 4 m and 8 m respectively ) . in addition , as the mass function will be sampled to lower mass limits , the absolute total number of detected clusters and groups will be comparable to those detected with the lsst . this is a very important result since the lsst will image more than twice the area of j - pas . additionally , since j - pas will cover a substantial part of the northern sky , whereas the des and lsst will focus on the southern hemisphere , the j - pas optical cluster sample will result in an exquisite sample to follow - up clusters detected with erosita , for instance . in addition , we have model the @xmath94 relation to a model in order to estimate the relevance of our cluster sample for cosmological purposes . we have considered a log - log normal model with a linear dependence with redshift . the results are compatible with a non - evolution of this relation with redshift . also , the main scatter obtained from the limited subsample of clusters down to @xmath139 and within @xmath140 is @xmath141 . these value is comparable to the results presented in other works limited to a local sample @xcite or very massive clusters @xcite . this results highlight then the enormous potential of j - pas for constraining cosmological parameters with galaxy clusters . finally , we have also looked into the precision with which we will be able to measure dark matter halo masses by using as observable the total stellar mass of the cluster . the results , based on simulations , suggest that we can recover galaxy clusters halo masses with an average scatter of @xmath142 down to @xmath143 . we note that this quantity becomes comparable to what other work found for samples of clusters five times more massive both in observations @xcite and in simulations @xcite when restricted to similar mass ranges . similarly , high accuracies were usually reached with other techniques such as wl @xcite , x - rays @xcite or cmb data @xcite . the impressive calibrations in measuring masses using large number of narrow - bands ( @xmath650 ) photometry provides a new technique that can reinforce the existing ones . a forthcoming paper ( ascaso et al . in prep ) will be devoted to investigate the impact of this selection function on the cosmological parameters , paying particular attention to the dark energy constraints . we thank the referee of this paper for his / her useful and insightful suggestions and comments that helped improving the original manuscript . ba acknowledges financial support for a postdoctoral fellowship from the observatory of paris . ec , gln , cmdo , am and ls acknowledge funding from papesp and cnpq . io acknowledges support from the european research council ( erc ) in the form of advanced grant , cosmicism . we acknowledge support from the spanish ministry for economy and competitiveness through grants ilink0862 and aya2013 - 48623-c02 - 1-p . ba also thanks the hospitality of the university of sao paulo and observatrio nacional for hosting her for a visit . ba dedicates this paper to the memory of the j - pas member javier gorosabel . adami , c. , durret , f. , benoist , c. , et al . 2010 , , 509 , a81 albrecht , a. , bernstein , g. , cahn , r. , et al . 2006 , arxiv : astro - ph/0609591 allen , s. w. , evrard , a. e. , & mantz , a. b. 2011 , , 49 , 409 andreon , s. 2010 , , 407 , 263 andreon , s. 2012 , , 548 , a83 andreon , s. , & berg , j. 2012 , , 547 , a117 andreon , s. , & congdon , p. 2014 , , 568 , a23 angulo , r. e. , springel , v. , white , s. d. m. , et al . 2012 , , 426 , 2046 arnalte - mur , p. , martnez , v. j. , norberg , p. , et al . 2014 , arxiv:1311.3280 ascaso , b. , wittman , d. , & bentez , n. 2012 , , 420 , 1167 ascaso , b. 2013 , highlights of spanish astrophysics vii , 115 ascaso , b. , wittman , d. , & dawson , w. 2014 , , 439 , 1980 ascaso , b. , bentez , n. , fernndez - soto , a. , et al . 2015 , , 452 , 549 ascaso , b. , mei , s. , & bentez , n. 2015 , mnras , in press , arxiv:1503.01113 austermann , j. e. , aird , k. a. , beall , j. a. , et al . 2012 , , 8452 bellagamba , f. , maturi , m. , hamana , t. , et al . 2011 , , 413 , 1145 bentez , n. 2000 , , 536 , 571 bentez , n. , gaztaaga , e. , miquel , r. , et al . 2009 , , 691 , 241 bentez , n. , moles , m. , aguerri , j. a. l. , et al . 2009 , , 692 , l5 benitez , n. , dupke , r. , moles , m. , et al . 2014 , arxiv:1403.5237 brammer , g. b. , van dokkum , p. g. , & coppi , p. 2008 , , 686 , 1503 cenarro , a. j. , moles , m. , cristbal - hornillos , d. , et al . 2013 , highlights of spanish astrophysics vii , 862 cenarro , a. j. , moles , m. , marn - franch , a. , et al . 2014 , , 9149 , 91491i cohn , j. d. , evrard , a. e. , white , m. , croton , d. , & ellingson , e. 2007 , , 382 , 1738 cole , s. , lacey , c. g. , baugh , c. m. , & frenk , c. s. 2000 , , 319 , 168 coupon , j. , ilbert , o. , kilbinger , m. , et al . 2009 , , 500 , 981 cucciati , o. , marinoni , c. , iovino , a. , et al . 2010 , , 520 , a42 eisenstein , d. j. , zehavi , i. , hogg , d. w. , et al . 2005 , , 633 , 560 erben , t. , hildebrandt , h. , lerchster , m. , et al . 2009 , , 493 , 1197 gerke , b. f. , newman , j. a. , davis , m. , et al . 2005 , , 625 , 6 gonzalez , a. h. , zaritsky , d. , & zabludoff , a. i. 2007 , , 666 , 147 hansson , k. s. a. , lisker , t. , & grebel , e. k. 2012 , , 427 , 2376 henriques , b. m. b. , white , s. d. m. , lemson , g. , et al . 2012 , , 421 , 2904 hilbert , s. , & white , s. d. m. 2010 , , 404 , 486 hildebrandt , h. , arnouts , s. , capak , p. , et al . 2010 , , 523 , a31 hildebrandt , h. , erben , t. , kuijken , k. , et al . 2012 , , 421 , 2355 huchra , j. p. , & geller , m. j. 1982 , , 257 , 423 huterer , d. , kirkby , d. , bean , r. , et al . 2015 , astroparticle physics , 63 , 23 ilbert , o. , arnouts , s. , mccracken , h. j. , et al . 2006 , , 457 , 841 ilbert , o. , capak , p. , salvato , m. , et al . 2009 , , 690 , 1236 ivezic , z. , tyson , j. a. , acosta , e. , et al . 2008 , arxiv:0805.2366 komatsu , e. , smith , k. m. , dunkley , j. , et al . 2011 , , 192 , 18 lagos , c. d. p. , lacey , c. g. , baugh , c. m. , bower , r. g. , & benson , a. j. 2011 , , 416 , 1566 laureijs , r. , amiaux , j. , arduini , s. , et al . 2011 , arxiv:1110.3193 le fvre , o. , vettolani , g. , garilli , b. , et al . 2005 , , 439 , 845 levi , m. , bebek , c. , beers , t. , et al . 2013 , arxiv:1308.0847 lima , m. , & hu , w. 2005 , , 72 , 043006 lin , y .- t . , mohr , j. j. , gonzalez , a. h. , & stanford , s. a. 2006 , , 650 , l99 liu , h. b. , hsieh , b. c. , ho , p. t. p. , lin , l. , & yan , r. 2008 , , 681 , 1046 lpez - sanjuan , c. , cenarro , a. j. , hernndez - monteagudo , c. , et al . 2014 , , 564 , a127 lsst science collaboration , abell , p. a. , allison , j. , et al . 2009 , arxiv:0912.0201 mantz , a. , allen , s. w. , rapetti , d. , & ebeling , h. 2010 , , 406 , 1759 marn - franch , a. , taylor , k. , santoro , f. g. , et al . 2015 , highlights of spanish astrophysics viii , 743 merloni , a. , predehl , p. , becker , w. , et al . 2012 , arxiv:1209.3114 merson , a. i. , baugh , c. m. , helly , j. c. , et al . 2013 , , 429 , 556 merson , a. i. , baugh , c. m. , abdalla , f. b. , et al . 2015 , arxiv:1502.04984 mitchell , p. d. , lacey , c. g. , baugh , c. m. , & cole , s. 2013 , , 435 , 87 moles , m. , bentez , n. , aguerri , j. a. l. , et al . 2008 , , 136 , 1325 molino , a. , bentez , n. , moles , m. , et al . 2014 , , 441 , 2891 mortonson , m. j. , weinberg , d. h. , & white , m. 2013 , arxiv:1401.0046 munari , e. , biviano , a. , borgani , s. , murante , g. , & fabjan , d. 2013 , , 430 , 2638 niemack , m. d. , ade , p. a. r. , aguirre , j. , et al . 2010 , , 7741 , oke , j. b. , & gunn , j. e. 1983 , , 266 , 713 olsen , l. f. , benoist , c. , cappi , a. , et al . 2007 , , 461 , 81 perlmutter , s. , aldering , g. , goldhaber , g. , et al . 1999 , , 517 , 565 planck collaboration , ade , p. a. r. , aghanim , n. , et al . 2011 , , 536 , a18 planck collaboration , ade , p. a. r. , aghanim , n. , et al . 2013 , arxiv:1303.5080 planck collaboration , ade , p. a. r. , aghanim , n. , et al . 2014 , , 571 , a29 planck collaboration , ade , p. a. r. , aghanim , n. , et al . 2015 , arxiv:1502.01597 postman , m. , lauer , t. r. , oegerle , w. , & donahue , m. 2002 , , 579 , 93 press , w. h. , teukolsky , s. a. , vetterling , w. t. , & flannery , b. p. 1992 , cambridge : university press , |c1992 , 2nd ed . , raichoor , a. , mei , s. , erben , t. , et al . 2014 , arxiv:1410.2276 rafelski , m. , teplitz , h. i. , gardner , j. p. , et al . 2015 , , 150 , 31 reid , b. a. , percival , w. j. , eisenstein , d. j. , et al . 2010 , , 404 , 60 riess , a. g. , filippenko , a. v. , challis , p. , et al . 1998 , , 116 , 1009 rozo , e. , rykoff , e. s. , evrard , a. , et al . 2009 , , 699 , 768 rozo , e. , wechsler , r. h. , rykoff , e. s. , et al . 2010 , , 708 , 645 rozo , e. , & rykoff , e. s. 2014 , , 783 , 80 rozo , e. , evrard , a. e. , rykoff , e. s. , & bartlett , j. g. 2014 , , 438 , 62 rykoff , e. s. , rozo , e. , busha , m. t. , et al . 2014 , , 785 , 104 snchez , a. g. , scccola , c. g. , ross , a. j. , et al . 2012 , , 425 , 415 saro , a. , bocquet , s. , rozo , e. , et al . 2015 , , 454 , 2305 skelton , r. e. , bell , e. f. , & somerville , r. s. 2012 , , 753 , 44 somerville , r. s. , gilmore , r. c. , primack , j. r. , & domnguez , a. 2012 , , 423 , 1992 spergel , d. n. , verde , l. , peiris , h. v. , et al . 2003 , , 148 , 175 springel , v. , white , s. d. m. , jenkins , a. , et al . 2005 , , 435 , 629 taylor , k. , marn - franch , a. , laporte , r. , et al . 2014 , journal of astronomical instrumentation , 3 , 1350010 the dark energy survey collaboration 2005 , arxiv : astro - ph/0510346 von der linden , a. , allen , m. t. , applegate , d. e. , et al . 2014 , , 439 , 2 weinberg , d. h. , mortonson , m. j. , eisenstein , d. j. , et al . 2013 , , 530 , 87 weinmann , s. m. , lisker , t. , guo , q. , meyer , h. t. , & janz , j. 2011 , , 416 , 1197 wittman , d. m. , et al . 2002 , , 4836 , 73 xavier , h. s. , abramo , l. r. , sako , m. , et al . 2014 , , 444 , 2313 zandivarez , a. , daz - gimnez , e. , mendes de oliveira , c. , et al . 2014 , , 561 , a71

we introduce the study of quantum protocols that probabilistically simulate quantum channels from a sender in the future to a receiver in the past . the maximum probability of simulation is determined by causality and depends on the amount and type ( classical or quantum ) of information that the channel can transmit . we illustrate this dependence in several examples , including ideal classical and quantum channels , measure - and - prepare channels , partial trace channels , and universal cloning channels . for the simulation of partial trace channels , we consider generalized teleportation protocols that take input copies of a pure state in the future and produce output copies of the same state in the past . in this case , we show that the maximum probability of successful teleportation increases with the number of input copies , a feature that was impossible in classical physics . in the limit of asymptotically large , the probability converges to the probability of simulation for an ideal classical channel . similar results are found for universal cloning channels from copies to approximate copies , exploiting a time - reversal duality between universal cloning and partial trace .

1112.1469

quantum theory is generally formulated relative to a given causal structure . this is the case both in quantum field theory , where the spacetime metric is given from the beginning , and in the operational framework of quantum information , where protocols and computations consist in sequences of operations performed at different times . relative to the given causal structure , quantum theory has to satisfy the _ causality principle _ @xcite , stating that the probability of a measurement outcome at a given time be independent of the choice of operations performed at later times . the causality principle forbids any form of signalling from the future to the past : a sender in the future can not deterministically transfer the state of his system to a receiver in the past . however , it is easy to imagine situations where the state of a system is transferred from the future to the past _ with some probability _ , without leading to signalling . this is what happens , for example , if we eliminate the communication of classical data in the original quantum teleportation protocol @xcite : the receiver in the past ( say , alice ) can prepare a maximally entangled state and the sender in the future ( say , bob ) can perform a bell measurement on the state to be teleported together with half of the entangled state , as in fig . [ fig : originaltele ] . ( color online ) probablilistic teleportation from the future to the past . today alice prepares a maximally entangled two - qubit state @xmath3 of the ancilla and the output system , while tomorrow bob performs a measurement on the ancilla and his input qubit . the measurement has four possible outcomes , for one of which ( here denoted by @xmath4 ) successful teleportation will occur . with probability @xmath5 alice obtains bob s input state today . however , she will not know whether teleportation has been successful until bob performs his measurement and communicates the outcome . ] in this way , upon observing the right measurement outcome , bob will know that the state has been teleported successfully to the past , and alice will not need to perform any correction operation on the output . of course , to know that teleportation has succeeded , bob needs to wait until he sees the outcome of the bell measurement . however , the important fact here is that , even before the measurement is performed , one half of the entangled state is already ready to be used by alice in a quantum circuit : alice can perform any desired computation on her system , and in the end , if teleportation is successful , bob will know that alice s computation has been performed on the input state he provided in the future . conditionally to the occurrence of the right outcome , everything behaves as though the input state has travelled from the future to the past through an ideal quantum channel . note that this fact is not in contradiction with causality : the probability of teleportation is small enough that no signal can be sent from bob to alice . we will refer to the above use of probabilistic teleportation as a _ probabilistic simulation of the identity channel from the future to the past_. the idea of probabilistic teleportation as a simulated time - travel was originally proposed by bennett and schumacher in an unpublished work @xcite , and , more recently , by svetlichny @xcite . in different terms , this idea also appeared implicitly in the context of the graphical language of categorical quantum mechanics @xcite , where the reinterpretation of probabilistic teleportation as information flow from the future to the past follows from a stretching of wires in the basic teleportation diagram . interestingly , similar ideas also appeared in the work of horowitz and maldacena @xcite in the context of black hole evaporation , as an attempt to reconcile the unitarity of the black hole @xmath6-matrix with hawking s semiclassical arguments ( see also the discussion by gottesmann and preskill @xcite ) . recently , the role of probabilistic teleportation in simulating closed timelike curves has been further explored by lloyd _ et al _ in refs . @xcite . in particular , ref . @xcite reports an experiment that uses probabilistic teleportation to simulate a quantum computation within a closed timelike curve . de silva , galvao and kashefi @xcite showed that some patterns in measurement - based quantum computation can be interpreted as deterministic simulations of closed timelike curves . all the works mentioned so far focused on the use of teleportation for the probabilistic simulation of an ideal quantum channel from the future to the past . however , there are many interesting scenarios where one needs to consider the simulation of more general quantum channels . for example , suppose that we have @xmath0 identical copies of a the same state and that we want to teleport just one copy to the past . does the probability of success increase with the number if input copies ? and , if it does , what is the the asymptotic value of the success probability in the limit @xmath7 ? to answer these questions we have to address the probabilistic simulation of channels that trace all systems but one . in this paper we will address the general problem of the probabilistic simulation of a given channel from the future to the past , showing how the causality principle determines the maximum probability of success . to find the maximum probability we will optimize over all generalized teleportation schemes where a bipartite state is prepared and the input of the channel is jointly measured along with half of the bipartite state , so that , for a particular outcome , the desired channel is simulated , as in fig . [ genchan ] . ( color online ) probabilistic simulation of a given quantum channel from the future to the past via a generalized teleportation scheme . the output system and an ancilla are first prepared in a bipartite state @xmath3 . when the input becomes available , the input state and the ancilla are measured jointly . for a particular measurement outcome @xmath4 the simulation will be successful : the effective transformation from input to output will be proportional to the desired quantum channel , with proportionality constant @xmath8 equal to the probability of successful simulation . ] the spirit of this work is similar to the spirit of the early works on the optimal cloning of non - orthogonal quantum states @xcite : in that case , one knew from the no - cloning theorem @xcite that it is impossible to produce perfect copies of the input state and the goal was to find the optimal physical process that approximates the impossible cloning transformation . in our case , we know from causality that it is impossible to have a channel that deterministically transfers information from the future to the past and our goal is to find the optimal process that achieves the desired channel with maximum probability of success . in the same way in which the study of optimal cloning shed light on the process of copying information in the quantum world , we expect that the study of optimal simulations of channels to the past will shed light the interplay between causal structure and quantum information flow . the main message of this paper is that the maximum probability of simulating a quantum channel from the future to the past is a decreasing function of the amount of information that the channel can transmit . this general feature will be illustrated in several examples . first , we will consider ideal classical channels ( those that perfectly transmit the states of an orthonormal basis ) and , second , ideal quantum channels ( those that perfectly transmit all states of a given quantum system ) . in the first case we find the maximum probability @xmath9 where @xmath10 is the dimension of the hilbert space , while in the second case we find @xmath11 which is exactly the probability of the outcome that does not require correction operations in the original teleportation protocol @xcite . we will then focus on the probabilistic simulation of measure - and - prepare channels , which are a class of channels that transmit only classical information . in this case , we will find that the maximum probability of success is at least equal to the probability @xmath12 for a classical channel with the same output hilbert space . the exact value of the probability of success will be computed in two relevant cases of measure - and - prepare channels : the channel associated to the optimal estimation of a pure state , and the universal not channel @xcite . finally , we will analyze in detail the case of trace channels , sending @xmath0 copies of a given pure state to @xmath1 copies of the same state , and the case of universal cloning channels @xcite , sending @xmath0 copies of a pure state to @xmath13 optimal approximate copies . in both cases we find that the maximum probability of success is given by @xmath14 where @xmath10 is the dimension of the one - particle hilbert space and the superscript in @xmath15 reminds that we are restricting ourselves to the symmetric subspaces , in which the @xmath0 input copies and the @xmath16 output copies of the given pure state live . in the particular case of @xmath17 eq . ( [ gennmbound ] ) yields the value @xmath18 which increases with the number of input copies , starting from @xmath19 and reaching the classical value @xmath20 in the asymptotic limit @xmath7 . this result has to be contrasted with the classical scenario , where having more input copies of a pure state can not lead to any improvement : since classical pure states can be perfectly cloned , there is no difference in having more input copies . the convergence of the probability of success to the classical value @xmath12 will be explained as a consequence of the convergence of the trace channel to a measure - and - prepare channel @xcite . concerning the probability @xmath21 in eq . ( [ gennmbound ] ) , it is also worth noting that it is symmetric in @xmath0 and @xmath16 : in other words , the probability of successful simulation for a trace channel from @xmath0 to @xmath22 copies is equal to the probability of successful simulation for a universal cloning channel from @xmath16 to @xmath23 copies . we will explain the symmetry of the probabilities as a consequence of a _ time - reversal duality _ between trace and universal cloning . as we anticipated , the message emerging from our results is that the maximum probability for a given channel is a decreasing function of the amount of information that the channel can transmit : the smallest value @xmath24 corresponds to the identity channel , while measure - and - prepare channels have probability @xmath25 , and the erasure channels @xmath26 ~ \rho_0 $ ] have probability @xmath27 . we will also make the connection more quantitative , providing a set of _ statistical information bounds _ , stating that the probability of success in the simulation of a channel is upper bounded by the inverse of the amount of information that the channel can transmit . the amount of information will be quantified here as the maximum payoff that two parties can achieve in a communication game . the structure of the paper is the following : in section [ sec : general ] we present the general method and , in particular , the _ causality bound _ , a necessary and sufficient condition for the existence of a probabilistic simulation with probability @xmath8 . then , we will analyze the probabilistic simulation of several channels : sections [ sec : ideal ] , [ sec : measprep ] , and [ sec : opttele ] will focus on ideal classical and quantum channels , on measure - and - prepare channels , and on partial trace channels from @xmath0 to @xmath28 copies , respectively . section [ sec : opttele ] will also discuss the case of universal cloning channels from @xmath0 to @xmath29 copies , exploiting the existence of a time - reversal duality between partial trace channels and cloning channels . finally , in section [ sec : summary ] we will derive the statistical information bound and use it to explain the asymptotic behaviour of the maximum probability for partial trace and cloning channels . the conclusions of the paper are drawn in section [ sec : conclusion ] . to find the maximum probability to simulate a given channel we will use formalism of the choi operators @xcite and the framework of quantum combs @xcite . in this section we present the basic notions used in the paper and we derive the _ causality bound _ , a necessary and sufficient condition for the simulation of a given channel with probability @xmath8 . any quantum circuit can be represented as a collection of preparations , measurements , transformations , and quantum wires along which states travel undisturbed . in this paper we only consider generalized teleportation circuits , which consist of one joint preparation of output and ancilla and one joint measurement on the input and ancilla , as in the circuit of fig . [ gentele ] . ( color online ) generalized teleportation circuit consisting of a joint state preparation and of a joint measurement . ] here the preparation is represented by a joint quantum state @xmath30 of the ancilla and the output system , and the quantum measurement is represented by a _ positive operator valued measure ( povm ) _ @xmath31 , namely a collection of positive operators @xmath32 acting on the hilbert space @xmath33 with the normalization property @xmath34 @xmath35 and @xmath36 representing the identity operators on the input hilbert space @xmath37 and on the ancilla hilbert space @xmath38 , respectively . each possible outcome @xmath39 corresponds to a different linear map @xmath40 , mapping density matrices on the input hilbert space @xmath41 to density matrices on the output hilbert space @xmath42 , and given by @xmath43 , \end{aligned}\ ] ] @xmath44 denoting the partial trace over the hilbert spaces @xmath37 and @xmath45 . by construction , each linear map @xmath46 is completely positive and trace non - increasing , namely it is a _ quantum operation _ note that , however , in this case the output of the quantum operation precedes the input in time , differing to the standard case originally considered by kraus @xcite . consider now the sum over all possible maps @xmath47 corresponding to different measurement outcomes . summing over all possible measurement outcomes is equivalent to ignoring the measurement outcome , thus obtaining the completely positive trace - preserving map ( also known as _ quantum channel _ ) @xmath48 given by @xmath49 \\ & = \sum_{i \in x } \tr_{in , a } [ \rho \otimes |\psi\>\<\psi| ] \\ & = \rho_0 ~ \tr_{in}[\rho ] \qquad \rho_0 : = \tr_a [ |\psi\>\<\psi|],\end{aligned}\ ] ] the third equality coming from the normalization of eq . ( [ eq : alloraaa ] ) . note that the trace - preserving map @xmath48 is an _ erasure channel _ : it always produces the same output state @xmath50 independently of the input . the fact that @xmath48 is an erasure channel is a manifestation of the causality principle stating that no signal can be sent from the future back to the past . in the following we will use the notation @xmath51 to mean that @xmath48 is the map defined by @xmath52 $ ] , for every quantum state @xmath53 on @xmath37 . since the linear maps @xmath54 are all completely positive , we must have @xmath55 where the inequality @xmath56 for two linear maps @xmath57 and @xmath58 means that the linear map @xmath59 is completely positive . we will refer to eq . ( [ causbound ] ) as the _ causality bound _ for a quantum operation @xmath46 in a generalized teleportation scheme . we can now focus on the a probabilistic simulation of a given quantum channel @xmath60 from the future to the past . in this case we want that for one particular measurement outcome @xmath61 the transformation @xmath62 is proportional to @xmath60 , namely @xmath63 . in this case , the proportionality constant @xmath8 represents the probability that the simulation succeeds . the maximum value of the success probability @xmath8 must be compatible with the causality bound @xmath64 we will refer to the above bound as the _ causality bound for the simulation of channel @xmath60_. an obvious consequence of the causality bound is the following : a channel @xmath60 can be simulated with probability @xmath65 in a generalized probabilistic teleportation protocol if and only if @xmath60 is an erasure channel , namely @xmath66 . * suppose that we have @xmath65 in the causality bound of eq . ( [ mapsinequality ] ) and define the quantum operation @xmath67 , so that we have @xmath68 . applying both sides of the equality to a generic state @xmath53 and taking the trace , we obtain @xmath69 + \tr[\map d(\rho ) ] = \tr[\rho]$ ] . since @xmath60 is a quantum channel , by definition @xmath70=\tr[\rho]$ ] . that is , @xmath71= 0 $ ] for every @xmath53 . hence , we conclude that @xmath72 , and , therefore @xmath73 . @xmath74 in the following section we will show that the validity of the causality bound is not only necessary , but also sufficient for the existence of a probabilistic simulation with probability @xmath8 . a very convenient tool to study quantum operations and channels is the choi correspondence , which associates linear operators on he hilbert space @xmath75 to linear maps sending operators on @xmath41 to operators on @xmath42 . the choi operator @xmath76 corresponding to a linear map @xmath77 is given by @xmath78 where @xmath79 indicates the unnormalized maximally entangled state @xmath80 , @xmath81 being an orthonormal basis for @xmath37 . we will often use the double ket " notation of ref.@xcite for bipartite states on a tensor product hilbert space @xmath82 : this notation is based on the one - to - one correspondence between operators @xmath83 from @xmath84 to @xmath85 and states @xmath86 in @xmath82 , given by @xmath87 where @xmath88 is an orthonormal basis for @xmath89 and @xmath90 is an orthonormal basis for @xmath84 . the double ket " notation has the useful properties @xmath91~,\end{aligned}\ ] ] for every pair of operators @xmath92 from @xmath84 to @xmath85 , and @xmath93 for every operator @xmath94 on @xmath85 , @xmath95 on @xmath84 , and @xmath83 from @xmath84 to @xmath85 . here latexmath:[$b^t : = \sum_{i , j } ~ \<\psi_i| b @xmath95 in the given basis @xmath90 . since the correspondence between linear maps and choi operators is one - to - one every property of a linear map corresponds to a property of the associated choi operator . in particular , we have choi s theorem , stating that a linear map @xmath97 is completely positive if and only if its choi operator @xmath76 is positive semi - definite @xcite . moreoever , a linear map is trace preserving if and only if its choi operator obeys @xmath98=i_{in}$ ] . let us now re - cast the causality bound of eq . ( [ mapsinequality ] ) in terms of choi operators . the choi operator of the erasure channel @xmath99 is given by @xmath100 denoting by @xmath83 the choi operator of the quantum channel @xmath101 , the causality bound of eq . ( [ mapsinequality ] ) becomes @xmath102 this operator form of the causality bound is much simpler to handle than the version with linear maps . in the following we will use eq . ( [ choiinequality ] ) to find the maximum probability @xmath8 for the simulation of a channel @xmath60 to the past . we now use the method of quantum combs @xcite to show that any value of @xmath8 compatible with the causality bound of eq . ( [ choiinequality ] ) can be achieved in a suitable generalized teleportation protocol . the key result that we use is theorem 4 of ref . @xcite , whose statement , specialized to the case considered in this paper , can be rephrased as [ lem : shortcomb ] let @xmath103 be a collection of positive operators on @xmath104 satisfying the property @xmath105 for some state @xmath50 on @xmath106 . then , there exists an ancillary system @xmath94 , a pure state @xmath107 , and a povm @xmath108 on @xmath109 such that , for every @xmath110 , @xmath111 is the choi operator of the quantum operation @xmath46 given by @xmath112 , \end{aligned}\ ] ] for every state @xmath53 on @xmath37 . a simple consequence of lemma [ lem : shortcomb ] is that every value of @xmath8 compatible with the causality bound is attainable in some generalized teleportation protocol : [ theo : existence ] let @xmath113 be the choi operator of the channel @xmath60 . if there exists a state @xmath50 on @xmath106 such that @xmath114 , then there exists a generalized teleportation scheme such that simulates @xmath60 with probability @xmath8 . * to prove the thesis it is enough to define @xmath115 and @xmath116 and to apply lemma [ lem : shortcomb ] to the collection of operators @xmath117 . the simulation of channel @xmath60 will succeed when the measurement outcome is @xmath118 . thanks to theorem [ theo : existence ] we can derive the maximum probability of successful simulation from the causality bound of eq . ( [ choiinequality ] ) : we will know automatically that there exists a generalized teleportation protocol succeeding with that maximum probability . in this section we will illustrate the general method of the causality bound in two simple examples . the first example is about the probabilistic simulation of an ideal classical channel , which preserves the elements of an orthonormal basis , but destroys all off - diagonal elements of the density matrix . in this case , we will show that the probability of success is given by @xmath119 where @xmath10 is the dimension of the hilbert space . the second example is about the probabilistic simulation of an ideal quantum channel , represented by the identity map . in this case , using the causality bound we will show that the probability of success is given by @xmath120 @xmath10 being again dimension of the hilbert space . this also proves that the original teleportation protocol @xcite is optimal for the probabilistic simulation of an ideal channel from the future to the past . [ subsect : idealclassical ] an ideal classical channel can be modelled in quantum theory as a von neumann measurement on an orthonormal basis , represented by the channel @xmath121 , where @xmath53 is the state of a @xmath10-dimensional quantum system . by the definition of choi operator in eq . ( [ correspondence ] ) , the choi operator of @xmath122 is given by @xmath123 inserting the above expression in the causality bound of eq . ( [ choiinequality ] ) , we obtain @xmath124 then , taking on both sides the expectation value on the state @xmath125 , we get @xmath126 , for every @xmath127 , which also implies @xmath128 . choosing @xmath129 we then achieve the maximum value @xmath130 a generalized teleportation protocol that simulates the ideal classical channel with maximum probability @xmath131 can be easily obtained by preparing the classically correlated state @xmath132 and by measuring the two - outcome povm @xmath133 given by @xmath134 and @xmath135 . another instructive application of the general method is the proof that the value of the probability of successful teleportation @xmath136 in the original teleportation protocol @xcite is the optimal value determined by causality . in the case of teleportation , the quantum channel that we want to simulate is just the identity : @xmath137 . by definition of choi operator ( [ correspondence ] ) , the choi operator of the identity channel is @xmath138 plugging this into the causality bound of eq . ( [ choiinequality ] ) , we then obtain @xmath139 to find the maximum value of the simulation probability @xmath8 we must ensure that the difference between the right hand side and the left hand side is a positive operator . in general , this is not a trivial issue because the operators on the two sides are not necessarily diagonalizable in the same basis . to resolve this problem , we notice a symmetry of the left hand side . defining the complex conjugate of an operator as @xmath140 and using eq . ( [ doubleketalg ] ) we have @xmath141 and , therefore , @xmath142 . applying the same transformation on both sides of the inequality ( [ originalineq ] ) , we obtain @xmath143 since the inequality must hold for every unitary @xmath144 , it also holds for the integral over the normalized haar measure @xmath145 , @xmath146 having used the schur s lemma for the last equality . now , both sides of the inequality are diagonalizable in the same basis . since the operator on the left hand side has eigenvalue @xmath147 and the operator on the right hand side has eigenvalue of @xmath148 , the inequality becomes @xmath149 . hence , the maximum probability of teleportation compatible with the causality bound is @xmath150 thus , we have shown from causality that the original teleportation protocol is the optimal probabilistic simulation of an identity channel from the future to the past . the probability of success for ideal channels provides a lower bound for the probability of success for arbitrary channels : indeed , if we have protocols that simulate ideal channels from the future to the past , then we can use these protocols to simulate a desired channel @xmath60 in two ways : _ i ) _ we can apply @xmath60 to the input in the future and then send the output to the past through the ideal channel , or _ ii ) _ we can send the input to the past through the ideal channel and apply the desired channel @xmath60 to the input in the past . this argument leads to the following : let @xmath60 be a quantum channel sending states on the hilbert space @xmath37 to states on the hilbert space @xmath106 . then , the maximum probability @xmath151 to simulate the action of @xmath60 from the future to the past satisfies the bound @xmath152 for classical channels , of the form @xmath153 and @xmath154 are orthonormal bases for @xmath37 and @xmath106 , respectively , the bound becomes @xmath155 in this section we focus on measure - and - prepare channels , namely channels @xmath156 of the form @xmath157 ~,\end{aligned}\ ] ] where @xmath158 is a povm on the input hilbert space @xmath37 , that is , @xmath159 and @xmath160 is a collection of normalized density matrices on the output hilbert space @xmath106 , that is , @xmath161 = 1 \quad \forall i \in x~.\end{aligned}\ ] ] measure - and - prepare channels are interesting as an example of channels that transmit only classical information : indeed , it is well known that the quantum capacity of a measure - and - prepare channel is always zero ( this follows , e.g. from the general upper bound presented by holevo and werner in ref . @xcite ) . in this section we will first show that the maximum probability to simulate a given measure - and - prepare channel from the future to the past satisfies the general lower bound @xmath162 where @xmath163 is the dimension of the output hilbert space . the bound is achieved for @xmath164 by von neumann measurements on an orthonormal basis , and in that case coincides with the bound for the simulation of an ideal classical channel ( cf . subsection [ subsect : idealclassical ] ) . ( [ mpbound ] ) is quite intuitive : if a channel can only transmit classical information , then the probability to simulate it is at least equal to the probability to simulate an ideal classical channel with the same output . then we will compute the exact value of the probability of success for two particular examples of measure - and - prepare channels : the first example will be the channel associated to the optimal estimation of a pure state , namely the linear map @xmath165 sending states on @xmath89 to states on @xmath89 given by @xmath166 \qquad & \rho_u : = u |0\>\<0| u^\dag\\ & p_u : = d ~ u |0\>\<0| u^\dag~,\end{aligned}\ ] ] where @xmath10 is the dimension of @xmath89 and @xmath167 is a fixed state . in this case the general bound is tight : the probability of success is given by @xmath168 the second example will consist in the _ universal not _ channel @xcite , namely the linear map @xmath169 sending states on @xmath89 to states on @xmath89 given by @xmath170 \qquad & \rho^{\perp}_u : = \frac{u ( i - |0\>\<0| ) u^\dag}{d-1 } ~~ \\ & p_u : = d ~ u |0\>\<0 | u^\dag ~~ \end{aligned}\ ] ] in this case we will show that the probability of success is given by @xmath171 well above the lower bound @xmath172 . the intuition that the probability of success depends on the amount of information that a channel can transmit suggests that the universal not channel can transmit less classical information than the state estimation channel . the choi operator of the generic measure - and - prepare channel @xmath173 in eq . ( [ eq : mp ] ) is given by @xmath174 let us consider the causality bound @xmath175 , where @xmath50 is a state on @xmath106 to be optimized . a ( possibly suboptimal ) choice for the state @xmath50 is @xmath176 , for which the bound becomes @xmath177 the bound will be satisfied as long as we choose @xmath178 , where @xmath179 is the largest eigenvalue of the choi operator @xmath180 . we now show that @xmath181 can not be larger than @xmath182 . to this purpose , we consider an arbitrary normalized state @xmath183 and show that @xmath184 . using eq . ( [ doubleketprod ] ) , the normalization of the state @xmath185 reads @xmath186=1 . \end{aligned}\ ] ] we then obtain @xmath187.\\ & = \sum'_i \tr [ \psi p_i \psi^{\dagger } ] \tr [ \rho_i \sigma_i ] \quad \sigma_i : = \frac { \psi p_i \psi^{\dagger } } { \tr[\psi p_i \psi^{\dagger}]}\\ & \le \sum_{i \in x } \tr [ \psi p_i \psi^{\dagger } ] \\ & = \tr[\psi^\dag \psi]\\ & = 1.\end{aligned}\ ] ] here we used eq . ( [ doubleketalg ] ) for the first equality and eq . ( [ doubleketprod ] ) for the second . the sum in the third equation is restricted to the values of @xmath188 such that @xmath189 \not = 0 $ ] . the inequality is due to the fact that the overlap between the density matrices @xmath190 and @xmath191 is bounded by @xmath182 , the fourth equality is due to the normalization condition @xmath192 , and the last equality is just eq . ( [ psinorm ] ) . since the bound @xmath193 holds for arbitrary @xmath194 , we conclude that the maximum eigenvalue of @xmath195 satisfies @xmath196 . this implies that the value @xmath197 is compatible with the causality bound , and therefore , it is achievable with a suitable generalized teleportation protocol . hence , for the maximum probability @xmath198 we have the lower bound @xmath199 . the choi operator for the pure state estimation channel is @xmath200 clearly , @xmath201 enjoys the symmetry @xmath202 for every @xmath203 , which implies that the causality bound has the form @xmath204 ( the proof is the same as the proof for the ideal quantum channel in subsection [ sec : original ] ) . since the maximum eigenvalue of @xmath201 is @xmath205 , we have that the maximum probability is @xmath206 . using eqs . ( [ cest ] ) and ( [ unot ] ) , the universal not channel can be written as @xmath207 hence , its choi operator is given by @xmath208 again , we observe that @xmath209 enjoys the symmetry @xmath210 for every @xmath203 , which implies that the causality bound has the form @xmath211 now , the maximum eigenvalue of @xmath209 is @xmath212 . hence , the maximum probability is given by @xmath213 . in this section we consider probabilistic protocols that take @xmath0 input copies of a pure state in the future and produce @xmath1 copies of the same state in the past . in this case , the channel that we want to simulate is the partial trace over @xmath214 systems , symmetrized over all possible choices of the systems that are traced : @xmath215 , \end{aligned}\ ] ] here the partial trace is over all input hilbert spaces except the first @xmath16 , @xmath216 is an element of the group @xmath217 of the permutations of @xmath0 objects and @xmath218 is the unitary operator that permutes the @xmath0 input hilbert spaces according to the permutation @xmath216 . the goal will be to see how the probability of success , denoted by @xmath219 , varies with the number of input and output copies . in particular , we will show that the probability of success increases with the number of available input copies . note that in the classical world there can not be any increase of the probability of success with the number of copies of an pure input state : since we can perfectly clone the classical pure states corresponding to an orthonormal basis , we must have @xmath220 the classical value @xmath221 is clearly an upper bound for the probability of success in the quantum case : indeed , if we have a probabilistic teleportation protocol that succeeds with probability @xmath15 in transforming @xmath0 copies of a pure quantum state into @xmath16 copies of the same quantum state , then the protocol will work in particular on the states of an orthonormal basis , thus simulating the classical trace channel from @xmath0 to @xmath16 copies . hence , we must have @xmath222 this bound is very intuitive : if there existed a quantum protocol that succeeded with probability @xmath223 , then two parties , alice and bob , could use the protocol to reliably win a lottery where the goal is to guess today a winning number @xmath224 between 1 and @xmath10 that will be extracted tomorrow . to win the lottery , alice and bob can agree to use an orthonormal basis @xmath225 such that the basis vector @xmath226 corresponds to the number @xmath224 . tomorrow , when the winning number is announced , bob inputs into the teleportation machine @xmath0 identical systems , each of which is prepared in the state @xmath226 corresponding to the winning number . today alice measures one of the @xmath16 copies in the agreed basis and with probability @xmath15 she obtains the winning number , just in time to submit her lottery ticket . if @xmath227 were larger than the classical value @xmath228 , then alice would win the lottery with probability larger than the probability @xmath229 to win the lottery by a random guess . in the following we will show that , unlike in the classical case , in the quantum case the probability of successful teleportation increases with the number of input copies . first , we will consider the simplest case @xmath230 , where the calculation of the maximum probability of success can be done explicitly even without constraining the input copies to be in a pure state . in this case , we find out that the probability increases from the value @xmath231 in the @xmath232 case to the value @xmath233 . this result suggests that the quantum improvement is a general feature that does not depend on the assumption of pure input states . we will then move to the case of general @xmath0 , keeping the number of output copies fixed to @xmath17 . in this case , we will that the success probability approaches the classical value @xmath234 as @xmath235 . an explanation of this asymptotic behaviour will be provided in sec . [ sec : summary ] . finally , we will extend the analysis to general @xmath16 , also including the case @xmath236 , where optimal clones of the input state are teleported to the past . surprisingly , in the case of cloning channels we will show that for fixed @xmath0 the probability of success is an increasing function of the number of output copies @xmath16 . also this fact will be explained in sec . [ sec : summary ] . for two copies of input and one copy of output , the input hilbert space is @xmath237 and the output hilbert space is @xmath238 . in this case , the quantum channel to be simulated is @xmath239 where the labels 1 and 2 label on the right hand side label the two input spaces . using eq . ( [ correspondence ] ) , we find that the choi operator of this channel is given by @xmath240 where we have used 0 to label the output hilbert space . plugging this into the causality bound of eq . ( [ choiinequality ] ) we obtain @xmath241 like in subsection [ sec : original ] , we can notice a symmetry of the choi operator : @xmath242 for every @xmath243 . applying this transformation to both sides of the inequality ( [ oramuoio ] ) and integrating over the haar measure we obtain @xmath244 it now remains to find the eigenvalues of the choi operator @xmath245 . to do this , we define the following set of orthonormal vectors in @xmath246 : @xmath247 in terms of these vectors , the choi operator can be rewritten as @xmath248 thus we see that the eigenvalues of this choi operator are @xmath249 and @xmath250 . to ensure that eq . ( [ oramuoio ] ) holds , we must have @xmath251 thus , the maximum teleportation probability from 2 copies of input is @xmath252 which is greater than the maximum probability of teleportation from 1 to 1 copy , @xmath253 . in other words , adding an extra copy of the input yields improvement in the teleportation probability . it is worth stressing that , since we are simulating the symmetrized trace channel @xmath254 , any bipartite state @xmath255 on @xmath37 with the property @xmath256 + \tr_{2 } [ \sigma ] ) /2 = \rho $ ] will be transformed with probability @xmath257 into @xmath53 . this holds in particular when @xmath258 , corresponding to two identical copies of the mixed state @xmath53 . we now calculate the maximum probability of teleportation from @xmath0 input copies to one output copy . in this case , we will only require the teleportation protocol to work perfectly when we are inputting @xmath0 copies of the same pure state . since the density matrix of @xmath0 identical copies of a pure state has support in the symmetric subspace of @xmath260 , henceforth denoted as @xmath261 , we will focus on the restriction of the trace channel to the symmetric subspace , namely on the channel @xmath262 given by @xmath263 where @xmath264 is the projector on @xmath261 and @xmath265 is the symmetrized partial trace channel defined in eq . ( [ symmtrace ] ) . note that the map @xmath262 is trace - preserving only for states with support in @xmath261 . it is useful to observe that , due to the projection on the symmetric subspace , the sum over all permutations in eq . ( [ symmtrace ] ) can be omitted : indeed , we have @xmath266 \\ & \qquad = \tr_{n } \dots \tr_{2 } \left [ p_+^{(n ) } \rho p_{+}^{(n ) } \right ] , \end{aligned}\ ] ] having used the property @xmath267 for every permutation @xmath268 . ( color online ) probabilistic teleportation from @xmath0 copies to @xmath17 copy of a pure state . here , @xmath101 represents the symmetrized partial trace over all but one input copy , restricted to the symmetric subspace . ] by the definition of choi operator in eq . ( [ correspondence ] ) we find that the choi operator @xmath269 is given by @xmath270 where the output hilbert space is labelled by @xmath118 , the input hilbert spaces are labelled with numbers from @xmath182 to @xmath0 , and the projector @xmath271 acts on the tensor product of all input hilbert spaces except the first . we want to find the maximum value @xmath272 compatible with the causality bound of eq . ( [ choiinequality ] ) , which now reads @xmath273 to this purpose , we notice that the choi operator @xmath269 has the symmetry @xmath274 for every @xmath275 . following again the group theoretic argument of subsection [ sec : original ] we can rewrite eq . ( [ eq : causaln1 ] ) as @xmath276 hence , it only remains to find the eigenvalues of @xmath269 . expanding the projector @xmath277 as @xmath278 where @xmath279 is an orthonormal basis for @xmath280 , we can express @xmath281 as @xmath282 : indeed , we have @xmath283 |\varphi_m\ > \\ & = \frac{d^{(n)}_+}{d^{(n-1)}_+ } ~ \ < \varphi_n| p^{(n-1)}_{+ } |\varphi_m\>\\ & = \frac{d^{(n)}_+}{d^{(n-1)}_+ } ~ \delta_{n , m } , \end{aligned}\ ] ] that is , the vectors @xmath284 are mutually orthogonal and have the same norm . therefore , @xmath269 only has one non - zero eigenvalue , given by @xmath285 . plugging this value into the causality bound of eq . we obtain @xmath286 . the maximum probability of success is then given by @xmath287 note that the probability of successful teleportation increases with the number @xmath0 of input copies , unlike in the classical case . for @xmath232 we retrieve the value @xmath288 of the standard teleportation protocol , while in the limit of @xmath0 going to infinity we observe that @xmath272 tends towards the classical limit @xmath289 . such an asymptotic behaviour will be explained in section [ sec : summary ] . here we calculate the maximum probability of teleportation from @xmath0 input copies to @xmath1 output copies of a generic pure state . again , since for every integer number @xmath290 the density matrix of @xmath290 identical copies of a pure state has support in the symmetric subspace @xmath291 , we will restrict the input of the partial trace channel @xmath292 to be in @xmath293 and the output to be in @xmath294 . in other words , we will focus on the probabilistic simulation of the channel @xmath295 given by @xmath296 p^{(m)}_+ ~ , \end{aligned}\ ] ] where @xmath297 is the symmetrized partial trace channel defined in eq . ( [ symmtrace ] ) . again , thanks to the projection on the symmetric subspace we can write @xmath298 \right ) p^{(m)}_+ ~. \end{aligned}\ ] ] ( color online ) probabilistic teleportation from with @xmath0 input copies in the future to @xmath1 output copies in the past . here , @xmath101 represents the symmetrized partial trace over all but @xmath16 of the @xmath0 input systems , where the input and output are constrained to be in the symmetric subspace . ] by the definition of choi operator in eq . ( [ correspondence ] ) , we then obtain @xmath299 where the projector @xmath300 acts on the tensor product of the last @xmath301 input hilbert spaces . our goal is to find the maximum probability of success compatible with the causality bound @xmath302 where @xmath303 is a suitable state on @xmath304 and @xmath305 is the identity on @xmath306 . now , it is easy to see that the choi operator @xmath307 has the symmetry @xmath308 for every @xmath243 . due to this symmetry , the state @xmath303 in the causality bound can be chosen without loss of generality to be the invariant state @xmath309 , so that the bound becomes @xmath310 to find the maximum probability it is enough to find the maximum eigenvalue of @xmath311 . expanding the projector @xmath312 as @xmath313 for some orthonormal basis @xmath314 for @xmath315 , we can rewrite eq . ( [ cnmlast ] ) as @xmath316 having defined @xmath317 and @xmath318 : indeed , we have @xmath319 & = \frac{d^{(n)}_+}{d^{(n - m)}_+ } ~ \ < \varphi_m| p^{(n - m)}_{+ } |\varphi_n\ > \\ & = \frac{d^{(n)}_+}{d^{(n-1)}_+ } ~ \delta_{mn } , \end{aligned}\ ] ] that is , the vectors @xmath320 are mutually orthogonal and have the same norm . therefore , @xmath321 only has one non - zero eigenvalue , given by @xmath322 . the causality bound of eq . ( [ boundnm ] ) is equivalent to @xmath323 , which implies that the maximum probability is given by @xmath324 note that @xmath325 converges to the value @xmath326 in the limit of @xmath0 going to infinity . such an asymptotic behaviour will be explained in section [ sec : summary ] . we now generalize the results of the previous subsection to the case @xmath236 . of course , in this case we can not expect to obtain @xmath16 perfect output copies of the input state , because that would violate the no - cloning theorem@xcite . what we can do is to consider the best possible approximation of the ideal cloning process . such an optimal approximation is given by the optimal _ universal cloning _ channel @xcite , here denoted by @xmath327 for consistency of notation . the cloning channel @xmath327 is given by @xmath328 note that the linear map @xmath295 is trace - preserving only for density matrices @xmath53 with support within the symmetric subspace @xmath329 . indeed , @xmath295 should be regarded as a quantum channel sending states on @xmath330 to states on @xmath331 . from eqs . ( [ cnm ] ) and ( [ clonnm ] ) , we can note the existence of an important duality between universal cloning and partial trace : for two arbitrary operators @xmath332 on @xmath333 and @xmath334 on @xmath335 one has @xmath336 = \frac { d_+^{(m ) } } { d_+^{(n ) } } ~ \tr [ \map c^+_{q , m \to n } ( \mu ) ~ \nu ] . \end{aligned}\ ] ] in other words , the map representing the universal cloning from @xmath0 to @xmath16 copies in the heisenberg picture is the map representing the partial trace from @xmath16 to @xmath0 copies in the schrdinger picture , up to a positive scaling factor . we will refer to this property as the _ time - reversal duality _ between partial trace and universal cloning . in terms of choi operators , the duality can be written as @xmath337 where @xmath338 is the operator that exchanges the @xmath0 output hilbert spaces with the @xmath16 input hilbert spaces for the channel @xmath339 . ( [ clontracechoiduality ] ) allows us to easily find the probability of success for the cloning channel @xmath295 . first of all , eq . ( [ clontracechoiduality ] ) implies that @xmath340 has the symmetry @xmath341 for every @xmath342 . using this symmetry the causality bound becomes @xmath343 now , eq . ( [ clontracechoiduality ] ) implies that the above equation is equivalent to @xmath344 which is nothing but the causality bound for the trace channel @xmath345 [ cf . ( [ boundnm ] ) ] . therefore , we have that the causality bound for the cloning channel @xmath346 is satisfied with probability @xmath8 if and only if the causality bound for the trace channel @xmath347 is satisfied with probability @xmath348 . hence , we have @xmath349 quite surprisingly , the above value increases with the number @xmath16 of output copies , reaching the asymptotic value @xmath350 in the limit @xmath351 . this asymptotic behaviour will be explained in section [ sec : summary ] . the results obtained so far indicate a dependence of the maximum probability of success on the amount and type ( quantum or classical ) of information that the channel can transmit : first , we noted that the only channels that can be simulated with unit probability are the erasure channels , that produce a fixed output state independently of the input , that is , they do not transmit any information . then , we saw that measure - and - prepare channels , that only transmit classical information , can be simulated with probability @xmath352 , @xmath163 being the dimension of the output hilbert space . finally , channels that transmit both quantum and classical information can be simulated with probability @xmath353 , as in eq . ( [ generallower ] ) in the following we will make a quantitative connection between the amount of information transmitted by a channel and the probability of simulating it with a generalized teleportation protocol : we will show that the probability of success is upper bounded by the inverse of a measure of information defined as the maximum payoff in a communication game . we will then use this result to explain the asymptotic behaviour of the probability of success for the simulation of a partial trace channel with a large number of input systems and of a universal cloning channel with a large number of output systems . in this subsection we relate probability of success in the simulation of a channel to the past with the amount of information transmitted by the channel . in this context , a convenient measure of information is the expected payoff obtained by two parties in a communication game . suppose that the channel @xmath60 is used to transfer states from alice s side to states on bob s side . alice can encode a set of messages @xmath354 in a set of states @xmath355 on @xmath356 and communicate a message to bob by sending the corresponding state through the channel @xmath60 . on his side , bob can try to decode alice s message from the output state @xmath357 . generally , the decoding will be described by a povm @xmath358 on the output hilbert space @xmath106 , with the outcome @xmath359 corresponding to the decoded message . usually , but not necessarily , the set of decoded messages @xmath354 coincides with the set of encoded messages @xmath360 . to assess the success of the communication , we can introduce a payoff function @xmath361 , that quantifies the gain ( or the loss ) associated to decoding @xmath362 while the encoded message was @xmath188 . for given @xmath188 , the expected payoff will be @xmath363 . \end{aligned}\ ] ] now , suppose that alice is in the future and bob is in the past , and that they are using a probabilistic simulation of channel @xmath60 . the causality bound of eq . ( [ mapsinequality ] ) imposes @xmath364 , and , therefore @xmath365 having defined @xmath366 $ ] . introducing a prior probability distribution @xmath367 on alice s messages we can then consider the average expected payoff @xmath368 hence , the bound of eq . ( [ fidboundi ] ) becomes @xmath369 note that the value @xmath370 depends only on the payoff function @xmath371 and on the prior probabilities @xmath367 . we have now arrived at the bound @xmath372 which holds for every given payoff function @xmath371 , prior probabilites @xmath373 , encoding @xmath374 , and decoding @xmath375 . maximizing the average payoff over all possible choices of encoding @xmath376 and decoding @xmath377 we then get the maximum value @xmath378 which quantifies the maximum payoff that can be achieved when using the channel @xmath60 in a communication game with given payoff function @xmath371 and prior probabilities @xmath379 . we then conclude that the maximum probability to simulate the channel @xmath60 , denoted by @xmath151 , has to satisfy the bound @xmath380 for every payoff function @xmath371 and for every prior @xmath379 . the above bound states that the probability of success can not be greater than the inverse of the maximum amount of statistical information ( average payoff ) transmitted by the channel . for this reason , we refer to this bound as the _ statistical information bound_. notice that the derivation of the information bound in eq . ( [ infobound ] ) can be extended to the case where alice and bob are allowed to communicate through an ideal quantum channel on a reference system with hilbert space @xmath381 , in addition to the channel @xmath60 . in this case , alice can encode her message @xmath382 in a bipartite state @xmath383 on @xmath384 . later , she will send the input system through the channel @xmath60 and the reference through the ideal channel . bob s task will then be to decode the message from the output state @xmath385 , through a bipartite povm @xmath386 on @xmath387 . following the same steps as above , in this case we obtain the _ entanglement - assisted statistical information bound _ @xmath388 having defined @xmath389 \right\}\\ \mathbb e^{(r)}_{ave } ( f ) : = \sum_{i\in x } \pi_i \left\ { \sum_{j\in y } f(i , j ) ~ \tr\left [ \rho_0 \otimes \tr_{in}[\rho_i ] ) p_j \right ] \right\}.\end{aligned}\ ] ] note , however , that both statistical information bounds are just upper bounds : in general there is no guarantee about their achievability . the aim of the following discussion is to provide an explanation for the asymptotic behaviour of the probability of success in the simulation of the partial trace channel @xmath391 in the limit @xmath7 . first , we apply the statistical information bound of eq . ( [ infobound ] ) to show that the probability of success satisfies the bound @xmath392 consider a communication scenario where alice encodes a message in state from the set @xmath393 and sends it to bob through the partial trace channel @xmath394 . suppose that bob s decoding is given by the povm @xmath395 where @xmath396 is the normalized invariant measure on the pure states of @xmath89 . as a payoff function , let us assume the dirac delta @xmath397 , so that the expected payoff for input @xmath398 is just the probability ( density ) to make the correct guess @xmath399 p^{(m)}_\psi \right\ } \\ & = \tr [ \left ( |\varphi\>\<\varphi| \right)^{\otimes m } p^{(m)}_\varphi ] \\ & = d_+^{(m ) } \qquad \forall |\varphi\ > \in { \mathcal{h}}. \end{aligned}\ ] ] since the expected payoff is the same for every state , we have @xmath400 for every prior distribution . on the other hand , choosing the uniform distribution @xmath401 for the input states , we have @xmath402 \\ & = 1 . \end{aligned}\ ] ] the statistical information bound of eq . ( [ infobound ] ) then gives @xmath403 . in subsection [ subsect : tracenm ] we observed that the probability of success @xmath15 asymptotically approaches the value @xmath404 in the limit @xmath7 . an explanation of this behaviour is the following : from ref . @xcite , we know that in the limit @xmath7 the partial trace channel @xmath405 converges to the estimation channel @xmath406 defined by @xmath407 . \end{aligned}\ ] ] with @xmath408 . hence , the probability of success @xmath15 must converge to @xmath409 , the probability of success for the estimation channel @xmath410 . on the other , hand , since @xmath411 is a measure - and - prepare channel , by the general bound of eq . ( [ mpbound ] ) we must have @xmath412 . in conclusion , we obtained @xmath413 that is , @xmath414 . in other words , the asymptotic behaviour of @xmath15 is dictated by the statistical information bound and by the convergence of the trace channel @xmath415 to a measure - and - prepare channel . an illustration of the situation for the case of @xmath17 is given in fig . [ fig : channelsgraph ] . it is worth stressing , once more , that no improvement with the number of input copies would be possible for a classical partial trace channel : since in that case the probability is already @xmath289 , any improvement would violate the causality bound for ideal classical channels . hence , the increase of the probability of teleportation with the number @xmath0 of input copies is intrinsically a quantum phenomenon . a graph of the maximum probability @xmath8 to simulate the partial trace channel from @xmath0 copies to @xmath17 copy of a pure input state . the maximum probabilities for erasure channels , measure - and - prepare channels , and ideal classical and quantum channels are also located in the plot . ] here we consider the universal cloning channel @xmath405 . as we did for the trace channels , we will now provide an explanation for the asymptotic behaviour of the probability of success in the limit @xmath351 . a direct explanation comes from the time - reversal duality between cloning and partial trace , which implies the relation @xmath417 and thus allows one to reduce the asymptotic behaviour of the probability of success of a cloning channel with many output copies to the asymptotic behaviour of the probability of success for a trace channel with many input copies . a less direct , but perhaps more insightful explanation can be obtained by following the same steps used in the previous subsection . let us start by showing that the statistical information bound of eq . ( [ infobound ] ) implies that the probability of success satisfies the bound @xmath418 consider a communication scenario where alice encodes a message in state from the set @xmath393 and sends it to bob through the cloning channel @xmath394 . suppose that bob s decoding is given by the povm @xmath419 , where @xmath396 is the normalized invariant measure on the pure states of @xmath89 . as a payoff function we assume again the dirac delta @xmath420 , so that the expected payoff for input @xmath398 is the probability ( density ) of the correct guess : @xmath421 p^{(m)}_\psi \right\ } \\ & = \tr\left\ { \map c_{q , n \to m}^+ \left [ \left ( |\varphi\>\<\varphi| \right)^{\otimes n } \right ] p^{(m)}_\varphi \right\ } \\ & = \frac { d_+^{(n ) } } { d_+^{(m ) } } \tr\left [ \left ( |\varphi\>\<\varphi| \right)^{\otimes n } \map c^+_{q , m \to n } \left ( p^{(m)}_\varphi \right ) \right ] \\ & = \tr [ \left ( |\varphi\>\<\varphi| \right)^{\otimes n } p^{(n)}_\varphi ] \\ & = d_+^{(n ) } \qquad \forall |\varphi\ > \in { \mathcal{h}}. \end{aligned}\ ] ] since the expected payoff is the same for every state , for every prior distribution we have @xmath422 . on the other hand , we already know that @xmath423 , so that the information bound of eq . ( [ infobound ] ) gives @xmath424 . on the other hand , from ref . @xcite we know that in the limit @xmath425 the cloning channel @xmath405 converges to the estimation channel @xmath411 . hence , the probability of success @xmath15 must converge to @xmath409 . again , we obtain the relation @xmath426 that is , @xmath427 . also in this case , the asymptotic behaviour of the probability @xmath428 is dictated by the statistical information bound and by the convergence of the cloning channel @xmath429 to a measure - and - prepare channel . this work showed how causality determines the maximum probability of simulating a given quantum channel from the future to the past . since causality states that it is not possible to signal from the future to the past , the probability of success in the simulation should be small enough to prevent any signal from being sent . as a consequence , channels that are able to transmit more information will have a smaller probability of being simulated , while channels that transmit less information will have a larger probability . in particular , erasure channels , that transmit no information at all , can be simulated with unit probability , while measure - and - prepare channels , that transmit only classical information , can be simulated with probability at least equal to @xmath430 the inverse of the output hilbert space dimension . the hardest channels to simulate are the ideal quantum channels , for which the probability is equal to @xmath431 , the inverse of the square of the hilbert space dimension . a quantitative connection between the amount of information transmitted by a channel and the probability of successful simulation is given by the statistical information bounds , which state that the probability of success is upper bounded by the inverse of the statistical information transmitted by the channel . in the paper we computed explicitly the maximum probability for the simulation of several channels , including partial trace channels from @xmath0 input systems to @xmath28 output systems , and universal cloning channels from @xmath0 input systems to @xmath432 output systems . we pointed out a time - reversal duality between trace and cloning and we exploited it to show that the value of the maximum probability of success is given by @xmath433 . note that for fixed @xmath16 , the probability of success is an increasing function of @xmath0 , the number of input copies . similarly , for fixed @xmath0 , the probability of success is an increasing function of @xmath16 , the number of output copies . these are genuinely quantum features , that are impossible in the classical world : indeed , classically the probability of success for all trace channels and for all cloning channels is given by @xmath434 , independently of @xmath0 and @xmath16 . in the case of a single output copy , @xmath435 the probability of success for a quantum partial trace channel tends to the classical value @xmath434 in the limit of @xmath0 going to infinity . this asymptotic behaviour can be explained from the fact that a trace channel with asymptotically large number of input copies converges to a measure - and - prepare channel @xcite . the same explanation can be provided for the asymptotic behaviour of the probability of success for cloning channels . also in this case the asymptotic behaviour is dictated by the fact that a universal cloning channel with asymptotically large number of output copies converges to a measure - and - prepare channel @xcite . note that in this paper we focused on determining the maximum probability of success for many channels , but we did not show explicitly the generalized teleportation schemes that achieve the simulation of those channels . in particular , it is interesting to ask which kind of entangled states and which kind of measurements allow one to achieve teleportation from @xmath0 input copies to @xmath16 output copies . moreover , it is worth asking whether these probabilistic teleportation protocols can be extended to deterministic protocols ( now , from the past to the future ) by reintroducing classical communication and correction operations , as in the case of tele - cloning @xcite . the fact that the probability of successful teleportation from @xmath0 copies to @xmath17 copy converges to @xmath436 suggests that our probabilistic protocol could be extended to a deterministic protocol that needs only @xmath437 bits of classical communication , instead of the @xmath438 bits of the original teleportation protocol . however , it is quite possible that in general there exists a trade - off between maximizing the probability of success for a particular outcome and being able to implement a correction for all outcomes . all these issues will be investigated in a forthcoming work @xcite . another interesting direction of future research is to consider the simulation of multipartite channels under the constraint that different input and output systems can arrive at different instants of time , and to show how the probability of simulation depends on the possible causal arrangements of the input / output systems . for example , one could consider the case of a cloning channel from @xmath232 input copy to @xmath439 approximate copies , one copy being sent to the past and the other being sent to the future . examples of this kind are expected to shed light on the interplay between causal structure and quantum information flow . 90 g. chiribella , g. m. dariano , and p. perinotti , _ probabilistic theories with purification " _ , phys . a * 81 * , 062348 ( 2010 ) g. chiribella , g. m. dariano , and p. perinotti , _ informational derivation of quantum theory " _ , phys . rev . a * 84 * , 012311 ( 2011 ) . bennett , g. brassard , c. crepeau , r. jozsa , a. peres , and w.k . wooters , `` _ _ teleporting an unknown quantum state via dual classical and einstein - podolski - rosen channels _ _ '' , phys . rev . lett . * 70 * , 1895 ( 1993 ) . s. lloyd , l. maccone , r. garcia - patron , v. giovannetti , y. shikano , s. pirandola , l. a. rozema , a. darabi , y. soudagar , l. k. shalm , a. m. steinberg _ `` closed timelike curves via post - selection : theory and experimental demonstration '' _ , phys . rev .lett . * 106 * , 040403 , ( 2011 ) g. m. dariano and c. macchiavello , _ optimal phase - covariant cloning for qubits and qutrits " _ , a * 67 * , 042306 ( 2003 ) . w. k. wooters and w. h. zurek , _ a single quantum can not be cloned _ , nature * 299 * , 802 ( 1982 ) g. chiribella , _ `` on quantum estimation , quantum cloning , and finite quantum de finetti theorems '' _ , theory of quantum computation , communication , and cryptography , lecture notes in computer science , * 6519/2011 * , 9 ( 2011 )

this paper studies infinite graphs produced from a natural unfolding operation applied to finite graphs . graphs produced via such operations are of finite degree and automatic over the unary alphabet ( that is , they can be described by finite automata over unary alphabet ) . we investigate algorithmic properties of such unfolded graphs given their finite presentations . in particular , we ask whether a given node belongs to an infinite component , whether two given nodes in the graph are reachable from one another , and whether the graph is connected . we give polynomial - time algorithms for each of these questions . for a fixed input graph , the algorithm for the first question is in constant time and the second question is decided using an automaton that recognizes reachability relation in a uniform way . hence , we improve on previous work , in which non - elementary or non - uniform algorithms were found .

0809.3330

we study the algorithmic properties of infinite graphs that result from a natural unfolding operation applied to finite graphs . the unfolding process always produces infinite graphs of finite degree . moreover , the class of resulting graphs is a subclass of the class of automatic graphs . as such , any element of this class possesses all the known algorithmic and algebraic properties of automatic structures . an equivalent way to describe these graphs employs automata over a unary alphabet ( see theorem [ thm : gsigma ] ) . therefore , we call this class of graphs _ unary automatic graphs of finite degree_. in recent years there has been increasing interest in the study of structures that can be presented by automata . the underlying idea in this line of research consists of using automata ( such as word automata , bchi automata , tree automata , and rabin automata ) to represent structures and study logical and algorithmic consequences of such presentations . informally , a structure @xmath0 is _ automatic _ if the domain @xmath1 and all the relations @xmath2 , @xmath3 , @xmath4 of the structure are recognized by finite automata ( precise definitions are in the next section ) . for instance , an automatic graph is one whose set of vertices and set of edges can each be recognized by finite automata . the idea of automatic structures was initially introduced by hodgson @xcite and was later rediscovered by khoussainov and nerode @xcite . automatic structures possess a number of nice algorithmic and model - theoretic properties . for example , khoussainov and nerode proved that the first - order theory of any automatic structure is decidable @xcite . this result is extended by adding the @xmath5 ( there are infinitely many ) and @xmath6 ( there are @xmath7 many mod @xmath8 ) quantifiers to the first order logic @xcite . blumensath and grdel proved a logical characterization theorem stating that automatic structures are exactly those definable in the following fragment of the arithmetic @xmath9 , where @xmath10 and @xmath11 have their usual meanings and @xmath12 is a weak divisibility predicate for which @xmath13 if and only if @xmath14 is a power of @xmath15 and divides @xmath16 @xcite . automatic structures are closed under first - order interpretations . there are descriptions of automatic linear orders and trees in terms of model theoretic concepts such as cantor - bendixson ranks @xcite . also , khoussainov , nies , rubin and stephan have characterized the isomorphism types of automatic boolean algebras @xcite ; thomas and oliver have given a full description of finitely generated automatic groups @xcite . some of these results have direct algorithmic implications . for example , isomorphism problem for automatic well - ordered sets and boolean algebras is decidable @xcite . there is also a body of work devoted to the study of resource - bounded complexity of the first order theories of automatic structures . for example , on the one hand , grdel and blumensath constructed examples of automatic structures whose first - order theories are non - elementary @xcite . on the other hand , lohrey in @xcite proved that the first - order theory of any automatic graph of bounded degree is elementary . it is worth noting that when both a first - order formula and an automatic structure @xmath17 are fixed , determining if a tuple @xmath18 from @xmath17 satisfies @xmath19 can be done in linear time . most of the results about automatic structures , including the ones mentioned above , demonstrate that in various concrete senses automatic structures are not complex from a logical point of view . however , this intuition can be misleading . for example , in @xcite it is shown that the isomorphism problem for automatic structures is @xmath20-complete . this informally tells us that there is no hope for a description ( in a natural logical language ) of the isomorphism types of automatic structures . also , khoussainov and minnes @xcite provide examples of automatic structures whose scott ranks can be as high as possible , fully covering the interval @xmath21 $ ] of ordinals ( where @xmath22 is the first non - computable ordinal ) . they also show that the ordinal heights of well - founded automatic relations can be arbitrarily large ordinals below @xmath22 . in this paper , we study the class of unary automatic graphs of finite degree . since these graphs are described by the unfolding operation ( definition [ dfn : unfolding ] ) on the pair of finite graphs @xmath23 , we use this pair to represent the graph . the size of this pair is the sum of the sizes of the automata that represent these graphs . in the study of algorithmic properties of these graphs one directly deals with the pair @xmath23 . we are interested in the following natural decision problems : * * connectivity problem*. given an automatic graph @xmath24 , decide if @xmath24 is connected . * * reachability problem*. given an automatic graph @xmath24 and two vertices @xmath14 and @xmath16 of the graph , decide if there is a path from @xmath14 to @xmath16 . if we restrict to the class of finite graphs , these two problems are decidable and can be solved in linear time on the sizes of the graphs . however , we are interested in infinite graphs and therefore much more work is needed to investigate the problems above . in addition , we also pose the following two problems : * * infinity testing problem*. given an automatic graph @xmath24 and a vertex @xmath14 , decide if the component of @xmath24 containing @xmath14 is infinite . * * infinite component problem*. given an automatic graph @xmath24 decide if @xmath24 has an infinite component . unfortunately , for the class of automatic graphs all of the above problems are undecidable . in fact , one can provide exact bounds on this undecidability . the connectivity problem is @xmath25-complete ; the reachability problem is @xmath26-complete ; the infinite component problem is @xmath27-complete ; and the infinity testing problem is @xmath25-complete @xcite . since all unary automatic structures are first - order definable in @xmath28 ( the monadic second - order logic of the successor function ) , it is not hard to prove that all the problems above are decidable @xcite . direct constructions using this definability in @xmath28 yield algorithms with non - elementary time since one needs to transform @xmath28 formulas into automata @xcite . however , we provide polynomial - time algorithms for solving all the above problems for this class of graphs . we now outline the rest of this paper by explaining the main results . we comment that these polynomial - time algorithms are based on deterministic input automata . section 2 introduces the main definitions needed , including the concept of automatic structure . section 3 singles out unary automatic graphs and provides a characterization theorem ( theorem [ thm : characterization ] ) . section 4 introduces unary automatic graphs of finite degree . the main result is theorem [ thm : gsigma ] that explicitly provides an algorithm for building unary automatic graphs of finite degree . this theorem is used throughout the paper . section 5 is devoted to deciding the infinite component problem . the main result is the following : * theorem [ thm : infinite component ] * _ the infinite component problem for unary automatic graph of finite degree @xmath24 is solved in @xmath29 , where @xmath8 is the number of states of the deterministic finite automaton recognizing @xmath24 . _ in this section , we make use of the concept of oriented walk for finite directed graphs . the subsequent section is devoted to deciding the infinity testing problem . the main result is the following : * theorem [ thm : inftest ] * _ the infinity testing problem for unary automatic graph of finite degree @xmath24 is solved in @xmath29 , where @xmath8 is the number of states of the deterministic finite automaton @xmath17 recognizing @xmath24 . in particular , when @xmath17 is fixed , there is a constant time algorithm that decides the infinity testing problem on @xmath24 . _ the fact that there is a constant time algorithm when @xmath17 is fixed will be made clear in the proof . the value of the constant is polynomial in the number of states of @xmath17 . the reachability problem is addressed in section 7 . this problem has been studied in @xcite,@xcite , @xcite via the class of * pushdown graphs*. a pushdown graph is the configuration space of a pushdown automaton . unary automatic graphs are pushdown graphs @xcite . in @xcite it is proved that for a pushdown graph @xmath24 , given a node @xmath30 , there is an automaton that recognizes all nodes reachable from @xmath30 . the number of states in the automaton depends on the input node @xmath30 . this result implies that there is an algorithm that decides the reachability problem on unary automatic graphs of finite degree . however , there are several issues with this algorithm . the automata constructed by the algorithm are not uniform in @xmath30 in the sense that different automata are built for different input nodes @xmath30 . moreover , the automata are nondeterministic . hence , the size of the deterministic equivalent automata is exponential in the size of the representation of @xmath30 . section 7 provides an alternative algorithm to solve the reachability problem on unary automatic graphs of finite degree uniformly . this new algorithm constructs a deterministic automaton @xmath31 that accepts the set of pairs @xmath32 there is a path from @xmath33 to @xmath34 . the size of @xmath31 only depends on the number of states of the automaton @xmath8 , and constructing the automaton requires polynomial - time in @xmath8 . the practical advantage of such a uniform solution is that , when @xmath31 is built , deciding whether node @xmath30 is reachable from @xmath33 by a path takes only linear time ( details are in section 7 ) . the main result of this section is the following : * theorem [ thm : reachability ] * _ suppose @xmath24 is a unary automatic graph of finite degree represented by deterministic finite automaton @xmath17 of size @xmath8 . there exists a polynomial - time algorithm that solves the reachability problem on @xmath24 . for inputs @xmath35 , the running time of the algorithm is @xmath36 . _ finally , the last section solves the connectivity problem for @xmath24 . * theorem [ thm : connectivity ] * _ the connectivity problem for unary automatic graph of finite degree @xmath24 is solved in @xmath29 , where @xmath8 is the number of states of the deterministic finite automaton recognizing @xmath24 . _ the authors would like to thank referees for comments on improvement of this paper . a * finite automaton * @xmath17 over an alphabet @xmath37 is a tuple @xmath38 , where @xmath39 is a finite set of * states * , @xmath40 is the * initial state * , @xmath41 is the * transition table * and @xmath42 is the set of * final states*. a * computation * of @xmath17 on a word @xmath43 ( @xmath44 ) is a sequence of states , say @xmath45 , such that @xmath46 and @xmath47 for all @xmath48 . if @xmath49 , then the computation is * successful * and we say that automaton @xmath17 * accepts * the word . the * language * accepted by the automaton @xmath17 is the set of all words accepted by @xmath17 . in general , @xmath50 is * fa recognizable * , or * regular * , if @xmath51 is the language accepted by some finite automaton . in this paper we always assume the automata are deterministic . for two states @xmath52 , the * distance * from @xmath53 to @xmath54 is the minimum number of transitions required for @xmath17 to go from @xmath53 to @xmath54 . to formalize the notion of a relation being recognized by an automaton , we define synchronous @xmath8-tape automata . such an automaton can be thought of as a one - way turing machine with @xmath8 input tapes . each tape is semi - infinite having written on it a word in the alphabet @xmath37 followed by a succession of @xmath55 symbols . the automaton starts in the initial state , reads simultaneously the first symbol of each tape , changes state , reads simultaneously the second symbol of each tape , changes state , etc . , until it reads @xmath55 on each tape . the automaton then stops and accepts the @xmath8-tuple of words if and only if it is in a final state . more formally , we write @xmath56 for @xmath57 where @xmath55 is a symbol not in @xmath37 . the * convolution * of a tuple @xmath58 is the string @xmath59 of length @xmath60 symbol is @xmath61 where @xmath62 is the @xmath63 symbol of @xmath64 if @xmath65 is the relation @xmath66 formed as the set of convolutions of all the tuples in @xmath67 . an @xmath8-ary relation @xmath68 is * fa recognizable * , or * regular * , if its convolution @xmath69 is recognizable by a finite automaton . a * structure * @xmath70 consists of a countable set @xmath51 called the * domain * and some relations and operations on @xmath51 . we may assume that @xmath70 only contains relational predicates since operations can be replaced with their graphs . we write @xmath71 where @xmath72 is an @xmath73-ary relation on @xmath51 . the relation @xmath74 are sometimes called basic or atomic relations . we assume that the function @xmath75 is always a computable one . a structure @xmath70 is * automatic over alphabet @xmath37 * if its domain @xmath50 is finite automaton recognizable , and there is an algorithm that for each @xmath76 produces an @xmath77-tape automaton recognizing the relation @xmath78 . a structure is called * automatic * if it is automatic over some alphabet . if @xmath79 is isomorphic to an automatic structure @xmath70 , then we call @xmath70 an * automatic presentation * of @xmath79 and say that @xmath79 is _ * automatically presentable*_. an example of an automatic structure is the word structure @xmath80 , where for all @xmath81 , @xmath82 , @xmath83 , @xmath84 if and only if @xmath85 , and @xmath86 is the lexicographical order . the configuration graph of any turing machine is another example of an automatic structure . examples of automatically presentable structures are @xmath87 , @xmath88 , @xmath89 , the group @xmath90 , the order on the rational @xmath91 , and the boolean algebra of finite and co - finite subsets of @xmath92 . consider the first - order logic extended by @xmath93 ( there exist infinitely many ) and @xmath6 ( there exist @xmath8 many mod @xmath7 , where @xmath8 and @xmath7 are natural numbers ) quantifiers . we denote this logic by @xmath94 . we will use the following theorem without explicit reference to it . @xcite let @xmath17 be an automatic structure . there exists an algorithm that , given a formula @xmath19 in @xmath95 , produces an automaton that recognizes exactly those tuples @xmath18 from the structure that make @xmath96 true . in particular , the set of all sentences of @xmath95 which are true in @xmath17 is decidable . we now turn our attention to the subclass of the automatic structures which is the focus of the paper . a structure @xmath17 is * unary automatic * if it has an automatic presentation whose domain is @xmath97 and whose relations are automatic . examples of unary automatic structures are @xmath98 and @xmath99 . some recent work on unary automatic structures includes a characterization of unary automatic linearly ordered sets , permutation structures , graphs , and equivalence structures @xcite . for example , unary automatic linearly ordered sets are exactly those that are isomorphic to a finite sum of orders of type @xmath100 , @xmath101 ( the order of negative integers ) , and finite @xmath8 . a * unary automatic graph * is a graph @xmath102 whose domain is @xmath97 , and whose edge relation @xmath103 is regular . we use the following example to illustrate that this class of graphs are the best possible . consider the class of graphs with all vertices being of the form @xmath104 for some alphabet @xmath105 . at first sight , graphs of this form may have an intermediate position between unary and general automatic graphs . however , the infinite grid @xmath106 can be coded automatically over @xmath104 by @xmath107 , and @xmath108 is not decidable @xcite . in particular , counter machines can be coded into the grid , so the reachability problem is not decidable . to eliminate bulky exposition , we make the following assumptions in the rest of the paper . * the automata under consideration are viewed as deterministic . hence , when we write automata , we mean deterministic finite automata . * all structures are infinite unless explicitly specified . * the graphs are undirected . the case of directed graphs can be treated in a similar manner . let @xmath109 be an automatic graph . let @xmath17 be an automaton recognizing @xmath103 . we establish some terminology for the automaton @xmath17 . the general shape of @xmath17 is given in figure [ fg : standard ] . all the states reachable from the initial state by reading inputs of type @xmath110 are called * @xmath110-states*. a * tail * in @xmath17 is a sequence of states linked by transitions without repetition . a * loop * is a sequence of states linked by transitions such that the last state coincides with the first one , and with no repetition in the middle . the set of @xmath110-states is a disjoint union of a tail and a loop . we call the tail the * @xmath110-tail * and the loop the * @xmath110-loop*. let @xmath111 be a @xmath110 state . all the states reachable from @xmath111 by reading inputs of type @xmath112 are called * @xmath112-states*. this collection of all @xmath112-states is also a disjoint union of a tail and a loop ( see the figure ) , called the * @xmath112-tail * and the * @xmath112-loop * , respectively . the * @xmath113-tails * and * @xmath113-loops * are defined in a similar matter . we say that an automaton is * standard * if the lengths of all its loops and tails equal some number @xmath114 , called the * loop constant*. if @xmath17 is a standard automaton recognizing a binary relation , it has exactly @xmath115 @xmath110-states . on each of these states , there is a @xmath116-tail and a @xmath117-tail of length exactly @xmath114 . at the end of each @xmath112-tail and @xmath113-tail there is a @xmath112-loop and @xmath118-loop , respectively , of size exactly @xmath114 . therefore if @xmath8 is the number of states in @xmath17 , then @xmath119 . [ lemma : standard ] let @xmath17 be an @xmath8 state automaton recognizing a binary relation @xmath103 on @xmath97 . there exists an equivalent standard automaton with at most @xmath120 states . let @xmath114 be the least common multiple of the lengths of all loops and tails of @xmath17 . an easy estimate shows that @xmath114 is no more than @xmath121 . one can transform @xmath17 into an equivalent standard automaton whose loop constant is @xmath114 . hence , there is a standard automaton equivalent to @xmath17 whose size is bounded above by @xmath122 . we can simplify the general shape of the automaton using the fact that we consider undirected graphs . indeed , we need only consider transitions labelled by @xmath123 . to see this , given an automaton with only @xmath118 transitions , to include all symmetric transitions , add a copy of each @xmath118 transition which is labelled with @xmath116 . we recall a characterization theorem of unary automatic graphs from @xcite . let @xmath124 and @xmath125 be finite graphs . let @xmath126 be subsets of @xmath127 , and @xmath128 be subsets of @xmath129 . consider the graph @xmath130 followed by @xmath100 many copies of @xmath79 , ordered as @xmath131 . formally , the vertex set of @xmath132 is @xmath133 and we write @xmath134 for @xmath135 and @xmath136 . the edge set @xmath137 of @xmath132 consists of all pairs @xmath138 such that @xmath139 . we define the infinite graph , * @xmath140 * , as follows : @xmath141 the vertex set is @xmath142 ; @xmath143 the edge set contains @xmath144 as well as the following edges , for all @xmath145 , @xmath146 , and @xmath147 : * @xmath148 when @xmath149 , and @xmath150 when @xmath151 , * @xmath152 when @xmath153 , and @xmath154 when @xmath155 . [ thm : characterization ] @xcite a graph is unary automatic if and only if it is isomorphic to @xmath140 for some parameters @xmath79 , @xmath130 , and @xmath156 . moreover , if @xmath17 is a standard automaton representing @xmath24 then the parameters @xmath157 can be extracted in @xmath158 ; otherwise , the parameters can be extracted in @xmath159 , where @xmath8 is the number of states in @xmath17 . a graph is of * finite degree * if there are at most finitely many edges from each vertex @xmath30 . we call an automaton @xmath17 recognizing a binary relation over @xmath160 a * one - loop automaton * if its transition diagram contains exactly one loop , the @xmath110-loop . the general structure of one - loop automata is given in figure [ fg : type1auto ] . we will always assume that the lengths of all the tails of the one - loop automata are not bigger than the size of the @xmath110-loop . the following is an easy proposition and we omit its proof . let @xmath109 be a unary automatic graph , then @xmath24 is of finite degree if and only if there is a one - loop automaton @xmath17 recognizing @xmath103 . by lemma [ lemma : standard ] , transforming a given automaton to an equivalent standard automaton may blow up the number of states exponentially . however , there is only polynomial blow up if @xmath17 is a one - loop automaton . if @xmath17 is a one - loop automaton with @xmath8 states , there exists an equivalent standard one - loop automaton with loop constant @xmath161 . let @xmath162 be the length of the loop in @xmath17 and @xmath163 be the length of the longest tail in @xmath17 . let @xmath114 be the least multiple of @xmath162 such that @xmath164 . it is easy to see that @xmath165 . one can transform @xmath17 into an equivalent standard one - loop automaton whose loop constant is @xmath114 . note that the equivalent standard automaton has @xmath115 ( 1,1)-states . from each of them there is a @xmath112-tail of length @xmath114 and a @xmath118-tail of length @xmath114 . hence the automaton has @xmath166 states . by the above lemma , we always assume the input automaton @xmath17 is standard . in the rest of the paper , we will state all results in terms of the loop constant @xmath114 instead of @xmath8 , the number of states of the input automaton . since @xmath161 , for any constant @xmath167 , an @xmath168 algorithm can also be viewed as an @xmath169 algorithm . given two unary automatic graphs of finite degree @xmath170 and @xmath171 ( where we recall the convention that the domain of each graph is @xmath97 ) , we can form the * union graph * @xmath172 and the * intersection graph * @xmath173 . automatic graphs of finite degree are closed under these operations . indeed , let @xmath174 and @xmath175 be one - loop automata recognizing @xmath176 and @xmath177 with loop constants @xmath178 and @xmath179 , respectively . the standard construction that builds automata for the union and intersection operations produces a one - loop automaton whose loop constant is @xmath180 . we introduce another operation : consider the new graph @xmath181 , where the set @xmath182 of edges is defined as follows ; a pair @xmath183 is in @xmath184 if and only if @xmath185 and @xmath186 . the relation @xmath182 is recognized by the same automaton as @xmath176 , modified so that all @xmath113-states that are final declared non - final , and all the @xmath113-states that are non - final declared final . thus , we have the following proposition : if @xmath187 and @xmath188 are automatic graphs of finite degree then so are @xmath189 , @xmath190 , and @xmath191 . now our goal is to recast theorem [ thm : characterization ] for graphs of finite degree . our analysis will show that , in contrast to the general case for automatic graphs , the parameters @xmath79 , @xmath130 , and @xmath156 for graphs of finite degree can be extracted in linear time . [ dfn : unfolding ] let @xmath192 and @xmath193 be finite graphs . consider the finite sets @xmath194 consisting of all mappings @xmath195 , and @xmath196 consisting of all mappings @xmath197 . any infinite sequence @xmath198 where @xmath199 and @xmath200 for each @xmath76 , defines the infinite graph @xmath201 as follows : * @xmath202 . * @xmath203 . thus @xmath204 is obtained by taking @xmath130 together with an infinite disjoint union of @xmath205 such that edges between @xmath130 and the first copy of @xmath205 are put according to the mapping @xmath206 , and edges between successive copies of @xmath205 are put according to @xmath62 . figure [ fg : gsigma ] illustrates the general shape of a unary automatic graph of finite degree that is build from @xmath130 , @xmath205 , @xmath206 , and @xmath207 , where @xmath208 is the infinite word @xmath209 . [ thm : gsigma ] a graph of finite degree @xmath109 possesses a unary automatic presentation if and only if there exist finite graphs @xmath210 and mappings @xmath211 and @xmath212 such that @xmath24 is isomorphic to @xmath213 . let @xmath109 be a unary automatic graph of finite degree . let @xmath17 be an automaton recognizing @xmath103 . in linear time on the number of states of @xmath17 we can easily transform @xmath17 into a one - loop automaton . so , we assume that @xmath17 is a one - loop automaton with loop constant @xmath114 . we construct the finite graph @xmath130 by setting @xmath214 , where @xmath53 is the starting state , @xmath215 are all states on the @xmath110-tail such that @xmath216 is reached from @xmath217 by reading @xmath110 for @xmath218 ; and for @xmath219 , @xmath220 iff there is a final state @xmath221 on the @xmath113-tail out of @xmath216 , and the distance from @xmath216 to @xmath221 is @xmath222 . we construct the graph @xmath205 similarly by setting @xmath223 where @xmath224 are all states on the @xmath110-loop . the edge relation @xmath225 is defined in a similar way as @xmath226 . the mapping @xmath227 is defined for any @xmath228 by putting @xmath229 in @xmath230 if and only if there exists a final state @xmath231 on the @xmath118-tail out of @xmath232 , and the distance from @xmath232 to @xmath231 equals @xmath233 . the mapping @xmath234 is constructed in a similar manner by reading the @xmath118-tails out of the @xmath110-loop . it is clear from this construction that the graphs @xmath24 and @xmath235 are isomorphic . conversely , consider the graph @xmath213 for some @xmath236 and @xmath237 . assume that @xmath238 , @xmath239 . a one - loop automaton @xmath17 recognizing the edge relation of @xmath240 is constructed as follows . the @xmath110-tail of the automaton is formed by @xmath241 and the @xmath110-loop is formed by @xmath242 , both in natural order . the initial state is @xmath53 . if for some @xmath243 , @xmath244 , then put a final state @xmath221 on the @xmath118-tail starting from @xmath216 such that the distance from @xmath216 to @xmath221 is @xmath222 . if @xmath245 , then repeat the process but make the corresponding distance @xmath246 . the set of edges @xmath225 and mapping @xmath234 are treated in a similar manner by putting final states on the @xmath118-tails from the @xmath110-loop . again , we see that @xmath17 represents a unary automatic graph that is isomorphic to @xmath213 . the proof of the above theorem also gives us the following corollary . if @xmath24 is a unary automatic graph of finite degree , the parameters @xmath130 , @xmath205 , @xmath234 and @xmath206 can be extracted in @xmath247 time , where @xmath114 is the loop constant of the one - loop automaton representing the graph . furthermore , @xmath248 . recall the graphs are undirected . a * component * of @xmath24 is the transitive closure of a vertex under the edge relation . the * infinite component problem * asks whether a given graph @xmath24 has an infinite component . [ thm : infinite component ] the infinite component problem for unary automatic graph of finite degree @xmath24 is solved in @xmath249 , where @xmath114 is the loop constant of the automaton recognizing @xmath24 . by theorem [ thm : gsigma ] , let @xmath250 . we observe that it is sufficient to consider the case in which @xmath251 ( hence @xmath252 ) since @xmath213 has an infinite component if and only if @xmath253 has one . let @xmath254 be the @xmath255 copy of @xmath205 in @xmath24 . let @xmath256 be the copy of vertex @xmath14 in @xmath257 . we construct a finite directed graph @xmath258 as follows . each node in @xmath259 represents a distinct connected component in @xmath205 . for simplicity , we assume that @xmath260 and hence use @xmath14 to denote its own component in @xmath205 . the case in which @xmath261 can be treated in a similar way . for @xmath262 , put @xmath263 if and only if @xmath264 for some @xmath265 and @xmath266 that are in the same component as @xmath14 and @xmath16 , respectively . constructing @xmath267 requires finding connected components of @xmath205 hence takes time @xmath247 . to prove the above theorem , we make essential use of the following definition . see also @xcite . [ dfn : edge - path ] an * _ oriented walk _ * in a directed graph @xmath268 is a subgraph @xmath269 of @xmath268 that consists of a sequence of nodes @xmath270 such that for @xmath271 , either @xmath272 or @xmath273 is an arc in @xmath268 , and for each @xmath274 , exactly one of @xmath275 and @xmath276 belongs to @xmath269 . an oriented walk is an * _ oriented cycle _ * if @xmath277 and there are no repeated nodes in @xmath278 . in an oriented walk @xmath269 , an arc @xmath279 is called a * forward arc * and @xmath280 is called a * backward arc*. the * net length * of @xmath269 , denoted @xmath281 , is the difference between the number of forward arcs and backward arcs . note the net length can be negative . the next lemma establishes a connection between oriented cycles in @xmath267 and infinite components in @xmath24 . [ lm : cycle ] there is an infinite component in @xmath24 if and only if there is an oriented cycle in @xmath282 such that the net length of the cycle is positive . suppose there is an oriented cycle @xmath269 from @xmath14 to @xmath14 in @xmath282 of net length @xmath283 . for all @xmath284 , @xmath269 defines the path @xmath285 in @xmath24 from @xmath286 to @xmath287 where @xmath285 lies in @xmath288 . therefore , for a fixed @xmath284 , all vertices in the set @xmath289 belong to the same component of @xmath24 . in particular , this implies that @xmath24 contains an infinite component . conversely , suppose there is an infinite component @xmath51 in @xmath24 . since @xmath205 is finite , there must be some @xmath14 in @xmath290 such that there are infinitely many copies of @xmath14 in @xmath51 . let @xmath286 and @xmath291 be two copies of @xmath14 in @xmath51 such that @xmath292 . consider a path between @xmath286 and @xmath291 . we can assume that on this path there is at most one copy of any vertex @xmath293 apart from @xmath14 ( otherwise , choose @xmath291 to be the copy of @xmath14 in the path that has this property ) . by definition of @xmath253 and @xmath267 , the node @xmath14 must be on an oriented cycle of @xmath267 with net length @xmath222 . by the equivalence in lemma [ lm : cycle ] , it suffices to provide an algorithm that decides if @xmath282 contains an oriented cycle with positive net length . notice that the existence of an oriented cycle with positive net length is equivalent to the existence of an oriented cycle with negative net length . therefore , we give an algorithm which finds oriented cycles with non - zero net length . for each node @xmath14 in @xmath267 , we search for an oriented cycle of positive net length from @xmath14 by creating a labeled queue of nodes @xmath294 which are connected to @xmath14 . ` alg : oriented - cycle ` 1 . pick node @xmath295 for which a queue has not been built yet . initially the queue @xmath294 is empty . let @xmath296 , and put @xmath14 into the queue . mark @xmath14 as _ unprocessed_. if queues have been built for each @xmath295 , stop the process and return _ let @xmath16 be the first _ unprocessed _ node in @xmath294 . if there are no _ unprocessed _ nodes in @xmath294 , return to ( 1 ) . 3 . for each of the nodes @xmath297 in the set @xmath298 , do the following . 1 . if @xmath299 , set @xmath300 ; if @xmath301 , set @xmath302 . ( if both hold , do steps ( a ) , ( b ) , ( c ) first for @xmath303 and then for @xmath304 . ) 2 . if @xmath305 , then set @xmath306 , put @xmath297 into @xmath294 , and mark @xmath297 as _ unprocessed_. 3 . if @xmath307 then 1 . if @xmath306 , move to next @xmath297 , 2 . if @xmath308 , stop the process and return _ yes_. 4 . mark @xmath16 as _ processed _ and go back to ( 2 ) . an important property of this algorithm is that when we are building a queue for node @xmath14 and are processing @xmath297 , both @xmath309 and @xmath310 represent net lengths of paths from @xmath14 to @xmath297 . we claim that the algorithm returns _ yes _ if and only if there is an oriented cycle in @xmath267 with non - zero net length . suppose the algorithm returns _ yes_. then , there is a base node @xmath14 and a node @xmath297 such that @xmath308 . this means that there is an oriented walk @xmath269 from @xmath14 to @xmath297 with net length @xmath309 and there is an oriented walk @xmath311 from @xmath14 to @xmath297 with net length @xmath310 . consider the oriented walk @xmath312 , where @xmath313 is the oriented walk @xmath311 in reverse direction . clearly this is an oriented walk from @xmath14 to @xmath14 with net length @xmath314 . if there are no repeated nodes in @xmath312 , then it is the required oriented cycle . otherwise , let @xmath16 be a repeated node in @xmath312 such that no nodes between the two occurrences of @xmath16 are repeated . consider the oriented walk between these two occurrences of @xmath16 , if it has a non - zero net length , then it is our required oriented cycle ; otherwise , we disregard the part between the two occurrences of @xmath297 and make the oriented walk shorter without altering its net length . conversely , suppose there is an oriented cycle @xmath315 of non - zero net length where @xmath316 . however , we assume for a contradiction that the algorithm returns _ no_. consider how the algorithm acts when we pick @xmath317 at step ( 1 ) . for each @xmath318 , one can prove the following statements by induction on @xmath76 . 1 . @xmath319 always gets a label @xmath320 2 . @xmath320 equals the net length of the oriented walk from @xmath317 to @xmath319 in @xmath269 . by the description of the algorithm , @xmath317 gets the label @xmath321 . suppose the statements holds for @xmath319 , @xmath322 , then at the next stage , the algorithm labels all nodes in @xmath323 . in particular , it calculates @xmath324 . by the inductive hypothesis , @xmath324 is the net length of the oriented walk from @xmath317 to @xmath325 in @xmath269 . if @xmath325 has already had a label @xmath326 and @xmath327 , then the algorithm would return _ yes_. therefore @xmath328 . by assumption on @xmath269 , @xmath329 . however , since @xmath330 , the induction gives that @xmath331 . this is a contradiction , and thus the above algorithm is correct . in summary , the following algorithm solves the infinite component problem . suppose we are given an automaton ( with loop constant @xmath114 ) which recognizes the unary automatic graph of finite degree @xmath24 . recall that @xmath114 is also the cardinality of @xmath290 . we first compute @xmath267 , in time @xmath247 . then we run ` oriented - cycle ` to decide whether @xmath267 contains an oriented cycle with positive net length . for each node @xmath14 in @xmath267 , the process runs in time @xmath247 . since @xmath267 contains @xmath114 number of nodes , this takes time @xmath332 . note that lemma [ lm : cycle ] holds for the case when @xmath333 deciding the infinity testing problem ------------------------------------- we next turn our attention to the * infinity testing problem * for unary automatic graphs of finite degree . recall that this problem asks for an algorithm that , given a vertex @xmath30 and a graph @xmath24 , decides if @xmath30 belongs to an infinite component . we prove the following theorem . [ thm : inftest ] the infinity testing problem for unary automatic graph of finite degree @xmath24 is solved in @xmath332 , where @xmath114 is the loop constant of the automaton @xmath17 recognizing @xmath24 . in particular , when @xmath17 is fixed , there is a constant time algorithm that decides the infinity testing problem on @xmath24 . for a fixed input @xmath286 , we have the following lemma . [ lm : infinite ] if @xmath286 is connected to some @xmath334 such that @xmath335 , then @xmath286 is in an infinite component . suppose such a @xmath334 exists . take a path @xmath336 in @xmath24 from @xmath286 to @xmath334 . since @xmath114 is the cardinality of @xmath290 , there is @xmath337 such that @xmath338 and @xmath339 appear in @xmath336 with @xmath340 . therefore all nodes in the set @xmath341 are in the same component as @xmath286 . let @xmath342 . to decide if @xmath286 and @xmath334 are in the same component , we run a breadth first search in @xmath24 starting from @xmath286 and going through all vertices in @xmath343 . the algorithm is as follows : ` alg : finitereach ` 1 . let @xmath344 . 2 . initialize the queue @xmath345 to be empty . put the pair @xmath346 into @xmath345 and mark it as _ unprocessed_. 3 . if there are no _ unprocessed _ pairs in @xmath345 , stop the process . otherwise , let @xmath347 be the first _ unprocessed _ pair . for arcs @xmath348 of the form @xmath304 or @xmath303 in @xmath349 , do the following . 1 . if @xmath348 is of the form @xmath304 , let @xmath350 ; if @xmath348 is of the form @xmath303 , let @xmath351 . 2 . if @xmath352 and @xmath353 is not in @xmath345 , then put @xmath353 into @xmath345 and mark @xmath353 as _ unprocessed_. 4 . mark @xmath347 as _ processed _ , and go to ( 2 ) . note that any @xmath334 is reachable from @xmath286 on the graph @xmath24 restricted on @xmath354 if and only if after running ` finitereach ` on the input @xmath286 , the pair @xmath355 is in @xmath345 . when running the algorithm we only use the exact value of the input @xmath76 when @xmath356 ( we set @xmath357 whenever @xmath284 ) , so the running time of ` finitereach ` is bounded by the number of edges in @xmath24 restricted to @xmath358 . therefore the running time is @xmath332 . let @xmath359 . [ lm : inftest - b ] let @xmath360 . @xmath286 is in an infinite component if and only if @xmath361 . suppose a vertex @xmath362 , then there is a path from @xmath286 to @xmath363 . by lemma [ lm : infinite ] , @xmath286 is in an infinite component . conversely , if @xmath286 is in an infinite component , then there must be some vertices in @xmath364 reachable from @xmath286 . take a path from @xmath286 to a vertex @xmath363 such that @xmath363 is the first vertex in @xmath364 appearing on this path . then @xmath362 . we assume the input vertex @xmath286 is given by tuple @xmath365 . the above lemma suggests a simple algorithm to check if @xmath286 is in an infinite component . ` alg : infinitetest ` 1 . run ` finitereach ` on vertex @xmath256 , computing the set @xmath366 while building the queue @xmath345 . 2 . for every @xmath362 , check if there is edge @xmath299 . return @xmath367 if one such edge is found ; otherwise , return @xmath368 . running ` finitereach ` takes @xmath332 and checking for edge @xmath304 takes @xmath247 . the running time is therefore @xmath332 . since @xmath14 is bounded by @xmath114 , if @xmath17 is fixed , checking whether @xmath286 belongs to an infinite component takes constant time . suppose @xmath24 is a unary automatic graph of finite degree represented by an automaton with loop constant @xmath114 . the * reachability problem * on @xmath24 is formulated as : given two vertices @xmath369 in @xmath24 , decide if @xmath286 and @xmath334 are in the same component . we prove the following theorem . [ thm : reachability ] suppose @xmath24 is a unary automatic graph of finite degree represented by an automaton @xmath17 of loop constant @xmath114 . there exists a polynomial - time algorithm that solves the reachability problem on @xmath24 . for inputs @xmath35 , the running time of the algorithm is @xmath370 . we restrict to the case when @xmath371 . the proof can be modified slightly to work in the more general case , @xmath372 . since , by theorem [ thm : inftest ] , there is an @xmath332-time algorithm to check if @xmath256 is in a finite component , we can work on the two possible cases separately . we first deal with the case when the input @xmath286 is in a finite component . by lemma [ lm : infinite ] , @xmath286 and @xmath334 are in the same ( finite ) component if and only if after running ` finitereach ` on the input @xmath286 , the pair @xmath355 is in the queue @xmath345 . [ cr : finite reach ] if all components of @xmath24 are finite and we represent @xmath373 as @xmath374 , then there is an @xmath332-algorithm deciding if @xmath286 and @xmath334 are in the same component . now , suppose that @xmath286 is in an infinite component . we start with the following question : given @xmath375 , are @xmath286 and @xmath376 in the same component in @xmath24 ? to answer this , we present an algorithm that computes all vertices @xmath375 whose @xmath255 copy lies in the same @xmath24-component as @xmath286 . the algorithm is similar to ` finitereach ` , except that it does not depend on the input @xmath76 . line(3b ) in the algorithm is changed to the following : ( 3b ) if @xmath377 and @xmath353 is not in @xmath345 , then put @xmath353 into @xmath345 and mark @xmath353 as _ unprocessed_. we use this modified algorithm to define the set @xmath378 . intuitively , we can think of the algorithm as a breadth first search through @xmath379 which originates at @xmath380 . therefore , @xmath381 if and only if there exists a path from @xmath380 to @xmath382 in @xmath24 restricted to @xmath379 . [ lm : reach_reach ] suppose @xmath286 is in an infinite component . the vertex @xmath376 is in the same component as @xmath286 if and only if @xmath376 is also in an infinite component and @xmath383 . suppose @xmath376 is in an infinite component and @xmath383 . if @xmath284 , then the observation above implies that there is a path from @xmath286 to @xmath376 in @xmath384 . so , it remains to prove that @xmath286 and @xmath376 are in the same component even if @xmath385 . since @xmath381 , there is a path @xmath336 in @xmath24 from @xmath380 to @xmath382 . let @xmath386 be the least number such that @xmath387 . if @xmath388 , then it is clear that @xmath286 and @xmath376 are in the same component . thus , suppose that @xmath389 . let @xmath297 be such that @xmath390 . then @xmath336 is @xmath391 where @xmath392 is a path from @xmath380 to @xmath393 and @xmath394 is a path from @xmath393 to @xmath382 . since @xmath286 is in an infinite component , it is easy to see that @xmath380 is also in an infinite component . there exists an @xmath395 such that all vertices in the set @xmath396 are in the same component . likewise , there is an @xmath397 such that all vertices in @xmath398 are in the same component . consider @xmath399 and @xmath400 . analogous to the path @xmath392 , there is a path @xmath401 from @xmath399 to @xmath402 . similarly , there is a path @xmath403 from @xmath402 to @xmath400 . we describe another path @xmath404 from @xmath380 to @xmath382 as follows . @xmath404 first goes from @xmath380 to @xmath399 , then goes along @xmath405 from @xmath399 to @xmath400 and finally goes to @xmath382 . notice that the least @xmath406 such that @xmath407 must be larger than @xmath386 . we can iterate this procedure of lengthening the path between @xmath380 and @xmath382 until @xmath408 , as is required to reduce to the previous case . to prove the implication in the other direction , we assume that @xmath286 and @xmath376 are in the same infinite component . then @xmath376 is , of course , in an infinite component . we want to prove that @xmath409 . let @xmath410 . suppose there exists a path @xmath336 in @xmath24 from @xmath286 to @xmath376 which stays in @xmath411 . then , indeed , @xmath381 . on the other hand , suppose no such path exists . since @xmath286 and @xmath376 are in the same component , there is some path @xmath336 from @xmath286 to @xmath376 . let @xmath412 be the largest number such that @xmath413 . let @xmath414 be the least number such that @xmath415 . we are in one of two cases : @xmath416 or @xmath417 . we will prove that if @xmath416 then there is a path @xmath404 from @xmath286 to @xmath376 such that @xmath418 and @xmath419 . the case in which @xmath417 can be handled in a similar manner . without loss of generality , we assume @xmath420 since otherwise we can change the input @xmath14 and make @xmath421 . let @xmath297 be a vertex in @xmath205 such that @xmath422 . then @xmath336 is @xmath391 where @xmath392 is a path from @xmath286 to @xmath423 and @xmath394 is a path from @xmath423 to @xmath376 . since @xmath424 , there must be some @xmath425 and @xmath426 in @xmath392 such that @xmath427 . for the same reason , there must be some @xmath428 and @xmath429 in @xmath394 such that @xmath430 . therefore , @xmath336 contains paths between any consecutive pair of vertices in the sequence @xmath431 . consider the following sequence of vertices : @xmath432 it is easy to check that there exists a path between each pair of consecutive vertices in the sequence . therefore the above sequence describes a path @xmath404 from @xmath286 to @xmath376 . it is easy to see that @xmath433 . also since @xmath421 , @xmath434 . therefore @xmath404 is our desired path . in the following , we abuse notation by using @xmath435 and @xmath234 on subsets of @xmath436 . we inductively define a sequence @xmath437 such that each @xmath438 is a subset of @xmath436 . let @xmath439 and for @xmath427 , we define @xmath440 . the following lemma is immediate from this definition . [ lm : reach_closure ] suppose @xmath286 is in an infinite component , then @xmath286 and @xmath334 are in the same component if and only if @xmath334 is also in an infinite component and @xmath441 . we can use the above lemma to construct a simple - minded algorithm that solves the reachability problem on inputs @xmath442 . ` alg : navereach ` 1 . check if each of @xmath286 , @xmath334 are in an infinite component of @xmath24 ( using the algorithm of theorem [ thm : inftest ] ) . 2 . if exactly one of @xmath286 and @xmath334 is in a finite component , then return _ if both @xmath286 and @xmath334 are in finite components , then run ` finitereach ` on input @xmath286 and check if @xmath355 is in @xmath345 . if both @xmath286 and @xmath334 are in infinite components , then compute @xmath443 . if @xmath441 , return _ yes _ ; otherwise , return _ no_. we now consider the complexity of this algorithm . the set @xmath444 can be computed in time @xmath332 . given @xmath445 , we can compute @xmath438 in time @xmath332 by computing @xmath446 for any @xmath447 . therefore , the total running time of ` navereach ` on input @xmath286 , @xmath334 is @xmath448 . we want to replace the multiplication with addition and hence tweak the algorithm . from lemma [ lm : inftest - b ] , @xmath286 is in an infinite component in @xmath24 if and only if ` finitereach ` finds a vertex @xmath363 connecting to @xmath286 . now , suppose that @xmath286 is in an infinite component . we can use ` finitereach ` to find such a @xmath16 , and a path from @xmath286 to @xmath363 . on this path , there must be two vertices @xmath449 with @xmath450 . let @xmath451 . note that @xmath452 can be computed from the algorithm . it is easy to see that all vertices in the set @xmath453 belong to the same component . [ lm : reach_repeat ] @xmath454 . by definition , @xmath455 if and only if @xmath380 and @xmath382 are in the same component of @xmath24 . suppose that there exists a path in @xmath24 from @xmath380 to @xmath382 . then there is a path from @xmath456 to @xmath457 . since @xmath380 and @xmath456 are in the same component of @xmath24 , @xmath380 and @xmath457 are in the same component . hence @xmath458 . for the reverse inclusion , suppose @xmath459 . then there exists a path from @xmath380 to @xmath457 . therefore , @xmath456 and @xmath457 are in the same component . since @xmath460 , @xmath380 and @xmath382 are in the same component . using the above lemma , we define a new algorithm ` reach ` on inputs @xmath286 , @xmath334 by replacing line ( 4 ) in ` navereach ` with \(4 ) if @xmath286 and @xmath334 belong to infinite components , then compute @xmath461 . if @xmath462 for @xmath463 such that @xmath464 , return _ yes _ ; otherwise , return _ no_. say input vertices are given as @xmath286 and @xmath334 . by lemma [ lm : reach_closure ] and lemma [ lm : reach_repeat ] , the algorithm ` reach ` returns _ yes _ if and only if @xmath286 and @xmath334 are in the same component . since @xmath460 , calculating @xmath461 requires time @xmath465 . therefore the running time of ` reach ` on input @xmath286 , @xmath334 is @xmath466 . notice that , in fact , the algorithm produces a number @xmath467 such that in order to check if @xmath286 , @xmath334 ( @xmath468 ) are in the same component , we need to test if @xmath469 and if @xmath470 . therefore if @xmath24 is fixed and we compute @xmath471 for all @xmath14 beforehand , then deciding whether two vertices @xmath33 , @xmath30 belong to the same component takes linear time . the above proof can also be used to build an automaton that decides reachability uniformly : [ cr : reach_aut ] given a unary automatic graph of finite degree @xmath24 represented by an automaton with loop constant @xmath114 , there is a deterministic automaton with at most @xmath472 states that solves the reachability problem on @xmath24 . the time required to construct this automaton is @xmath473 . for all @xmath474 , @xmath475 , let string @xmath476 represent vertex @xmath286 in @xmath24 . suppose @xmath477 , we construct an automaton @xmath31 that accepts @xmath478 if and only if @xmath286 and @xmath334 are in the same component in @xmath24 . 1 . @xmath31 has a @xmath110-tail of length @xmath479 . let the states on the tail be @xmath480 , where @xmath53 is the initial state . these states represent vertices in @xmath481 . 2 . from @xmath482 , there is a @xmath110-loop of length @xmath114 . we call the states on the loop @xmath483 . these states represent vertices in @xmath484 . 3 . for @xmath485 , there is a @xmath118-tail from @xmath486 of length @xmath487 . we denote the states on this tail by @xmath488 . these states represent vertices in @xmath489 . 4 . for @xmath490 , if @xmath286 is in an infinite component , then there is a @xmath118-loop of length @xmath491 from @xmath492 . the states on this loop are called @xmath493 . these states represent vertices in @xmath494 . 5 . for @xmath495 , if @xmath380 is in a finite component , then there is a @xmath118-tail from @xmath496 of length @xmath479 . these states are denoted @xmath497 and represent vertices in @xmath498 . if @xmath380 is in an infinite component , from @xmath496 , there is a @xmath118-loop of length @xmath491 . we write these states as @xmath499 . the final ( accepting ) states of @xmath31 are defined as follows : 1 . states @xmath500 are final . 2 . for @xmath385 , if @xmath286 is in a finite component , run the algorithm ` finitereach ` on input @xmath286 and declare state @xmath501 final if @xmath502 . 3 . for @xmath385 , if @xmath286 is in an infinite component , compute @xmath503 . 1 . make state @xmath501 final if @xmath504 is in an infinite component and @xmath505 . 2 . make state @xmath506 final if @xmath505 4 . if @xmath380 is in a finite component , run the algorithm ` finitereach ` on input @xmath380 and make state @xmath507 final if @xmath502 . if @xmath380 is in an infinite component , compute @xmath503 . declare state @xmath508 final if @xmath505 . one can show that @xmath31 is the desired automaton . to compute the complexity of building @xmath31 , we summarize the computation involved . 1 . for all @xmath286 in @xmath509 , decide whether @xmath286 is in a finite component . this takes time @xmath473 by theorem [ thm : inftest ] . 2 . for all @xmath286 in @xmath509 such that @xmath286 is in a finite component , run ` finitereach ` on input @xmath286 . this takes time @xmath473 by corollary [ cr : finite reach ] . 3 . for all @xmath360 such that @xmath380 is in an infinite component , compute the sets @xmath510 @xmath511 . this requires time @xmath473 by theorem [ thm : reachability ] . therefore the running time required to construct @xmath31 is @xmath473 . finally , we present a solution to the * connectivity problem * on unary automatic graphs of finite degree . recall a graph is * connected * if there is a path between any pair of vertices . the construction of @xmath31 from the last section suggests an immediate solution to the connectivity problem . ` alg : naveconnect ` 1 . construct the automaton @xmath31 . 2 . check if all states in @xmath31 are final states . if it is the case , return @xmath367 ; otherwise , return @xmath512 . the above algorithm takes time @xmath473 . note that @xmath31 provides a uniform solution to the reachability problem on @xmath24 . given the `` regularity '' of the class of infinite graphs we are studying , it is reasonable to believe there is a more intuitive algorithm that solves the connectivity problem . it turns out that this is the case . [ thm : connectivity ] the connectivity problem for unary automatic graph of finite degree @xmath24 is solved in @xmath332 , where @xmath114 is the loop constant of the automaton recognizing @xmath24 . observe that if @xmath24 does not contain an infinite component , then @xmath24 is not connected . therefore we suppose @xmath24 contains an infinite component @xmath513 . [ lm : infinite - connect ] for all @xmath514 , there is a vertex in @xmath257 belonging to @xmath513 . since @xmath513 is infinite , there is a vertex @xmath286 and @xmath515 such that all vertices in @xmath516 belong to @xmath513 and @xmath76 is the least such number . by minimality , @xmath517 . take a walk along the path from @xmath518 to @xmath256 . let @xmath519 be the first vertex in @xmath520 that appears on this path . it is easy to see that @xmath521 must also be in @xmath513 . therefore @xmath513 has a non - empty intersection with each copy of @xmath205 in @xmath24 . pick an arbitrary @xmath522 and run ` finitereach ` on @xmath523 to compute the queue @xmath345 . set @xmath524 . suppose @xmath24 contains an infinite component , then @xmath24 is connected if and only if @xmath525 . suppose there is a vertex @xmath526 . then there is no path in @xmath24 between @xmath523 to @xmath521 . otherwise , we can shorten the path from @xmath523 to @xmath521 using an argument similar to the proof of lemma [ lm : reach_reach ] , and show the existence of a path between @xmath523 to @xmath521 in the subgraph restricted on @xmath527 . therefore @xmath24 is not connected . conversely , if @xmath528 , then every set of the form @xmath529 for @xmath530 equals @xmath290 . by lemma [ lm : infinite - connect ] , all vertices are in the same component . by the above lemma the following algorithm decides the connectivity problem on @xmath268 : ` alg : connectivity ` 1 . use the algorithm proposed by theorem [ thm : infinite component ] to decide if there is an infinite component in @xmath268 . if there is no infinite component , then stop and return @xmath512 . 2 . pick an arbitrary @xmath522 , run ` finitereach ` on @xmath523 to compute the queue @xmath345 . 3 . let @xmath531 . if @xmath532 , return @xmath367 ; otherwise , return @xmath512 . solving the infinite component problem takes @xmath332 by theorem [ thm : infinite component ] . running algorithm ` finitereach ` also takes @xmath332 . therefore ` connectivity ` takes @xmath332 . in this paper we addressed algorithmic problems for graphs of finite degree that have automata presentations over a unary alphabet . we provided polynomial - time algorithms that solve connectivity , reachability , infinity testing , and infinite component problems . in our future work we plan to improve these algorithms for other stronger classes of unary automatic graphs . we also point out that there are many other algorithmic problems for finite graphs that can be studied for the class of unary automatic graphs . these , for example , may concern finding spanning trees for automatic graphs , studying the isomorphism problems , and other related issues . j. r. bchi , _ on a decision method in restricted second - order arithmetic_. proc . international congress on logic , methodology and philosophy of science ( e. nagel , p. suppes , a. tarski , eds . ) , stanford university press , 1 - 11 , 1960 .

one of the fundamental theorems of uniform distribution theory states that the fractional parts of the sequence are uniformly distributed modulo one ( u.d . mod 1 ) for every irrational number . another important result of weyl states that for every sequence of distinct positive integers the sequence of fractional parts of is u.d . mod 1 for almost all . however , in this general case it is usually extremely difficult to classify those for which uniform distribution occurs , and to measure the speed of convergence of the empirical distribution of towards the uniform distribution . in the present paper we investigate this problem in the case when is the thue morse sequence of integers , which means the sequence of positive integers having an even sum of digits in base 2 . in particular we utilize a connection with lacunary trigonometric products , and by giving sharp metric estimates for such products we derive sharp metric estimates for exponential sums of and for the discrepancy of furthermore , we comment on the connection between our results and an open problem in the metric theory of diophantine approximation , and we provide some explicit examples of numbers for which we can give estimates for the discrepancy of .

1502.06738

throughout the rest of this paper , let @xmath2 denote the sequence of positive integers which have an even sum - of - digits function in base @xmath9 , sorted in increasing order . in other words , @xmath2 is the sequence of thue morse integers @xmath10 . furthermore , we write @xmath11 for the sequences of those numbers which are _ not _ contained in @xmath2 , sorted in increasing order ; thus @xmath12 . the numbers @xmath2 are frequently called _ evil numbers _ , while the numbers @xmath11 are called _ odious numbers_. + the thue morse integers are characterized by the thue morse sequence @xmath13 which has been discovered several times in the literature . for an extensive survey , see @xcite . in our notation we have @xmath14 if and only if @xmath15 + in this paper we analyze exponential sums of the form @xmath16 for reals @xmath17 , and what is intimately connected products of the form @xmath18 , as well as distribution properties of the sequence @xmath19 . to quantify the regularity of the distribution of a finite set of real numbers in @xmath20 $ ] we use the notion of the _ star - discrepancy _ @xmath21 . for given numbers @xmath22 , their star - discrepancy is defined by @xmath23 } \left| \frac{1}{n } \sum_{n=1}^n \mathbf{1}_{[0,a ] } ( x_n ) - a \right|.\ ] ] an infinite sequence @xmath24 whose discrepancy @xmath21 tends to zero as @xmath25 is called _ uniformly distributed modulo one _ ( u.d . mod 1 ) . informally speaking , the star - discrepancy is a measure for the deviation between uniform distribution on @xmath20 $ ] and the empirical distribution of a given point set ; in probabilistic terminology this corresponds to the kolmogorov smirnov statistic . discrepancy theory is a rich subject , which has close links to number theory , probability theory , ergodic theory and numerical analysis . for more information on discrepancy theory , we refer to the standard monographs @xcite . + sequences of the form @xmath26 are called _ kronecker sequences_. one of the fundamental results of discrepancy theory states that such a sequence is u.d . mod 1 if and only if @xmath1 is irrational . it is also well - known that the discrepancy of such a sequence depends on diophantine approximation properties of @xmath1 . more precisely , we have @xmath27 as @xmath25 , where @xmath28 are the continued fraction coefficients of @xmath1 and where @xmath29 is defined by @xmath30 with @xmath31 denoting the best approximation denominators of @xmath1 ( see for example ( * ? ? ? * corollary 1.64 ) ) . hence @xmath32 if @xmath1 has bounded continued fraction coefficients , and , as a consequence of metric results of khintchine @xcite , for every @xmath33 we have @xmath34 for almost all @xmath35 . + + it is known ( and it will be re - proved implicitly in this paper ; see section 6 ) that the sequence @xmath36 ( which we will call _ thue morse kronecker sequence _ ) is also uniformly distributed in the unit interval if and only if @xmath1 is irrational . however , it turns out to be a very difficult task to give sharp estimates for the discrepancy of this sequence for concrete values of @xmath1 . as we will see , the discrepancy of a thue morse kronecker sequence @xmath37 depends on diophantine approximation properties _ and _ properties of the digit representation of @xmath1 in base 2 . until now there are only few ( non - trivial ) cases of @xmath1 where we have enough information about both of these aspects . + exponential sums and discrepancy theory are intimately connected . one such connection is weyl s criterion , two others are the erds turn inequality and koksma s inequality . the erds turn inequality ( see for example @xcite ) states that for points @xmath38 $ ] we have @xmath39 where @xmath40 is an arbitrary positive integer . koksma s inequality says that @xmath41 } f ) d_n^*(x_1 , \dots , x_n),\ ] ] for any function @xmath42 having bounded variation on @xmath20 $ ] . when combined , the erds turn inequality and koksma s inequality show that exponential sums can be used to obtain both upper and lower bounds for the discrepancy . + as an explicit lower bound from we get ( compare for example @xcite ) : @xmath43 where @xmath40 is an arbitrary positive integer . koksma s inequality and its multi - dimensional generalization are also the cornerstone of the application of low - discrepancy point sets in numerical integration ( so - called _ quasi - monte carlo integration _ ; see for example @xcite ) . + consequently , in this paper we will mainly be concerned with the problem of investigating exponential sums of the form @xmath16 . it turns out that this investigation relies on studying lacunary products of the form @xmath18 . furthermore we study the discrepancy of @xmath36 . for all three topics we obtain sharp metric results . the investigation of the lower bound for the discrepancy leads to a challenging open problem in diophantine approximation . finally , we consider two concrete non - trivial special examples for @xmath1 . + the main results of this paper are the following . ( throughout the rest of this paper , we write @xmath44 for @xmath45 . ) + [ th_thue ] let @xmath2 be the sequence of thue morse integers , and let @xmath46 be an integer . let @xmath47 be arbitrary . then for almost all @xmath48 we have @xmath49 for all @xmath50 , and @xmath51 for infinitely many @xmath52 . + note that the exponential function in grows more slowly than any ( fixed ) power of @xmath52 ; but faster than any ( fixed ) power of @xmath53 . in other words , as a consequence of theorem [ th_thue ] for every @xmath47 and every @xmath54 we have @xmath55 for almost all @xmath1 . + it will turn out that theorem [ th_thue ] is an almost immediate consequence of the following result on lacunary trigonometric products . + [ th_theorem2 ] let @xmath56 be arbitrary . then for almost all @xmath57 we have @xmath58 for all @xmath59 , and @xmath60 for infinitely many @xmath61 . + this result is a consequence of a more general result , theorem [ lac_co1 ] , which will be formulated later in section [ sect_lac ] since it needs some technical prerequisites . + from the lower bound in theorem [ th_thue ] and formula we immediately obtain a metric lower bound for the discrepancy @xmath62 of the thue kronecker sequence . however in theorem [ th_thue2 ] it turns out that the true metric order of the discrepancy @xmath62 is much larger . + [ th_thue2 ] let @xmath63 be the sequence of thue morse integers . let @xmath47 be arbitrary . then for almost all @xmath57 we have @xmath64 and @xmath65 for infinitely many @xmath52 . here @xmath66 is a real constant defined below for which it is known that @xmath67 + the number @xmath66 in theorem [ th_thue2 ] appears in a result of fouvry and mauduit @xcite , which states that @xmath68 for @xmath69 , with constants @xmath70 and @xmath66 with @xmath71 in lemma [ lem7 ] , which is contained in section [ sect_proof_th2 ] , we will improve the estimate for @xmath66 to . note that as a consequence of and we have @xmath72 for almost all @xmath1 . this should be compared with the general metric discrepancy bound @xmath73 for almost all @xmath1 , which holds for every strictly increasing sequence of positive integers @xmath74 ( see @xcite ) . it is known that in the general setting the upper bound given by is optimal ( up to powers of logarithms ; see @xcite ) . thus the upper bound given in theorem [ th_thue2 ] is significantly stronger than the general metric discrepancy bound given by . furthermore we want to emphasize the fact that the precision of theorem [ th_thue2 ] is quite remarkable , in view of the fact that good bounds for the typical order of the discrepancy are only known for a very small number of classes of parametric sequences . + one of the main objectives of theorems [ th_thue ] and [ th_thue2 ] is to examine the degree of pseudorandomness of the parametric sequences @xmath75 , and consequently also of the thue morse sequence @xmath2 of integers itself . by classical probability theory , for a sequence @xmath76 of independent , identically distributed ( i.i.d . ) random variables having uniform distribution on @xmath20 $ ] we have the law of the iterated logarithm ( lil ) @xmath77 and the chung smirnov lil for the kolmogorov smirnov statistic ( that is , for the discrepancy ) @xmath78 in other words , for a random sequence of points exponential sums are typically of asymptotic order roughly @xmath79 , and the discrepancy is typically also of the corresponding asymptotic order . furthermore , similar results usually hold for exponential sums of @xmath80 and for the discrepancy of @xmath81 when @xmath82 is a `` random '' increasing sequence of integers . in the simplest model , when for every number @xmath83 we decide independently and with fair probability whether it should be contained in @xmath82 or not , then holds almost surely ( with respect to the probability space over which the @xmath84 s are defined ) for almost all @xmath1 . in a similar fashion both results and essentially remain valid when the random sequence @xmath82 is constructed in a more complicated fashion ( see for example @xcite ) . + thus theorem [ th_thue ] and theorem [ th_thue2 ] show that the typical we mean a result which is valid for a set of full lebesgue measure . ] asymptotic order of exponential sums and of the discrepancy of @xmath75 for the thue morse integers @xmath2 does _ not _ match with the corresponding order in the random case , by this means showing an interesting deviation from `` pseudorandom '' behavior of the sequence @xmath2 itself . on the other hand , the behavior of @xmath75 also does _ not _ match with the behavior of @xmath85-sequences for typical values of @xmath1 . more precisely , as already mentioned above , as a consequence of metric results of khintchine @xcite and due to the fact that the discrepancy of @xmath86 can be expressed in terms of the continued fractions expansion of @xmath1 , we have @xmath87 for almost all @xmath1 . consequently , by theorem [ th_thue2 ] , the typical asymptotic order of the discrepancy of parametric sequences @xmath75 is significantly larger than that of typical @xmath85-sequences , and by theorem [ th_thue ] this is also true for exponential sums . thus , with respect to exponential sums as well as with respect to the discrepancy , parametric sequences @xmath3 generated by the thue morse integers @xmath2 occupy a position somewhere between @xmath85-sequences and truly random sequences . we also want to comment on the fact that there is a huge difference between the order of the exponential sums in theorem [ th_thue ] and the order of the discrepancy in theorem [ th_thue2 ] . this is a very surprising phenomenon , which is related to problems from metric diophantine approximation ( which are implicit in the proof of theorem [ th_thue2 ] , and are briefly discussed in the concluding section [ sec_conc ] ) . + as already mentioned earlier , it is rather difficult to give the right order for the exponential sums in theorem [ th_thue ] , the trigonometric products in theorem [ th_theorem2 ] , and the discrepancy of @xmath19 for concrete non - trivial examples of @xmath1 . what do we mean by a `` non - trivial '' example ? in the first part of section [ sect_proof_th4 ] we will point out the following facts : * the order of the discrepancy of the pure kronecker sequence @xmath88 never is significantly larger than the order of the discrepancy of the thue kronecker sequence @xmath19 . * if the order of the discrepancy @xmath89 of the pure kronecker sequence satisfies @xmath90 then the discrepancy @xmath91 of the thue kronecker sequence is essentially of the same order as the discrepancy of the pure kronecker sequence . * if the order of the discrepancy @xmath89 of the pure kronecker sequence satisfies @xmath92 then @xmath91 satisfies @xmath93 . + thus an `` interesting non - trivial '' example means for us an example where @xmath1 is a `` natural '' real number such as @xmath94 ( it seems to us that there is no chance to handle these numbers since we do not have enough information on their digit representation ) , or where @xmath1 is such that @xmath91 and hence @xmath89 is small ( say @xmath95 however we can not give such examples ) or where the quality of the distribution of the sequences @xmath26 and @xmath19 differ strongly . two such examples are given in theorem [ th_thue4 ] . especially in the first example the difference between @xmath89 and @xmath91 is of the maximal possible form . + [ th_thue4 ] 1 . let @xmath96 . then for the star - discrepancy @xmath97 of the pure kronecker sequence @xmath26 we have @xmath98 whereas for the star - discrepancy @xmath91 of the thue kronecker sequence @xmath19 we have @xmath99 @xmath100 for every @xmath101 2 . let @xmath102 the thue morse real in base 2 . then for the star - discrepancy @xmath97 of the pure kronecker sequence @xmath103 we have @xmath104 whereas for the star - discrepancy @xmath91 of the thue kronecker sequence @xmath105 we have @xmath106 + we would like to point out here that there is an intimate connection between distribution properties of @xmath19 and of certain types of hybrid sequences . for some information on the analysis of hybrid sequences see for example @xcite , @xcite and @xcite . + as already mentioned , the proofs of theorems [ th_thue ] and [ th_thue2 ] are based on a connection between exponential sums of @xmath3 and the lacunary trigonometric products studied in theorem [ th_theorem2 ] . we will establish this connection in the following lines , and exploit it in section [ sect_lac ] in more detail . for the time being , we assume that @xmath52 is of the form @xmath107 for some positive integer @xmath61 . + to analyze the exponential sums appearing in theorem [ th_thue ] and on the right - hand side of , we define @xmath108 by the assumption that @xmath109 we have @xmath110 which yields @xmath111 and @xmath112 a similar analysis for the sequence @xmath11 shows that @xmath113 where again we assume that @xmath109 . + by taking logarithms , we can convert the trigonometric products appearing in and into so - called _ lacunary sums _ ; these sums have been intensively investigated in fourier analysis , and a wide range of mathematical methods is available for studying them ( see the following section [ sect_lac ] ) . thus the theory of lacunary sums allows us to obtain an estimate for the size of the exponential sums @xmath114 in the case when @xmath52 is a power of 2 ; however , it will turn out that we may also drop the condition that @xmath52 is an integral power of @xmath9 by applying a dyadic decomposition method . + note that by and by the fact that the thue morse integers have asymptotic density 1/2 it is easy to show that all the conclusions of theorem [ th_thue ] and theorem [ th_thue2 ] remain valid if we replace the sequence @xmath2 by the sequence @xmath11 ( of those numbers which are _ not _ thue morse integers ) . + the outline of the remaining part of this paper is as follows . in section [ sect_lac ] we explain the main principles of the theory of lacunary ( trigonometric ) sums , and state several lemmas as well as theorem [ lac_co1 ] , which we require for the proofs of theorem [ th_thue ] and theorem [ th_theorem2 ] . in section [ sect_lac2 ] we give the proofs for the results stated in section [ sect_lac ] , and in section [ sect_proof ] we give the proofs of theorem [ th_thue ] and theorem [ th_theorem2 ] . in section [ sect_proof_th2 ] , we prove theorem [ th_thue2 ] , and in section [ sect_proof_th4 ] we prove theorem [ th_thue4 ] . finally , in section [ sec_conc ] , we briefly mention a problem from metric diophantine approximation , which was posed by leveque in @xcite and is related to the proof of theorem [ th_thue2 ] . it is a well - known fact that so - called _ lacunary _ systems of trigonometric functions , that is , systems of the form @xmath115 or @xmath116 for rapidly increasing @xmath117 , exhibit properties which are typical for sequences of independent random variables . this similarity includes the central limit theorem , the law of the iterated logarithm , and kolmogorov s `` three series '' convergence theorem . the situation is particularly well understood when @xmath117 satisfies the _ hadamard gap condition _ @xmath118 to a certain degree this almost - independence property extends to systems @xmath119 for a function @xmath42 which is periodic with period one and satisfies certain regularity properties ; however , in this case the number - theoretic properties of @xmath117 play an important role , and the almost - independent behavior generally fails when is relaxed to a weaker growth condition . the case which has been investigated in the greatest detail is that when @xmath42 has bounded variation on @xmath20 $ ] , since this case is ( by koksma s inequality ) closely connected to the discrepancy of the sequence of fractional parts @xmath120 , which in turn can be interpreted as the ( one - sided ) kolmogorov smirnov statistic adopted to the case of the uniform measure on @xmath20 $ ] . + to estimate the trigonometric products appearing in and we will use the equalities @xmath121 and @xmath122 respectively , to transform the problem of lacunary trigonometric products into a problem concerning lacunary sums . however , the functions @xmath123 and @xmath124 do _ not _ have bounded variation in @xmath20 $ ] ( see the figures below ) . consequently , the known results are not applicable in this situation , and we have to adopt the proof techniques in such a way that they can handle this kind of problem . is neither required to have bounded variation , nor to be lipschitz- or hlder - continuous , nor to have a modulus of continuity of a certain regularity ; see for example @xcite . however , for these results the growth requirements for @xmath117 are much stronger than , which means that they are not applicable in our case , since by and we have to deal with lacunary sequences growing exactly with the speed presumed in , and not faster . ] + + figure 1 : the functions @xmath125 ( left ) and @xmath126 ( right ) . in the following we will assume that @xmath42 is a measurable real function satisfying @xmath127 and that @xmath117 is a sequence satisfying . we write @xmath128 for the fourier series of @xmath42 , and we will assume that the fourier coefficients of @xmath42 satisfy latexmath:[\[\label{f_coeff_b } the inequalities in line appear frequently in the theory of lacunary series , since the upper bound stated there describes precisely ( up to multiplication with a constant ) the maximal asymptotic order of the fourier coefficients of a function of bounded variation ( see for example @xcite ) . however , even if the function is _ not _ of bounded variation the estimates in may still be true ; this can be seen by the fact that for the ( unbounded ) functions @xmath125 and @xmath126 from lines and , respectively , we have @xmath130 and @xmath131 by the way , we note that both @xmath125 and @xmath126 are even functions , and that both of them satisfy . + throughout the remaining part of this paper , we will write @xmath132 for the lebesgue measure on the unit interval . note that the unit interval , equipped with borel sets and lebesgue measure , is a probability space , and that accordingly every measurable function on @xmath20 $ ] can be seen as a random variable over this probability space . we also write @xmath133 for the @xmath134 norm and @xmath135 for the supremum norm of a function , respectively . + the main technical tool in this section is the following exponential inequality ( lemma [ lac_lemma1 ] ) . together with the subsequent lemmas it will allow us to give an upper bound for the measure of those @xmath1 for which and are large ( stated in lemma [ lac_lemma3 ] ) . we state lemmas [ lac_lemma1][lac_lemma3 ] in a slightly more general form than necessary for the proofs of theorem [ th_thue ] and [ th_theorem2 ] , and we will use them to prove an additional new theorem , namely theorem [ lac_co1 ] below . [ lac_lemma1 ] assume that @xmath42 is an even measurable function satisfying , whose fourier coefficients satisfy . furthermore , let @xmath117 be a sequence of positive integers satisfying for some number @xmath136 . then there exists a number @xmath137 , such that the following holds . let @xmath138 be given , and write @xmath139 for the @xmath140-th partial sum of the fourier series of @xmath42 . then for all @xmath141\ ] ] we have @xmath142 and @xmath143 the same two conclusions hold if @xmath42 is an odd function instead of an even function . we emphasize the fact that the number @xmath144 in the statement of lemma [ lac_lemma1 ] depends _ only _ on the growth parameter @xmath145 ; it does not depend on the function @xmath42 or the sequence @xmath117 . the same will be true for the numbers @xmath146 in lemmas [ lac_lemma1a ] and [ lac_lemma3 ] below . + from lemma [ lac_lemma1 ] we will deduce the following lemma [ lac_lemma1a ] , which is a large deviations bound for the maximal partial sum of a lacunary sums . [ lac_lemma1a ] let @xmath42 and @xmath117 be as in lemma [ lac_lemma1 ] . then there exists a number @xmath137 , such that the following holds . let @xmath138 be given , and assume that @xmath61 is an integral power of 2 . write @xmath139 for the @xmath140-th partial sum of the fourier series of @xmath42 . then we have @xmath147 and , under the additional assumption that @xmath148 , we also have @xmath149 the same two conclusions hold if @xmath42 is an odd function instead of an even function . [ lac_lemma2 ] let @xmath42 and @xmath117 be as in lemma [ lac_lemma1 ] . let @xmath61 be given , and write @xmath150 for the remainder term of the @xmath140-th partial sum of the fourier series of @xmath42 . then we have @xmath151 from lemmas [ lac_lemma1a ] and [ lac_lemma2 ] we will deduce the following lemma [ lac_lemma3 ] . [ lac_lemma3 ] let @xmath42 and @xmath117 be as in lemma [ lac_lemma1 ] . then there exists a number @xmath137 , such that the following holds . let @xmath138 be given , and assume that @xmath61 is an integral power of 2 . then we have @xmath152 and , under the additional assumption that @xmath148 , we also have @xmath153 the same two conclusions hold if @xmath42 is an odd function instead of an even function . as a consequence of lemma [ lac_lemma3 ] we obtain the following theorem , which is a bounded law of the iterated logarithm and is of some interest in its own right . as far as we know , this is the first law of the iterated logarithm for hadamard lacunary function series which can be applied to a class of unbounded functions @xmath42 . + [ lac_co1 ] assume that @xmath42 is an even measurable function satisfying , whose fourier coefficients satisfy . furthermore , let @xmath117 be a sequence of positive integers satisfying for some number @xmath136 . then we have @xmath154 for almost all @xmath48 , where we can choose @xmath155 + we note in passing that from our proofs it seems that the conclusion of theorem [ lac_co1 ] remains true if the conditions @xmath156 in are relaxed to @xmath157 for some fixed @xmath33 ; however , in this case the constant @xmath158 has to be replaced by some other constant @xmath159 which may also depend on @xmath160 . we will not pursue this possible generalization any further in the present paper . + for the proofs of theorem [ th_thue ] and [ th_theorem2 ] we will also need the following result . it has first been stated by fortet @xcite ; a concise proof can be found in @xcite . this result can be seen as a special case of the more general results in @xcite . [ lac_lemma4 ] let @xmath42 be a function satisfying , which additionally satisfies a hlder continuity condition of order @xmath161 for some @xmath162 . then @xmath163 where @xmath164 the proof of lemma [ lac_lemma1 ] , as well as the proofs of lemmas [ lac_lemma1a ] , [ lac_lemma2 ] , [ lac_lemma3 ] and theorem [ lac_co1 ] , uses methods of takahashi @xcite and philipp @xcite . + assuming that @xmath42 is even , the @xmath140-th partial sum of the fourier series of @xmath42 is of the form @xmath165 where by assumption the coefficients @xmath166 satisfy the inequality on the left - hand side of . we note that implies that @xmath167 we divide the set of integers @xmath168 into blocks @xmath169 of consecutive numbers , for some appropriate @xmath170 , such that every block contains @xmath171 numbers ( the last block may contain less ) . is sufficiently large such that all appearing logarithms are well - defined and positive . ] more precisely , we set @xmath172 and @xmath173 we set @xmath174 and @xmath175 then by the cauchy schwarz inequality we have @xmath176 writing @xmath177 and using the inequality @xmath178 we have @xmath179 where we used the fact that by , and we have @xmath180 for @xmath181 + using the classical trigonometric identity @xmath182 we have @xmath183 here we write @xmath184 for the sum of all those cosine - functions having frequencies in the interval @xmath185 $ ] , where @xmath186 and @xmath187 denote the smallest resp . largest element of @xmath188 , and we write @xmath189 for the sum of those cosine - functions having frequencies smaller than @xmath190 . it is easy to check that no other frequencies can occur in . we note that all the frequencies of the cosine - functions in @xmath191 are also contained in the interval @xmath192 $ ] , and write @xmath193 using this notation we have @xmath194 from minkowski s inequality and we deduce that @xmath195 another way of continuing from line is to use the cauchy schwarz inequality , which leads to @xmath196 now assume that @xmath197 are two indices from the set @xmath198 , and that both @xmath199 and @xmath200 are even . then by construction the frequency of any trigonometric function in @xmath201 is at least twice as large as the frequency of any trigonometric function in @xmath202 . to see why this is the case , we recall that the frequency of the largest trigonometric function in @xmath201 is at most @xmath203 , that the frequency of the smallest trigonometric function in @xmath202 is at least @xmath204 , and that by @xmath205 as a consequence for every set of distinct indices @xmath206 ( where the cardinality @xmath207 is arbitrary ) , all of which are even and are contained in @xmath198 , the functions @xmath208 are orthogonal , i.e. @xmath209 ( this argument is explained in more detail in @xcite ) . thus by , , , and we have @xmath210 in the same way we can get an upper bound for @xmath211 , and thus by we finally obtain @xmath212 this proves the first conclusion of the lemma . in the same way we can use ( and the corresponding upper bound for @xmath211 ) to obtain @xmath213 which proves the second conclusion of the lemma . + thus we have proved both parts of lemma [ lac_lemma1 ] in the case when @xmath42 is even ; the proof in the odd case can be carried out in exactly the same way . + by assumption @xmath61 is an integral power of 2 . we set @xmath214 . by classical dyadic decomposition , we can write every subset @xmath215 of @xmath168 as the disjoint sum of at most one set of cardinality @xmath216 , at most one set of cardinality @xmath217 , at most one set of cardinality @xmath218 , and so on , at most one set of cardinality @xmath219 , and additionally at most one set of cardinality _ at most _ @xmath219 , where all these sets contain consecutive positive integers . to be able to represent every sets @xmath215 in this way , we need @xmath220 sets of cardinality @xmath221 , for @xmath222 , and all the sets of cardinality at most @xmath219 starting at an integer multiple of @xmath219 . more precisely , the sets of cardinality @xmath223 are of the form @xmath224 and the sets of cardinality at most @xmath225 are of the form @xmath226 for @xmath227 and @xmath228 , we set @xmath229 using the first part of lemma [ lac_lemma1 ] with @xmath230 ( note that by our construction this value of @xmath66 is admissible in lemma [ lac_lemma1 ] , provided that @xmath61 is sufficiently large ) we have @xmath231 and consequently @xmath232 thus we have @xmath233 now we set @xmath234 by , minkowski s inequality , and the carleson hunt inequality ( see for example @xcite ) , we have @xmath235 for some absolute constant @xmath236 . estimating the integral in can be reduced ( via ) to the problem of counting the number of solutions @xmath237 of the diophantine equation @xmath238 we have @xmath239 for some constant @xmath158 depending only on @xmath145 ( see , for example , @xcite ) . as a consequence by markov s inequality we obtain @xmath240 and @xmath241 for some constant @xmath242 depending only on @xmath145 . we set @xmath243 then by and we have @xmath244 for sufficiently large @xmath61 . + by the dyadic decomposition described at the beginning of this proof , for every @xmath245 ( where @xmath246 denotes the complement of the set @xmath247 ) we have @xmath248 for all possible values @xmath249 . thus we have shown that @xmath250 for sufficiently large @xmath61 . note that whenever the function @xmath42 satisfies the assumptions of lemma [ lac_lemma1 ] , then the function @xmath251 also satisfies these assumptions . thus applying exactly the same arguments as above to the functions @xmath251 and @xmath252 instead of @xmath42 and @xmath253 we also obtain @xmath254 for sufficiently large @xmath61 , which , together with , proves the first conclusion of lemma [ lac_lemma1a ] . + the proof of the second conclusions of lemma [ lac_lemma1a ] is very similar to that of the first conclusion of the lemma . we use the same dyadic decomposition , but now we define @xmath255 and use the second part of lemma [ lac_lemma1 ] with @xmath256 note that this choice of @xmath66 is admissible , due to the restrictions that @xmath257 and @xmath148 . we obtain that @xmath258 and consequently @xmath259 the sets @xmath260 can be defined in the same way as in the proof of the first part of the lemma . using similar calculations we obtain that @xmath261 except for a set of measure at most @xmath262 ( provided that @xmath61 is sufficiently large ) . this proves the second part of lemma [ lac_lemma1a ] . + the lemma follows from a simple application of minkowski s inequality . we have @xmath263 by we have @xmath264 which implies , together with , that @xmath265 this proves the lemma . + assume that we have decomposed @xmath266 as in lemmas [ lac_lemma1 ] , [ lac_lemma1a ] and [ lac_lemma2 ] . by lemma [ lac_lemma2 ] and markov s inequality we have @xmath267 together with lemma [ lac_lemma1a ] this yields @xmath268 for sufficiently large @xmath61 , which is the first part of lemma [ lac_lemma3 ] . + in the same way we can deduce the second conclusion of lemma [ lac_lemma3 ] from a combination of the second conclusion of lemma [ lac_lemma1a ] and lemma [ lac_lemma2 ] . + from lemma [ lac_lemma1a ] and lemma [ lac_lemma2 ] we can easily deduce theorem [ lac_co1 ] , using standard methods . let us first assume that @xmath42 is either even or odd . for @xmath269 , let @xmath270 denote the sets defined by @xmath271 then by lemma [ lac_lemma4 ] we have @xmath272 which implies that @xmath273 thus by the borel cantelli lemma with probability 1 only finitely many events @xmath270 happen ; in other words , for almost all @xmath48 we have @xmath274 as a consequence for almost all @xmath48 we have @xmath275 which proves theorem [ lac_co1 ] in the case when @xmath42 is either even or odd . for general @xmath42 we apply to the even and odd part separately , which results in an additional multiplicative factor of 2 . note that @xmath276 . + in the same way we can use the second conclusion of lemma [ lac_lemma3 ] to obtain @xmath277 for almost all @xmath48 . theorem [ lac_co1 ] now follows from a combination of and . + theorem [ th_thue ] and theorem [ th_theorem2 ] will follow easily from the following lemma , which is a version of lemma [ lac_lemma4 ] in the case when the function @xmath42 only satisfies ( instead of being hlder - continuous ) . we state it only for the special case of the two functions @xmath125 and @xmath126 from and . [ lemmath1 ] we have @xmath278 and @xmath279 for almost all @xmath1 . as already noted , the functions @xmath125 and @xmath126 satisfy conditions and . let a number @xmath280 be given , and write @xmath281 for the @xmath280-th partial sum of the fourier series of @xmath125 , and @xmath282 for the remainder term . then by lemma [ lac_lemma4 ] we have @xmath283 where @xmath284 is defined according to ( for the function @xmath281 ) . note that lemma [ lac_lemma4 ] is applicable since @xmath253 is a trigonometric polynomial ( and consequently also is lipschitz - continuous ) . note furthermore that by we have @xmath285 consequently by theorem [ lac_co1 ] we have @xmath286 replacing @xmath125 by @xmath126 and replacing @xmath281 and @xmath282 by @xmath287 and @xmath288 , respectively , we obtain and with @xmath281 and @xmath282 replaced by @xmath287 and @xmath288 , respectively . some standard calculations show that @xmath289 furthermore , the expression on the right - hand side of clearly tends to zero as @xmath290 . thus , overall we have @xmath291 and @xmath292 where @xmath293 and @xmath294 are defined according to . calculating the values of @xmath293 and @xmath294 is a simple exercise , using the fourier series expansion of @xmath125 and @xmath126 , respectively ; it turns out that in our specific setting we have @xmath295 and @xmath296 by applying the same arguments to @xmath297 and @xmath298 instead of @xmath125 and @xmath126 we can get absolute values in and , if we wish . this proves lemma [ lemmath1 ] . + + _ part 1 : upper bounds . _ by periodicity it is obviously sufficient to prove theorem [ th_thue ] for @xmath299 . let @xmath47 and @xmath1 be given , and set @xmath300 . we will assume that @xmath1 is an element of the set of full measure for which the conclusion of lemma [ lemmath1 ] holds . then we have @xmath301 and @xmath302 for all @xmath303 . equation already gives the upper bound in theorem [ th_theorem2 ] . to obtain the upper bound in theorem [ th_thue ] , let @xmath52 be given , and assume that @xmath304 . we can write @xmath305 for numbers @xmath306 , where we assume that @xmath307 is chosen in such a way that @xmath308 ; this is simply the binary representation of @xmath52 . for simplicity of writing we set @xmath309 and @xmath310 then clearly we have @xmath311 note that for the `` odious numbers '' @xmath312 we have , for every @xmath313 , that @xmath314 furthermore , from the special structure of the thue morse sequence we see that for @xmath315 we have @xmath316 which together with , , and implies that @xmath317 thus by and we have @xmath318 using the fact that the assumption @xmath304 implies that @xmath319 , and also using the inequalities and , we obtain @xmath320 and @xmath321 combining , and we obtain @xmath322 as a consequence we have @xmath323 for all sufficiently large @xmath52 . this proves the upper bound in theorem [ th_thue ] . + _ part 2 : lower bounds . _ now we prove the lower bound in theorems [ th_thue ] and [ th_theorem2 ] . again we assume that @xmath299 , that @xmath1 and @xmath47 are fixed , and that @xmath1 is from the set of full measure for which the conclusion of lemma [ lemmath1 ] holds . again we set @xmath300 . then by lemma [ lemmath1 ] there exist infinitely many values of @xmath61 for which both inequalities @xmath324 and @xmath325 hold simultaneously . the first of the two relations already gives the lower bound in theorem [ th_theorem2 ] . by and , we have @xmath326 thus for infinitely many @xmath61 we have @xmath327 consequently we also have @xmath328 for infinitely many @xmath52 . this proves the lower bound in theorem [ th_thue ] . for given @xmath329 , we set @xmath330 and @xmath331 integrals of this type have been studied in great detail in @xcite . for the integral @xmath332 it is proved there that @xmath333 where @xmath334 are positive constants with @xmath335 ( thereby improving an earlier result of minyan @xcite ) . + hence for every @xmath33 for l large enough we have @xmath336 we will improve the estimate given for @xmath66 by fouvry and mauduit in the following . [ lem7 ] let @xmath66 be defined as in . then @xmath337 by the formula above of equation @xmath338 in @xcite we have @xmath339 where @xmath340 and @xmath341 furthermore , it was shown in @xcite that the functions @xmath342 are symmetric around @xmath343 on @xmath344 $ ] , and that they are concave on @xmath344 $ ] . hence @xmath345}{\min } \phi_{j } ( \alpha)$ ] and @xmath346}{\max } \phi_{j } ( \alpha).\ ] ] let @xmath347 and @xmath348}{\min } q_{j } ( \alpha ) , m_{j } : = \underset{\alpha \in \left[0,1\right]}{\max } q_{j } ( \alpha).$ ] + note that @xmath349 of course also is symmetric around @xmath350 in @xmath344 $ ] . we have for every @xmath351 $ ] @xmath352 therefore @xmath353}{\max } q_{j } ( \alpha)\leq m_j$ ] . + analogously we obtain @xmath354 for all @xmath355 $ ] . altogether @xmath356 and @xmath357 , and therefore for every @xmath313 fixed we have @xmath358 similarly we get @xmath359 hence @xmath360 and consequently @xmath361 by considering the function @xmath362 , in the following we will prove that @xmath363 and @xmath364 . + + figure 2 : the function @xmath362 . wherever @xmath365 is differentiable we have @xmath366}{\max } \left|{\phi_{6 } } ' ( \alpha)\right|}{\phi_{5}(0 ) } + \frac{\phi_{6 } \left(\frac{1}{2}\right)}{\left(\phi_{5 } ( 0)\right)^{2 } } \underset{\alpha \in \left[0,1\right]}{\max } \left|{\phi_{5}}'(\alpha)\right| . \label{phiequ}\end{aligned}\ ] ] it can easily be checked for example by differentiating @xmath367 and @xmath368 with the help of mathematica that @xmath369 is the sum resp . difference of + 32 products of absolute values of sines and cosines , each product weighted by a factor @xmath370 and further + 32 such products weighted by @xmath371 , further + 32 with weight @xmath372 , further + 32 with weight @xmath373 and finally further + 32 products with factor @xmath374 . + hence @xmath375 @xmath376 in the same way we show that also @xmath377 . by combining these estimates with the values of @xmath378 and @xmath379 in , we finally obtain @xmath380 . + now we calculate @xmath381 with the help of mathematica and obtain @xmath382 @xmath383 hence @xmath384 and @xmath385 by this implies lemma [ lem7 ] . let us remark that numerical experiments with @xmath386 suggest that @xmath387 it is tempting to conjecture that the precise value of @xmath66 can be expressed in a simple way in terms of the `` usual '' mathematical constants such as @xmath388 , etc . however , we do not know what such an expression could look like , and can not even make a reasonable guess ( the numerical argument in the proof of lemma [ lem7 ] does not give any hints ) . + + _ part 1 : upper bound _ + for the integral @xmath389 we can use the equality @xmath390 which holds for @xmath391 and which implies that @xmath392 ( this is essentially a variant of the classical bound for the @xmath393-norm of the dirichlet kernel ) . + for @xmath394 and @xmath33 we set @xmath395 by , and we have @xmath396 for sufficiently large @xmath397 . consequently we have @xmath398 which implies that by the borel cantelli lemma with probability one only finitely many events @xmath399 occur . we can show the same result if we replace the sequence @xmath2 in by @xmath11 . in other words , for almost all @xmath1 we have @xmath400 and the same estimate for @xmath11 instead of @xmath2 , for all @xmath401 . + now assume that @xmath402 and @xmath52 are given . furthermore we assume that for these values of @xmath1 and @xmath160 the estimate and the corresponding estimate for @xmath11 instead of @xmath2 hold for @xmath403 , and that @xmath404 . we apply the same dyadic decomposition of @xmath52 as in the proof of the upper bound of theorem [ th_thue ] in section [ sect_proof ] . in the same way as we obtained , together with the koksma - erds - turn inequality we can now obtain with @xmath405 @xmath406 for the last term in by and we have @xmath407 since @xmath408 , together with we have shown that @xmath409 for all sufficiently large @xmath52 , which proves the upper bound in theorem [ th_thue2 ] . + _ part 2 : lower bound . _ + by and for the discrepancy @xmath97 of the sequence @xmath410 with @xmath411 for each positive integer @xmath40 we have @xmath412 let @xmath413 we will show below that for any given @xmath414 for almost all @xmath1 there are infinitely many @xmath61 such that there exists a positive integer @xmath415 with @xmath416 and @xmath417 it is a well - known fact in metric diophantine approximation that for almost all @xmath1 we have @xmath418 for all @xmath419 large enough . hence if is true for almost all @xmath1 then there are infinitely many @xmath61 such that for @xmath411 we have @xmath420 and the desired result follows ( note that @xmath421 ) . it remains to show the existence of the numbers @xmath416 which satisfy . + let @xmath33 be given . from the definition of @xmath422 it is easily seen that @xmath423 this follows from the fact that the derivative of the function @xmath424 is bounded uniformly by @xmath425 . now let @xmath426 be the function defined by @xmath427 this definition means that @xmath428 is constant on intervals of length @xmath429 which lie between two integer multiples of @xmath429 , and coincides with @xmath430 on the left endpoint of such intervals . by we have @xmath431 which means that it is sufficient to prove with @xmath430 replaced by @xmath428 ( remember that the value of @xmath47 was arbitrary and @xmath432 ) . the reason for using the functions @xmath428 instead of @xmath430 is that every function @xmath428 can be written as a sum of at most @xmath433 different indicator functions of intervals ; consequently , we know that the set of values of @xmath1 where @xmath434 is `` large '' can be written as the union of at most @xmath433 intervals , which implies an upper bound for the size of the fourier coefficients of the indicator function of this set ( see below for details ) . + let @xmath435 be a positive integer which will be chosen in dependence on @xmath160 ( we assume that @xmath436 is `` large '' ) . we define real numbers @xmath437 furthermore , we define @xmath438 for @xmath439 . + then by and we have @xmath440 for sufficiently large @xmath61 , where @xmath441 and @xmath66 are the numbers from , and where we used the fact that @xmath442 . hence we have @xmath443 for sufficiently large @xmath61 . consequently for every @xmath61 large enough there is an @xmath444 with @xmath445 which implies that @xmath446 note that , as a consequence of the construction of @xmath428 , the set @xmath447 always is a union of at most @xmath433 disjoint intervals . it is easily seen that by trimming the sets @xmath447 appropriately we can always find a set @xmath448 such that @xmath449 , such that @xmath448 also is the union of at most @xmath433 intervals , and such that for the measure of the sets @xmath448 we have the exact equality @xmath450 let @xmath451 be a `` small '' number . let @xmath452 denote the largest integer such that @xmath453 note that the right - hand side of can be written as @xmath454 and consequently we have @xmath455 ; that means , @xmath452 grows exponentially in @xmath61 . note also that it is easily seen that @xmath456 for sufficiently large @xmath61 ( provided that @xmath457 is chosen sufficiently small ) , which is important for . + we will show that for almost all @xmath1 for infinitely many @xmath61 there is a @xmath458 for these @xmath419 then we have @xmath459 for @xmath436 large enough and @xmath457 small enough in dependence on @xmath160 . together with this will establish , as desired . + it remains to show . let @xmath460 denote the indicator function of the set @xmath448 , extended with period one . then we know that @xmath461 setting @xmath462 we clearly have @xmath463 and @xmath464 } ~ \mathbb{i}_l \leq 4^l.\ ] ] from we can easily calculate that @xmath465 we write @xmath466 for the fourier series of @xmath467 ( note that it has no constant term , since @xmath467 has integral zero ) . in the sequel , we want to show that the sum @xmath468 is large in comparison with the sum @xmath469 for almost all @xmath1 and infinitely many @xmath61 . since @xmath470 such an estimate will show that the sum on the left - hand side of is large ( for almost all @xmath1 , for infinitely many @xmath61 ) , which in turn implies that many of the events described in will occur . a lower bound for is easy to obtain ; to find an asymptotic upper bound for , we will calculate the @xmath471 norm of these sums , and apply the borel - cantelli lemma . + from we directly obtain @xmath472 next we estimate @xmath473 which is relatively difficult . as a consequence of and a classical inequality for the size of the fourier coefficients of functions of bounded variation ( see for example @xcite ) we have latexmath:[\[\label{furk } we split the function @xmath467 into an even and an odd part ( that is , into a cosine- and a sine - series ) . in the sequel , we consider only the even part ; the odd part can be treated in exactly the same way . let @xmath475 denote the @xmath476-th partial sum of the fourier series of the even part of @xmath467 , and let @xmath477 denote the remainder term . then by minkowski s inequality we have @xmath478 furthermore , , minkowski s inequality , and parseval s identity imply that @xmath479 to estimate the first term on the right - hand side of , we expand @xmath480 into a fourier series and use the orthogonality of the trigonometric system . then we obtain @xmath481 to estimate the size of the sum on the right - hand size of , we assume that @xmath482 and @xmath483 are fixed . in the case @xmath484 and @xmath485 , we clearly have @xmath486 whenever @xmath487 ; thus the cardinality of the set on the right - hand side of is @xmath452 . if @xmath484 and @xmath488 , then we have to count the number of pairs @xmath489 for which @xmath490 ; this number is @xmath491 . for the values @xmath492 and @xmath493 we also have to count the number of pairs @xmath489 for which @xmath490 ; so this cardinality is also @xmath491 . the last example shows that the greatest common divisor of @xmath482 and @xmath483 plays a role in this calculation . using similar considerations , in the case of general ( fixed ) values of @xmath482 and @xmath483 it turns out that we have @xmath486 whenever @xmath494 as a consequence we have @xmath495 combining this estimate with we obtain @xmath496 the sum on the right - hand side of the last equation is called a _ gcd sum_. it is well - known that such sums play an important role in the metric theory of diophantine approximation ; the particular sum in probably appeared for the first time in leveque s paper @xcite ( see also @xcite and @xcite ) . a precise upper bound for these sums has been obtained by hilberdink @xcite . hilberdink s result implies that there exists an absolute constant @xmath236 such that @xmath497 combining this estimate with and ( and using parseval s identity ) we have @xmath498 and , together with and , and with a similar argument for the odd part of @xmath467 , we obtain @xmath499 by chebyshev s inequality we have @xmath500 and since @xmath501 grows exponentially in @xmath61 these probabilities give a convergent series when summing over @xmath61 . thus by the borel cantelli lemma with probability one only finitely many events @xmath502 happen , which by implies that @xmath503 for some absolute constant @xmath504 . comparing this upper bound with and using we conclude that @xmath505 for almost all @xmath1 . in particular we have @xmath506 for almost all @xmath1 , which means that for almost all @xmath1 infinitely many events occur . as noted after equation , this proves the theorem . it is known ( see @xcite , formula ( 2.10 ) ) that for all @xmath1 we have @xmath507 for all @xmath61 , where @xmath508 . thus from , and the weyl criterion it follows that @xmath36 is u.d . mod 1 iff @xmath1 is irrational . + hence by , , , and for @xmath52 of the form @xmath411 we have @xmath509 where we write @xmath91 for the star - discrepancy of the first @xmath52 terms of the thue kronecker sequence . + from the left - hand side of this inequality it is not difficult but we do not want to go into the details here to show that @xmath510 for all @xmath52 , where @xmath97 denotes the star - discrepancy of the pure kronecker sequence . + furthermore it is easy to show we again do not go into the details that @xmath511 hence for all @xmath52 we have @xmath512 from this we conclude that the order of the discrepancy @xmath513 of the thue kronecker sequence always is essentially ( up to logarithmic factors ) larger or equal to the order of the discrepancy @xmath514 of the pure kronecker sequence . in fact , the order of @xmath515 essentially equals the order of @xmath514 plus an expression which is at most of order @xmath516 . the order of the additional expression is controlled by lacunary products of sine - functions . + hence we conclude what we have already announced in section [ sect_int ] : * if the order of @xmath89 satisfies @xmath90 then @xmath91 essentially is of the same order as @xmath89 * if @xmath89 satisfies @xmath92 then @xmath91 satisfies @xmath517 hence the two examples given in theorem [ th_thue4 ] show interesting non - trivial cases where we have ( almost ) best possible distribution for the pure kronecker sequence with bad distribution for the thue kronecker sequence . indeed especially the first example gives essentially the extremal values for @xmath89 and for @xmath91 , and shows that the right - hand side of is also essentially optimal . + it remains an open problem to give concrete examples @xmath1 where the corresponding thue morse kronecker sequence has `` small '' discrepancy @xmath89 , e.g. , a discrepancy of the metric order given in theorem [ th_thue2 ] or smaller . of course it also remains an open problem to give good estimates for @xmath91 in the case of `` natural '' examples of @xmath1 like @xmath518 . + + 1 . by @xcite we know that the continued fraction coefficients of the number @xmath519 are bounded . this is equivalent with the existence of some @xmath520 such that @xmath521 for all @xmath522 i.e. , @xmath523 for all such @xmath253 and @xmath145 . + this implies especially @xmath524 for all @xmath525 , i.e. , + @xmath526 for all such @xmath527 and hence @xmath528 has bounded continued fraction coefficients . so the star - discrepancy of the pure kronecker sequence @xmath26 satisfies @xmath529 + on the other hand we already know that for the star - discrepancy @xmath91 of the thue kronecker sequence @xmath37 with @xmath530 we have @xmath531 we give a suitable lower bound for @xmath532 here @xmath533 and we have used @xmath534 + note that the base 2 representation of @xmath535 has one of the following ten possible forms : @xmath536 @xmath537 @xmath538 @xmath539 @xmath540 @xmath541 @xmath542 @xmath543 @xmath544 @xmath545 hence @xmath546 always and therefore @xmath547 always . + because of @xmath548 and @xmath549 for @xmath550 we have @xmath551 therefore , noting that @xmath552 for @xmath553 , we also have @xmath554 so @xmath555 and the lower bound for @xmath556 follows . the upper bound for @xmath556 follows from and from the upper bound for @xmath97 . it was shown in @xcite that @xmath557 has approximation degree 1 , hence for the star - discrepancy of the sequence @xmath558 we have @xmath559 for every @xmath560 to prove the lower bound for the star - discrepancy @xmath91 of the sequence @xmath105 , like in the proof of part a ) we have to estimate @xmath561 from below . we will give in the following as an additional information also an upper estimate for @xmath562 in order to show that our lower estimate is rather sharp . + we may restrict ourselves to @xmath61 of the form @xmath563 . then @xmath564 in the following we use some well - known facts on properties of the thue morse sequence : + the base 2-representation of @xmath565 consists of 8-blocks @xmath566 of the form @xmath567 or @xmath568 + four such consecutive 8-blocks can occur in the following ten combinations : @xmath569 @xmath570 @xmath571 @xmath572 @xmath573 @xmath574 @xmath575 @xmath576 @xmath577 @xmath578 let , for example , @xmath579 be such that @xmath580 ( @xmath581 is the block of 32 digits defined + above ) + for @xmath582 let @xmath583 and @xmath584 then @xmath585 @xmath586 in the same way we determine @xmath587 further it is well known that the frequencies @xmath588 of the occurrence of a quadruple @xmath589 of @xmath590-blocks in the thue morse sequence are given by @xmath591 and @xmath592 hence we get @xmath593 @xmath594 @xmath595 which leads to @xmath596 for @xmath597 small and @xmath61 large enough . this finishes the proof . in conclusion , we mention an open problem from the theory of diophantine approximation which is related to our proof of the lower bound in theorem [ th_thue2 ] . in metric diophantine approximation , one is often interested in finding conditions on @xmath598 which guarantee that @xmath599 has infinitely many integer solutions @xmath600 for almost all @xmath1 . two instances of this problem , either under the additional requirement that @xmath600 are coprime ( _ duffin schaeffer conjecture _ ) or without this additional requirement ( _ catlin conjecture _ ) , constitute probably the two most important open problems in metric number theory . for the origin of the duffin schaeffer conjecture see @xcite , for the catlin conjecture see @xcite . problems of this type are discussed in great detail in glyn harman s monograph on _ metric number theory _ @xcite . for a recent survey , see @xcite . + the problem without the requirement of coprime solutions can also be written in the following form : let @xmath601 be intervals of length @xmath602 , which are symmetric around 0 . let @xmath603 denote the lebesgue measure ( that is , the length ) of these intervals . under which conditions on @xmath603 do we have @xmath604 for almost all @xmath1 ? here @xmath605 denotes the indicator function of @xmath606 , extended with period one . + now in a first step this problem can be generalized to the case when the intervals @xmath601 are not necessarily symmetric around @xmath607 , which leads to a problem in _ inhomogeneous _ diophantine approximation . this type of question is also quite well - investigated . + perpetuating this line of thought , it is natural to ask what happens if we do nt assume that @xmath601 are _ intervals _ , but if they may denote any _ measurable sets _ in @xmath20 $ ] . writing @xmath603 for the measure of these sets , the question is under which conditions on @xmath603 we have @xmath604 for almost all @xmath1 . note that a necessary condition is the divergence of the sum of the measures , by the borel - cantelli lemma . it seems that hardly anything is known about this general problem . as far as we know , this problem was first stated by leveque in @xcite . in this paper he had answered a conjecture of erds , and he formulated a generalized version of erds conjecture . we consider this as a very interesting open problem , and we re - state it below . + * open problem : * let @xmath601 be measurable sets in @xmath20 $ ] , and let @xmath603 denote their measure . under which conditions on @xmath608 is it true that for almost all @xmath1 the fractional part @xmath609 is contained in the set @xmath610 for infinitely many indices @xmath611 ; equivalently , under which conditions is it true that @xmath612 where the indicator functions are extended with period one . + a problem quite similar to this one emerged during the proof of the lower bound of theorem [ th_thue2 ] . however , the situation was comparatively simple there , for example since we could assume there that the sets @xmath610 can be written as the sum of a moderate number of intervals . the general problem seems to be much more complicated ; leveque wrote that this general problem `` seems rather intractable '' . k. fukuyama . a central limit theorem and a metric discrepancy result for sequences with bounded gaps . in _ dependence in probability , analysis and number theory _ , pages 233246 . kendrick press , heber city , ut , 2010 .

the theoretical description of interacting fermions in one spatial dimension is simplified by the fact that the low energy spectrum of noninteracting fermions is identical to the one for a harmonic chain . this fermion - boson transmutation allows to describe interacting fermions as a system of coupled oscillators . tomonaga s model of interacting nonrelativistic fermions on a ring is presented and first discussed in low order perturbation theory . after introducing the concept of two independent species of right and left moving fermions the exact solution of the tomonaga - luttinger model is discussed in detail . the momentum distribution and spectral functions are calculated using the method of the bosonization of the field operator . the general luttinger liquid phenomenology is shortly discussed .

cond-mat9710330

the effects of the coulomb interaction of electrons in metals can only be described approximately @xcite . with the progress in producing artificial low dimensional structures the theoretical work on _ one - dimensional _ interacting fermions has gained importance @xcite . special features of the spectrum of low energy excitations in one dimension allow _ exact _ solutions of models of _ interacting _ fermions . the main idea can already be understood by working with noninteracting fermions , which have the same spectrum of excitations as a harmonic chain . this `` fermion - boson - transmutation '' ( fbt ) was discussed in a recent publication on different levels of sophistication @xcite . in the following this paper is referred to as i. the basic idea can be understood on the level of elementary quantum statistical mechanics . only in chapter v. of i the method of second quantization was introduced to calculate the momentum distribution of _ noninteracting _ fermions in the _ canonical _ ensemble . it was pointed out that the technique presented can also be used in the description of _ interacting fermions_. from the response we obtained to i it is clear that many readers would have liked to see more explicitly the application to this problem . the purpose of this paper is to fill this gap on a level which requires only a basic knowledge of the method of second quantization but none of relativistic quantum field theory . the exact solution to the `` tomonaga - luttinger ( tl- ) model '' is presented starting with tomonaga s model of spinless interacting nonrelativistic fermions on a ring @xcite . in section ii the momentum distribution in the interacting ground state is calculated in leading order perturbation theory . this already demonstrates one of the important ideas towards the exact solution and on the quantitative level shows a sign of the infrared singularities later found in the exact solution . tomonaga s concept of `` right '' and `` left '' moving electrons later extended by luttinger @xcite is introduced and the commutation relations of the fourier components of the operator of the particle density are derived . in order to avoid mathematical subtleties we work with _ two _ cut - offs one for the number of momentum states ( `` band cut - off '' ) and one for the inverse range of the two - body interaction ( `` interaction cut - off '' ) . this allows to smoothly go from tomanaga s original model to the model discussed by luttinger . in section iii the exact low energy eigenvalues of the tl - model are obtained using the bosonization technique . as the most important `` luttinger - liquid '' ( ll ) feature the momentum distribution in the ground state is calculated . its derivative at the fermi points diverges in a power law fashion determined by the important constant called `` anomalous dimension '' . one effort to experimentally verify ll - properties is photoemission from quasi - one dimensional conductors @xcite . as the theoretical description of photoemission involves spectral functions they are discussed in section iv . the model including spin is introduced in section v. after the definition of `` charge '' and `` spin '' bosons the hamiltonian of the model can be written as a sum of two _ commuting _ terms of the spinless type leading to the phenomenon of `` spin - charge - separation '' . in section vi . the general ll - concept is presented and it is shortly discussed how the ll parameters which completely determine the low energy physics also for more complicated @xmath0 lattice models can be extracted from finite size numerical data . in appendix a the proof of the kronig - identity which provides the bosonization of the kinetic energy is given . the bosonization of the field operator is discussed in appendix b. a novel straightforward approach is presented to determine the particle number changing contribution . we put @xmath1 in this paper . the method of bosonization which was presented in i in the context of noninteracting electrons in one dimension is the key concept to understand ground state properties and the spectrum of excited states with low excitation energy also for _ interacting _ fermions . in a seminal paper tomonaga @xcite studied nonrelativistic fermions in the _ high density _ limit , in which the spatial range of the two - body interaction is much larger than the interparticle distance . he studied _ fermions on a ring , i.e. using _ periodic boundary conditions_. the hamiltonian @xmath2 is a sum of the kinetic energy @xmath3 and a two - body interaction @xmath4 @xmath5 where in the second line we have switched to the language of second quantization , with @xmath6 the field operator which annihilates a fermion at the position @xmath7 . it can be expressed in terms of the annihilation operators @xmath8 of momentum states in the standard way @xcite @xmath9 where the sum runs over momenta @xmath10 compatible with the periodic boundary condition . the @xmath11 obey canonical anticommutation relations @xmath12_+ = 0 \ ; , [ c_m , c_n^\dagger ] _ + = \delta_{mn}\hat 1.\ ] ] in order to be consistent with the periodic boundary condition @xmath13 it is useful to expand the two body potential @xmath14 in a fourier series @xmath15 special choices of the fourier transformed potential @xmath16 will be discussed later . in case @xmath17 exists , we can express @xmath4 in terms of the operator of the particle density @xmath18 where the @xmath19 are the position operators of the particles . using @xmath20_+ = \delta(x - y)$ ] for @xmath21 $ ] one obtains @xmath22 where we have introduced the particle number operator @xmath23 . as in tomonaga s paper we present the main results for the spinless model and discuss the modifications due to the electron spin later . it is useful to fourier decompose the density operators @xmath24 if we use the second quantized version @xmath25 the @xmath26 can be expressed as follows @xmath27 in terms of the @xmath26 the interaction term is given by @xmath28 we have used @xmath29 as @xmath30 is real and @xmath31 = 0 $ ] , which follows most easily from the fact that only the commuting position operators @xmath19 enter in the definition of @xmath32 in eq . ( 5 ) . now following tomonaga we assume that the spatial range of the two - body interaction is much larger than the mean particle separation which is inversely proportional to the fermi momentum @xmath33 . two examples for @xmath16 with a finite value of @xmath34 which will be used later in the paper are shown in fig . 1 . [ hbt ] the ground state of the noninteracting system with @xmath35 fermions ( @xmath35 odd ) is the slater determinant @xmath36 ( fermi sea ) in which the momentum states from @xmath37 to @xmath33 are occupied , where @xmath38 and @xmath33 is the fermi momentum . a first insight into the nature of the _ interacting _ ground state can be obtained using first order perturbation theory in @xmath4 @xcite @xmath39 where @xmath40 and @xmath41 is the kinetic energy of the fermi sea . due to the projector @xmath42 only the first term on the right hand side ( rhs ) of eq . ( 9 ) contributes to the first order correction in the above equation . as we assume @xmath43 to be zero for @xmath44 ( much ) larger than a cut - off @xmath45 only states @xmath46 with @xmath47 of the order @xmath48 or smaller occur in @xmath49 . the second quantized form of @xmath26 eq . ( 8) shows that @xmath46 is a _ two particle - two hole state_. for @xmath50 the density operator @xmath51 creates a particle - hole pair around the right fermi point @xmath52 , while @xmath26 creates a particle - hole pair with opposite momentum around the left fermi point @xmath53 @xmath54 the kinetic energy of the two - particle - two hole state @xmath55 is given by @xmath56 where @xmath57 . as the particles and the holes are in the _ vicinity _ of the fermi points one can linearize the energy dispersion around the two fermi points in order to calculate @xmath58 and @xmath59 . with the fermi energy @xmath60 and the fermi velocity @xmath61 one uses instead of the _ one _ parabolic dispersion _ two _ linear branches @xmath62 as will be discussed later the use of the linear dispersions , also shown in fig . 2 , leads to an error which vanishes in the high density limit @xmath63 . it simplifies the expression for the kinetic energy of the two particle - two hole state to @xmath64 and leads to @xmath65 higher order perturbation theory yields additional terms with multiple particle - hole pairs . + an important way to characterize the interacting ground state is to calculate its momentum distribution @xmath66 , i.e. to study how the fermi step distribution of the noninteracting electrons is modified by the interaction . it follows from eq . ( 13 ) that the leading correction to the fermi function is quadratic in the two - body potential . for @xmath67 one obtains e.g. @xmath68 where we have used that the two particle - two hole states are eigenstates of @xmath69 . the sum over @xmath70 can be trivially performed and yields a factor @xmath47 . the sum over @xmath47 is especially simple for the step potential @xmath71 where @xmath72 is the unit step function . if the sum over @xmath73 is approximated by an integral one finally obtains @xmath74 .\ ] ] the logarithmic singularity for @xmath75 is an artefact of the perturbation expansion , as the occupancies have to be smaller than one . the exact solution for the momentum distribution @xmath76 will be derived in detail later . in the thermodynamic limit @xmath77 one obtains a continuous function @xmath78 which in the neighbourhood of @xmath33 is given by @xmath79 the constant @xmath80 is called the _ anomalous dimension_. for weak two - body potentials @xmath80 is small and one can use @xmath81 for fixed @xmath7 . by comparison with the perturbational result eq . ( 14 ) one finds @xmath82 and @xmath83 as the lowest order perturbational result for the anomalous dimension . a quantitative comparison of @xmath84 with the exact result will be given later . + perturbation theory shows that there are three types of electrons : those with momenta @xmath85 , which move with velocities @xmath86 which will be called right movers , those with momenta @xmath87 and velocities @xmath88 ( left movers ) and the inert electrons deep in the fermi sea , which play no role in the low energy physics . this basic insight of tomonaga was used by luttinger @xcite to further modify the model . in addition to the linearization eq . ( 12 ) he extended the range of the possible @xmath89-values for _ both _ the right and left movers to @xmath90 with @xmath47 positive and negative integers . this further simplifies the dispersion of the electrons , but introduces infinitely many negative energy electron states , which have to be filled in the ground state . as this infinite dirac sea leads to various mathematical subtleties we prefer to work with a _ finite band cut - off_. we introduce the momentum @xmath91 and allow states with momenta @xmath92 for right movers and @xmath93 for left movers . in fig.2 we have chosen @xmath94 and show the states added as the dot - dashed lines . [ hbt ] later we also consider the limit @xmath95 . the kinetic energy of the model solved exactly in the next section reads @xmath96 where @xmath97 and @xmath98 . the creation and annihilation operators for the right @xmath99 and left @xmath100 movers obey the fermion anticommutation relations @xmath101 @xmath102_+ = 0\ ; \quad \left[c_{n,\alpha } , c^\dagger _ { m,\beta}\right]_+ = \delta_{n , m}\delta_{\alpha,\beta}\hat 1.\ ] ] in the interaction term @xmath4 eq . ( 9 ) we replace @xmath103 by @xmath104 where @xmath105 with @xmath106 are the density operators of the right and left movers . because @xmath107 is symmetric in @xmath47 and @xmath108 the relation @xmath109 holds . it is not more difficult to work with the following generalization @xmath110 where the @xmath111 are the particle number operators @xmath112 and we have dropped the terms proportional to the particle number operators as they can be later included in the chemical potential . the second line in eq . ( 21 ) can also be expressed in terms of the total particle number operator @xmath113 and the current operator @xmath114 @xcite . for the interaction functions in eq . ( 21 ) we have used the standard notation @xcite . if we put @xmath115 the tomonaga - luttinger hamiltonian @xmath116 yields in perturbation theory in @xmath117 the _ same result _ for @xmath118 as in eq . ( 15 ) . in @xmath119 only the terms proportional to @xmath120 contribute for @xmath121 . concerning the low energy physics the original model with the quadratic dispersion is equivalent to the _ tomonaga - luttinger ( tl ) model _ described by the hamiltonian @xmath122 in the high density limit @xmath123 . there is an essential difference in the commutation relations of the @xmath124 and the @xmath125 . while as discussed previously @xmath126 = 0 $ ] , which can be seen most easily in the first quantized version , the @xmath124 do _ not _ all commute . from the anticommutation relations for the fermions on different branches it follows that @xmath127 = 0 $ ] . the commutators @xmath128 $ ] on the other hand are nontrivial @xmath129 & = & \sum_{m',n ' } [ c^\dagger _ { m',\alpha } \ ; c_{m'+ m,\alpha},\ ; c^\dagger _ { n'\alpha } \ ; c_{n'+ n,\alpha } ] w^\alpha_{m',m'+m } w^\alpha_{n',n'+n}\\ \nonumber & = & \sum_{m'n ' } \left(\delta_{n',m'+m}\ ; c^\dagger _ { m',\alpha}\ ; c_{n'+n,\alpha } -\delta_{m',n'+n}\ ; c^\dagger _ { n'\alpha}\ ; c_{m'+m,\alpha}\right ) w^\alpha_{m',m'+m}\ ; w^\alpha_{n',n'+n}\\ \nonumber & = & \sum_{m ' } \left ( w^\alpha_{m',m'+m } - w^{\alpha}_{m',m'+n } \right ) w^\alpha_{m ' , m'+m+n}\ ; c^\dagger _ { m',\alpha } \ ; c_{m'+m+n}.\end{aligned}\ ] ] in going to the last equation the summation index was changed from @xmath130 to @xmath131 in the second term . it is easy to see that the rhs of eq . ( 23 ) _ vanishes _ if @xmath108 and @xmath47 have the _ same _ sign . take , for example , right movers and @xmath108 and @xmath47 positive . then due to the factor @xmath132 the rhs of eq . ( 21 ) can only be nonzero for @xmath133 and @xmath134 . but then also @xmath135 _ and _ @xmath136 and the commutator vanishes because of the factor @xmath137 . next we consider the case @xmath138 shortly discussed in section v. in i. for @xmath139 one obtains for the right movers @xmath140 = \sum^{-m_0+m-1}_{m ' = -m_0 } c^\dagger _ { m',+ } \ ; c_{m',+ } \equiv { \cal n}_+(m).\ ] ] we see that the commutator is given by the particle number of the lowest @xmath108 right moving one - particle states . as explained earlier an interaction @xmath16 which drops off rapidly for @xmath141 does not produce holes deep in the fermi sea in the interacting ground state and eigenstates with low excitation energy . in this subspace of the total hilbert space it is an excellent approximation which becomes asymptotically exact for @xmath142 to replace @xmath143 by @xmath144 on the rhs of eq . ( 24 ) . in the interaction term @xmath145 eq . ( 20 ) only @xmath146 for @xmath147 contribute . therefore it is sufficient to discuss the commutation relations eq . ( 23 ) for @xmath108 and @xmath47 with different sign for @xmath148 . for @xmath149 one obtains for right movers @xmath150 = \sum^{-m_0-n-1}_{m ' = -m_0 } c^\dagger _ { m',+ } \ ; c_{m'+m - n,+ } = -\sum^{-m_0-n-1}_{m'= -m_0 } c_{m'+m - n,+}\ ; c^\dagger _ { m',+}.\ ] ] the operator @xmath151 tries to create an electron deep in the fermi - dirac sea , which is not possible in the space of low lying excited states , as all states at the bottom of the band are occupied . therefore the commutator vanishes in this subspace . analogous arguments apply to the left movers and we finally obtain @xmath152 = \alpha m \,\hat 1 \;\delta_{\alpha \beta}\;\delta_{m ,- n}.\ ] ] the importance of this relation was first realized by tomonaga @xcite who worked with @xmath153 . luttinger @xcite made an error with the commutation relation for @xmath154 later corrected by mattis and lieb @xcite . the commutation relations eq . ( 26 ) for the operators @xmath124 apart from a proper normalization factor , look like boson commutation relations . if we define the operators @xmath155 and the corresponding adjoint operators @xmath156 using @xmath157 eq . ( 26 ) implies the usual bosonic commutation relations @xmath158 = 0 , \quad [ b_n , b^\dagger _ m ] = \delta_{mn}\hat 1.\ ] ] the kinetic energy @xmath159 in eq . ( 18 ) consists of two commuting terms , which both are of the form discussed for fixed boundary condition in i. therefore they both can be expressed in terms of the boson operators with the help of the kronig identity @xcite which was stated in i without giving an explicit proof . this is therefore presented in appendix a. it yields @xmath160 as we dropped terms proportional to @xmath161 in @xmath117 which only lead to a renormalization of the chemical potential we also drop the term proportional to @xmath161 in the kinetic energy in the following . neglecting also a constant term in the total hamiltonian it reads @xcite @xmath162 \equiv h_b + h_{{\cal n } , \hat { \cal j}}\end{aligned}\ ] ] with the velocities @xmath163 here @xmath164 determines the energy change for adding particles while @xmath165 enters the energy change when the difference in the number of right and left movers is changed . for our original model with @xmath166 the velocity @xmath165 equals the fermi velocity @xmath167 of the noninteracting system . as the particle number operators @xmath168 commute with the boson operators @xmath169 the two terms @xmath170 and @xmath171 in the hamiltonian commute and can be treated separately . the boson part @xmath170 of the @xmath172-hamiltonian is bilinear in the boson creation and annihilation operators . as shown below it can be brought into the form @xmath173 by solving an eigenvalue problem , and the @xmath174 are also boson creation ( annihilation ) operators . as @xmath170 in eq . ( 30 ) is a sum of commuting terms @xmath175 only the coupling of the four operators @xmath176 has to be considered at a time . momentum conservation further reduces the @xmath177 eigenvalue problem to a @xmath178 problem . the eigenvalue problem is obtained by looking at the heisenberg equations of motion for @xmath179 . this involves the commutators @xmath180\\ \lbrack b^\dagger _ { -n},h_b \rbrack \\ \end{array } \right ) = \left ( \begin{array}{ccc } k_n(v_f+g_4(k_n)/(2\pi ) ) & k_ng_2(k_n)/(2\pi ) \\ -k_ng_2(k_n)/(2\pi ) & -k_n(v_f+g_4(k_n)/(2\pi))\\ \end{array } \right ) \left ( \begin{array}{ccc } b_n\\ b^\dagger _ { -n}\\ \end{array } \right ) % \end{displaymath}\end{aligned}\ ] ] the matrix @xmath181 on the rhs of eq . ( 33 ) has the eigenvalues @xmath182 as @xmath181 is not symmetric one has to distinguish left and right eigenvectors . we denote the left eigenvector to @xmath183 as @xmath184 where @xmath8 and @xmath185 can be chosen as real quantities as @xmath181 itself is real . the formal solution of the heisenberg equations of motion @xmath186 $ ] etc . reads @xmath187 if we multiply this equation form the left with @xmath188 we can replace the matrix by its eigenvalue @xmath189 which shows that the operator @xmath190 has the simple time evolution @xmath191 . with a proper normalization of the left eigenvector the @xmath192 are therefore the boson operators which diagonalize @xmath170 to the form eq . the @xmath192 should obey the boson commutation relation @xmath193 = \hat 1 $ ] . ( 36 ) yields @xmath194 = c^2_n - s^2_n$ ] the left eigenvector @xmath184 has to be normalized by @xmath195 . elementary calculation of the eigenvector then yields @xmath196 = s^2_{-n}\ ] ] in the noninteracting limit @xmath197 the weights @xmath185 vanish and the bogoljubov transformation in eq . ( 36 ) reduces to @xmath198 . the dispersion relation @xmath199 of the bosons in the noninteracting case ( see i ) is in the interacting case replaced by @xmath200 where the @xmath89-dependent sound velocity usually called the charge velocity follows from eq . ( 34 ) as @xmath201 for our original model @xmath202 it is given by @xmath203 . the corresponding dispersion is shown in fig . 3 for the two potentials @xmath204 presented in fig . 1 . [ hbt ] for @xmath205 the boson dispersion is approximately linear @xmath206 with @xmath207 . from eqs . ( 31 ) and ( 38 ) we obtain the relation @xmath208 first pointed out by haldane @xcite . for large momenta @xmath209 the dispersion of the noninteracting case is approached . in the figure we have used a _ purely _ repulsive interaction @xmath210 which leads to @xmath211 . for attractive interaction there is an instability if @xmath212 gets larger than @xmath213 @xcite , leading to phase separation . in the following we concentrate on repulsive interactions . the interacting ground state @xmath214 with the integers @xmath215 is determined by @xmath216 . it can be constructed by a unitary transformation out of the noninteracting ground state @xmath217 here we have allowed the number of right and left movers to be different . our original model corresponds to @xmath218 . in the following we calculate expectation values in the interacting ground state by algebraic methods using @xmath219 . therefore the explicit form of this state @xcite is not needed . the excited states for given particle numbers @xmath220 are constructed as for harmonic oscillators @xcite @xmath221 their energy is given by @xmath222 . the nature of the interacting ground state shows up clearly in the momentum distribution @xmath223 . in the following we calculate @xmath224 and show how the fermi step of the noninteracting limit is modified near the right fermi point @xmath33 . as in i it can be obtained by first calculating @xmath225 , where @xmath226 is the auxiliary field operator defined in ( b.6 ) and then using the inversion formula ( b.24 ) to calculate the momentum distribution . with the use of the bosonization formula ( b.25 ) the calculation is quite simple . as we want to use @xmath227 we express in the operator function @xmath228 defined in ( b.9 ) the @xmath229 in terms of the @xmath230 . from @xmath231 in eq . ( 36 ) one obtains the inversion @xmath232 this yields @xmath233 as the @xmath192 and @xmath234 commute , @xmath235 $ ] can be written as a product of two exponentials with the annihilation part to the right . if applied to the interacting ground state this factor yields unity , i.e. @xmath236 the expectation value of @xmath237 as seen from eq . ( b.25 ) involves the overlap of the above state with itself . using the baker hausdorff formula @xmath238}$ ] we obtain @xmath239)\right\}.\ ] ] this is exactly of the form as the _ finite temperature _ canonical expectation value for _ noninteracting _ fermions ( eq.(50 ) in i ) but with the bose function @xmath240 replaced by @xmath241 which depends on the interaction as shown in eq.(37 ) . for arbitrary potentials therefore @xmath118 can be calculated by the recursive method outlined in the appendix of i. the sum in the exponent on the rhs of eq . ( 45 ) is especially simple , if @xmath241 decays exponentially @xmath242 the corresponding potential @xmath43 can be obtained using eq . ( 37 ) by solving a quadratic equation . the factor @xmath243 in the exponent on the rhs of eq . ( 46 ) was introduced because @xmath43 is proportional to @xmath185 for large momenta . it is shown as the full line in fig . 1 for @xmath244 . if the cosine is written as a sum of two exponentials sums of the type @xmath245 appear in the exponent on the rhs of eq . this yields @xmath246^{{s^2}(0)}.\ ] ] the calculation of @xmath118 simplifies if we go to the thermodynamic limit @xmath247 then it is appropriate to switch to the physical field operators @xmath248 defined as in eq . ( 2 ) but with @xmath8 replaced by @xmath249 . they are related to the auxiliary field operators defined in ( b.6 ) by @xmath250 the momentum distribution @xmath251 follows from the inversion formula ( b.22 ) and the fact that @xmath252 is a function of @xmath253 only as @xmath254 if one now takes the limit @xmath255 one obtains for fixed @xmath7 from eq . ( 47 ) @xmath256 in contrast to the noninteracting case this zero temperature time - independent green s function for large @xmath7 does not decay as @xmath257 , where @xmath258 is the spatial dimension , but as @xmath259 . therefore the quantity @xmath260 is called the anomalous dimension . the momentum distribution in the thermodynamic limit will be denoted @xmath261 . the behaviour of @xmath262 near the right fermi point is most easily studied by looking at its derivative @xmath263 where @xmath264 . in the noninteracting case @xmath265 and @xmath266 , where the negative delta function is due to the step at @xmath267 . for finite values of @xmath268 this singularity for @xmath269 is weakened . for @xmath270 the function of @xmath7 fourier transformed in eq . ( 52 ) decays faster than @xmath271 and @xmath272 is _ finite_. for @xmath273 one obtains the behaviour mentioned in section ii . as @xmath274 is an even function of @xmath275 , it is sufficient to consider @xmath276 . with the substitution @xmath277 one obtains @xmath278 for @xmath279 the limit @xmath280 in the integrand can be performed and the integral is finite , but @xmath281 diverges at @xmath269 because of the prefactor of the integral . integration with respect to @xmath275 finally yields eq . ( 16 ) . at finite temperatures the derivative of @xmath282 at @xmath33 diverges like @xmath283 for @xmath284 . the special case @xmath285 requires a separate treatment not given here . + the power law behaviour eq . ( 16 ) of @xmath262 has been derived here for the special potential shown in fig.@xmath2861 leading to the exponential decay eq . ( 45 ) of @xmath287 . different potentials with a different decay of @xmath288 lead for @xmath289 to the _ same _ qualitative behaviour as for our special potential if @xmath16 goes to constant value for @xmath290 . this implies a constant value @xmath268 and the anomalous dimension is given by eq . the anomalous dimension can be expressed in terms of the velocities @xmath164 and @xmath165 using eqs . ( 31 ) and ( 37 ) @xmath291 where @xmath292 is less than one for repulsive interactions and larger than one for attractive interactions . + the power law behaviour of @xmath262 for interactions @xmath16 with a finite limit @xmath34 in one spatial dimension is different from the fermi liquid type behaviour of a usual three dimensional conductors in which the momentum distribution of the interacting ground states has a finite discontinuity of size @xmath293 at the fermi momentum @xcite . we note that for a _ finite _ one - dimensional system and small values of @xmath80 the distinction between the power - law behaviour and a finite discontinuity is not straightforward . in fig . 4 we show the momentum distribution for a step potential ( see fig . 1 ) with @xmath294 and 0.3 corresponding to @xmath295 and @xmath296 for different system sizes . [ hbt ] the size of the system is specified by the dimensionless ratio @xmath297 where @xmath298 is the spatial range of the interaction . for the smaller interaction we also compare the exact result with the lowest order perturbation theory result eq . ( 15 ) in the range @xmath299 . as the dimensionless coupling constant @xmath300 is not really much smaller than one the deviations are already well visible . + power laws occur for the tl - model also in other physical quantities . in the next section we present a short discussion of one - particle spectral functions which largely determine the current in a photoemission experiment @xcite . for noninteracting electrons the density of one - particle states per unit volume ( dos ) @xmath301 plays an important role for the calculation of thermodynamic properties @xcite . in one dimension it is defined as @xmath302 in the thermodynamic limit @xmath303 and for a system of noninteracting right movers @xmath304 one obtains @xmath305 , i.e. an energy independent dos . for photoemission only the occupied states matter and one defines the occupied dos @xmath306 by multiplying the delta function in eq . ( 55 ) with the fermi function @xmath307 , which at @xmath308 is the step function at the fermi energy @xmath309 . this simple dos concept has to be generalized properly when the fermions interact . for a translational invariant system as discussed in the previous sections one defines @xmath310 as the generalization of the occupied dos relevant for photoemission @xcite . in the _ noninteracting _ limit @xmath311 is the normalized slater determinant where the state with momentum @xmath44 has been removed relative to the n - electron ground state . as its energy is given by @xmath312 the spectral function @xmath313 reduces to the sum of delta functions as in eq . ( 55 ) restricted to the occupied momenta . the individual terms in the sum of the rhs of eq . ( 56 ) enter angular resolved photoemission@xcite , while @xmath313 is needed for the description of angular integrated photoemission @xcite . for translational invariant systems expectation values with @xmath314 replaced by @xmath315 with @xmath316 in eq . ( 56 ) vanish because of momentum conservation and one can write @xmath313 in the form @xmath317 where @xmath318 is the field operator in the heisenberg representation . if we switch from the original nonrelativistic model to the tl - model the spectral function acquires an additional @xmath80 label . in the following we sketch the calculation of @xmath319 . for that purpose it is convenient to express the field operator @xmath320 eq . ( b.8 ) in terms of the @xmath321 instead of the @xmath322 because the former have a simple time dependence . in @xmath323 the operators @xmath324 and @xmath325 commute . therefore @xmath326 ( and @xmath327 ) in eq . ( b.8 ) can be written as a product of two exponentials with the annihilation operators to the right . after using the baker - hausdorff formula once in order to complete the process of normal ordering ( bringing all annihilation operators to the right of the creation operators ) one obtains @xmath328 and @xmath329 . + this is one of the most important formulas for the physics of one - dimensional interacting fermions . the time dependent operator @xmath330 follows from eq . ( 58 ) by replacing @xmath324 and @xmath331 by @xmath332 and @xmath333 and @xmath334 in @xmath335 ( see eq . ( b.23 ) ) by @xmath336 . various kinds of time dependent correlation functions can quite simply be calculated using this result @xcite . here we restrict ourselves to @xmath337 in order to obtain @xmath338 . + as @xmath334 commutes with the bosonic operators we first show that the particle number changing operators @xmath339 and @xmath336 lead to a simple time dependent factor . using ( b.17 ) one obtains @xmath340 as @xmath341 in eq.(58 ) is normal ordered in the @xmath80 s one has to use the baker - hausdorff formula only once to normal order @xmath342 . this yields @xmath343}\\ \nonumber & = & \frac{a^2(n_c)}{l } \exp\left\ { \sum^\infty_{n=1 } \frac{1 + 2s^2_n}{n } e^{i\omega_nt}\right\ } \end{aligned}\ ] ] where @xmath344 is the chemical potential , which we `` eliminate '' by discussing @xmath345 as a function of @xmath346 . the spectral function can be calculated analytically in the universal _ low energy regime _ @xmath347 for the step potential which has a strictly linear boson dispersion @xmath348 for @xmath349 ( see fig . 2 ) and a different linear dispersion @xmath350 for @xmath351 . this implies that we can write @xmath352 as a product of two power series in @xmath353 and @xmath354 @xmath355 now the fourier integral can be performed analytically . for @xmath356 the @xmath357 and for @xmath358 the @xmath359 with @xmath360 do not contribute and one obtains in the universal regime @xmath361 @xmath362 the @xmath363 for @xmath364 can be given analytically as the sum in the exponent in eq . ( 60 ) is of the form @xmath365 with @xmath366 and @xmath367 . for the calculation of the @xmath363 for @xmath364 one makes no error to extend the sum to infinity , which yields @xmath368 = ( 1 - z)^{-(1 + \alpha)}$ ] . now one can use the well known power expansion of this function . alternatively one can show that the recursion relation in ( a4 ) of i can simply be solved explicitly . if we also use @xmath369 , where @xmath370 = 0.57721 ... is euler s constant @xcite we find the _ exact _ weights of the delta peaks in eq . ( 62 ) @xmath371 for @xmath372 and @xmath373 . for @xmath374 one can use the product representation of euler s gamma function @xcite to write the weights as @xmath375 now the thermodynamic limit can be performed in eq . ( 62 ) and one obtains with @xmath376 for @xmath377 in the universal regime @xmath378 the spectral weight near ( below ) the chemical potential @xmath379 is suppressed by the same power law behaviour as in @xmath380 . this type of power law occurs also for potentials with a finite @xmath34 but a different decay for momenta large compared to @xmath381 . but then the prefactor is different and the power law behaviour does not hold _ exactly _ in a finite range as eq . ( 65 ) but only _ asymptotically _ in the limit @xmath382 @xcite . this power law suppression of spectral weight was claimed to be experimentally verified in photoemission experiments with an organic compound which consists of weakly coupled chains @xcite . reliable results for more complicated models of interacting fermions can often only be obtained using numerical methods for systems of rather small size . as our result eq . ( 63 ) is exact also for finite systems we shortly discuss how one can extract the anomalous dimension even for small values of @xmath383 . as eq . ( 64 ) only holds for @xmath384 it is _ not _ useful to fit the weights to the power law form . it is much easier to look at the scaling behaviour of the peak at @xmath385 with weight @xmath386 with system size . from a log - log - plot one can directly infer the value of @xmath80 . for @xmath387 the spectral function @xmath388 approaches the noninteracting value @xmath389 . the crossover from the power law behaviour to this high energy limit is discussed in @xcite . finally we want to point out that the low energy power law behaviour of @xmath390 given by eq . ( 65 ) holds for _ arbitrary _ values of @xmath80 . _ no _ additional linear contribution occurs for @xmath391 as in the case of the momentum distribution @xmath392 . this difference is easily lost without a proper treatment of the cut - offs . electrons are spin one - half particles and for their description it is necessary to include the spin degree of freedom in the model . we choose a fixed quantization axis and denote the two spin states by @xmath393 . then the field operator @xmath394 acquires a spin index as the @xmath8 . the density operator of spin projection @xmath395 is defined as @xmath396 and in the two - body interaction in eq . ( 6 ) @xmath397 is replaced by @xmath398 . usually the interaction @xmath399 is independent of the spin indices @xcite . after the linearization and introduction of the right and left movers the @xmath146 in eq . ( 20 ) obtain an additional spin index , as well as the boson operators in eq . the tl - hamiltonian for spin 1/2 then reads @xmath400\\ & & \hspace*{1.5 cm } \left . + ~\frac{\pi}{l}~ \left [ \left ( v_f \delta_{\sigma \sigma'}+ \frac { g_4^{\sigma \sigma ' } ( 0)}{2\pi}\right ) ~\sum_\alpha { { \cal n}}_{\alpha , \sigma } { { \cal n}}_{\alpha , \sigma ' } + \frac { g_2^{\sigma \sigma'}(0)}{\pi } { { \cal n}}_{+,\sigma } { { \cal n}}_{-,\sigma ' } \right ] \right\}~.\nonumber\end{aligned}\ ] ] the interaction matrix elements @xmath401 are taken as depending on the relative spin only , i.e. @xmath402 and @xmath403 are independent of the spin label @xmath395 . the interaction term in eq . ( 66 ) couples the electrons with different spin . with the assumption on the @xmath404 it is possible to switch to new boson operators @xmath405 with @xmath406 @xmath407 which obey @xmath408 = 0~;~ \left [ b_{a , n } , b_{a ' , n'}^\dagger \right ] = \delta_{a a ' } \delta_{n n ' } { \hat 1}$ ] . the kinetic energy can also be expressed in terms of `` charge '' ( @xmath409 ) and `` spin '' ( @xmath410 ) bose operators using @xmath411 . if we define the interaction matrix elements @xmath412 via @xmath413 and define @xmath414 we can write @xmath415 as @xmath416 where the @xmath417 are of the form eq . ( 30,31 ) but the interaction matrix elements have the additional label @xmath418 . the two terms on the rhs of eq . ( 69 ) _ commute _ , i.e. the `` charge '' and `` spin '' excitation are completely independent . this is usually called `` spin - charge separation '' @xcite . the `` diagonalization '' of the two separate parts proceeds exactly as in section iii and the low energy excitations are `` massless bosons '' @xmath419 with the _ charge velocity _ @xmath420 and the _ spin velocity _ @xmath421 usually different . for a spin independent interaction @xmath422 in eq . ( 68 ) vanishes and @xmath421 equals the fermi velocity @xmath167 . as discussed in the next section the hamiltonian @xmath423 with properly chosen interaction matrix elements describes correctly the low energy physics of more complicated @xmath0 models e.g. on a lattice . then @xmath421 and @xmath167 are generally different even if the interaction on the lattice is spin independent . the bosonization of the field operator @xmath424 proceeds exactly as in appendix b. in order to calculate quantities like the momentum distribution @xmath425 it is then useful to express the @xmath426 in @xmath427 like in ( b.9 ) in terms of the charge and spin bosons @xmath428 where the plus ( minus ) sign holds for the up ( down ) spin . therefore e.g. the function @xmath429 is a _ product _ of a charge and a spin part . the separate factors are calculated as in the spinless case . the only difference is a factor @xmath430 in the exponent due to the additional factor @xmath431 in eq . the final result is exactly of the form eq . ( 45 ) but with @xmath432 replaced by @xmath433 . this leads to the power law behaviour presented in eq . ( 16 ) as in the spinless case but the anomalous dimension given by @xmath434 the individual contributions can be expressed in terms of the @xmath435 as @xmath436 similar to eq . also the result for @xmath437 needed in the calculation of the integrated spectral function @xmath438 takes the form of a product of a `` spin '' and a `` charge '' factor . in each factor the weight @xmath439 in the exponent of eq . ( 60 ) is replaced by @xmath440 . therefore the fourier transform of the _ individual factors _ are in the low energy regime of the power law form eq . ( 65 ) with @xmath80 replaced by @xmath441 . the convolution of the two power laws finally for @xmath377 yields @xmath442 in the asymptotic regime as in the spinless model . alternatively this follows more directly from the fact that @xmath437 decays as @xmath443 for large times . in order to see a drastic difference between the model with and without spin , one has to calculate the individual terms in the sum in eq . the delta peaks of the noninteracting model are broadened into _ one _ `` infrared '' power law threshold in the model without spin @xcite and _ two _ power law singularities in the model including spin @xcite . we have seen that the low energy physics of the @xmath172-model is determined by two parameters @xmath420 and @xmath444 or @xmath164 and @xmath165 in the spinless model and four parameters @xmath445 or @xmath446 in the model including spin . they also determine the low temperature thermodynamics of the model . we shortly discuss the model including spin . the low temperature specific heat of a linear chain is given by @xmath447 , where the constant @xmath448 is inversely proportional to the sound velocity . this is the @xmath449 version of debye s law @xcite . the ratio of @xmath448 and the corresponding value @xmath450 for the noninteracting fermions is therefore @xmath451 at zero temperature we can express the pressure @xmath452 as @xmath453 and the compressibility @xmath454 . because of @xmath455 the @xmath35-dependence of the ground state energy of the @xmath172-model using eq . ( 30 ) is given by @xmath456 leading to @xmath457 the compressibility ratio is therefore given by @xmath458 the factor @xmath430 in eq . ( 73 ) is absent in the spinless case . we leave it as an exercise to show that one obtains similarly for the spin susceptibility @xmath459 these relations show that the @xmath460 which determine the contributions @xmath461 to the anomalous dimension @xmath80 and therefore the power law behaviour of the green s functions also enter the low temperature thermodynamics . it was an important insight by haldane @xcite to realize that the low energy physics of the @xmath172-model is _ generic _ for interacting fermions in one dimension . in the language of the renormalization group @xcite the tl - hamiltonian is the _ fixed point _ hamiltonian for 1d fermions with a repulsive interaction . this is the main reason for the importance of the tl - model for the physics in one dimension . usually the ll parameters @xmath462 of the effective tl - model which describes the low energy physics of a given model can only be calculated approximately . an example is the original tomonaga model with the quadratic energy dispersion ( with or without spin ) where it is _ not _ assumed that the interaction cutoff is much smaller than the fermi momentum , i.e. an _ arbitrary _ repulsive interaction is used . an exception is the extreme short range limit @xmath463 . for the spinless model @xmath4 then vanishes because the pauli principle forbids two fermions at the same space point ( @xmath464 in eq . ( 1 ) ) . for the corresponding model including spin ground state properties can be calculated exactly using the bethe - ansatz method @xcite . the ratios @xmath465 and @xmath466 of the effective tl - model which describes the low energy physics of the model can be inferred by calculating the compressibility and spin susceptibility using eqs . ( 74 ) and ( 75 ) . the charge and spin velocities can be obtained as @xmath467 , where @xmath468 is the lowest energy for charge ( spin ) excited states with total momentum @xmath469 . from this information one can obtain the anomalous dimension , which determines the decay of the green s function without the need to calculate this function . for 1d models on a lattice , which can not be solved exactly by the bethe - ansatz method one can determine the ll parameters approximately by numerically calculating the quantities discussed above for small systems . in the case of weak interactions perturbation theory can be used . we show this to leading order for the spinless tomonaga model with a short range interaction . the ground state energy up to first order in the interaction is given by the expectation value of @xmath2 in the fermi sea @xmath36 @xmath470 the interaction contributes the hartree and the fock term . in the large system limit the sums can be replaced by integrals in the standard way and one obtains @xmath471 in order to obtain the charge velocity to linear order in the interaction we have to calculate the expectation value of @xmath2 in the state @xmath472 . taking the difference with @xmath473 and dividing by @xmath469 yields @xmath474 the contribution of the interaction is due to the different fock energies in the states @xmath475 and @xmath36 . combining the results from eqs . ( 77 ) and ( 78 ) gives @xmath476 which leads to @xmath477 if one generalizes the calculation of @xmath478 in section ii to the case of a short range interaction one obtains the same result for the anomalous dimension . this shows to leading order perturbation theory that the low energy physcis of the tomonaga model with an arbitrary range interaction is that of the tl model with @xmath479 . + because of the limited length of the paper the rapidly expanding field of electronic transport in one - dimensional interacting fermi systems was not discussed . important insights can rather quickly be obtained from the simple exercise of normal ordering of @xmath480 with the use of eq.(58 ) and the corresponding expression for the left movers @xcite . the interested reader should look up the relevant references in the review article cited @xcite , especially the important work by kane and fisher @xcite . i would like to thank l.bartosch , p.kopietz and v.meden for a critical reading of the manuscript and the latter also for the pleasant collaboration on several issues discussed in this paper . the now more extended discussion of commutation relation of the phase operator and the particle number operator was stimulated by a recent preprint ( cond - mat/9805275 ) by j.von delft and h.schoeller . in this appendix we generalize an identity found by kronig in the context of the ( now long forgotten ) `` neutrino theory of light '' @xcite to the case of a finite cut - off @xmath481 . for the right movers it reads @xmath482 for the proof we simply evaluate the lhs using eqs . ( 20 ) and ( 27 ) @xmath483 where we have split the double sum over @xmath47 and @xmath130 in the contributions from @xmath484 and @xmath485 . we will show that the operator @xmath486 yields the rhs of eq . ( a.1 ) while @xmath487 vanishes @xmath488 the second equality follows from a change of summation variables and using @xmath489 . for the discussion of @xmath487 we split the double sum into the contributions from @xmath490 and @xmath491 @xmath492 in the first term we write @xmath493 and use @xmath494 as the new summation variable and in the second term we write @xmath495 . this yields @xmath496 as the products of the four operators are both _ antisymmetric _ in @xmath497 and @xmath498 the operator @xmath487 vanishes after performing the summations , which completes the proof . if we multiply eq . ( a.1 ) with @xmath499 and use the analogous relation to ( a.1 ) for the left movers we obtain eq . ( 29 ) for the kinetic energy @xmath500 . we note that in the proof _ no _ use is made of the commutation relations of the @xmath501 and @xmath502 . the kronig identity holds as an _ exact _ operator identity without performing the limit @xmath503 . in this appendix we generalize chapter v. of i on the bosonization of fermionic operators to the case of a _ single _ fermion operator . a new straightforward construction of the particle number changing part is given which simplifies earlier presentations @xcite . we start with a simple relation which holds for operators @xmath504 obeying @xmath505 = \hat 1 $ ] . it reads @xmath506 = \lambda e^{\lambda b^{\dagger}}~.\ ] ] to prove it one differentiates @xmath507 with respect to @xmath508 and obtains @xmath509 . therefore @xmath510 . multiplying from the left with @xmath511 yields ( b.1 ) . next we define the following operators linear in the boson operators @xmath512 @xmath513 with arbitrary constants @xmath514 and @xmath515 . using ( b.1 ) and elementary generalizations one can prove the commutation relations @xmath516 & = & \lambda_m e^{b_{- } } e^{a_{+}}\\ \nonumber [ b_m^\dagger , e^{b_{- } } e^{a_{+ } } ] & = & -\mu_m e^{b_{- } } e^{a_{+}}~.\end{aligned}\ ] ] in the following we consider fermionic operators @xmath517 which obey @xmath518 = -\lambda_m \hat s$ ] and @xmath519 = \mu_m \hat s$ ] . then the operator @xmath520 commutes with all @xmath521 and @xmath522 . we therefore write @xmath523 and subsequently construct @xmath524 such that both sides of eq . ( b.4 ) yield identical matrix elements . + in the following we present the bosonization of the field operator for the right movers described by the @xmath525 and just mention the corresponding result for the left movers . the @xmath525 obey for @xmath526 the following commutation relations @xmath527 = - \frac{1}{\sqrt{m}}c_{l+m,+}\quad , \quad [ b^\dagger _ m , c_{l,+ } ] = - \frac{1}{\sqrt{m}}c_{l - m,+},\ ] ] the auxiliary field operator @xmath320 defined as @xmath528 obeys commutation relations as the operator @xmath517 discussed above @xmath529 = - \frac{1}{\sqrt{m}}e^{-imv}\tilde \psi_+(v)\ ; ; \ ; [ b^\dagger_m , \tilde\psi_+ ( v ) ] = - \frac{1}{\sqrt{m } } e^{imv } \tilde \psi_+ ( v)~.\ ] ] therefore it is of the form @xmath530 where the operator @xmath531 is given by @xmath532 as @xmath533 reduces the number of right movers by one , the operator @xmath534 , which commutes with all @xmath535 , also must have this property . in order to determine @xmath536 we work with the eigenstates of the noninteracting system ( compare eq . ( 41 ) ) @xmath537 after the limit @xmath142 the @xmath538 can run from minus to plus infinity . it is easy to see that @xmath539 has no overlap to excited states @xmath540 as @xmath541 commutes with the @xmath501 the rhs of eq . ( b.11 ) vanishes unless all @xmath542 are zero . this implies @xmath543 where @xmath544 is a c - number . in order to determine the @xmath545 we calculate @xmath546 using ( b.6 ) as well as ( b.8 ) . in the calculation of the matrix element with the fermionic form ( b.6 ) we use eq . ( 40 ) which yields @xmath547 the factor @xmath548 occurs because we have to commute @xmath525 through the product of @xmath549 fermionic operators of the left movers if we assume @xmath550 to be odd . we note that _ no _ such factor occurs for the corresponding matrix element of the left movers . the calculation of the ground state to ground state matrix element of @xmath320 using ( b.8 ) is simple as both exponentials involving the boson operators can be replaced by the unit operator and the matrix element is just @xmath545 . the comparison therefore yields @xmath551 and @xmath552 . together with ( b.12 ) and the definition @xmath553 this implies @xmath554 to the states in eq . ( b.10 ) and use again that @xmath555 commutes with the boson operators we obtain @xmath556 this shows that the operator @xmath557 is _ independent _ of @xmath204 and given by @xmath558 it follows immediately that @xmath334 is _ unitary _ , i.e. @xmath559 . from ( b.15 ) one can infer that for arbitrary functions @xmath560 of the number operator @xmath561 one has @xmath562 . concerning the @xmath563-dependence the operator @xmath334 resembles a one - dimensional tight - binding nearest neighbour hopping operator @xcite . therefore the eigenstates of @xmath334 are just the corresponding `` blochstates '' @xmath564 and the eigenvalue of @xmath334 is given by @xmath565 . if one defines the operator latexmath:[\[\hat{k}_+ \equiv~\sum_{\{m_l\}}~\sum_{\tilde{n}_-}~\int^\pi_{-\pi}~ the operator @xmath334 can be written as @xmath567 in the `` @xmath89-representation '' the particle number operator @xmath568 acts like the differential operator @xmath569 , which seems to imply the commutation relation @xmath570 = i { \hat 1}$ ] . this is _ not _ correct as an operator identity in the full hilbert space . in the @xmath89-representation the operator @xmath571 acts as a multiplication operator which destroys the @xmath572-periodicity in @xmath89 unless the state @xmath573 fulfills @xmath574 . in the corresponding expression @xmath578 _ no _ factor @xmath579 appears and therefore @xmath580 and @xmath581 anticommute , which is necessary to enforce anticommutation relations between @xmath582 and @xmath583 . the bosonization of @xmath525 itself is obtained by the inversion formula to ( b.6 ) @xmath584 we finally show how the bosonization of a _ single _ fermion operator described in this appendix relates to the bosonization of the _ product _ @xmath585 described in i. using ( b.8 ) and ( b.21 ) we can write the product as @xmath586 } e^{-i \tilde{{\cal n}}_+(u - v)}~.\end{aligned}\ ] ] here we have used the baker - hausdorff formula @xmath587}$ ] , valid if [ a , b ] commutes with a and b. the commutator in the exponent is given by @xmath588 & = & ~\sum^\infty_{n , n'=1}~\frac{e^{inu}}{\sqrt{n}}~\frac{e^{-i n ' v}}{\sqrt{n'}}~ [ b_n , b_{n'}^\dagger ] = ~\lim_{\eta \rightarrow 0 } \sum^\infty_{n=1 } ~e^{in ( u - v+i \eta)}/n \nonumber \\ & = & - \lim_{\eta \rightarrow 0 } \left[\ln \left ( 1 - e^{i(u - v + i \eta ) } \right ) \right]~.\end{aligned}\ ] ] in order to make the infinite series convergent we have added a convergence factor @xmath589 with @xmath590 going to zero in the end of the calculation . this is necessary as we had to perform the limit @xmath591 in eq . ( b.6 ) , i.e. we work with an _ infinite _ dirac see as in luttinger s paper @xcite . combining eqs . ( b.23 ) and ( b.24 ) gives @xmath592 which corresponds to eqs . ( 47 ) and ( 49 ) of i. a convergence factor also has to be introduced explicitely in the bosonization formula for the field operator ( b.8 ) if _ no _ normal ordering is used . if one uses the baker - hausdorff formula and ( incorrectly ) assumes the commutation relation @xmath593 = i { \hat 1}$ ] one obtains @xmath594}(u)},\ ] ] where the `` phase field '' operator @xmath595}(u)\equiv \hat k_{\alpha } + \alpha ( \tilde { { \cal n}}_{\alpha } -\frac { 1}{2})u + \phi_{\alpha}^{[\eta]}(u)+ ( \phi_{\alpha}^{[\eta]}(u))^{\dagger}+\frac{\pi}{2}(1+\alpha ) \tilde { { \cal n}}_{-\alpha},\ ] ] with @xmath596}(u ) = -i\sum_{n \neq 0}\theta ( \alpha n)\frac{e^{inu}}{\sqrt{| n|}}~b_n~e^{-|n| \eta/2}\ ] ] contains the boson as well as the particle number changing contribution . it is used in various applications @xcite e.g the discussion of transport . it is obviously more rigorous to use of the normal ordered form ( b.8 ) , as always done in the present paper . 99 n.w . ashcroft and n.d . mermin , solid state physics ( holt , rinehart and winston , new york , 1976 ) g.d . mahan , many - particle physics ( plenum , new york , 1990 ) for an excellent recent theoretical review see : j. voit , `` one - dimensional fermi liquids '' , rep . 58 * , 977 - 1116 ( 1995 ) k. schnhammer and v. meden , fermion - boson transmutation and comparison of statistical ensembles in one dimension , am . j. physics * 64 * , 1168 - 1176 ( 1996 ) s. tomonaga , remark on bloch s method of sound waves applied to many fermion problems , phys . * 5 * , 544 - 569 ( 1950 ) j.m . luttinger , an exactly soluble model of a many - fermion system , j. math . phys . * 4 * , 1154 - 1162 ( 1963 ) d.c . mattis and e.h . lieb , exact solution to a many - fermion system and its associated boson field , j. math . * 6 * , 304 - 312 ( 1965 ) f.d . haldane , luttinger liquid theory of one dimensional quantum fluids : i. properties of the luttinger model and their extension to the general 1d interacting spinless fermi gas , j. phys . * c14 * , 2585 - 2919 ( 1981 ) g. baym , lectures on quantum mechanics , benjamin / cunnings publishing , reading ( 1969 ) r.l . kronig , `` zur neutrinotheorie des lichtes iii '' physica * 2 * , 968 - 980 ( 1935 ) ; see also c.b . dover `` properties of the luttinger model '' , annals of physics * 50 * , 500 - 533 ( 1968 ) a. luther and i. peschel : single particle states , kohn anomaly and pairing fluctuations in one dimension , phys . rev . * b9 * , 2911 - 2919 ( 1974 ) v. meden and k. schnhammer , spectral functions for the tomonaga - luttinger model , phys . rev . * b46 * , 15753 - 15760 ( 1992 ) j. voit , charge - spin separation and spectral properties of luttinger liquids , phys . rev . * b47 * , 6740 - 6743 ( 1993 ) k. schnhammer and v. meden , nonuniversal spectral properties of the luttinger model , phys . rev . * b47 * , 16205 - 16215 ( 1993 ) b. dardel et al . , possible observation of luttinger - liquid behaviour from photoemission spectroscopy of one - dimensional organic conductors , europhys . lett . * 24 * , 687 - 697 ( 1993 ) we should express a warning : many authors in the present field , e.g. @xcite , use a different sign convention in the exponent ( @xmath597 ) . this is in contradiction to the usual sign convention for a plane wave in quantum mechanics . v. meden , phd - thesis : spektrale eigenschaften niedrigdimensionaler korrelierter elektronensysteme , 1996 ( unpublished ) m. abramowitz and i.a . stegum , handbook of mathematical functions ( dover , new york , 1970 ) d.c . mattis , `` new wave - operator identity applied to the study of persistent currents in 1d '' , j.math.phys.*15 * , 609 - 612 , 1974 c.l . kane and m.p.a . fisher , transmission through barriers and resonant tunneling in an interacting one - dimensional electron gas , phys.rev.b*46 * , 15233 - 15262 , 1992

the self - energy - functional approach proposed recently is applied to the single - band hubbard model at half - filling to study the mott - hubbard metal - insulator transition within the most simple but non - trivial approximation . this leads to a mean - field approach which is interesting conceptually : trial self - energies from a two - site single - impurity anderson model are used to evaluate an exact and general variational principle . while this restriction of the domain of the functional represents a strong approximation , the approach is still thermodynamically consistent by construction and represents a conceptual improvement of the `` linearized dmft '' which has been suggested previously as a handy approach to study the critical regime close to the transition . it turns out that the two - site approximation is able to reproduce the complete ( zero and finite - temperature ) phase diagram for the mott transition . for the critical point at , the entire calculation can be done analytically . this calculation elucidates different general aspects of the self - energy - functional theory . furthermore , it is shown how to deal with a number of technical difficulties which appear when the self - energy functional is evaluated in practice .

cond-mat0306278

the correlation - driven transition from a paramagnetic metal to a paramagnetic insulator ( mott - hubbard transition @xcite ) is one of the most interesting problems in condensed - matter physics . as a prime example for a quantum - phase transition , the mott - hubbard transition is important from the physical point of view but also for the development and test of general theoretical methods to treat correlated electron systems . the minimum model required to study the mott - hubbard transition is the single - band hubbard model . @xcite inherent to this model is the competition between the electrons kinetic energy which tends to delocalize the electrons and favors a metallic state and the on - site coulomb interaction which tends to localize the electrons to avoid double occupancies and thereby favors an insulating state at half filling . except for the one - dimensional case , @xcite however , exact results with regard to the nature of the transition and the critical interaction strength @xmath1 are not available even for this highly simplified model system . a direct numerical solution using exact - diagonalization or quantum monte - carlo methods @xcite suffers from the difficulty to access the thermodynamic limit or the low - temperature , low - energy regimes . considerable progress has been made in recent years due to the development of the dynamical mean - field theory ( dmft ) @xcite which focuses on the opposite limit of infinite spatial dimensions @xmath2 . @xcite within the dmft the problem is simplified by mapping the original lattice model onto an impurity model the parameters of which must be determined by a self - consistency condition . different techniques to solve the effective impurity model have been employed to study the mott transition within the dmft , iterative perturbation theory , @xcite exact diagonalization , @xcite renormalization - group methods , @xcite and quantum monte - carlo . @xcite one of the most important characteristic of the transition is the value of the critical interaction strength @xmath1 at zero temperature . roughly , the different techniques to solve the mean - field equations predict @xmath3 where @xmath4 is the width of the free density of states . recently , a self - energy - functional approach ( sfa ) has been put forward . @xcite the sfa is a general variational approach to correlated lattice models where the grand potential @xmath5 is considered as a functional of the self - energy @xmath6 . as this functional is constructed from an infinite series of renormalized skeleton diagrams , it is not known in an explicit form and the variational principle @xmath7 = 0 $ ] can not be exploited _ directly_. usually , one replaces the exact but unknown functional with an explicitly known but approximate one this is essentially the standard diagrammatic approach @xcite which leads to weak - coupling approximations in the end . @xcite opposed to this weak - coupling perturbational approach , the functional dependence @xmath8 $ ] is not approximated at all in the sfa . the key observation is that the functional , though unknown explicitly , can be evaluated on a restricted domain of trial self - energies @xmath9 . the evaluation of the functional is exact , while the approximation is due to the fact that the self - energy in the variational principle is no longer considered as arbitrary . in this way , depending on the choice for the space @xmath9 , some well - known but also some novel approximations can be realized . here we are interested in the single - band hubbard model with hamiltonian @xmath10 . as argued in ref . a useful trial self - energy has to be constructed as the exact self - energy of a different model ( `` reference system '' ) with hamiltonian @xmath11 . the reference system can be chosen arbitrarily it must , however , share the same interaction part with the original model @xmath10 . the variational parameters at one s disposal are therefore the one - particle parameters of the reference system @xmath12 . the trial self - energy is parameterized as @xmath13 , and the variational principle reads @xmath14 / \partial { \bm t } ' = 0 $ ] . to provide trial self - energies is the only purpose of the reference system @xmath11 . whenever one is able to compute @xmath6 for the reference system @xmath11 , an exact evaluation of @xmath15 $ ] is possible . choosing @xmath11 to be a system of decoupled sites , yields a hubbard - i - type approximation . an improved approximation is obtained when @xmath11 consists of decoupled clusters with a finite number of sites @xmath16 per cluster as has been considered in refs . . this approach not only recovers the so - called cluster - perturbation theory ( cpt ) @xcite but also gives a variational improvement ( v - cpt ) which e.g. allows to describe phases with spontaneously broken symmetry . @xcite another possibility is to take @xmath17 , which implies the trial self - energy to be local , but to include a coupling to a number of @xmath18 additional uncorrelated ( `` bath '' ) sites . in this case the reference system consists of a decoupled set of single - impurity anderson models ( siam ) with @xmath19 sites each . as has been shown in ref . , this approach not only recovers the dmft ( namely in the limit @xmath20 ) but also provides a new variant of the exact - diagonalization approach , namely for any finite @xmath18 . as compared to previous dmft - exact - diagonalization approaches , @xcite the construction gives a thermodynamically consistent approximation even for small @xmath18 . more complicated reference systems may be taken for the construction of consistent approximations , for example a system of decoupled clusters of size @xmath21 where each site in the cluster is coupled to @xmath18 additional bath sites . it has been shown @xcite that in the limit @xmath20 the cellular dmft ( c - dmft ) is obtained , @xcite while approximations with finite @xmath18 represent cluster approximations which `` interpolate '' between the cpt ( @xmath22 ) and the c - dmft ( @xmath23 ) . in the present paper the mott transition is studied within the most simple but non - trivial approximation : the variational principle @xmath7 = 0 $ ] is exploited using a local trial self - energy from a reference system with @xmath17 and a single additional bath site only , @xmath24 . as @xmath25 this approximation will be referred to as the two - site dynamical impurity approximation ( @xmath26-dia ) in the following . the paper is organized as follows : a brief review of the self - energy - functional approach will be given in the next section [ sec : sfa ] . the general aspects of the evaluation of the sfa are discussed in sec . [ sec : eval ] while sec . [ sec : local ] focuses on local approximations ( @xmath17 , @xmath27 arbitrary ) in particular . in sec . [ sec : twosite ] the further specialization to the case @xmath28 is considered . the @xmath26 dynamical - impurity approximation is motivated ( i ) by the fact that at the critical point for the mott transition the entire calculation can be done _ analytically _ , ( ii ) by the conceptual simplicity of the approach which rests on a _ single _ approximation only and ( iii ) by making contact with a linearized dmft ( l - dmft ) @xcite developed previously . this is discussed in detail in sec . [ sec : twosite ] while sec . [ sec : lin ] then presents the analytical calculation for the critical regime . the results are discussed in sec . [ sec : results ] . the complete phase diagram for @xmath0 and finite temperatures is addressed in sec . [ sec : temp ] and the conclusions are given in sec . [ sec : con ] . useful information on correlated electron systems can be gained by exact - diagonalization or quantum monte - carlo methods applied to a lattice of finite size . @xcite this approach , however , suffers from the difficulty to access the thermodynamic limit and is therefore of limited use to describe phases with long - order and phase transitions . on the other hand , using an _ embedding approximation _ , one can directly work in the thermodynamic limit and describe phase transitions while the actual numerical treatment has to be done a system of finite size only . in the context of an embedding technique we have to distinguish between the original model of infinite size @xmath10 and an ( e.g. spatially ) truncated _ reference system _ @xmath11 . the reference system must not necessarily be finite but it may consist of an infinite number of decoupled subsystems with a finite number of degrees of freedom each . in any case , @xmath11 must be exactly solvable . if one is interested not only in the equilibrium thermodynamics but also in the elementary one - particle excitations , an embedding technique should focus on a dynamical quantity , such as the frequency - dependent self - energy @xmath6 . the knowledge of @xmath6 then allows to derive a thermodynamic potential @xmath5 as well as different static and dynamic quantities via general relations . the main steps are the following : ( i ) truncate the original model @xmath10 to obtain a simpler model @xmath11 which is tractable numerically . ( ii ) calculate the self - energy @xmath29 of the reference system . ( iii ) use @xmath30 as an approximation for the self - energy of @xmath10 and determine the grand potential @xmath5 as well as further quantities of interest . ( iv ) to get a self - consistent scheme , optimize the parameters of @xmath11 by a feedback from the approximate solution at hand . ideally , the last step should be based on a general variational principle . this is exactly the strategy of the self - energy - functional approach ( sfa ) proposed recently . @xcite a brief review of the essentials of this approach is given in the following . consider a system of fermions on an infinite lattice with on - site coulomb interaction at temperature @xmath31 and chemical potential @xmath32 . its hamiltonian @xmath33 consists of a one - particle part which depends on a set of hopping parameters @xmath34 and an interaction part with coulomb - interaction parameters @xmath35 : @xmath36 the grand potential @xmath5 can be obtained from the stationary point of a self - energy functional @xmath37 \equiv { \rm tr } \ln ( - ( { \bm g}_0^{-1 } - { \bm \sigma})^{-1 } ) + f[{\bm \sigma } ] \label{eq : omega}\ ] ] as has been discussed in ref . . here the subscript @xmath34 indicates the parametric dependence of the functional on the hopping . this dependence is _ exclusively _ due to @xmath38 , the free green s function of @xmath10 . further , @xmath39 $ ] is the legendre transform of the luttinger - ward functional @xmath40 $ ] . as the latter is constructed as an infinite series of renormalized skeleton diagrams , @xcite the self - energy functional is not known explicitely . nevertheless , the _ exact _ evaluation of @xmath41 $ ] and the determination of the stationary point is possible @xcite on a _ restricted _ space @xmath9 of trial self - energies @xmath42 . due to this restriction the procedure becomes an approximation . generally , the space @xmath43 consists of @xmath12 representable self - energies . @xmath6 is termed @xmath12 representable if there are hopping parameters @xmath12 such that @xmath13 is the exact self - energy of the model @xmath44 ( `` reference system '' ) . note that both the original system @xmath10 and the reference system @xmath11 must share the same interaction part . for any @xmath6 parameterized as @xmath45 we then have : @xcite @xmath46 & = & \omega'({\bm t } ' ) \nonumber \\ & + & { \rm tr } \ln ( - ( { \bm g}_0({\bm t})^{-1 } - { \bm \sigma}({\bm t}'))^{-1 } ) \nonumber \\ & - & { \rm tr } \ln ( - ( { \bm g}_0({\bm t}')^{-1 } - { \bm \sigma}({\bm t}'))^{-1 } ) \ : , \label{eq : om}\end{aligned}\ ] ] where @xmath47 , @xmath48 , and @xmath45 are the grand potential , the free green s function and the self - energy of the reference system @xmath11 while @xmath49 is the free green s function of @xmath10 . for a proper choice of @xmath12 , a ( numerically ) exact computation of these quantities is possible . hence , the self - energy functional ( [ eq : om ] ) can be evaluated exactly for this @xmath13 . a certain approximation is characterized by a choice for @xmath43 . as @xmath6 is parameterized by @xmath12 , this means to specify a space of variational parameters @xmath12 . any choice will lead by construction to a non - perturbative approach which is thermodynamically consistent as an explicit expression for a thermodynamical potential is provided . it turns out that a stationary point of the self - energy functional is a saddle point in general . as in different standard variational methods , such as in the time - dependent density - functional approach , @xcite in the greens - function approach @xcite and also in a recently considered variant , @xcite this implies that there is no strict upper bound for the grand potential . for a further discussion of the general concepts of the sfa see ref . . ( a ) and a possible reference system @xmath11 ( b ) . @xmath11 is a set of decoupled single - impurity anderson models with one correlated ( @xmath50 ) impurity site and a number of @xmath51 uncorrelated ( @xmath52 ) bath sites each . in the figure @xmath53 . note that the interaction part is the same for ( a ) and ( b ) . variational parameters are the one - particle parameters of @xmath11 . , width=207 ] so far the discussion is completely general . in sec . [ sec : local ] we will consider @xmath10 to be the hubbard model and @xmath11 to be a system of decoupled single - impurity anderson models ( siam ) . each siam consists of @xmath27 sites , one correlated site ( with @xmath50 ) and @xmath51 uncorrelated `` bath '' sites ( @xmath52 ) . this is illustrated by fig . [ fig : refsystem ] for @xmath53 . note that for any choice of @xmath27 , the original system and the reference system share the same interaction part as required by the general theory . the one - particle parameters of @xmath11 are the variational parameters , i.e.the one - particle energies of the original sites and the bath sites and the hopping ( `` hybridization '' ) between them . as noted in ref . , the dynamical mean - field theory ( dmft ) is recovered in the limit @xmath54 . for @xmath55 one obtains a new variant of the dmft - exact - diagonalization approach . @xcite for a finite number of bath sites , the dmft self - consistency condition can not be strictly satisfied . in the dmft - ed method one therefore has to introduce a certain measure which allows to minimize the error due to the discretization of the bath . the conceptual advantage of the sfa consists in the fact that this measure is replaced by a variational procedure which is based on a physical variational principle . as shown in ref . , a very good quantitative agreement with results from full dmft calculations can be achieved for the quasi - particle weight even with @xmath53 . with @xmath28 a much simpler approach is considered here ( secs . [ sec : twosite ] and [ sec : lin ] ) which , however , is still thermodynamically consistent and allows for simple and systematic investigations of the mott transition . the evaluation of the self - energy - functional theory can be done by solving the euler equation @xcite corresponding to the variational principle . while such an approach is possible in principle , it appears complicated as dynamical two - particle quantities of the reference system are required . an attractive alternative consists in the direct calculation of the grand potential along eq . ( [ eq : om ] ) . the numerical computation of @xmath56 $ ] for a given set of one - particle parameters @xmath12 is straightforward for a reference system @xmath11 of finite size . there are , however , a few technical difficulties which appear in the practical calculation and which shall be discussed in the following . the problem of finding a stationary point of the function @xmath57 $ ] is not addressed here as this is a standard numerical problem very similar to the problem of finding a minimum of a real single - valued function of several arguments . @xmath56 $ ] consists of three parts as given by eq . ( [ eq : om ] ) . the grand potential of the reference system can be calculated as @xmath58 from the many - body eigenenergies @xmath59 of @xmath60 where @xmath61 is the total particle number operator . direct numerical diagonalization or ( at @xmath0 ) the lanczos technique @xcite may be used . next , the second term on the r.h.s . of eq . ( [ eq : om ] ) is discussed . in the following the dependence of @xmath6 on @xmath12 will be suppressed for convenience and its dependence on @xmath62 is made explicit in the notations . the diagonalization of @xmath63 yields ( via the lehmann representation ) the green s function @xmath64 and the free green s function @xmath65 of the reference system . the self - energy is then obtained from the dyson equation of the reference system @xmath66 . using the self - energy @xmath67 , one obtains the ( approximate ) green s function of the original model via @xmath68 . the remaining task is to calculate @xmath69 for a lattice consisting of a finite number of sites @xmath70 . the thermodynamic limit @xmath71 is performed in the end . translational symmetry is not necessarily required . it is important to note that @xmath72 is causal , i.e. @xmath73 with @xmath74 , @xmath75 hermitian and @xmath75 positive definite for any real @xmath62 ( @xmath76 is a positive infinitesimal ) . the causality of the green s function @xmath72 is ensured by the causality of @xmath67 and @xmath77 , see appendix [ sec : causal ] . the latter are causal as these are exact quantities . let @xmath78 be the ( real , first - order ) poles of @xmath72 . for @xmath79 we have @xmath80 where due to the causality of @xmath72 the matrix @xmath81 is positive definite . therefore , the frequency - dependent diagonalization of @xmath82 with unitary @xmath83 ( for real @xmath62 ) yields a diagonal green s function @xmath84 with elements @xmath85 that are real for real @xmath62 and have first - order poles at @xmath86 with _ positive _ residues . as the green s function can be written as @xmath87 , the unitary transformation @xmath83 also diagonalizes the real and symmetric matrix @xmath88 , i.e. @xmath89 and @xmath90 . the @xmath91 are real for real @xmath62 and have first - order poles at @xmath92 with @xmath93 being the poles of the self - energy . for @xmath94 we have @xmath95 with positive definite @xmath96 . consequently , the residues of @xmath91 at @xmath92 are _ positive_. one can write @xmath97 and @xmath98 with @xmath99 and where @xmath100 and @xmath101 are analytical in the entire @xmath62 plane . as @xmath102 and @xmath103 for @xmath104 , one has @xmath105 and @xmath106 and real . consequently , @xmath107 and @xmath108 . this will be used in the following . the trace `` tr '' in eq . ( [ eq : om ] ) consists of a sum @xmath109 over the fermionic matsubara frequencies @xmath110 ( @xmath111 integer ) and a trace `` tr '' with respect to the quantum numbers @xmath112 ; see eq . ( [ eq : ham ] ) . the convergence of the frequency sum is ensured by the usual factor @xmath113 from the diagram rules . the calculation then proceeds as follows : @xmath114 with @xmath115 in eq . ( [ eq : trlng ] ) @xmath116 denotes the principal branch of the logarithm , @xmath117 is the fermi function , and @xmath118 is a contour in the complex @xmath62 plane enclosing the first - order poles of the fermi function in counterclockwise direction . in step ( a ) the transformation @xmath83 is performed under the trace . convergence of the integral for @xmath119 is ensured by the fermi function and by the factor @xmath120 , respectively . step ( b ) results from analytical continuation to real frequencies . in step ( c ) @xmath107 has been used ( see appendix [ sec : fom ] ) . at this point the causality of the green s function is essential , as discussed above . in step ( d ) the fermi function is written as a derivative with respect to @xmath62 and integration by parts is performed . step ( e ) uses the results of appendix [ sec : theta ] for the derivative of the step function . as the different diagonal elements of the green s function , @xmath85 , have the same set of poles and zeros , and @xmath91 with vanishing residue . note that adding a pole with zero weight to @xmath85 can only be accomplished if simultaneously a pole with zero weight is added to @xmath91 , and @xmath121 is unchanged . ] the sum over @xmath122 becomes trivial and yields a factor @xmath123 only where @xmath70 is the dimension of the hopping matrix , i.e. the number of orbitals . the factor 2 accounts for the two spin directions . the contribution from the poles of the self - energy ( appendix [ sec : theta ] ) is denoted by @xmath124 . apart from this correction term , @xmath125 turns out to be the grand potential of a system of _ non - interacting _ quasi - particles with _ unit _ weight and energies given by the poles of @xmath68 . consider now the third term on the r.h.s . of eq . ( [ eq : om ] ) . a calculation completely analogous to eq . ( [ eq : trlng ] ) results in : @xmath126 here @xmath127 are the poles of @xmath128 . again , the first term in eq . ( [ eq : trlngs ] ) is the grand potential of a non - interacting system of fermions with one - particle energies given by the poles of the green s function @xmath128 . the same holds for the second term , but with energies given by the poles of the self - energy . by construction , the self - energy is the same for both , the original system and the reference system . hence , the same correction term @xmath124 appears in eqs . ( [ eq : trlng ] ) and ( [ eq : trlngs ] ) and cancels out in eq . ( [ eq : om ] ) . note that a pole of @xmath72 at @xmath86 with residue @xmath129 implies a pole of @xmath64 at the same frequency @xmath78 ( with residue @xmath130 ) . hence , contributions due to poles with vanishing residues cancel out in eq . ( [ eq : om ] ) . the reason is the following : suppose that @xmath131 for @xmath62 close to @xmath78 with residue @xmath132 . for the diagonal elements of @xmath88 this implies that @xmath133 near @xmath78 . a zero of @xmath91 with infinite positive coefficient @xmath134 must be due to @xmath67 . therefore , for the diagonal elements of @xmath135 this means that @xmath136 . consequently , @xmath137 for @xmath62 close to @xmath78 with @xmath132 . the argument also shows that although the residues do not appear in eqs . ( [ eq : trlng ] ) and ( [ eq : trlngs ] ) explicitly , one can state that poles with small residues will give a small contribution in eq . ( [ eq : om ] ) . based on the causality of the respective green s functions , an efficient algorithm can be set up to find @xmath78 and @xmath127 numerically which are then needed in eqs . ( [ eq : trlng ] ) and ( [ eq : trlngs ] ) . the poles of @xmath64 are directly obtained from the diagonalization of the reference system . the problem consists in finding the poles of @xmath72 . allowing for poles with vanishing ( very small ) residue , one can assume the function @xmath138 for fixed but arbitrary @xmath122 to display all the poles of @xmath64 . since @xmath138 is of the form @xmath139 with @xmath140 , it is monotonically decreasing . hence , there is exactly one zero of @xmath138 located in the interval between two adjacent poles @xmath127 and @xmath141 . as @xmath138 is monotonous , the zero @xmath142 can easily be found numerically by an iterative bisection procedure . the zeros of @xmath138 are the poles of @xmath143 and the poles of @xmath143 are the same as the poles of @xmath91 . now , since @xmath90 , @xmath85 and @xmath138 must have the same set of zeros . the function @xmath85 is monotonically decreasing . therefore , the poles of @xmath85 can be found between the @xmath142 by using the same iterative bisection procedure once again . in the following we consider @xmath10 to be the single - band hubbard model : @xmath144 for the reference system @xmath11 shown in fig . [ fig : refsystem ] , the self - energy is local : @xmath145 . clearly , the approximation is the better the more degrees of freedom are included in @xmath11 . the optimal local approximation is obtained with the most flexible ( but local ) trial self - energy . this is the dmft which is recovered for an infinite number of uncorrelated bath sites ( per original correlated site ) , i.e. for @xmath146 . on the other hand , @xmath147 corresponds to a hubbard - i - type approximation . here it will be shown that , for arbitrary @xmath27 , the evaluation of the self - energy functional reduces to a one - dimensional integration only . this is an important simplification for any practical numerical ( or even analytical ) calculations . in case that the self - energy is local it is advantageous to start from step ( c ) in eq . ( [ eq : trlng ] ) . assuming translational symmetry , the matrix @xmath88 is diagonalized by fourier transformation to reciprocal space . its eigenvalues @xmath91 are given by @xmath148 where @xmath149 and @xmath150 is the tight - binding bloch dispersion . the self - energy is taken to be spin - independent and independent of the site index , i.e. a paramagnetic homogeneous phase is assumed for simplicity . the second term on the r.h.s . of eq . ( [ eq : om ] ) then becomes : @xmath151 here @xmath70 is the number of lattice sites , the factor 2 stems for the spin summation , @xmath152 is the non - interacting density of states , and @xmath153 is an antiderivative . note that there is no `` correction term '' @xmath124 , as the derivative in eq . ( [ eq : trlng1 ] ) is with respect to @xmath154 . the reference system @xmath11 is a set of decoupled single - impurity anderson models with one correlated ( `` impurity '' ) site and @xmath51 bath sites each . the hamiltonian is @xmath155 with @xmath156 the parameters @xmath157 , @xmath158 , and @xmath159 for @xmath160 are the on - site energies of the impurity site and of the bath sites , and the hybridization between them , respectively . these are the variational parameters of the theory . the third term on the r.h.s . of eq ( [ eq : om ] ) reads : @xmath161 where the first equation is derived in appendix [ sec : hur ] and @xmath162 is the impurity green s function , @xmath163 is the ( free ) conduction - band green s function , and @xmath164 is the hybridization function . the final expressions ( [ eq : trlng1 ] ) and ( [ eq : trlngs1 ] ) involve one - dimensional integrations only and can therefore be calculated numerically without serious problems . in the following we will focus on the case @xmath26 , i.e. on the two - site dynamical - impurity approximation ( @xmath26-dia ) . there are different intentions which are followed up : \(i ) for any approximation within the context of the sfa , one has to compute the self - energy of the reference system @xmath11 . as the interaction part is the same for both , @xmath10 and @xmath11 , this still constitutes a non - trivial many - body problem which can only be treated by numerical means in most cases . the reference system characterized by @xmath17 and @xmath28 , represents an exception : for a special point in the space of model parameters ( zero temperature , half - filling and @xmath165 , the critical interaction for the mott transition ) , the entire calculation can be done _ analytically_. this not only includes the diagonalization of the reference system which actually is a simple dimer model but also and more important here , the exact evaluation of the self - energy functional for the trial self - energies considered and the subsequent variational optimization . therefore , the study of the @xmath26-dia is ideally suited to elucidate different technical points which are relevant for any @xmath21 and @xmath27 and which must be considered carefully . \(ii ) the @xmath26-dia must be considered as _ inferior _ when compared to approximations with higher @xmath27 and when compared to @xmath166 ( the dmft ) in particular . on the other hand , one has to keep in mind that the dmft must always be supplemented by an additional ( numerical ) method to solve the mean - field equations which necessarily involves additional approximations . even if the additional approximations can be controlled , dmft results always depend on the accuracy of the numerical method employed . as concerns the @xmath26-dia , there is no such difficulty : the theory rests on a _ single _ approximation only , namely the restriction of the space @xmath9 to self - energies representable by the two - site reference system the rest of the calculation is rigorous . it is this conceptual simplicity which makes the approximation attractive . \(iii ) that an approach referring to an @xmath26-site siam is able to give reasonable results has been shown beforehand by the linearized dmft ( l - dmft ) . @xcite the l - dmft is a well - motivated but ad - hoc simplification of the full dmft and maps the hubbard model self - consistently onto the @xmath26-site siam just at the critical point for the mott transition . the l - dmft can be also be considered to represent the lowest - order realization of a more general projective self - consistent method ( pscm ) . @xcite as compared to the full dmft , the linearized theory yields surprisingly good estimates for the critical interaction @xmath1 in the single - band model , on translation invariant lattices @xcite as well as for lattice geometries with reduced translational symmetry . @xcite the approach can also be extended beyond the critical regime . @xcite the main disadvantage of the l - dmft is that it not consistently derived from a thermodynamical potential . another intention of the present paper is therefore to suggest a two - site method that conceptually improves upon the l - dmft in this respect . in fact , as the parameters of the effective @xmath26-site impurity model are determined via a physically meaningful variational principle , the two - site approximation within the sfa should be regarded as an _ optimal _ two - site approach . the interesting question is , of course , whether or not this improvement of the method also implies improved results . to this end the results from the analytical evaluation of the @xmath26-dia have to be compared with those of the l - dmft and with available numerical results for @xmath167 ( full dmft ) . this represents a good check of the practicability of the new method . as the @xmath26-site dia still represents a very handy method , it can also be employed to investigate overall trends . it will be interesting to study the critical interaction @xmath1 for a variety of different geometries ( i.e. for different free densities of states ) . ( per site ) as calculated from eq . ( [ eq : om ] ) and eqs . ( [ eq : trlng1 ] ) and ( [ eq : trlngs1 ] ) for the reference system with @xmath26 as a function of the hybridization strength @xmath168 ( only the difference @xmath169 is plotted ) . calculations for the hubbard model at zero temperature @xmath0 and half - filling ( @xmath170 ) . the free density of states is taken to be semi - elliptical with a band width @xmath171 . the interaction strength is varied from @xmath172 to @xmath173 as indicated . arrows indicate stationary points of the self - energy functional . the mott transition takes place at a critical interaction strength @xmath174 . , width=283 ] let us first consider a numerical evaluation of the theory . [ fig : omega ] shows the results of a numerical calculation along the lines discussed above for the paramagnetic phase of the hubbard model at half - filling and zero temperature . the free density of states ( dos ) @xmath175 is taken to be semi - elliptical with a band width @xmath171 . the calculations are performed using the @xmath26-dia , i.e. there is one bath site ( per correlated site ) only . due to manifest particle - hole symmetry , two of the variational parameters , the on - site energies , are already fixed : @xmath176 and @xmath177 . as a function of the remaining variational parameter , the hybridization strength @xmath178 , the grand potential @xmath179 $ ] shows two ( non - equivalent ) stationary points for @xmath172 ( see fig . [ fig : omega ] ) : a minimum at a finite @xmath180 and a maximum at @xmath181 ( as @xmath182 , there is another minimum at @xmath183 which can be ignored here ) . the two stationary points correspond to two physically different phases : for @xmath184 the interacting local density of states is finite at @xmath185 while it vanishes for @xmath181 . so there is a metallic and an insulating phase coexisting . due to the lower @xmath5 at the respective stationary point , the metallic phase is stable as compared to the insulating one . with increasing @xmath186 the optimal @xmath187 and the energy difference @xmath188 decrease . for @xmath189 there is a metal - insulator ( mott - hubbard ) transition which is characterized by a coalescence of the stable metallic with the metastable insulating phase . for @xmath190 there is the insulating phase only . qualitatively , this continuous transition is completely consistent with the preformed - gap scenario @xcite ( however , see also refs . ) . for @xmath191 the self - energy is a two - pole function . this leads to a three - peak structure in the interacting local density of states : similar as in the full dmft , there are two hubbard `` bands '' separated by an energy of the order of @xmath186 , and a quasi - particle resonance at @xmath185 . on approaching the critical interaction @xmath1 from below , the weight @xmath154 of the resonance vanishes linearly @xmath192 leaving a finite gap for @xmath190 . as shown in ref . the quasi - particle weight calculated from the self - energy at the respective optimal @xmath193 is in a very good quantitative agreement with results from full dmft calculations in the whole range from @xmath52 to @xmath165 for @xmath53 and even for @xmath26 which is the case considered here . in the following we will concentrate on the critical regime @xmath194 . it will be shown that the critical interaction strength can be calculated analytically for @xmath26 . the independent analytical result can be compared with the numerical one of the preceeding section . this represents a strong test of the numerics . consider the function @xmath179 $ ] . as @xmath182 there must be a stationary point of @xmath195 at @xmath181 for any @xmath186 . this implies that the linear term in an expansion around @xmath181 is missing , i.e. : @xmath196 the coefficient @xmath197 depends on @xmath186 . assuming that the metal is stable against the insulator for @xmath191 , we must have @xmath198 for @xmath191 and @xmath199 for @xmath200 . this is a necessary condition for a continuous ( `` second - order '' ) transition and consistent with the numerical results displayed in fig . [ fig : omega ] . therefore , the critical point for the mott transition within the two - site model is characterized by @xmath201 the task is to calculate the three contributions to the grand potential following eq . ( [ eq : om ] ) , to expand in @xmath168 up the the second - order term and to find the interaction strength satisfying the condition ( [ eq : cond ] ) . consider the grand potential of the reference system first . with @xmath202 , @xmath203 and @xmath170 the ground state of the two - site system , eq . ( [ eq : siam2 ] ) , lies is the invariant subspace with total particle number @xmath204 . the ground - state energy @xmath205 is readily calculated : @xmath206 at @xmath0 the grand potential of the reference system is @xmath207 . therefore , @xmath208 actually , this is the grand potential _ per site_. we proceed with the third term on the r.h.s . of eq . ( [ eq : om ] ) . for the analytical calculation , it is convenient to start from eq . ( [ eq : trlngs ] ) : @xmath209 where it has been used that @xmath210 for @xmath0 . the factor 2 is due to the spin degeneracy . @xmath124 will cancel out later . the green s function of the two - site model is easily calculated . there are four excitation energies , labeled by @xmath211 , given by the four poles of the impurity green s function at : @xmath212 this yields : @xmath213 finally , for the second term on the r.h.s . ( [ eq : om ] ) we have : @xmath214 here , eq . ( [ eq : trlng ] ) has been used with @xmath149 , and the @xmath215 sum has been replaced by an integration over @xmath154 weighted by the free dos @xmath175 . for a given @xmath216 , the quasi - particle energies are obtained as the poles of the lattice green s function in reciprocal space , i.e. from the solutions @xmath217 of the equation @xmath218 . to find the roots , the self - energy of the reference system is needed : @xmath219 this leads to a cubic equation : @xmath220 the solutions for @xmath181 are easily obtained . for small @xmath221 we find one root near @xmath185 : @xmath222 and another one near @xmath223 : @xmath224 because of the step function in eq . ( [ eq : uclatt ] ) , the third root near @xmath225 is not needed here . this yields : @xmath226 inserting the results , eqs . ( [ eq : ucie ] ) , ( [ eq : ucimp ] ) , and ( [ eq : uclatt1 ] ) , into eq . ( [ eq : om ] ) and using the symmetry @xmath227 , we find : @xmath228 now , the condition ( [ eq : cond ] ) gives the critical @xmath186 for the mott transition : @xmath229 this implicit analytical equation for @xmath1 is the final result . for an arbitrary free dos no further simplifications are possible . it should be stressed once more that eq . ( [ eq : uc ] ) results from an exact variational principle simply by the restriction that the trial self - energies be representable by the two - site reference system . the two - site model generally yields a two - pole self - energy which is the minimal requirement for a three - peak structure of the single - particle excitation spectrum . therefore , it may also be stated that eq . ( [ eq : uc ] ) gives the optimal result for a two - pole self - energy . ( [ eq : uc ] ) turns out to be more complicated as compared to the result of the linearized dmft : @xcite @xmath230 interestingly , the first term on the r.h.s . of eq . ( [ eq : uc ] ) resembles brinkman and rice ( gutzwiller ) result @xcite for the critical interaction : @xmath231 a rough approximation of the second term in ( [ eq : uc ] ) using @xmath232 then yields @xmath233 . this reduction of the critical interaction as compared to the br result is the dominating effect . for a semi - elliptical free dos with variance @xmath234 ( band width @xmath171 ) , one has @xmath235 and @xmath236 . the numerical solution of eq . ( [ eq : uc ] ) is straightforward and yields @xmath237 . as expected this is fully consistent with the numerical determination of @xmath1 ( sec . [ sec : twosite ] ) . the result is close to the one predicted by the l - dmft . compared to numerical estimates from full dmft calculations @xmath238 ( nrg , ref . ) and @xmath239 ( pscm , ref . ) , one can state that the prediction of the l - dmft is improved . as a function of the inverse band width @xmath240 . calculations for different ( normalized ) free densities of states @xmath175 as indicated ( where @xmath241 and @xmath242 in the insets ) . ] . critical interaction @xmath1 as obtained from the sfa within the @xmath26 dynamical impurity approximation for different free densities of states . each dos is normalized , @xmath243 , symmetric , @xmath227 , and has unit variance , @xmath244 . @xmath245 is a cutoff function . @xmath4 is the band width . the critical interaction from the gutzwiller ( brinkman - rice ) approach @xcite and the available full dmft results are shown for comparison . the linearized dmft @xcite yields @xmath246 for any dos with unit variance . [ tab : uc ] [ cols="^,^,^,^,^",options="header " , ] results for different free densities of states are displayed in fig . [ fig : uc ] and tab . [ tab : uc ] . for a meaningful comparison , each dos has unit variance @xmath234 where @xmath247 . the band width @xmath4 varies . the dos with the smallest @xmath4 ( but @xmath234 ) consists of two @xmath248-peaks while @xmath249 for a gaussian dos . clearly , there is no true mott transition in the former case ( but also in the latter this is questionable ) . however , the inclusion of these extreme cases is instructive when studying the trend of @xmath1 as a function of @xmath4 . note that for @xmath234 the l - dmft yields @xmath250 irrespective of the form of the dos . the critical interaction from the two - site dia is always close the l - dmft result but considerably lower than the gutzwiller value . the two - site dia confirms the central prediction of the l - dmft that it is the variance of the dos that is crucial for the critical interaction . however , there is also a weak trend superimposed , namely a systematic increase of @xmath1 with decreasing band width @xmath4 ( with the exception of the gaussian dos ) . this is the same trend that is also present in the gutzwiller results . it would be interesting to see whether or not this trend is confirmed by full dmft calculations . comparing the full dmft results for the gaussian , for the rectangular and the semi - elliptic dos shows the mentioned trend . however , the comparison of the two - site dia and of the l - dmft with the full dmft for the available numerical data is not fully conclusive . so far only the zero - temperature limit has been considered . the mott transition at finite temperatures , however , is particularly interesting as there is a comprehensive physical picture available with a comparatively complex phase diagram . this phase diagram in the @xmath186-@xmath31 plane was first suggested by the iterative perturbation theory . @xcite the nature of the transition and the topology of the phase diagram have been established entirely using analytic arguments @xcite and has been worked out quantitatively using different numerical methods . @xcite the critical regime is accessible by the projective self - consistent method . @xcite as this phase diagram represents a valuable benchmark for any approximation , it is interesting to see whether or not it can be rederived within the most simple two - site dia . the application of the theory for finite @xmath31 is straightforward but can no longer be done analytically . as for the derivation of the @xmath0 results in the non - critical regime ( see sec . [ sec : twosite ] ) , calculations are performed along the lines of sec . [ sec : local ] . it turns out that for finite temperatures the mott transition is predicted to be discontinuous . this is demonstrated in fig . [ fig : omtemp ] which shows @xmath195 for different @xmath31 and fixed @xmath251 ( @xmath26 ) . at low but finite @xmath31 there are three stationary points ( see arrows ) corresponding to three different phases of the system . the metallic phase has the largest value @xmath193 . with increasing temperature @xmath187 decreases and @xmath252 for low @xmath31 this gives a linear entropy @xmath253 and a linear specific heat @xmath254 as it is characteristic for a fermi liquid ( @xmath170 is fixed ) . the insulating phase has the smallest value @xmath255 , and the grand potential @xmath256 for low @xmath31 . for @xmath257 the entropy approaches @xmath258 reflecting a @xmath259-fold ground - state degeneracy of the insulator which is known to be an artifact of mean - field theory . @xcite the specific heat @xmath260 vanishes exponentially for @xmath261 . for @xmath251 and low @xmath31 , however , the insulating phase is metastable as compared to the metallic phase since @xmath262 . due to the different behavior at low @xmath31 , there must be a temperature @xmath263 where @xmath264 . in fact , for @xmath265 the insulator is stable as compared to the metal . at @xmath263 or , conversely , at a critical interaction @xmath266 , the @xmath26-dia thus predicts a first - order transition with a discontinuous jump in the entropy . as can be seen in fig . [ fig : omtemp ] , the metallic phase ceases to exist for still higher temperatures , and only the insulating phase is left . this is due to a coalescence of the metallic with a third phase at another critical temperature @xmath267 ( or , conversely , at a critical interaction @xmath268 . this third phase turns out to be less stable as compared to the metal and to the insulator in the entire parameter regime . similarly , one can define a critical temperature @xmath269 ( at smaller @xmath186 ) and thus a critical interaction @xmath270 where the insulating phase coalesces with the third phase . for different @xmath31 as a function of @xmath168 . the arrows indicate the stationary points . ] calculations for different @xmath186 and @xmath31 have been performed to obtain the entire phase diagram within the two - site approach . the result is shown in fig . [ fig : pd ] . for @xmath271 there is a metallic phase which is smoothly connected to the @xmath52 limit . on the other hand , for @xmath272 there is an insulating phase which is smoothly connected with the mott insulator for @xmath273 . metallic and insulating phase are coexisting for @xmath274 . at zero temperature , the metal is stable as compared to the insulator in the entire coexistence region , and the transition is continuous at @xmath275 . for finite temperatures the transition is discontinuous at a critical interaction @xmath266 with @xmath276 . with increasing temperatures the coexistence region @xmath274 shrinks and disappears above a critical temperature @xmath277 which is defined via @xmath278 . above @xmath277 the metallic phase is smoothly connected with the insulating phase . . reference system @xmath11 : two - site model . @xmath279 : critical interaction strength up to which there is a ( metallic ) solution smoothly connected with the metal at @xmath52 . @xmath280 : critical interaction strength down to which there is an ( insulating ) solution smoothly connected with the mott insulator for @xmath273 . @xmath281 : first - order transition line in the coexistence region , terminating at the second - order critical point at @xmath277 . ] qualitatively , this is exactly the same result that is obtained within the full dmft . @xcite it is very remarkable that the rather complex topology of the phase diagram can be reproduced with a comparatively simple two - site model as a reference system . this shows that it is essential to perform the mapping onto the reference system in a way which is thermodynamically consistent and which is controlled by a physical variational principle . note that the l - dmft , or its extension away from the critical point at @xmath0 , @xcite fails to reproduce the discontinuous transition for @xmath282 and the critical temperature @xmath277 due to the ad - hoc character of the approximation . quantitatively , one should expect some deviations from the results of the full dmft due to the simplicity of the two - site model . comparing with the nrg result @xcite for @xmath283 , the two - site approximation overestimates the critical interaction by a few per cent . the determination of the critical temperature is difficult in any numerical approach . @xmath284 is estimated from the qmc and nrg results of refs . . thus , the two - site approximation underestimates @xmath277 by more than a factor 2 . it is worth mentioning that the numerical effort to obtain the entire phase diagram in the @xmath285 plane is of the order of a few minutes on a standard workstation which is negligible as compared to a dmft - qmc calculation . our present understanding of the mott - hubbard transition is mainly based on the exact solution of the one - dimensional case on the one hand and on the @xmath2 mean - field picture provided by the dmft on the other hand . for the physically more relevant two- or three - dimensional hubbard model , however , neither the analytical concepts developed for @xmath286 nor the mean - field theory can be expected to give an essentially correct and comprehensive description . direct numerical approaches , such as qmc , are able to give essentially exact results for a @xmath287 dimensional lattice of finite spatial extent . however , the relevance of the results for the thermodynamic limit and , in many cases , for the low - temperature , low - energy regime remains an open question . in this situation , a combination of a direct numerical approach for systems of finite size with the mean - field concept appears to be advantageous . this is more or less the direction that is followed up by the different cluster extensions of the dmft . @xcite via a generalized mean - field concept , the original lattice problem given by the hamiltonian @xmath10 is mapped onto a cluster problem described by a hamiltonian @xmath11 . in fact , due to the presence of strong short - range antiferromagnetic correlations a considerable revision of the mean - field picture is probably necessary . @xcite however , the cluster extensions of the dmft suffer from the fact that the mean - field formulation requires that the sites of the finite cluster are coupled to uncorrelated baths with an infinite number of degrees of freedom each . this circumstance complicates the practical solution of the @xmath11 problem ( which must be solved repeatedly during the self - consistency cycle ) so much that additional approximations are required and/or stochastic numerical methods . the self - energy - functional approach offers an interesting alternative as the reference system @xmath11 , the original model @xmath10 is mapped onto , is by no means completely predetermined . the sfa gives a very general prescription how this mapping can be performed while keeping the thermodynamical consistency of the approach as an explicit expression for the grand potential is provided . in this way the dmft and the cellular dmft are recovered as certain limits for special choices of @xmath11 , namely for a decoupled system of clusters of size @xmath17 or @xmath16 including couplings to @xmath288 bath degrees of freedom . there is , however , the additional possibility to construct approximations with @xmath289 which are consistent in themselves in the same way as are the dmft and the c - dmft . now , the question is whether or not an approximation with a finite number of bath sites @xmath289 can be tolerated . note that even in the ( c-)dmft it becomes necessary to reduce the problem posed by @xmath11 with @xmath288 to a numerically tractable one with a _ finite _ number of degrees of freedom . while this additional approximation is controlled within qmc or ed approaches , for example , it nevertheless violates thermodynamical consistency . within the sfa , on the contrary , the approximation is derived from a thermodynamical potential for any @xmath18 , infinite or finite or even so small as @xmath24 . depending on the quantity and accuracy one is interested in , there can be a rapid convergence with respect to @xmath18 ( cf . ref . ) so that a small number might be sufficient . for cluster approximations , the best choice is by no means clear as it must be balanced with the choice of the cluster size @xmath21 . this strongly depends on the lattice dimension . in fact , it has been shown @xcite that for the one - dimensional hubbard model a larger @xmath21 is to be preferred as compared to a larger @xmath18 . in this context , it is interesting to see where one are led to with the most simple reference system conceivable . this is a model @xmath11 characterized by @xmath290 and @xmath24 which yields the so - called two - site dynamical impurity approximation ( dia ) . the answer is given with the present paper : even in this approximation the mott transition shows up . at zero temperature the transition turns out to be continuous at a finite critical interaction @xmath1 where there is a coalescence of the metallic with a coexisting insulating phase . for finite temperatures , on the other hand , the transition is discontinuous . the first - order line @xmath266 terminates at a second - order critical point ( @xmath291 ) , and for @xmath292 there is a smooth crossover only . this is qualitatively the same picture that has been found beforehand in the full dmft . furthermore , the two - site dia yields a @xmath1 at @xmath0 that is surprisingly close to the result of the full dmft . in this respect the two - site dia is of similar quality as the linearized dmft , @xcite another approach that is based on a mapping onto the two - site siam . whether or not the @xmath1-results of the two - site dia improve on those of the l - dmft is difficult to decide in view of the existing ( full ) dmft data . more important , however , the conceptual improvement gained is substantial : while the mapping procedure on the two - site model , though physically motivated , is done in an ad - hoc way in the l - dmft , the two - site dia is derivable from a thermodynamical potential and can be characterized as an optimal two - site approach in fact . the fact that a reasonable mean - field description of the mott transition is possible even for the simple two - site dia , motivates further sfa studies of the transition with improved approximations in the future ( e.g. using finite clusters , @xmath293 , and small @xmath18 ) . in case of larger and more complex reference systems , however , it becomes more and more important to have a numerically efficient and accurate method for the evaluation of the self - energy functional at hand . the detailed analysis of the different contributions to the functional has been another major intention of the present study . using the causality properties of the green function , it has been shown that the important @xmath294 term can be written as the grand potential of a system of non - interacting quasi - particles with unit weight ( apart from a correction term that cancels out in the functional eventually ) . while this result is also interesting by itself , it is very well suited for an efficient numerical implementation . as the energies of the ficticious quasi - particles are given by the poles of the green function , they can be found in a comparatively simple way by exploiting general causality properties . in particular , there is no need for a small but finite lorentzian broadening , @xmath295 , which must be introduced in a more direct way to evaluate the functional . @xcite this will become important when studying the critical regime of the mott transition , where an accurate computation of small energy differences is vital , using an approximation with several variational parameters . studies in this direction are intended for the future . it is a pleasure to acknowledge valuable discussions with b. michaelis , e. arrigoni and w. hanke . this work is supported by the deutsche forschungsgemeinschaft ( sfb 290 `` metallische dnne filme : struktur , magnetismus und elektronische eigenschaften '' and sfb 410 `` ii - vi - halbleiter : wachstumsmechanismen , niederdimensionale strukturen und grenzflchen '' ) . here is will be shown that the green s function @xmath296 is causal : for any @xmath12 the self - energy @xmath297 is causal since it is defined to be the exact self - energy of the reference system @xmath298 . likewise , the green s function @xmath49 is causal . it has to be shown that @xmath299 is causal if @xmath6 and the free green s function @xmath49 are known to be causal . causality of @xmath296 means ( i ) that @xmath300 is analytic in the entire complex @xmath62 plane except for the real axis and ( ii ) that @xmath301 for real @xmath62 with @xmath302 , @xmath303 hermitian and @xmath303 positive definite . ( i ) is easily verified . to show ( ii ) we need the following _ lemma : _ for hermitian matrices @xmath304 , @xmath305 with @xmath305 positive definite , one has @xmath306 with @xmath307 , @xmath308 hermitian and @xmath308 positive definite . the _ proof _ of the lemma is straightforward : @xmath309 with @xmath310 , @xmath311 hermitian and @xmath312 positive definite . let @xmath313 with @xmath35 unitary and @xmath314 diagonal . then @xmath315 with @xmath307 hermitian and @xmath316 hermitian and positive definite . consider a fixed ( real ) frequency @xmath62 . since @xmath317 is causal , @xmath318 with @xmath319 , @xmath320 hermitian and @xmath320 positive definite . using the lemma , @xmath321 with @xmath322 , @xmath323 hermitian and @xmath323 positive definite . since @xmath6 is causal , @xmath324 with @xmath325 , @xmath326 hermitian and @xmath326 positive definite . therefore , @xmath327 with @xmath328 hermitian and @xmath329 hermitian and positive definite . using the lemma once more , shows @xmath296 to be causal . in a polar representation @xmath334 with @xmath335 . on the principal branch of the logarithm and for real @xmath62 one has @xmath336 . for any @xmath337 we have @xmath338 from eqs . ( [ eq : pos1 ] ) and ( [ eq : pos2 ] ) and thus @xmath339 for @xmath340 , and @xmath341 for @xmath342 . consequently , @xmath343 where @xmath344 is the step function . for @xmath86 , the imaginary part of @xmath345 diverges . however , @xmath335 ( in fact @xmath346 and @xmath347 ) . hence , @xmath348 remains finite at @xmath349 and can be ignored in an integration over real @xmath62 as the poles of @xmath350 are isolated . consider a function @xmath351 which is analytical except for isolated first - order poles on the real axis and which is real for real @xmath62 . @xmath352 is constant almost everywhere , and thus the derivative @xmath353 vanishes almost everywhere . a non - zero derivative may either occur if @xmath354 , i. e. at the zeros @xmath78 . this gives a contribution : @xmath355 where @xmath356 . a second contribution arises from the first - order poles of @xmath350 at @xmath93 . the poles are the zeros of the function @xmath357 . note that @xmath358 since the sign of the argument is unchanged . thus the contribution due to the poles is : @xmath359 we have : @xmath360 with @xmath361 and @xmath362 depending on the sign of the slope of @xmath363 at the zeros @xmath78 and the sign of the residue of @xmath363 at the poles @xmath93 , respectively . consider the function @xmath364 . the zeros of @xmath363 are the poles of the diagonalized green s function @xmath365 which has positive residues . this implies a positive @xmath366 at @xmath78 and @xmath367 . the poles of @xmath363 are the poles of the self - energy which has positive residues . thus the residues of @xmath363 at the poles are negative and @xmath368 is negative at @xmath93 and @xmath369 . hence : @xmath370 the reference system @xmath11 is a set of decoupled single - impurity anderson models with one - particle energy of the impurity site @xmath157 , conduction - band energies @xmath158 with @xmath160 and hybridization strengths @xmath159 . we first calculate the eigenvalues of the hopping matrix @xmath12 . the matrix is block diagonal with respect to the site index @xmath371 . each block is labeled by the `` orbital '' index @xmath372 . there are non - zero elements of the matrix for @xmath373 , @xmath374 , and @xmath375 only : @xmath376 furthermore , @xmath377 . using @xmath378 where @xmath379 and the general relation @xmath380 , we find : @xmath381 with @xmath382 and @xmath383 for @xmath160 as defined by eqs . ( [ eq : g1def ] ) and ( [ eq : grdef ] ) . the factor @xmath384 accounts for the two spin directions .

we have used the xafs ( x - ray - absorption fine structure ) technique to investigate the local structure about the cu , ba , and tl atoms in orthorhombic with a superconducting transition temperature = 60 k. our results clearly show that the o(1 ) , o(2 ) , cu , and ba atoms are at their ideal sites as given by the diffraction measurements , while the tl and o(3 ) atoms are more disordered than suggested by the average crystal structure . the tl - tl distance at 3.5 between the tlo layers does not change , but the tl - tl distance at 3.9 within the tlo layer is not observed and the tl - ba and ba - tl peaks are very broad . the shorter tl - o(3 ) distance in the tlo layer is about 2.33 , significantly shorter than the distance calculated with both the tl and o(3 ) atoms at their ideal sites ( or ) . a model based on these results shows that the tl atom is displaced along the directions from its ideal site by about 0.11 ; the displacements of neighboring tl atoms are correlated . the o(3 ) atom is shifted from the site by about 0.53 roughly along the directions . a comparison of the tl l-edge xafs spectra from three samples , with = 60 k , 76 k , and 89 k , shows that the o environment around the tl atom is sensitive to while the tl local displacement is insensitive to and the structural symmetry . these conclusions are compared with other experimental results and the implications for charge transfer and superconductivity are discussed .

cond-mat9404061

the tl@xmath6ba@xmath7ca@xmath8cu@xmath9o@xmath10 system is a model family of high - temperature superconductors . the structures consist of copper perovskite - like blocks containing @xmath11 planes ( @xmath11 up to 6 has been observed ) separated by one or two tlo layers . each cao layer is sandwiched between two layers while the bao layer is sandwiched between the and tlo layers . for tl - monolayer ( @xmath121 ) phases , the superconducting transition temperature ( ) increases to 122 k with the number of planes from @xmath13 to @xmath14 and then decreases with @xmath155 to 6 ; for tl - bilayer phases , increases up to 125 k with the number of planes @xmath11 from 1 to 3 and then decreases with @xmath11=4 , 5 . the fully stoichiometric tl - bilayer tl@xmath7ba@xmath7ca@xmath8cu@xmath9o@xmath16 compounds would contain no carriers if the formal ionic valence states , tl@xmath17ba@xmath18ca@xmath19cu@xmath20o@xmath21 , are assigned , and thus would not be superconductors . however , there is a hybridization of tl @xmath22 and o @xmath23 orbitals which either touch the fermi surface or cross it@xcite , so the tl does not act as a completely @xmath24 ion and charge transfer might occur between the and tlo layers . other effects can also change the hole - doping in the system including : ( i ) excess or missing o ; ( ii ) ca@xmath25 and/or cu@xmath25 substitution for tl@xmath26 ; ( iii ) tl deficiency ; ( iv ) ordering of the o and/or tl atoms in the tlo layers . which doping mechanism primarily controls is still an unsolved problem . structural studies on these compounds will be helpful in understanding the mechanism of superconductivity in this system . ( tl-2201)@xcite is particularly suitable for this study because it has a simple structure ( calcium - free and only 1 layer per formula ) ; the highest is about 90 k. the structure of tl-2201 , in fact , has been extensively studied since its first discovery in 1988@xcite . reasonable agreement for the general structure has been achieved from diffraction studies@xcite . it contains tl@xmath27o@xmath28 double layers and a plane separated by a bao layer , as shown in fig . 1 . for convenience of description , we define the @xmath0 site ( @xmath1 0 or @xmath2 ) as the ideal site for tl and o(3 ) in tetragonal and treat orthorhombic to be a pseudo - tetragonal structure . the atomic positions of tl and o in the tl@xmath27o@xmath28 double layers , however , are not well defined , due to a large anisotropic thermal factor and/or various static displacements from their ideal sites . most of the refined structures suggest a distorted site for the o(3 ) with a displacement of @xmath290.31 - 0.42 along the @xmath4 direction@xcite or @xmath290.22 - 0.28 along the @xmath3 direction@xcite , while some of them show that the tl atom is also displaced from its ideal site by @xmath290.11 - 0.14 along the @xmath3 direction@xcite or @xmath290.16 along the @xmath4 direction@xcite . the refined thermal factors in these studies are usually quite large in the ab - plane for the tl and o(3 ) atoms . recently , it has been made clear that the thallium content has a strong influence on the lattice symmetry and thus on the average atomic position in the tlo layer@xcite furthermore , significant distortions have recently been observed in both and tl@xmath7ba@xmath7cacu@xmath7o@xmath30 ( tl-2212 ) by means of x - ray - absorption fine structure ( xafs)@xcite . similar distortions are found in tl-2212 by the pair - distribution - function ( pdf ) analysis of neutron scattering data@xcite . it appears that these distortions not only have short - range correlations , but may in some cases also have long - range correlations , as evidenced by the presence of the modulational structure for selected sections , observed in high - resolution electron microscopy and electron diffraction@xcite . in addition , the temperature dependence of the local structure in the plane has been studied by xafs in tl-2201@xcite and by pdf in tl-2212@xcite , respectively . the splitting of the cu - o(axial ) in tlba@xmath7ca@xmath31cu@xmath32o@xmath33 has also been investigated by xafs in the range 10@xmath34t@xmath34156 k@xcite . in this study , we have focused on the static distortion of the , especially in the tlo layer . we have used the xafs technique@xcite to probe the local structure around _ all _ the metal atoms , cu , ba , and tl in , to revise the earlier results based on the tl l@xmath5 edge studies@xcite . both the sample preparation and the xafs data analysis are slightly different from the previous study@xcite ; they are described in sec . ii . the xafs data analysis ( presented in sec . iii ) uses constrained fits@xcite to reduce the number of independent parameters and to check the previous model . the xafs data and the data analysis for the cu and ba k - edges , and the tl l@xmath5-edge are discussed in secs . iv and v , respectively . a discussion and a comparison of our results with other experimental observations are given in sec . vi , and the results summarized in sec . vii . samples of about 10 grams weight were prepared from a mixture of tl@xmath7o@xmath31 , ba(oh)@xmath35h@xmath7o and cu(no@xmath31)@xmath35nh@xmath7o , always in the ratio of the cations desired in the final material with no excess of any element . the powders , milled in ethanol and thereafter dried in an inert atmosphere ( n@xmath7 ) , were heat treated by a two - step sintering process . the heating rate in both steps was 180@xmath36 c / hour and the temperature was regulated by a controller , eurotherm 818p , with a temperature stability better than @xmath370.5@xmath36 c. in sintering step i the mixtures were heated in al@xmath7o@xmath31 crucibles with al@xmath7o@xmath31 covers at 760@xmath36 c for 3 hours with only a limited loss , if any at all , of thallium . in sintering step ii pellets were placed in al@xmath7o@xmath31 crucibles sealed with gold gaskets and cover . they were subsequently centered in small pieces of ceramic tube and fixed with steel wedges ( a procedure that prevents evaporation from the crucible ) and heated at 860@xmath36 c for 6 hours . the loss of weight in sintering step ii , if attributed solely to thallium evaporation , is below 0.3% of the total amount of thallium . there is no sign of substitution of gold into the material or of incorporation of thallium into the gold gasket . from edx ( energy dispersive x - ray spectroscopy ) analysis , no traces of al or au can be detected in the sample . for estimated errors in barium and copper content see ref . sample purity and lattice parameters were determined from guinier film data using silicon ( nbs 640b ) as an internal standard . a computerized photoscanning system in combination with programs for indexing and unit cell refinements were used for the extraction of lattice parameters ( a=5.4881(2 ) , b=5.4532 , c=23.216(1 ) ) . the impurity content was below 2% . neutron powder diffraction data were collected at the pulsed neutron source isis , ral , uk using the high flux , medium resolution powder diffractometer polaris . a least - squares profile refinement program , tf12ls , based on the rietveld technique was used to analyze the tof ( time - of - flight ) neutron powder diffraction data . the refinement was carried out in space group abma and selected bond distances are found in tables i - iii . the superconducting properties were studied by ac susceptibility measurements using a quantum design squid magnetometer at temperatures down to 4 k in a field of 50 gauss . was determined to be 60 k. for further details concerning sample preparation and characterization consult ref . 12 . the samples were prepared by brushing a fine powder ( @xmath38 30 @xmath39 ) onto scotch tape . each layer was selected to be free of pinholes to minimize the xafs glitches@xcite and several layers were stacked to obtain a sample with a thickness of approximately one absorption length . samples were measured at 80 k using a liquid - nitrogen dewar . several data sets were collected for each sample to check for consistency and averaged to improve the signal - to - noise ratio . the x - ray absorption data were collected on beamline 10 - 2 at the stanford synchrotron radiation laboratory using si ( 400 ) monochromator crystals . the double monochromator was detuned on its rocking curve to approximately one half of maximum transmitted x - ray intensity to reduce the harmonic content of the beam . the standard procedures for the reduction and analysis of the data are described elsewhere @xcite . the extracted xafs function , @xmath40 , is given by @xmath41exp(-2k^{2 } \sigma^{2}_{j}-2r_{j}/\lambda)\ ] ] where the sum is taken over shells with @xmath42 atoms at a distance @xmath43 from the absorbing atom , @xmath44 is the momentum of the photoelectron given by @xmath45^{1/2}/\hbar$ ] where @xmath46 is the energy at one - half the absorption edge height , @xmath47(k ) is the back - scattering amplitude , @xmath48 is a phase shift of the photoelectron due to its interaction with the back scattering and absorbing atoms , @xmath49 is the effective electron mean free path , and @xmath50 ( debye - waller factor ) is the average variation ( both static and dynamic ) of @xmath43 . examples of the k - space data are given in fig . 2 for the cu and ba k - edges and tl l@xmath5-edge . a fourier transform of these k - space data exhibits peaks in r - space which correspond to neighbors at different radial distances from the absorbing atom . to obtain numerical values for @xmath43 , @xmath42 and @xmath50 , we perform iterated least - square fits of the real and imaginary parts of the fourier transform of @xmath51 to a sum of standards over a particular range in real ( r ) space . identical k - space transforms are used for the sample and the standards . in these fits , the overlaps of the radial distributions of other neighbors in r - space are included . it is difficult to extract all the isolated , single pair standards needed for this analysis from other experimental standards . in this study all the standards are generated by the feff5 code@xcite , using a cluster as close to the real lattice as possible with a radius of about 7.5 from the central absorbing atom to the boundary scattering atoms . the multiple scattering paths are included in the fit when it is necessary . in earlier studies we have used experimental atom - pair standards extracted from data collected for standard compounds . we have compared our experimental standards with feff5 and have obtained very good agreement for about twenty pairs@xcite . since a complete set of experimental standards was not available , we decided , based on our recent study , to use feff5 standards . one of the problems in using the standards from feff5 is the uncertainty of the amplitude - reduction - factor , @xmath52 . we use two different values of @xmath52 which we have obtained from our detailed comparisons@xcite ; 0.8 when the back - scattering atom is o , and 0.9 when the back - scattering atom is a metal atom . the standard background removal procedure which uses a simple spline or polynomial for @xmath53 , does not work well when the background consists of some features due to multielectron excitations@xcite and ramsauer - townsend - like effects@xcite . these features have been found at the br ( in rbbr ) , kr , rb ( in rbbr ) , ba ( in bao ) , ce ( in ceo@xmath7 ) , and pr ( in prba@xmath7cu@xmath31o@xmath54 ) k - edges@xcite and the au , tl ( in tl@xmath7o@xmath31 ) , pb ( in @xmath55pbo@xmath7 ) , and bi ( babio@xmath31 ) l@xmath5-edges@xcite in our studies . similar features have also been observed and confirmed by other groups@xcite . an iterative background extraction and removal procedure has been proposed to obtain and remove these features in the previous work@xcite . when the simple standard background removal procedure was applied to the ba k - edge and tl l@xmath5-edge xafs data of , spurious humps appear in the low - r region of the fourier transformed data as shown in fig . 3 . instead of performing the iterative background removal procedure to these complicated data directly , we utilized the extracted background from the bao@xcite and tl@xmath7o@xmath31@xcite data ; the new background was obtained by fitting the corresponding extracted background plus simple spline functions to the data . these new backgrounds are plotted in fig . they contain more wiggles than the background obtain from the standard procedure but result in smoother r - space data in the low - r region than those from the standard procedure . this simple method works very well . we note that using more spline sections in the standard procedure may also minimize the humps in the low - r region but that usually reduces the amplitude of the first r - space peak , simultaneously . one of the common problems in the xafs data analysis is that there are too many fitting parameters if the fit includes further shells in a complicated system . for example , if the fitting range is 3.0 @xmath29 13.0 in k - space and 1.0 @xmath29 4.0 in r - space , the _ estimated _ allowed number of fitting parameters is @xmath2921 , according to the formula @xmath56@xcite . in a complicated system , such as the cu edge data in @xcite , the number of atomic pairs could exceed 8 , which needs 24 or more fitting parameters , even if all the e@xmath57 are fixed and not considered as fitting parameters . therefore , a constrained fit is required in the xafs data analysis for a complicated system . these constraints include using the lattice constants and the known average locations of the atoms in the unit cell from diffraction measurements , combining the xafs results viewed by different atoms , and using equations which relate a parameter to one or more other parameters ( for examples , keeping the ratio of some parameters constant ; setting a parameter to the weighted sum of other parameters ) to check possible models . the constraints force the fit towards a particular minimum in parameter space , and the fit is not easily trapped in some other local minima . in this way , the constrained fits can help reject results which are physically inconsistent and to check particular models for the local structure . this method has been used to a limited extent in our earlier work@xcite and more extensively in our recent studies@xcite . from diffraction studies@xcite , the average local environment about a cu atom is as follows : 4 o(1 ) neighbors at r=1.93 in the plane , 2 o(2 ) neighbors at 2.71 along the c - axis , 8 ba neighbors at 3.35 above and below the plane , and 4 cu neighbors at 3.87 in the plane . in xafs data , the fourier transform of the xafs function @xmath40 to real ( r ) space is a sum of peaks corresponding to each of these different neighbors ; however , each distance is the average local value , not necessarily the value calculated from the average unit cell . the question considered in this section is how well does the cu local environment agree with diffraction results . the cu k - edge xafs data in r - space are shown in fig . 4 . the transform range is a square window from 3.1 to 16.5 , gaussian broadened by 0.3 . both the real part and the magnitude of the complex transform are plotted . the distances between the absorbing atom and the back - scattering atoms are determined by the phase of the xafs function , which is well represented by the positions of the zero - crossing points of the real and imaginary parts of the fourier transform . however , there is a shift , @xmath58 , of the xafs peaks from the real positions due the photoelectron phase - shifts . to compare the local structure which determines the cu k - edge xafs with the average structure obtained from diffraction , we generated the theoretical xafs function using feff5@xcite and the refined crystal structure obtained from diffraction measurements@xcite . in the calculation , all the atoms within a distance of 7.5 from the central absorbing atom , and all the scattering paths with a contribution of 4% or more , relative to the maximum contribution , are included . to roughly simulate both the static and thermal disorder in the calculation , we have set the sample temperature to 100 k and the debye temperature to 700(k )/(0.5r@xmath59 ) which increases @xmath60 with distance r@xmath59 for more distant shells . ( the same thermal parameters are used for the other edges in this paper unless noted ) . the fourier transform of the theoretical xafs ( without any fitting ) is shown in fig . 4 by dotted lines , with the same k - space window as for the experimental data , an overall amplitude - reduction - factor of 0.9 , and an overall e@xmath57 shift of @xmath61 ev . 4 shows that the phase of the theoretical xafs agrees very well with that of the experimental data up to 6.5 . this means that the o(1 ) , o(2 ) , ba , and cu atoms are well ordered about a given cu atom , in excellent agreement with the diffraction results . we point out that the cu edge xafs data is not sensitive to the distortion of the distant tlo layers if there is any , especially when the distortion is in the ab - plane , because the tlo and layers are separated by 4.7 and a small splitting of the cu - tl and/or o(3 ) distances can be absorbed by a large debye - waller factor . we note that the subtle differences in zero - crossing points are due mainly to the errors in e@xmath57 , especially in the region below 2.3 where the o atoms make a major contribution . the nearest o neighbors directly touch and partially screen the excited absorbing atom which has a hole in the core - shell level , thus the e@xmath57 shift for the nearest o neighbors should be different from the distant neighbors . since the same e@xmath57 shift is used for all neighbors in the simulation , an error in the phase - shift can be introduced in the first o peak . the discrepancy in the magnitude is mainly from the rough estimation of the debye - waller factors and the amplitude - reduction - factors of the xafs function in the theoretical calculation and is not strongly related to the separations of the atomic pairs . we emphasize that in the theoretical calculation we did not adjust the xafs parameters , n@xmath59 , r@xmath59 , @xmath62 , and e@xmath63 , individually ; instead , we only adjusted the overall amplitude and e@xmath57 and simulated the debye temperature of the different pairs by a simple formula . this method makes it possible and easy to compare the experimental xafs data with a theoretical function calculated for a particular model up to 7 ( for example , from the refined average unit cell obtained from diffraction ) . however , it is still difficult to make a detailed fit to the experimental xafs data much beyond 4 . the ba atom is in a plane between the and tlo layers . from diffraction measurements@xcite , it has 4 o(1 ) neighbors at 2.73 , 4 o(2 ) neighbors at 2.84 , 1 o(3 ) neighbor at 3.00 , 4 cu neighbors at 3.35 , 5 ba neighbors ( 4 in the ab - plane and 1 along the c - axis ) near 3.86 , and 4 tl around 3.9 @xcite . again we ask the question if the local structure about the ba atoms is the same as that given by the diffraction results assuming no distortions of the tl atoms . the fourier transform of @xmath51 for the ba k - edge data is shown in fig . 5 ; window is from 3.5 to 15.5 , gaussian broadened by 0.3 . we have generated the theoretical ba xafs function following the procedure described earlier for the cu edge ; it is plotted as dotted lines in fig . 5a . the phase of the theoretical and experimental xafs function agreed very well up to 3.7 ( a small discrepancy in the phase of the first peak can be easily minimized by using a different e@xmath57 shift ) . this again indicates that the o(1 ) , o(2 ) , cu and ba atoms are well ordered . similarly , there is excellent agreement in the phase for a large peak around 5.2 between the simulation and experimental xafs data . the dominant contributions to this peak arise from the collinear , forward , multiple scattering of the ba - o(1)-ba and the multiple scattering of intraplanar ba - o(2)-ba pairs . the phase of the real part of the transform within the 5.2 peak is shifted by nearly @xmath64 relative to the phase that would be observed for a single scattering ba - ba peak ( _ e.g. _ as occurs near 3.5 in fig . 5 ) due to the forward multiple scattering contributions . the good agreement between the calculations and the experimental data for this peak strongly suggests that the ba atom is well ordered . more importantly , however , there is a significant difference between the calculations and the experimental data on the high - r side of the peak near 3.6 (see fig . 5c in which a small fraction of the data is shown with an enlarged @xmath65axis ) , where the ba - tl pair ( r= 3.90 ) should make a major contribution . the experimental data in this region of r - space can be simulated much better by feff5 when the ba - tl pairs are excluded ; _ i.e. , _ the net ba - tl peak is small . this means that the ba - tl peak is either very broad due to disorder of the tl atoms or is split into multi - peaks due to a displacement of the tl atom from its ideal site . there are two adjacent tlo layers , which are surrounded by two bao layers , one above and one below . the average local structure around the tl atoms obtained from diffraction@xcite consists of 1 o(2 ) and 1 o(3 ) atoms near 2.0 , 4 o(3 ) within 2.3 - 3.2 ; 4 interplanar tl atoms near 3.5 and 4 intraplanar tl atoms near 3.9 ; and 4 ba atoms near 3.9 . however as outlined in the introduction and the above discussion at the ba edge , there is considerable evidence that the tlo layers are distorted . here we investigate this disorder using tl edge xafs data . first , we performed a multiple - scattering - cluster - simulation for the tl l@xmath5-edge xafs ( as described above for the cu k - edge data ) assuming the structure of ref . 4 in which the tl is at the ideal @xmath0 site while the o(3 ) is displaced along the @xmath4 direction by 0.37 from its ideal @xmath0 site . both the simulated and experimental xafs , @xmath51 , were fourier transformed to r - space using a k - space range of 3.2 - 15.5 , gaussian broadened by 0.3 , as shown in fig . 6a . in the simulation , we used two different sets of debye - waller temperatures , 300 ( k )/(0.5@xmath66 ) for tl - tl pairs and 700 ( k )/(0.5@xmath66 ) for other pairs , to better simulate both the static and thermal disorder . this gives a good simulation above 4.5 in r - space . significant differences in the phase between the simulated and experimental data are observed in fig . thus at least some of the distances derived from the average , crystallographic structure are not consistent with these xafs data . for the first o neighbor peak ( 1.32.3 in fig . 6 ) of the simulated data , the phase of the low - r side ( 1.31.8 ) , where the tl - o(2 ) and tl - o(3 ) pairs at r=2.0 make their major contributions , agrees reasonably well with the experimental data ; however , at the high - r side of this peak ( 1.82.3 ) , where the tl - o(3 ) pair at r=2.5 makes its major contribution , the phase is not consistent with the experimental data and a contribution in the simulated data is missing in the experimental data . a comparable inconsistency is observed in the peak centered near 3.6 . the phase and the shape of the simulation for this peak agree well with the actual data in the low - r side , where the interplanar tl - tl pair makes its major contributions ; however , it does not agree in the high - r side ( 3.74.0 ) , where the intraplanar tl - tl and also the tl - ba pairs make their major contributions . furthermore , the amplitude of the simulated peak in r - space near 3.6 is almost twice as big as that in the data . this discrepancy is too large to be explained just by the errors in the amplitude - reduction - factor and the debye - waller factors . this implies a significant distortion and/or disorder in the tl - tl and tl - ba distances from the average distances obtained from the refined structure in diffraction . a simulation without the intraplanar tl - tl pairs bears a much closer resemblance to the experimental spectra near 3.5 . these comparisons again suggest that the tl atoms , and probably also the o(3 ) atoms , occupy more distorted sites in the ab - plane than those obtained from the average diffraction experiments . from the qualitative analyses of the cu , ba , and tl edge xafs data shown above , we conclude that the positions of the cu , o(1 ) , o(2 ) , and ba atoms in the unit cell are consistent with the average crystallographic structure obtained from diffraction studies . however , the tl and o(3 ) atoms are significantly more distorted than the average structure obtained in many diffraction studies . the xafs data indicate that these distortions are located in the ab - plane of the tlo layer as suggested in some diffraction investigations . these qualitative features are model independent and give a sound guidance for further structural analysis . in this section , we will perform nonlinear - least - square fits to the xafs data to further check the qualitative features shown above and gain quantitative information on the local structure about the cu , ba , and tl atoms . the r - space cu k - edge xafs data have been fit up to 3.9 using individual standard atomic pairs generated by feff5 . five pairs are included in the fit : cu - o(1 ) near 1.9 , cu - o(2 ) near 2.7 , cu - o(1 ) near 4.3 , cu - ba near 3.4 , and cu - cu near 3.9 in which the multiple - scattering - paths for cu - o(1)-cu are included . the experimental data and the fit are shown in fig . 7 , where the same fourier transform range is used as in fig . 4 and the fitting range is between 1.0 - 3.9 . overall , a high quality fit has been achieved . a double - well potential was used in ref . 23 to improve the fit for the first cu - o(1 ) peak ; however , in this study , the refinement of the first peak is not important and we use only a single cu - o(1 ) peak . the structural parameters extracted from the xafs curve - fitting are tabulated in table i , together with the results from diffraction refinement . our xafs results are in excellent agreement with the diffraction results , as expected from our earlier comparison of the simulated and experimental xafs r - space spectra , in which there was an excellent phase agreement up to 6.7 . these xafs results are also very close to the previous results@xcite . in sec . iv , we showed that the tl is disordered or distorted . so in the fits to the ba k - edge data , we considered three situations for comparison : ( 1 ) no ba - tl peak , ( 2 ) a single ba - tl peak , and ( 3 ) a split ba - tl peak . for the first case , we fit the ba k - edge xafs data with 3 o peaks ( at r=2.7 , 2.8 and 3.0 , respectively ) , 1 cu and 1 ba peak . the tl peak near @xmath67=3.9 was not included . the debye - waller factor of the ba - cu peak was fixed to have the same value as that of the cu - ba peak from the cu k - edge fit . a reasonably good fit is achieved and all the extracted parameters are reasonable , which means that the ba - tl pairs do not make a significant contribution to the ba edge xafs , in contradiction to what one would expect for the undistorted tl layers . secondly , we included the ba - tl peak in the fit and fixed the amplitude of that peak to 4 , the number predicted by diffraction . the quality of the fit was improved by about 25% and the extracted distances and the amplitudes for the ba - o , ba - cu and ba - ba peaks remained essentially unchanged . the ba - tl distance is about 3.90 , very much the same as the average distance given by diffraction . the debye - waller factor of the ba - tl peak is about 0.094 , much bigger than that of the ba - ba peak ( 0.066 ) . finally , we tested the model proposed earlier@xcite . here we require consistency between the xafs spectra collected from three different absorbing atoms , achieved by using constraints in the fitting procedure . following our previous discussion , we assumed that the tl atoms are shifted along @xmath3 so that the 4 ba - tl pairs split into three distances , with roughly 1 pair at @xmath68 , 2 pairs at @xmath67 and 1 pair at @xmath69 . therefore , we included 3 o , 1 cu , 1 ba and three ba - tl peaks in the fit . to minimize the number of fitting parameters and to check the model , we fixed the average distance @xmath67 and the amplitudes of these three peaks to the values predicted by the model and only allow @xmath70 to vary for these ba - tl peaks . the debye - waller factors of the three ba - tl peaks and the e@xmath57 shifts for the same type of back - scattering atoms were also kept equal . in this way , the number of the variables was reduced to 18 , well below 23 , the estimated number of allowed variables obtained from ( @xmath71)@xcite . the quality of the fit and the extracted parameters for o , cu and ba peaks are essentially the same as the fit with a broader ba - tl peak but here @xmath72@xmath67=0.10 . as a final step , we also allowed the average ba - tl distance and the amplitudes of the ba - ba and ba - tl peaks to vary . the final fitting parameters remained essentially the same as the ones with the strong constraints . the experimental spectra and the final fitting curves are plotted in fig . 8a ; the contributions from o , cu and ba atoms are shown in fig . 8b , while the contributions from the split ba - tl pairs are shown in fig . these results clearly show that the contributions of the ba - tl pairs are quite small compared to the ba - cu and ba - ba pairs . the detailed results are listed in table ii , together with the results from diffraction and the model for comparison . we would like to point out that the ba - o(3 ) distance is not sensitive to the displacement of the o(3 ) atom in the ab - plane . for ba - tl peaks , @xmath70 is slightly coupled with the debye - waller factor @xmath60 and the coordination number @xmath73 . it varies from 0.08 to 0.11 depending on @xmath60 and @xmath73 ( or @xmath52 ) . we also used different k weighting ( _ i.e. _ , @xmath74 with n=1 , 2 , 3 ) and different k ranges for the fourier transform in the fit . the derived parameters are essentially the same within the estimated uncertainty . in all cases , the splitting distance , @xmath70 , for the ba - tl pair is significantly shorter than 0.2 , the distance predicted by the previous model . this means the displacement of the tl atom in the previous model ( 0.28 ) is too large and has to be modified to fit the ba edge xafs data . we also simulated the ba k - edge xafs data with a distorted model for the tlo layer . here we modified the previous model by assuming that the tl atom is displaced 0.11 from its ideal site along the @xmath3 direction . the simulated data are plotted in fig . the agreement with the data is much improved for the r - space range 3.54.5 compared to the earlier simulation ( fig . 5a ) , as illustrated in fig . there is also improved agreement in phase near 6 . from the analysis of the cu and ba k - edge xafs spectra , we conclude that the o(1 ) , o(2 ) , cu , and ba atoms are at their ideal sites in the unit cell just as diffraction shows , but the tl atom is shifted from its ideal site . however , it is difficult , from the cu and ba edge xafs spectra only , to distinguish between a randomly disordered tl site ( which gives a broad peak ) and a distorted tl site ( which gives multiple peaks ) and hence to derive the actual displacement of the tl atom if it is at a distorted site . further information from the tl edge is desirable . the tl l@xmath5-edge xafs data are more sensitive to the tl distortion , but the data analysis is more complicated by the fact that there are two adjacent tlo layers . in the fits to the tl edge data , we considered the tl displacements along the @xmath3 direction with two different magnitudes : ( 1 ) @xmath290.3 , similar to the previous model@xcite , and ( 2 ) @xmath290.1 , as suggested by the ba edge xafs data . in comparisons with diffraction studies , we also tried other models , see sec . vi.a . in order to enable a meaningful analysis of the tl l@xmath5-edge data we treat the tl atoms as if they all see the same local distortions however , we know that in the related bi - based cuprate superconductors , there exist strong periodic modulations@xcite which result in superstructure with long range order , a buckling in the ab - plane , and periodically extra o atoms . because diffraction results show little disorder for tl along the c - axis , we are motivated to use a simple model for the local environment . as discussed later , it is still compatible with a superstructure unit cell . we first performed a rough fit to the data in the region of [email protected] , with 2 o , 1 tl and 1 ba peaks , similar to the model used in the previous paper@xcite . the extracted structural parameters are the following : 1.8 o atoms at 2.04 , 2.9 o atoms at 2.35 , 2.8 ( which is too small compared to 6)@xcite tl atoms at 3.51 , and 7.6 ba atoms ( but very broad , with @xmath75 ) at 3.89 . this shows that the interplanar tl - tl pair ( at r=3.5 ) remains but the intraplanar tl - tl pairs ( with an average distance of 3.9 ) do not make a large contribution to the tl edge xafs spectra as would occur for an undistorted site . this result suggests that the tl atom is distorted ( not randomly disordered ) in a particular way so that the interplanar tl - tl peak remains but the intraplanar tl - tl peak seems to disappear . the fitting result we obtain here is quite close to the earlier result@xcite . however , the model based on that result requires a large shift for the tl atom which does not fit the ba edge data as discussed above , and the tl - tl amplitude is too low . therefore , the previous model must be modified to fit both the ba and tl edge xafs data . secondly , we followed the previous model but with a smaller shift for the tl atom . assuming that the tl atoms are shifted along the @xmath3 direction as shown in fig . 9 , the interplanar tl - tl distance remains approximately unchanged while the intraplanar tl - tl distance splits into two distances . in addition , the tl - ba distance splits into three distances , with the average distance unchanged to first order . in order to check this model more carefully , we explicitly included all the split peaks and the long tl - o peak at 4.3 in the fit using the constrained - parameter technique . the numbers of neighbors for all the tl and ba peaks are fixed to the values predicted by the model . the average tl - ba and intraplanar tl - tl distances are also fixed to the average distances given by the diffraction measurement , and only the split distances are allowed to vary . the debye - waller factors of the three tl - ba peaks are set to be equal to those of the ba - tl peaks at the ba edge , to further reduce the number of independent variables and to check the consistency of the fitting results . the e@xmath57 shifts for the same type of back - scattering atoms are kept equal except for the nearest o atoms . thus the total number of variables is 16 , far below the estimated number of allowed variables , 27@xcite . the fit was carried out in the region of 1.2 @xmath764.2 , with the fourier transform range of 3.3 @xmath29 15.5 , gaussian broadened by 0.3 . a high quality fit was achieved . finally , we allowed the average tl - ba and intraplanar tl - tl distances , the amplitudes , and the debye - waller for the three tl - ba pairs to vary . the fitting results remained essentially the same as those with the constraints . we also fit the data with different k weighting ( @xmath77 , n=1 , 2 , 3 ) and k range in the fourier transform . consistent results were achieved within the estimated uncertainty . the final fit is shown in fig . the contributions from o , tl and ba atoms are plotted in fig . 10 , b - d , respectively . the contributions from the three individual tl - tl peaks are plotted below the sum of their contributions . the contribution of the tl - tl at r=4.02 is small , and the tl - tl peak at r=3.72 is out of phase with the tl - tl peak at r=3.50 , resulting in partial destructive interference . thus the sum of the tl - tl peaks has almost the same phase as that of the tl - tl peak at r=3.50 but a smaller amplitude , which explains why the fit with only one tl - tl peak gives quite good agreement , but with a small amplitude . the three tl - ba peaks are also plotted below the sum of their peaks . the sum of the three split tl - ba peaks is a broad peak , with a similar phase to that of the tl - ba peak at the average diffraction distance , 3.90 , which again is consistent with one broad tl - ba peak in the previous fit . the extracted structural parameters are listed in table iii , together with the corresponding parameters calculated from the model . we emphasize that the tl - ba distances obtained in this new fit of the tl l@xmath5-edge data are in good agreement with the ba - tl distances extracted from the ba edge xafs data within the estimated uncertainty . in the model , we only shifted the tl atom along the @xmath3 direction to match the tl - tl , tl - ba and ba - tl distances extracted from our xafs data . the best overall agreement is achieved when the tl atom is displaced by about 0.11 . we have also shifted the o(3 ) atom along the @xmath4 direction from the @xmath0 site by 0.53 in the model to match the observed average interlayer tl - o distance at 2.33 . the splitting of the two shorter interlayer tl - o distances ( @xmath67=2.29 and 2.38 ) predicted by the model is consistent with the broad tl - o peak ( @xmath60=0.131 ) at @xmath67=2.33 in the data . we point out that absolute position for the o(3 ) is not well defined in the data because the tl - o peak at 2.33 is quite broad and the tl atom itself is displaced . the long interlayer tl - o(3 ) peak , predicted to be near @xmath783.123.24 by the model , is not clearly observed in the xafs data , probably because that peak is considerably broader than the shorter one , and/or smeared out due to multiple tl - o peaks in that range . instead of the well defined pair distances predicted by the model , the actual pdf of the tl - o(3 ) pairs within the tlo layer may be spread over the entire range 2.33.2 ( refer to table iii ) due to the variations of the tl and o(3 ) displacements in both magnitudes and directions . we again simulated the tl edge xafs , using the feff5 code and the distorted local structural model ( see fig . 9 ) . the simulation is shown in fig . the sample temperature and debye - waller temperature used here are the same as those used in the simulation shown in fig . 6a , except the parameters for the first two tl - o pairs ( @xmath67=2.00 and 2.09 ) in which the debye - waller factors were set to 0.03 because the static disorder had already partially been taken into account by the two separated peaks . the new simulation in fig . 6b shows a great improvement over the old one shown in fig . 6a , and the phase of the new simulation agrees very well with the actual data over the whole region . the extracted distances from our xafs data analyses for the undistorted atomic pairs , such as cu - o , cu - ba ( ba - cu ) , cu - cu , ba - o(1,2 ) , ba - ba , are in excellent agreement with those obtained from diffraction studies , with a typical error of 0.01 ( see table i and ii ) . this shows the high quality of the xafs data and the data analyses . the extracted distances for the distorted atomic pairs , such as ba - o(3 ) , ba - tl ( tl - ba ) , tl - o , are within the range of the distances given by diffraction investigations ( see table ii and iii ) . diffraction studies@xcite show that the refined thermal factors for the tl and o(3 ) atoms along the c - axis are only one third or less of those in the ab - plane . therefore , we did not attempt to vary the c - components of the tl and o(3 ) positions in our model ; they are fixed at the average positions given by diffraction studies . furthermore , the mean - square - root of the tl displacement is about 0.15 in the ab - plane ( @xmath79 or @xmath80 ) and 0.09 along the c - axis ( @xmath81 ) in the refinement of diffraction studies@xcite , which are quantitatively consistent with our xafs results . given the tl displacement of 0.11 in the ab - plane and the tl debye - waller factor of 0.09 ( see table iii ) , the approximately estimated mean - square - root of the tl displacement in the ab - plane is 0.14 ( @xmath82 ) , very colse to that of the diffraction results . the model ( fig . 9 ) based on our xafs data analyses suggests that the tl displacements are correlated over a short range . the partial - pair - distribution - function ( ppdf ) for the tl - tl pairs extracted from our data shows a dip near the average distance as shown in fig . 11a , which is significantly different from the tl - tl pdf for the tl in the ideal sites as shown as the dotted curve in fig . 11b ( with a debye - waller factor 0.07 ) . the second peak in the ideal structure is split into two peaks in the extracted pdf , one near 3.7 and another one near 4.0 , thus a dip is introduced near the average distance 3.87 . when we fit the data using the ideal structural model ( 4 tl - tl at 3.51 , 4 tl - tl at 3.87 and 4 tl - ba at 3.89 ) , the longer tl - tl pair is moved to 3.75 ( solid curve in fig . 11b ) which physically does not make sense because the average tl - tl distance in the ab - plane should equal to the average lattice , 3.87 . our model is similar to that suggested by the pdf analysis of the pulsed - neutron scattering spectrum in tl-2212@xcite , but more specific about the relative shift for the tl atoms between the two consecutive tlo layers . since the interlayer tl - tl distance remains unchanged , the displacement of the nearest tl - tl pairs between the tlo layers should be either in the same direction or perpendicular to each other , but not in opposite directions . the shifts of the tl and o(3 ) atoms and the pair distances obtained from xafs , however , are different from those obtained by pdf even if the small changes in the lattice constants between tl-2201 and tl-2212 are taken into account . the pdf analysis for tl-2212 suggests that the tl atom shifts by 0.32 along the @xmath3 direction , which results in three split tl - ba distances with @xmath830.23 . this is not consistent with our ba k - edge xafs data ( @xmath830.1 ) . we point out that xafs probes the local environment around each selected element separately , and thus determines the average local structure around every different element . the combination of the information from different types of absorbing atoms gives a more constrained result for the local structure . on the other hand , pdf analysis uses all the atomic pairs in the material together , so that identifying individual partial pair - distribution - functions ( ppdfs ) is a much more involved process . consequently , xafs offers an advantage in studying the near neighbor local structure around an absorbing atom in complicated materials . some diffraction studies also revealed a tl displacement from its ideal site@xcite . an x - ray structural study of tetragonal single crystals ( tl@xmath84cu@xmath85)ba@xmath7cuo@xmath86 with = 110 k suggested that the tl is displaced along the @xmath3 direction by 0.11 @xcite . another x - ray diffraction study on tetragonal single crystals tl@xmath87ba@xmath7cu@xmath88o@xmath89 with = 12 k showed a tl displacement of 0.16 along the @xmath4 direction@xcite . a neutron powder diffraction investigation of orthorhombic tl@xmath7ba@xmath7cuo@xmath89 with = 90 k inferred a tl displacement of 0.14 along the @xmath90-axis@xcite ( which is along the @xmath3 direction in a tetragonal unit cell ) . in all these diffraction studies , the tl atoms are assumed to be statistically distributed in the displaced sites . this leads to an uncorrelated tl displacement with the tl - tl pdf having its largest amplitude at the average distance ( see dotted lines in fig . 11b and c ) , which is significantly different from the ppdf proposed by our xafs study . to further check the models suggested by diffraction studies , we have tried to fit the xafs data with the uncorrelated tl displacement model for displacements along both the @xmath3 and @xmath4 directions . in the @xmath3 direction tl displacement model suggested by one diffraction study@xcite , the four - fold displaced sites of the tl are randomly occupied . thus , to check that model in the xafs data analysis , eight tl - tl pairs must be included in the fit with distances @xmath91 , @xmath92 , @xmath93 , @xmath94 , @xmath95 , @xmath96 , @xmath97 , @xmath98 and amplitudes 0.25 , 1 , 1.5 , 1 , 0.25 , 1 , 2 , 1 , respectively . here , @xmath93 and @xmath97 are the average tl - tl distances between and within the tlo layers , respectively ; the splitting of the tl - tl distances from their average distances are proportional to @xmath99 to first order . correspondingly , three tl - ba pairs must be included in the fit with distances @xmath100 , @xmath101 , @xmath102 and amplitudes 1 , 2 , 1 , respectively . three tl - o pairs near 2.0 , 2.3 and 4.4 were also included . the fit was started with the tl displacement of 0.11 from its ideal site ( which corresponds to @xmath99=0.085 and @xmath103=0.08 ) and debye - waller factors of 0.07 for the tl - tl and tl - ba pairs . the starting tl - tl pdf is shown in fig . 11b as dotted line . to reduce the number of fitting parameters and to check the proposed model , we fixed @xmath93 , @xmath97 , and @xmath101 at the average distances ( 3.51 , 3.87 , and 3.89 , respectively ) given by diffraction study , and the amplitudes of the tl and ba peaks to the numbers predicted by the model . the e@xmath57 shifts for the same type of back - scattering atoms are kept equal except for the nearest o atoms . with these constraints , the total number of variables is 25 for this 14-pair - fit . the fitting range is exactly the same as that used for the correlated model ( sec . the extracted ppdf is plotted in fig . 11c as the solid line . the quality of this fit with 25 parameters is comparable to the one obtained from the constrained fit using the correlated model ( fig . 9 ) with only 16 variables . the distances changed a little from the starting values to @xmath104=0.095 and @xmath103=0.07 . since the weighted number of the tl - tl pairs are fixed , their debye - waller factors are varied to adjust the height of the peaks . the extracted debye - waller factors are less than 0.04 for the tl - tl peak at 3.51 and larger than 0.2 for the tl - tl peak at 3.87 , which results in a sharp peak at 3.51 and a dip near 3.87 in the tl - tl pdf as shown in fig . 11c by the solid line . for the tl - ba pdf , there is no difference between the @xmath3 direction correlated and uncorrelated tl displacements , and the extracted tl - ba pdfs are very similar . although the quality of the fit for the uncorrelated model ( 25 variables in the fit ) is comparable to that of the correlated model ( 16 variables in the fit ) , the extracted tl - tl pdf ( solid line in fig . 11c ) from the constrained fit is significantly different from the starting tl - tl pdf ( dotted line in fig . 11c ) which corresponds to the uncorrelated model . furthermore , the extracted tl - tl pdf appears to be very close , within the imposed constraints , to that obtained from the correlated displacement model ( solid line in fig . 11a ) ; _ a dip shows up at the average distance 3.87 . for the uncorrelated model , a peak must show up at the average distance in the tl - tl pdf_. clearly , our xafs data do not support the uncorrelated @xmath3 tl displacement model . for the @xmath4 direction uncorrelated , four - fold tl displacement model suggested by diffraction study@xcite , eight tl - tl pairs must be included in the xafs fit with distances @xmath92 , @xmath93 , @xmath94 , @xmath105 , @xmath106 , @xmath97 , @xmath107 , @xmath108 and amplitudes 1.0 , 2.0 , 1.0 , 0.25 , 1.0 , 1.5 , 1.0 , 0.25 , respectively . two split tl - ba pairs are included in the fit with distances @xmath102 , @xmath100 . three tl - o pairs are also included . the fit was started with the tl displacement obtained from diffraction , 0.16 @xcite ( which corresponds to @xmath109 and @xmath103=0.07 ) . as was done for the fit to the @xmath3 uncorrelated tl displacement model , @xmath93 , @xmath97 , @xmath101 , and the amplitudes of the tl and ba peaks are fixed . the total number of variables is also 25 . the quality of the constrained fit is about 30% worse than that of the fit for the @xmath3 direction correlated displacement . the tl - tl peak at 3.87 is depressed by a large debye - waller factor , 0.2 , in the fitting process . the obtained distances , with @xmath99=0.25 and @xmath103=0.06 , deviate significantly from the starting model and are no longer consistent with the starting model . the extracted tl - tl pdf is plotted in fig . 11d ( solid line ) , which clearly moves away from the starting tl - tl pdf suggested by diffraction and towards the correlated tl - tl pdf ( solid line in fig . the extracted tl - ba pdf is somewhat similar to that obtained from the @xmath3 direction displacement , but it is difficult to distinguish between one broad peak , two split peaks , and three split peaks . the above analyses clearly show that the tl - tl peak at the average lattice distance , 3.87 , is essentially destroyed by the large debye - waller factor in the uncorrelated tl displacement model , for displacements along either the @xmath3 or the @xmath4 directions , and the extracted ppdf from the uncorrelated model tends to move away from the ppdf generated by the starting model and towards the ppdf obtained by the correlated model . therefore , we conclude that our xafs data do not support the uncorrelated tl displacement along either the @xmath3 direction or the @xmath4 direction proposed by the diffraction studies@xcite . we point out that xafs is more sensitive than diffraction to small local structural distortions which have no long range order . this can be seen from the following example . rietveld refinements were performed for tl-2212 using 7 distorted models with different directions of displacement including two correlated pdf models@xcite . the results of these refinements show relatively little sensitivity to the choice of model used for the tlo layers . three models , _ i.e. _ , the uncorrelated four - fold displacement of both o(3 ) and tl along the @xmath4 directions , the uncorrelated four - fold displacement of o(3 ) along the @xmath4 and tl along the @xmath3 directions , and the correlated two - fold displacement of both o(3 ) and tl along the @xmath3 directions ( with the space group pcca)@xcite , result in essentially the same quality of the fit to the data . in the xafs data analysis , these three models can easily be distinguished as shown above . however , the model we proposed may not be unique . more complicated models may generate a similar ppdf to that shown in fig . the general picture presented in this paper for the distortions of the tlo layers is similar to that in ref . 15 , but with more extensive xafs data for and more constraints on possible models . for the xafs data at the tl edge , both the tl - tl and tl - ba peaks are split into several peaks which all overlap ; at the ba edge , only the ba - tl peak is split . therefore , the xafs data analysis at the tl edge is more complicated and technically involved than at the ba edge . although the ba edge xafs data are not quite as sensitive to the tl distortion as the tl edge xafs data , they help rule out some possible models which fit the tl edge xafs data . we also collected tl l@xmath110edge xafs data from another sample@xcite , tl@xmath84ba@xmath7cuo@xmath111 , with = 76 k. this sample is made in a similar way as described in sec . it has a tetragonal structure with lattice constants @xmath112 , @xmath113 . we also have the tl l@xmath110edge xafs data used in the previous study@xcite , which were collected at 78 k from a tetragonal sample with = 89 k and lattice constants @xmath114 , @xmath115 . all the data were reduced in exactly the same way as described in sec . the tl l@xmath110edge xafs data are plotted in fig . 12 for the three samples with different t@xmath116 ; solid lines , = 60 k ; dotted lines , = 76 k ; dashed lines , = 89 k. all the data can be fit with the model shown in fig . the difference in the extracted distances from the constrained fit , among three samples , is within 0.01 , although the amplitude of the tl - o peak near 2 varies significantly among the samples . this clearly shows that the local tl displacement in the tlo layer is not sensitive to the structural symmetry and the transition temperature . this local distortion is more likely intrinsic to the tl-2201 compound , which will be further discussed in the following section . there is a significant difference in the amplitude of the first peak at @xmath117 ( c - axis tl - o(2,3 ) ) . the amplitude was measured in two different ways . first , the amplitudes were obtained from the constrained fit using the feff5 standards , as described earlier assuming a single peak for the first tl - o pair . second , the data were fit one to another in the range of @xmath118 , and the relative amplitudes were obtained . the average amplitude is plotted in fig . 13 as a function of ; it is very close to a linear function . the higher the tl - o amplitude of the peak near @xmath119 , the lower is . this trend is consistent with other experimental results@xcite ; a decrease in oxygen content of about [email protected] per formula unit changed from 0 to about 80@xmath2990 k@xcite although the absolute oxygen content was not well determined@xcite . the data plotted in fig . 13 are from three samples with different tl and o contents which are not precisely determined . it has been reported that both the tl and o contents can change significantly@xcite . the amplitude of the tl - o peak plotted in fig . 13 is from the contributions of the o(2 ) and the o(3 ) along the c - axis , and possibly an o(4)@xcite if it exists . this amplitude can be affected in several ways : ( a ) a change in the o content at the o(2 ) , o(3 ) and o(4)@xcite sites , ( b ) the o movement between the o(3 ) and o(4 ) sites , ( c ) changes in the tl - o(2 ) and tl - o(3 ) ( along the c - axis ) distances , ( d ) a variation in the tl content . from the limited data we have , it is not clear what is the main reason for the variation of the _ measured _ tl - o amplitude which seems much too large to be explained solely by a variation in o content . however , fig . 13 clearly shows that is very sensitive to the o environment around the tl atom . it is generally believed that the layer is the superconducting layer , thus this result also implies that a charge transfer occurs between the and tlo layers and it is mainly controlled by the o arrangement around the tl , not the cu substitution for tl . the atomic valence ( v ) can be estimated from the obtained local structure using the brown - altermatt - zachariasen formulae@xcite , which describe the relationship between the bond length ( @xmath67 ) and the bond valence ( @xmath120 ) @xmath121 or @xmath122,\ ] ] where @xmath123 , @xmath73 and @xmath124 are empirically determined parameters@xcite . for tl@xmath26-o , we used eq . ( 2 ) with the most recent available parameters , @xmath123=2.04 and @xmath73=6.4@xcite and the structural model ( fig . 9 ) constructed based on our xafs results for the sample with the highest o concentration . the calculated tl valence is ( [email protected])+ , very close to the formal valence , 3 + ( tl@xmath125ba@xmath126cu@xmath25o@xmath127 ) . this is also consistent with the photoemission@xcite and the x - ray - absorption near - edge spectra@xcite measurements which indicate a tl valence slightly less than 3 + . if both the tl and o(3 ) atoms were at their ideal sites ( @xmath0 ) , the calculated effective valence for tl would be 2.77 + , which is too small compared to the formal valence and not stable for the tl ion ( the tl ion should be either tl@xmath128 or tl@xmath26 ) . this is probably part of the reason why the tlo layer is distorted . we also calculated the cu valence , using eq . ( 3 ) with @xmath129 , @xmath130 @xcite and the local structure derived from xafs ( table i ) . the calculated cu valence is ( [email protected])+ , which is very close to the cu(2 ) valence in ( 2.20+)@xcite and cu in tl@xmath7ba@xmath7cacu@xmath7o@xmath30 ( 2.11+)@xcite calculated using the same formula . our estimated effective valences for cu and tl are qualitatively in agreement with the charge transfer ( or local charge ) picture between the tlo and layers@xcite,_i.e . _ , tl@xmath7o@xmath31 + cuo @xmath131 [ tl@xmath7o@xmath31]@xmath133 + [ cuo]@xmath134 . it is generally believed that the extra holes in the are predominantly at the o site@xcite . however , it is not clear where the extra electrons are in the tlo layers . we note that the extra holes in the layer are not solely introduced by the reduction of the tl valence . the variation in o content can also change the hole concentration in the layer . a decrease in o content of about 0.1 per formula unit , which corresponds to a decrease in hole concentration of 0.2 , increases from @xmath290 to @xmath2990 k@xcite . however , there is a controversy in the literature about the absolute , optimum o content for samples with the highest @xcite . long range order modulations have been observed in bi - sr - ca - cu - o superconductors and related materials@xcite . some long range order in the tlo layers is also observed in tl-2201 and tl-2212 by electron diffraction and high - resolution transmission electron microscopy@xcite . the modulation wave vector is about @xmath135 for tetragonal tl-2201 ( @xmath136 3.8 ) , @xmath137 for orthorhombic tl-2201 ( @xmath1385.4 ) , and @xmath139 for tetragonal tl-2212 . the superlattice for the first case would have a large unite cell , @xmath140 in the ab - plane . however , these superlattices@xcite are usually observed only with selected - area electron diffraction and are much weaker in intensity than in the bi - sr - ca - cu - o superconductors which exhibit strong supercell diffraction structure@xcite and c - axis displacemnets in the tlo layers are small ( see sec . such superstructures are not always observed in the tetragonal tl-2201 samples@xcite . consequently , this long range order of the lattice distortions in the tl - based superconductors seems weak , sample dependent , and may not be intrinsic . since both the tl and o(3 ) atoms can be shifted in two equivalent directions in the model we have proposed , material with this distorted local structure might well have a superstructure over a limited range , which is qualitatively consistent with the above electron diffraction measurements . there are many ways to construct superstructures with periodicity of @xmath140 using the distorted @xmath141 unit cell shown in fig . 9 for the tlo layers . two possible examples that are probably over simplified are shown in fig . the tl displacements ( not shown ) are the same from one @xmath141 cell to another , but the direction of the o displacements varies . in the upper panel , the o positions are not periodic ; some o atoms move together or apart , while others move in the same direction . in the lower panel , the direction of the o displacements is rotated by 90@xmath36 in the right half of the supercell . the local structure of , with an orthorhombic phase and = 60 k , has been studied at the cu k@xmath142 , ba k@xmath142 , and tl l@xmath110edges using the xafs technique . several new approaches have been implemented , _ e.g. _ , direct structural simulations up to 7 using the feff5 code , constrained fits , and comparisons of ppdfs for different models . both the qualitative and quantitative data analyses clearly show that the cu , o(1 ) , o(2 ) , and ba atoms are at their ideal sites in the unit cell as given by the diffraction studies , while the tl and o(3 ) atoms are displaced from the sites suggested by the average crystal structure . the tl - tl distance at r=3.5 between the tlo layers remains , but the tl - tl distance at 3.9 in the tlo layer is not observed and the tl - ba and ba - tl peaks are very broad . the shorter tl - o(3 ) distance in the tlo layer is about 2.33 , significantly shorter than the distance calculated with both the tl and o(3 ) atoms at their ideal @xmath0 sites ( @xmath10 or @xmath2 ) . an excellent fit to the xafs data can be achieved with a correlated displacement model shown in fig . the fitting results show that the tl atom is displaced along the @xmath3 direction from its ideal site by about 0.11 and the o(3 ) atom is shifted from the @xmath0 site by about 0.53 roughly along the @xmath4 direction . this model also fits very well to the xafs data collected from two tetragonal samples with different t@xmath143s . the xafs data do not support the uncorrelated displacement models proposed by diffraction investigations@xcite . our model is similar to that proposed by the pdf analyses of the neutron scattering data on tl-2212@xcite , but is more specific about the relative shift of the tl atoms between the two consecutive tlo layers . however , the tl displacement , 0.11 , obtained in our study , is only one third of the value ( 0.32 ) suggested by the pdf . the estimated cu and tl valences imply a charge transfer between the and tlo layers which is consistent with other experimental results@xcite . the estimated tl valence also supports the distorted structure of the tlo layer obtained from our xafs analysis . the estimated tl valence ( 2.95 + ) in the distorted tlo layer is much closer to its formal valence ( 3 + ) than that ( 2.77 + ) estimated from the ideal tlo layer . a comparison of the xafs data at the tl l@xmath110 edge from three samples clearly shows a correlation between the transition temperature and the local o environment around the tl atoms . this implies that the hole concentration in the layer is mainly controlled by the o arrangement about the tl in those samples , and not by cu substitution for the tl . we thank corwin booth for help in the data collection . the experiments were performed at the stanford synchrotron radiation laboratory , which is operated by the u.s . department of energy , division of chemical sciences , and by the nih , biomedical resource technology program , division of research resources . the experiment is partially carried out on uc / national laboratories prt beam time . the work is supported in part by nsf grant dmr-92 - 05204 . b. h. toby , w. dmowksi , t. egami , j. d. jorgensen , m. a. subramanian , j. gopalakrishnan , and a. w. sleight , in _ high temperature superconductors : relationships between properties , structure , and solid - state chemistry _ , edited by j. d. jorgensen _ et al _ ( materials research society , pittsburgh , pennsylvinia , 1989 ) , p.309 . in this paper , we have used @xmath52=0.9 for the tl - tl pair . in ref . 9 , @xmath52 was set to be about 0.5 which seems too small according to our recent detailed comparison of feff5 calculation with experimental xafs@xcite . .cu local structure obtained from xafs and diffraction . nbrs " indicates the weighted number of neighbors . the estimated errors are : number of neighbors , @xmath144 15 % ; distances , @xmath144 0.02 . [ cols="^,^,^,^,^,^,^ " , ]

the voltammetry for the reduction of oxygen at a microdisc electrode is reported in two room temperature ionic liquids : 1-butyl-1-methylpyrrolidinium _ bis_(trifluoromethylsulfonyl)imide + ( [ bmpyrr][ntf ) and trihexyltetradecylphosphonium _ bis_(trifluoromethylsulfonyl)imide + ( [ p[ntf ) at 298 k. simulated voltammograms using butler - volmer theory and symmetric marcus - hush ( smh ) theory were compared with experimental data . butler - volmer theory consistently provided experimental parameters with a higher level of certainty than smh theory . a value of solvent reorganisation energy for oxygen reduction in ionic liquids was inferred for the first time as 0.4 - 0.5 ev , which is attributable to inner - sphere reorganisation with a negligible contribution from solvent reorganisation . the developed butler - volmer and symmetric marcus - hush programs are also used to theoretically study the possibility of kinetically limited steady state currents , and to establish an approximate equivalence relationship between microdisc electrodes and spherical electrodes resting on a surface for steady state voltammetry for both butler - volmer and symmetric marcus - hush theory .

1503.01654

current understanding of interfacial electron transfer ( et ) kinetics is dominated by two theories , butler - volmer@xcite and marcus - hush@xcite , which both describe a one - electron transfer : @xmath2 the first , butler - volmer theory ( bv , equations [ bv red ] and [ bv ox ] ) , is well - reported and reliable for parameterisation@xcite , but is phenomenological , and therefore is unable to provide any detailed physical insights of a predictive nature . bv theory relates the electrode kinetics to the heterogeneous rate constants ( @xmath3 ) , the formal potential ( @xmath4 ) , and the transfer coefficient , @xmath5 , which empirically describes how product or reactant - like the transition state is@xcite . @xmath6 \label{bv red}\\ k_\mathrm{ox } = k^0\text{exp}\left[\left(1-\alpha\right)\frac{f\left(e - e_\mathrm{f}^\minuso\right)}{rt}\right ] \label{bv ox}\end{aligned}\ ] ] an alternative theory , symmetric marcus - hush ( smh ) , offers greater microscopic physical insight into the system under study . mh theory relates the heterogeneous rate constant to the reaction gibbs energy , @xmath7 , and reorganisation energy , @xmath8 : @xmath9\end{aligned}\ ] ] as is shown in equation [ inner plus outer ] , the total reorganisation energy is the sum of contributions from inner sphere bond reorganisation energy ( @xmath10 ) and solvent reorganisation energy ( @xmath11 ) . these can be estimated separately , with the outer - sphere solvent reorganisation energy from the born - marcus solvation equation@xcite , and inner - sphere contributions from computational calculations ( for example , dft etc@xcite ) of changes in bond lengths and angles of the molecule . @xmath12 these two models have been used to analyse a range of electrochemical processes by cyclic voltammetry in aqueous@xcite and non - aqueous@xcite solvents . a comprehensive review by henstridge _ _ et al__@xcite concluded that smh theory struggles to adequately describe the voltammetric response of solution - phase redox couples whose transfer coefficients deviate significantly from 0.5 . a fourth parameter , @xmath13 , was introduced , which accounts for any differences between the inner - shell force constant of the reactant and the product , to create asymmetric marcus - hush theory ( amh ) . this refinement was explored in aqueous@xcite and non aqueous@xcite solvents and surface bound systems@xcite by laborda _ et al _ , and it was concluded that amh was able to parameterise the experimental voltammetry at least as well as bv theory , and more accurately than smh theory . the electroreduction of oxygen has been extensively studied in a variety of solvents@xcite . the product of the initial one - electron reduction , superoxide , is a highly nucleophilic ion that initiates further reactions with a proton source , if one is present in solution . however , in absence of a proton source , the superoxide ion is stable , and the reaction proceeds as a simple one electron transfer : @xmath14 et kinetics of irreversible oxygen reduction has been studied in - depth in aqueous solvents@xcite and values for the total reorganisation energy for the reduction of oxygen to superoxide have been estimated to lie between 1.26 - 4.51 ev@xcite . this has been attributed to rearrangement of the polar water molecules upon formation of the superoxide ions . room temperature ionic liquids ( rtils ) are molten salts with melting points below room temperature@xcite , and normally consist of a bulky , asymmetric organic cation , and an inorganic anion@xcite . ionic liquids are interesting solvents in which to examine the reversible one - electron reduction of oxygen , because the superoxide has been known to be stable , leading to quasi - reversible voltammetry in a range of ils.@xcite . lynden - bell has conducted a thorough study into the applicability of marcus theory to rtil solvents@xcite . it was suggested _ via _ simulation that the outer sphere reorganisation energy for redox processes in rtils will be similar to that for the same process in a non ionic polar solvent . in a polar solvent , the molecules will re - orient themselves in response to the addition of charge , and in an ionic liquid the ions will translate themselves . it was predicted that these two processes produce similar reorganisation energies . the mechanism and et kinetics have been elucidated for oxygen reduction to hydrogen peroxide in protic ionic liquids,@xcite and in hydrophobic rtils in the presence of water.@xcite as yet , the et kinetics of the one electron reduction of oxygen to superoxide in dry ionic solvents has not been studied , and no values for reorganisation energy for this process have been established . this study is the first to examine smh theory of one electron oxygen reduction in ils at microdisc electrodes . the structures of the ions in the ils used in this study are shown in fig . [ il structures ] . the present work aims to develop the theory of smh voltammetry at microdisc electrodes and experimentally determine a value for solvent reorganisation , @xmath8 , for oxygen reduction in two ionic liquids . the present work also aims to compare the ability of smh theory and bv theory to simulate experimental reduction of oxygen and quantify a goodness of fit for each . this will allow conclusions to be reached about the relative ability of each theory to parameterise experimental data . we consider a one electron reduction of an electroactive species : @xmath2 at a microdisc electrode , using both butler - volmer theory and symmetric marcus - hush theory to describe the electron transfer kinetics . cyclic voltammetry is simulated using both models , and both are subsequently compared to experimental data ( see section [ results and discussion ] ) . this study utilises microdisc electrodes , which offer several advantages over macroelectrodes . firstly , as the capacitance of the double layer is proportional to the size of the electrode , a consequence of the reduced size of the electrode will be naturally lower capacitance . secondly , the small currents passed reduce the ohmic drop and allow a two - electrode system to be employed@xcite . lastly , the smaller area enables a very small volume of ionic liquid to be studied ( in a t - cell system , as described previously@xcite ) , which is desirable for logistical reasons ( expense , ability to dry the aliquot of il more quickly ) . cylindrical coordinates are employed to model the two dimensional microdisc electrode and simulate cyclic voltammetry . [ electrode schematic ] shows a schematic diagram of the microelectrode ( a ) and the two dimensional cylindrical space used in our model ( b ) . also defined are the spatial coordinates @xmath15 and @xmath16 , as well as the electrode radius @xmath17 . in cylindrical space , fick s second law for diffusional mass transport is : @xmath18 all symbols are defined in table [ dimensional ] . the mass transport is considered to take place under full support ( as is usually assumed when ionic liquids are used as solvents ) , and therefore electrical migration can be neglected . further , the absence of mechanical stirring and any significant temperature gradients eliminate the presence of forced and natural convection respectively , allowing a diffusion only model to be used . the electron transfer reaction is only considered to occur at the electrode surface ( @xmath19 , @xmath20 ) . the rate of this reaction is dependent on the potential applied to the electrode , @xmath21 ( v ) . in cyclic voltammetry , the potential is given by the equation : @xmath22 where @xmath23 and @xmath24 are respectively the initial applied potential and the vertex potential ( v ) , @xmath25 is the scan rate ( v s@xmath26 ) and @xmath27 is the time since the start of the experiment ( s ) . the flux of species a across the electrode due to electrolysis is then described as : @xmath28 where the rate constants @xmath29 and @xmath30 are functions of the applied potential @xmath21 , and @xmath31 and @xmath32 are the concentrations , at the electrode surface , of species a and b respectively . the mathematical forms of @xmath29 and @xmath30 depend on whether butler - volmer or marcus - hush kinetics are employed , as detailed below . within the butler - volmer formalism of electron transfer kinetics , @xmath29 and @xmath30 are considered to be exponentially dependent on the applied potential:@xcite @xmath33 \\ k_\mathrm{ox } = k^0\mathrm{exp}\left[\left(1-\alpha\right)\theta\right]\end{aligned}\ ] ] where @xmath3 is the standard heterogeneous electrochemical rate constant ( m s@xmath26 ) , @xmath5 is the transfer coefficient and @xmath34 is a dimensionless potential , defined as : @xmath35 where @xmath36 is the formal potential of the a / b couple and @xmath37 , @xmath38 , and @xmath39 have their usual meanings . the transfer coefficient @xmath5 , as described in the introduction , has a value of between zero and one , and is considered to reflect the structure of the transition state of the a / b redox couple@xcite . in ( symmetric ) marcus - hush theory@xcite , @xmath29 and @xmath30 are functions of the applied electrode potential and the reorganisation energy , @xmath8 . mathematically : @xmath40 @xmath41 is given by the following integral : @xmath42}{1 + \text{exp}\left[\mp x\right]}}\text{d}x\ ] ] @xmath43 is the activation energy of reduction / oxidation , and is given by : @xmath44 where @xmath45 is the dimensionless reorganisation energy , given by : @xmath46 @xmath47 is an integration variable , defined as : @xmath48 this accounts for the continuum of energy levels in the electrode of energy @xmath49 . to simulate cyclic voltammetry , the mass transport equation and electron transfer kinetic equations are normalised to remove scaling factors from the simulation . substituting dimensionless parameters as defined in table [ dimensionless ] into the mass transport equation , we obtain : @xmath50 the boundary condition at the electrode surface ( @xmath51 , @xmath52 ) becomes : @xmath53 where @xmath54 \\ k_\mathrm{ox } = k^0\text{exp}\left[\left(1-\alpha\right)\theta\right]\end{aligned}\ ] ] for butler - volmer kinetics , and @xmath55 for marcus hush kinetics . in both cases the dimensionless standard heterogeneous electrochemical rate constant is defined as : @xmath56 at the electrode surface , equal and opposite flux acts as the boundary condition for species b : @xmath57 the outer boundaries of the simulation space are set at : @xmath58 where @xmath59 and @xmath60 are , respectively , the largest dimensionless diffusion coefficient in the system and the total dimensionless time of the experiment . these boundaries have been shown to be sufficiently distant from the electrode to be well outside the depletion zone@xcite . at these boundaries , as well as at the symmetry boundary at @xmath61 , all @xmath62 and at the insulating surface ( @xmath63 , @xmath52 ) , a zero flux condition is imposed : @xmath64 where @xmath65 is one of the spatial coordinates @xmath38 or @xmath62 as appropriate . initial conditions are @xmath66 and @xmath67 throughout the simulation space . at @xmath68 the experiment starts and the electrode surface boundary conditions are imposed . the total dimensionless flux @xmath69 at the electrode is calculated as : @xmath70 this dimensionless flux is transformed into a real current @xmath71 by : @xmath72 in more complex reaction mechanisms , the single electron transfer thus far simulated may be followed by homogeneous kinetics . for example , the reduction of oxygen in the ionic liquid [ p@xmath1[ntf@xmath0 ( see section [ p14666ntf2 results ] ) follows the following mechanism : @xmath73 where the homogeneous reaction involves proton abstraction by the superoxide ion from the phosphonium cation in the ionic liquid . this irreversible , pseudo first order reaction forms a product which can then be oxidised , at less negative potentials than the initial electron transfer to oxygen . the homogeneous step is considered to have a fast rate constant . this being the case , if only the forward , reductive , wave of a cyclic voltammogram is measured and simulated , species x and y may be neglected and only the first 2 steps in the above mechanism considered . to incorporate the homogeneous step into the simulations , the dimensionless mass transport equation is amended to : @xmath74 where @xmath75 is a dimensionless homogeneous rate constant , defined as : @xmath76 and @xmath77 is a first order homogeneous rate constant for the reaction of species b. the above model is discretised using the crank - nicolson method@xcite and solved numerically over discrete spatial and temporal grids . the temporal grid is defined initially in terms of the applied dimensionless potential , @xmath34 . a parameter @xmath78 is defined as the number of discrete time steps per unit @xmath34 . the value of @xmath34 at each time step @xmath77 is then defined : @xmath79 where the minus sign is used for the reductive sweep , and the plus sign for the oxidative sweep . the dimensionless time , @xmath80 , at each time step is then calculated as : @xmath81 a non regular , exponentially expanding spatial grid is used to ensure computational efficiency . in the z direction , the first spatial point ( at the electrode / insulating surface ) is defined as @xmath82 . the first step size is defined as @xmath83 , with subsequent step sizes growing increasingly large until the simulation boundary is reached . each @xmath62 direction spatial point is defined as : @xmath84 the @xmath38 direction spatial grid is defined in the same way , except that rather than expanding from a simulation boundary , the grid expands away from the edge of the electrode at @xmath85 in both directions until @xmath61 and @xmath86 are reached . [ grid ] shows a schematic diagram of some of the spatial grid points used in simulations . in order to achieve convergence to within 0.5% of a fully converged result , the following grid parameters were used : @xmath87 , @xmath88 and @xmath89 . the alternating direction implicit ( adi)@xcite method was employed to simulate the voltammetry , in conjuction with the thomas algorithm@xcite to solve the mass transport equation and boundary conditions simultaneously in matrix form . the model was coded in c++ with multithreaded parallel programming employed to increase efficiency . typical running times were _ c.a . _ 10 minutes on a 2.26 ghz intel(r ) xenon(r ) cpu with 2.25 gb of ram . electrochemical reactions can occur at the surface of nanoparticles impacting upon a conductive , but electrochemical inert surface . a schematic diagram of a spherical nanoparticle in contact with a conducting but otherwise inert surface is shown in fig . [ sphere electrode schematic ] ( a ) . upon contact , the nanoparticle becomes electrochemically active , and a species in solution can be reduced / oxidised on its surface , if a suitable potential is applied to the inter surface . this is modelled in the same way as for a microdisc described above , but with changed boundary conditions@xcite . a cylindrical coordinate system is still used , as defined in fig . [ sphere electrode schematic ] . part ( a ) shows a schematic of the spherical electrode on the surface . part ( b ) shows the 2 dimensional simulation space . the electrode surface boundary condition is now applied at over the line @xmath90 where the flux normal to the electrode surface is given by : @xmath91 where @xmath65 is a coordinate normal to the electrode surface , and @xmath29 and @xmath30 are as defined above for either butler - volmer or marcus - hush kinetics . other boundary conditions were zero flux conditions at the @xmath61 , @xmath92 , @xmath52 and @xmath93 . the dimensionless flux is then defined as : @xmath94\text{cos}\phi\text { d}\phi\ ] ] the current is then defined , as for a microdisc above , as : @xmath95 the temporal grid for the simulation of voltammetry at a spherical electrode is the same as defined above for a microdisc electrode . to define the spatial grid , the geometry of the electrode is taken into account . the electrode surface is divided into @xmath96 equal spaced points , with a grid vertex at each of these points . this defines the grid points in the region @xmath51 and @xmath97 , as shown schematically in fig . [ sphere electrode schematic ] . in the regions @xmath63 and @xmath98 , the grid expands exactly as outlined for the microdisc above . convergence studies found @xmath99 sufficient to ensure convergence . computational methods were identical to those employed for microdisc simulations . simulation running times were _ c.a . _ 10 mins on a 2.26 ghz intel(r ) xenon(r ) cpu with 2.25 gb of ram . the shoup and szabo equation is used to calculate expected chronoamperometric responses at microdisc electrodes to within 0.6% error@xcite . for a reduction : @xmath100 \\ \tau & = & \frac{d_\mathrm{a}}{r_\mathrm{e}^2}t\end{aligned}\ ] ] in order to examine the relative merits of butler - volmer theory and marcus - hush theory in simulating cyclic voltammetry , a quantitative measure of error is desirable . this study uses the mean scaled absolute deviation ( msad ) to quantify the error between experiment and theory . the msad value for a simulated fit is given by : @xmath101 where @xmath65 is the number of experimental data points . since it is unlikely that each experimental data point will have an exactly corresponding simulated data point , a linear interpolation is used between the two simulated data points surrounding the experimental data point in question . this interpolated simulated current is then used in the calculation of the msad value . small values of the experimental current can lead to extremely large msad values despite the absolute difference between experiment and theory being small . for this reason , all experimental data points with a value less than 5% of the maximum absolute value of the current are neglected . this includes the initial part of the reductive wave as well as a narrow part of the oxidative wave , where the current passes from negative to positive values . previous work by feldberg@xcite and henstridge _ _ et . al.__@xcite showed that in the marcus - hush model of electron transfer kinetics , steady state currents can be kinetically rather than diffusionally limited for small values of the reorganisation energy , @xmath8 . in feldberg s study , an approximate equation of oldham and zoski@xcite describing steady state voltammetry at a microdisc was invoked and modified to use marcus - hush , rather than butler - volmer kinetics . henstridge used numerical simulation to generate marcus - hush steady state responses at a conductive nanosphere resting on an electrochemically inert but conductive surface . in this study , we simulate steady state voltammetry numerically using the symmetric marcus - hush formalism of electron transfer , and compare the circumstances in which a kinetically limited current is observed at a microdisc to those at a nanosphere . we also extend previous work showing the equivalence of microdiscs and spheres - on - surfaces under the butler - volmer formalism ( subject to a mass transport correction factor)@xcite , and show that the approximate equivalence is retained when marcus - hush kinetics are used . steady state voltammetry at a microdisc electrode was simulated using the model outlined above . various values of dimensionless reorganisation energy , @xmath45 , and dimensionless heterogeneous rate constant , @xmath102 , were simulated , and the results are shown in fig . [ vary k disc ] . part ( a ) shows results for @xmath103 , ( b ) shows @xmath104 and ( c ) shows @xmath105 . also shown in each case is the steady state voltammetry simulated using butler - volmer kinetics with the same values of @xmath102 , and @xmath106 , and marcus - hush results are shown relative to this butler - volmer result . for all three values of @xmath102 , it is seen that large @xmath45 values produce the same limiting current as the butler - volmer model . as @xmath45 becomes smaller , the limiting current becomes less than predicted by the butler - volmer model as kinetic control takes over . this effect is more extreme for the smallest values of @xmath102 , where even a @xmath45 value of 200 has not reached the butler - volmer limit at all points on the wave . for @xmath107 , unfeasibly small @xmath45 values would be required to show significant deviation form the butler - volmer result . this is in qualitative agreement with the previous studies of this nature@xcite . the definition of @xmath102 in our dimensionless model is : @xmath108 lowering @xmath102 will therefore mean that either @xmath3 or @xmath17 is lowered , or @xmath109 is raised . lowering @xmath3 trivially means the electrochemical kinetics become more likely to be the limiting factor in the steady state current as opposed to diffusion . either lower @xmath17 or increasing @xmath109 also both lower @xmath102 and lead to a more kinetically limited current . this is due to the fact that either of these changes will increase mass transport efficiency to the electrode surface , thus making electrochemical kinetics the more limiting factor . therefore , in order to see a kinetically limited current , it is desirable to have a small @xmath3 , a small @xmath17 and a large @xmath109 . recent work by molina _ _ et . al.__@xcite shows that steady state voltammetry at an isolated spherical electrodes alone in solution , a spherical electrode supported on a conductive but inert surface and a microdisc electrode ( all shown schematically in fig . [ 3 geometries ] ) may all be approximately described ( under the butler - volmer formalism ) by a single equation : @xmath110 } \label{steady state equation}\ ] ] where @xmath111 is a dimensionless heterogeneous rate constant scaled by a geometry dependent mass transport factor . these parameters are defined for isolated spheres ( is ) spheres supported on surfaces ( sos ) and microdiscs ( md ) as : @xmath112 therefore , steady state voltammetry at a microdisc of electrochemical rate constant @xmath113 produces approximately the same result as steady state voltammetry at a sphere on a surface of heterogeneous rate constant @xmath114 , as was shown . we now investigate if this relation holds under the marcus hush formalism over a wide range of @xmath102 and @xmath45 values . to do so , steady state voltammetry was simulated using the marcus - hush model described above for both a microdisc and a sphere on a surface . for each microdisc heterogeneous rate constant ( @xmath113 ) that was used , a corresponding sphere on a surface simulation was run with @xmath114 , with @xmath45 the same in each case . [ compare ] shows simulated steady state voltammograms at a microdisc electrode ( solid line ) and a sphere on a surface electrode ( circles ) at varying values of @xmath45 ( 20 , 24 , 30 , 40 , 60 , 100 and 200 ) . @xmath102 is equal to @xmath115 for the microdisc and @xmath116 for the sphere . qualitatively , the best agreement between the two geometries is seen at the smallest values of @xmath45 where an almost exact fit is observed . less good fits are seen at intermediate values of @xmath45 . [ error ] ( a ) shows the percentage error between the fits in fig . [ compare ] , as well as for three other values of @xmath113 : @xmath117 ( b ) , @xmath118 ( c ) , and @xmath119 ( d ) . to obtain the corresponding values of @xmath120 , @xmath113 is multiplied by @xmath121 . the percentage error is defined as : @xmath122 it is seen that as @xmath102 increases , the error becomes less and less dependent on the values of @xmath45 . this is because at high @xmath102 the marcus - hush model of electron transfer approaches the butler - volmer limit , and becomes independent of @xmath45 . over the whole range of @xmath102 and @xmath45 values here studied , the error in the equivalence between a microdisc and a sphere on a surface never exceeds 2.55% in regions of appreciable current . it is thus reasonable to say that the correspondence between the two geometries holds within the marcus - hush model of electron transfer , as well as within the butler - volmer model . we now apply our butler - volmer and marcus - hush models for voltammetry at a microdisc to the reduction of oxygen in two ionic liquids , and compare the effectiveness of the models in simulating experimental data . ferrocene ( fe(c@xmath123h@xmath123)@xmath124 , aldrich , 98% ) , tetra - n - butylammonium perchlorate ( tbp , fluka , puriss electrochemical grade , 99% ) and acetonitrile ( mecn , fischer scientific , hplc grade , 99% ) were used as received . ionic liquids trihexyltetradecylphosphonium _ bis_(trifluoromethylsulfonyl)-imide ( [ p@xmath1[ntf@xmath0 ) and 1-butyl-1-methylpyrrolidinium _ bis_(trifluoromethylsulfonyl)imide + ( [ bmpyrr][ntf@xmath0 ) were kindly donated by professor c. hardacre of queen s university , belfast and were used as received . oxygen ( 99.5% ) was purchased from boc , surrey , uk . electrochemical experiments ( cyclic voltammetry ( cv ) and chronoamperometry ) were conducted using a @xmath125-autolab potentiostat ( eco - chemie , netherlands ) . all experiments were conducted inside a temperature controlled faraday cage.@xcite the working microdisc electrode ( either gold or platinum , ij cambria scientific ltd , uk ) , 10 @xmath125 m nominal diameter , was polished prior to use using a water - alumina slurry ( 1 , 0.3 , 0.05 @xmath125 m , five minutes on each grade ) on soft lapping pads ( buehler , illinois).@xcite precise radii were determined through calibration of the electrode with a 2.0 mm solution of ferrocene in acetonitrile containing 0.1 m tbap ( silver wire as both a counter and quasi - reference electrode ) ; chronoamperometry was recorded at 298 k , and assuming a diffusion coefficient of @xmath126 m@xmath127 s@xmath26,@xcite the data was analysed with respect to the shoup and szabo equation@xcite ( described in section [ chrono section ] ) . this gave electrode radii of @xmath128 @xmath125 m ( gold ) and @xmath129 @xmath125 m ( platinum ) . a 0.5 mm silver wire was used both as a counter and a quasi - reference electrode . the experiments conducted with a gold electrode were arranged in a glass t - cell , as described previously,@xcite with a plastic collar attached to the working electrode . a 10 @xmath125l aliquot of ionic liquid was transferred into the t - cell , which was left under vacuum ( 0.2 mbar ) overnight , to ensure the complete removal of volatile substances , including atmospheric gases and any aqueous impurities . in the case of the platinum working electrode , a three electrode ( with two silver wires acting as a counter and a reference electrode ) system was used in a sealed sample vial with a pre - vacuumed ionic liquid . oxygen was then introduced to the t - cell or sample vial via a drying column of molecular sieves ( activated at 200 @xmath130c overnight , 4 , sigma aldrich ) . the gas was left on at a flow rate of 200 - 300 @xmath131 min@xmath26 for at least half an hour to reach saturation , at which point cyclic voltammograms were recorded to ensure equilibrium had been reached and the reductive current in successive experiments was stable . a wait time of minimum 15 minutes was imposed between each experiment to ensure a stable concentration of oxygen at the electrode surface . this section reports the comparative outcome of the use of bv theory and smh theory to model experimental cyclic voltammetry of the reduction of oxygen in the ionic liquids [ bmpyrr][ntf@xmath0 and [ p@xmath1[ntf@xmath0 : @xmath132 this is followed by the inference of physical implications from the application of smh and the determination of a value for solvent reorganisation energy , @xmath8 , for both rtils . the first system under study is the reduction of oxygen in [ bmpyrr][ntf@xmath0 . this ionic liquid was chosen for its relatively low viscosity ( 70 cp at 298 k)@xcite and previously observed well characterised voltammetry for oxygen reduction.@xcite cyclic voltammetry ( cv ) was performed on saturated solutions of oxygen in [ bmpyrr][ntf@xmath0 over a range of scan rates ( 100 - 400 mv s@xmath26 ) . a blank cv , from -3 v - 2 v vs ag , was recorded prior to the introduction of oxygen to ensure the ionic liquid was free from impurities , and to estimate solution capacitance for each scan rate . the capacitative current was subtracted from the experimental current to generate corrected voltammograms prior to further analysis . in all further experimental cases , the potential was swept from 0.00 v ( vs ag ) , a voltage at which _ ca . _ zero faradaic current flows , to a potential cathodic of the reduction of oxygen ( -1.50 v vs ag ) , then back to 0.00 v ( vs ag ) . at a scan rate of 200 mv s@xmath26 , the oxidative peak current at -0.8 v was 1.7 na , the limiting current was -4.4 na , and the half peak potential was -0.9 v vs ag wire . the peak corresponding to the reduction of oxygen appeared on the forward scan at _ ca . _ -1.3 v vs ag , whilst the oxidative peak appeared on the reverse scan at _ ca . _ -0.8 v vs ag . the resultant voltammograms ( fig . [ bmpyrrntf2 ] , a - c ) show more steady - state - like behaviour in the forward wave , which corresponds to the reduction of oxygen to superoxide , than in the back wave , corresponding to the reoxidation of superoxide to oxygen . the disparity in diffusion coefficients of the reactant and product in the ionic solvent results in different types of diffusion being observed at the electrode surface . in the case of oxygen , we see convergent fickian diffusion of oxygen , whereas for the more slowly diffusing superoxide we observe planar fickian diffusion.@xcite cvs were modelled using a previously reported program@xcite that utilises butler - volmer kinetics , as outlined above . as this is a well - reported system,@xcite the starting point for the diffusion coefficient and the concentration of oxygen were taken from literature . optimised values for @xmath3 , @xmath5 , @xmath133 , @xmath134 , @xmath135 , and @xmath136 ( tabulated in table [ bmpyrr parameters ] ) were obtained to give the `` best fit '' , which was quantified with the calculation of a mean scaled absolute deviation ( msad , as described in section [ msad ] ) . these values were used across the scan rates with all values remaining constant , with the exception of @xmath136 , which was allowed to deviate slightly ( @xmath137 v ) to account for the use of a quasi - reference electrode.@xcite the optimised values and the mean msads across the scan rates are reported in table [ bmpyrr parameters ] . the value of @xmath135 , the diffusion coefficient of the superoxide ion , is an order of magnitude smaller than that of the diffusion coefficient for oxygen , @xmath134 . this is due to the charged superoxide ion s increased interaction with the ionic solvent , which reduces its movement through the solution . this behaviour has been reported in other ionic liquids by buzzeo _ et al_,@xcite evans _ et al_,@xcite and huang _ et al_.@xcite the value for the transfer coefficient , @xmath5 , of 0.29 , suggests an early , reactant - like transition state.@xcite another notable outcome is the value for @xmath3 , of @xmath138 cm s@xmath26 , which is comparable to @xmath3 for oxygen reduction in other ionic solvents , as seen in table [ lit values ] , which details transfer coefficients and rate constants for oxygen reduction in a range of ionic liquids . symmetric marcus - hush ( smh ) theory was next used to model the electrode kinetics , and the same procedure as outlined above was followed to establish independent , optimal values for the parameters of concern . in this case , the transfer coefficient , @xmath5 , is no longer a parameter that is taken into consideration . instead , the reorganisation energy , @xmath8 , is used . the standard electrochemical rate constant , @xmath3 , also takes on different values in this model . the best fit values and the mean msad values are outlined in table [ bmpyrr parameters ] . the experimental value for @xmath8 was determined to be 0.4 @xmath139 0.1 ev in this solvent . this was determined by establishing best - fit parameters for the smh model with experimental data . the bv program reproduces experimental data with a greater level of accuracy that the smh program , as evidenced by lower msad values and a better fit , as seen in table [ bmpyrr parameters ] and fig . [ bmpyrrntf2 ] . this suggests bv kinetics parameterise the data better than the smh model . a similar conclusion has been reached with respect to the modelling of redox processes in molecular solvents@xcite . the second system under consideration is the chemically irreversible reduction of oxygen in [ p@xmath1[ntf@xmath0 . this ionic liquid has high hydrophobicity@xcite and previously reported well defined voltammetry.@xcite cyclic voltammetry ( cv ) was performed on saturated solutions of oxygen in [ p@xmath1[ntf@xmath0 over a range of scan rates ( 100 - 400 mv s@xmath26 ) ( see fig . [ p14666ntf2 ] , a - c ) . a blank cv was recorded prior to the introduction of oxygen to ensure the ionic liquid was free from impurities . the baselines of the raw experimental cvs were adjusted to generate corrected voltammograms prior to further analysis . in all cases , the potential was swept from 0.00 v vs ag , to a potential cathodic of the reduction of oxygen ( -2.3 v vs ag ) . at a scan rate of 200 mv s@xmath26 , the limiting current was -6.5 na , and the half peak potential was -1.6 v vs ag . the reduction of oxygen occurred on the forward scan at -2.20 v vs ag , whilst an oxidative feature appeared on the reverse scan at -0.86 v vs ag . unlike the case discussed in section [ o2 in bmpyrr ] , in this situation the highly reactive superoxide that is generated by the initial reduction encounters a proton source in the form of the cation of the solvent . evans _ _ et . al__@xcite detailed the mechanism of this process ( shown in scheme 1 ) and suggested that the observed back peak in the experimental voltammetry could be an oxygen - derived species created in follow - up reactions , most likely ho@xmath140 . given the documented chemical irreversibility of this system , only the forward wave was considered for further analysis . both the butler - volmer and marcus - hush programs were modified to account for a homogeneous chemical step in the mass transport equations , as detailed in the theory section . for the purposes of determining a starting point for @xmath134 and concentration , potential step chronoamperometry was performed on this system by stepping from a potential at which no current flows , 0.0 v , to -2.0 v for 2s ( see fig . [ chrono ] ) . this data was then analysed with respect to the shoup and szabo@xcite equation ( as detailed in the section 2.4 ) , to simultaneously determine the concentration and diffusion coefficient of oxygen ( see table [ p14666 parameters ] ) . optimised values for @xmath3 , @xmath5 , @xmath133 , @xmath134 , and @xmath136 were obtained to give the `` best fit '' using the modified butler - volmer program described above , which was quantified with the calculation of a mean scaled absolute deviation ( msad , as described in the section [ msad ] ) . these values were used across the scan rates with all values remaining constant , with the exception of @xmath136 , which was allowed to deviate slightly ( @xmath139 0.02 v ) to account for the use of a quasi - reference electrode . the optimised values and the mean msads across the scan rates are reported in table [ p14666 parameters ] . the values obtained for @xmath134 and @xmath133 are consistent with previous literature reports and are in agreement with reported diffusion coefficients and saturated concentrations of oxygen in a range of ionic liquids ( table [ lit values ] ) . the value for the transfer coefficient , @xmath5 , of 0.31 , suggests an early , reactant - like transition state . this is congruent with the result found in section [ o2 in bmpyrr ] , and a similar @xmath3 of @xmath141 cm s@xmath26 is reported , which is again in agreement with the literature values reported in table [ lit values ] . smh theory was used independently to generate the same set of parameters as generated in the bv theory with the transfer coefficient , @xmath5 , once again being replaced by a representation for reorganisation energy , @xmath8 . the experimental value for @xmath8 was determined to be 0.5 @xmath139 0.1 ev in this solvent . once again , the bv program provides parameters with a significantly greater level of certainty that the smh program . and , although the smh theory gives greater physical insight compared with the phenomenological bv theory , this greater uncertainty leaves bv as the preferred theory , in terms of attaining more reliable parameterisation . both systems under study , oxygen reduction in either a pyrrolidinium or phosphonium based il , display similarly low diffusion coefficients ( @xmath134 = 2.05 and 3.95 @xmath142 m@xmath127 s@xmath26 respectively ) and solubilities of oxygen ( @xmath133 = 10 and 7.3 mm ) . in addition to this , both the transfer coefficients ( @xmath5 = 0.29 @xmath139 0.01 , and 0.31 @xmath139 0.01 respectively ) and the reorganisation energies ( @xmath8 = 0.4 @xmath139 0.1 , and 0.5 @xmath139 0.1 respectively ) are similar . as outlined in the introduction , @xmath8 encompasses contributions from both inner - sphere reorganisation of the molecule itself , and outer - sphere solvent reorganisation about the electroactive species . the inner - sphere reorganisation is due to the change in bond length on going from o@xmath124 to o@xmath143 ( 0.12 )@xcite , with outer - sphere solvent reorganisation dominating in aqueous systems . literature reports suggest the overall reorganisation energy for the oxygen self - exchange reaction in water is 1.26 - 4.51 ev.@xcite the inner - sphere contribution was estimated to be 0.4 - 0.6 ev.@xcite given the close agreement with previously reported values for inner sphere reorganisation@xcite , the experimental values for @xmath8 in both of the systems considered in this work are consistent with a process dominated by an inner sphere contribution . previous reports have indicated that oxygen itself had weak interactions with an ionic liquid,@xcite and the result reported in the present work suggests that the ionic liquid does not undergo any significant reorganisation on moving from oxygen to superoxide . this may be related to the very small size of the o@xmath124/o@xmath143 species compared to the component ions of the rtil . the reported best fit values for the rate constant given by smh theory are similar to those established by bv theory , however there is less agreement with the experimental data than is seen when the bv program is used . this is reflected both in the larger mean msad values , and qualitatively observing the fit as seen in figs . [ bmpyrrntf2 ] and [ p14666ntf2 ] . further studies may involve consideration of asymmetric mh theory , which relaxes the requirement that the potential energy parabola are of equal curvature , and introduces a new parameter to describe the disparity in curvature between the reactant and product . this has been considered in a number of molecular systems,@xcite and has been determined to fit experimental data more accurately than smh . models were developed to simulate cyclic voltammetry at a microdisc electrode and a sphere on a surface electrode using both butler - volmer and symmetric marcus - hush kinetics . the microdisc model was used to investigate the possibility of kinetically limited steady state currents at a microdisc electrode , and it was found that this phenomenon is more likely to be observed at nano - sized sphere on a surface electrodes ( _ e.g. _ an impacting nanoparticle ) because of the smaller size . it was also found that the steady state equivalence ( subject to a mass transport correction ) between microdisc electrodes and spherical electrodes resting on a surface also hold when marcus - hush kinetics are employed . the reduction of oxygen at a microdisc electrode in two room temperature ionic liquids , [ bmpyrr][ntf@xmath0 and [ p@xmath1[ntf@xmath0 , has been studied with cyclic voltammetry at 298 k. bv theory and smh theory were used to create simulation programs , and the simulated voltammograms were compared with experimental data . bv theory , although empirical , provided voltammograms with lower msad values than those generated using smh theory . a reorganisation energy , from the smh program , was determined to be 0.4 - 0.5 ev for oxygen reduction in the two rtils studied . this low value likely corresponds to a dominant inner sphere reorganisation , and limited solvent reorganisation . further studies to investigate whether amh theory is better than smh in its ability to parameterise experimental data in ionic solvents will likely be valuable . for funding , eelt thanks the clarendon fund and st john s college , oxford , lx thanks honeywell analytics , and eob thanks epsrc and st john s college , oxford . l l l parameter & description & units + @xmath5 & transfer coefficient & unitless + + @xmath144 & concentration of species i & mol m@xmath145 + + @xmath146 & bulk solution concentration of species i & mol m@xmath145 + + @xmath147 & electrode surface concentration of species i & mol m@xmath145 + + @xmath148 & diffusion coefficient of species i & m@xmath127 s@xmath26 + + @xmath21 & applied potential & v + + @xmath149 & formal potential of a / b couple & v + + @xmath37 & faraday constant = 96485 & c mol@xmath26 + + @xmath71 & current & a + + @xmath77 & ( pseudo ) first order homogeneous rate constant & s@xmath26 + + @xmath3 & electrochemical rate constant & m s@xmath26 + + @xmath150 & reduction / oxidation rate constant & unitless + + @xmath8 & reorganisation energy & ev + + @xmath25 & scan rate & v s@xmath26 + + @xmath38 & gas constant = 8.314 & j k@xmath26 mol@xmath26 + + @xmath15 & radial coordinate & m + + @xmath151 & radius of electrode & m + + @xmath39 & temperature & k + + @xmath27 & time & s + + @xmath16 & @xmath16 coordinate & m + c c dimensionless parameter & definition + + @xmath152 & @xmath153 + + @xmath154 & @xmath155 + + @xmath69 & @xmath156 + + @xmath75 & @xmath157 + + @xmath102 & @xmath158 + + @xmath45 & @xmath159 + + @xmath38 & @xmath160 + + @xmath34 & @xmath161 + + @xmath80 & @xmath162 + + @xmath62 & @xmath163 + + c c c parameter & butler - volmer & marcus - hush + + @xmath134 / m@xmath127 s@xmath26 & @xmath164 & @xmath164 + + @xmath135 / m@xmath127 s@xmath26 & @xmath165 & @xmath166 + + @xmath133 / mm & 10 & 10 + + @xmath3 / cm s@xmath26 & 0.0023 & 0.0015 + + @xmath136 / v & @xmath167 & @xmath167 + + @xmath5 & 0.29 & - + + @xmath8 / ev & - & 0.4 + + msad ( average ) / % & 9.52 & 13.9 + c c c c c c ionic liquid & @xmath133 / mm & @xmath134 / m@xmath127 s@xmath26 & @xmath135 / m@xmath127 & @xmath3 / cm s@xmath26 & @xmath5 + + [ emim][bf@xmath168@xcite & @xmath169 & @xmath170 & - & @xmath171 & @xmath172 + + [ pmim][bf@xmath168@xcite & @xmath173 & @xmath174 & - & @xmath175 & @xmath176 + + [ bmim][bf@xmath168@xcite & @xmath177 & @xmath178 & - & @xmath179 & @xmath180 + + [ mopmpip][ntf@xmath0@xcite & @xmath181 & @xmath182 & - & - & @xmath183 + + [ hmpyrr][ntf@xmath0@xcite & @xmath184 & @xmath185 & - & - & @xmath186 + + [ emim][ntf@xmath0@xcite & @xmath187 & @xmath188 & @xmath189 & - & - + + [ n@xmath190[ntf@xmath0@xcite & @xmath191 & @xmath192 & @xmath193 & - & - + + [ c@xmath194dmim][ntf@xmath0@xcite & @xmath195 & @xmath196 & @xmath197 & @xmath171 & 0.4 + + [ n@xmath190[ntf@xmath0@xcite & @xmath198 & @xmath199 & @xmath200 & @xmath201 & @xmath176 + c c c parameter & butler - volmer & marcus - hush + + @xmath134 / m@xmath127 s@xmath26 & @xmath202 & @xmath202 + + @xmath133 / mm & 7.3 & 7.3 + + @xmath3 / cm s@xmath26 & 0.001 & 0.001 + + @xmath136 / v & @xmath203 & @xmath204 + + @xmath5 & 0.30 & - + + @xmath8 / ev & - & 0.5 + + msad ( average ) / % & 2.80 & 10.2 + butler , j. a. v. _ trans . faraday soc . _ * 1924 * , _ 19 _ , 729 733 erdey - grz , t. ; volmer , m. _ z. physik . chem . _ * 1930 * , _ 150 _ , 203213 marcus , n. , r. a. ; sutin _ biochim . * 1985 * , _ 811 _ , 265 322 hush , n. s. _ j. electroanal . chem . _ * 1999 * , _ 470 _ , 170 195 laborda , e. ; henstridge , m. c. ; compton , r. g. _ j. electroanal . chem . _ * 2012 * , _ 667 _ , 48 53 compton , r. g. ; banks , c. e. _ understanding voltammetry _ , world scientific : singapore , 2010 born , m. _ z. phys . _ * 1920 * , _ 1 _ , 45 48 hartnig , c. ; koper , m. t. m. _ j. electroanal . chem . _ * 2002 * , _ 532 _ , 165 170 henstridge . , m. c. ; wang , y. ; limon - petersen , j. g. ; laborda , e. ; compton , r. g. _ chem . let . _ * 2011 * , _ 517 _ , 29 35 wang , y. ; laborda , e. ; henstridge , m. c. ; martinez - ortiz , f. ; molina , a. ; compton , r. g. _ j. electroanal . * 2012 * , _ 668 _ , 7 12 suwatchara , d. ; rees , n. v. ; henstridge , m. c. ; laborda , e. ; compton , r. g. _ j. electroanal . * 2012 * , _ 665 _ , 38 44 suwatchara , d. ; henstridge , m. c. ; rees , n. v. ; compton , r. g. _ j. phys . c _ * 2011 * , _ 115 _ , 14876 14882 laborda , e. ; wang , y. ; henstridge , m. c. ; martnez - ortiz , f. ; molina , a. ; compton , r. g. _ chem . _ * 2011 * , _ 512 _ , 133 157 henstridge , m. c. ; laborda , e. ; rees , n. v. ; compton , r. g. _ electrochim . acta _ * 2012 * , _ 84 _ , 12 20 henstridge , m. c. ; laborda , e. ; y. , w. ; suwatchara , d. ; rees , n. v. ; molina , a. ; martnez - ortiz , f. ; compton , r. g. _ j. electroanal . chem . _ * 2012 * , _ 672 _ , 45 52 laborda , e. ; suwatchara , d. ; rees , n. v. ; henstridge , m. c. ; molina , a. ; compton , r. g. _ electrochim . acta . _ * 2013 * , _ 110 _ , 772 779 suwatchara , d. ; henstridge , m. c. ; rees , n. v. ; laborda , e. ; compton , r. g. _ j. electroanal . chem . _ * 2012 * , _ 677 _ , 120 126 suwatchara , d. ; rees , n. v. ; henstridge , m. c. ; laborda , e. ; compton , r. g. _ j. electroanal . chem . _ * 2012 * , _ 685 _ , 53 62 henstridge , m. c. ; laborda , e. ; compton , r. g. _ j. electroanal . chem . _ * 2012 * , _ 674 _ , 90 96 buzzeo , m. c. ; klymenko , o. v. ; wadhawan , j. d. ; hardacre , c. ; seddon , k. r. ; compton , r. g. _ j. phys . chem . * 2003 * , _ 107 _ , 8872 8878 katayama , y. ; sekiguchi , k. ; yamagata , m. ; miura , t. _ j. electrochem . soc . _ * 2005 * , _ 152 _ , e247 e250 zhang , l. ; li , h. ; zhang , j. _ power sources _ * 2014 * , _ 255 _ , 242 250 wadhawan , j. d. ; welford , p. j. ; mcpeak , h. b. ; hahn , c. e. w. ; compton , r. g. _ sens . actuators b _ * 2003 * , _ 88 _ , 40 52 alnashef , i. m. ; leonard , m. l. ; kittle , m. c. ; matthews , m. a. ; weidner , j. w. _ electrochem . solid - state . * 2001 * , _ 4 _ , d16 d18 hayyan , m. ; mjalli , f. s. ; hashim , m. a. ; alnashef , i. m. ; tan , x. m. _ j. electroanal . _ * 2011 * , _ 657 _ , 150 157 gotti , g. ; fajerwerg , k. ; evrard , d. ; gros , p. _ int . j. electrochem . _ * 2013 * , _ 8 _ , 12643 12657 eberson , l. _ electron transfer reactions in organic chemistry _ , 1st ed . ; springer - verlag : heidelberg , 1987 wasserscheid , p. ; keim , w. _ angew . _ * 2000 * , _ 39 _ , 3772 3789 hussey , c. l. _ pure . chem . _ * 1988 * , _ 60 _ , 1763 1772 huang , x. ; rogers , e. i. ; hardacre , c. ; compton , r. g. _ the journal of physical chemistry b _ * 2009 * , _ 113 _ , 8953 8959 evans , r. g. ; klymenko , o. v. ; saddoughi , s. a. ; hardacre , c. ; compton , r. g. _ j. phys . b _ * 2004 * , _ 108 _ , 7878 7886 zhang , d. ; okajima , t. ; matsumoto , f. ; ohsaka , t. _ j. electrochem . soc . _ * 2004 * , _ 151 _ , d31 d37 lynden - bell , r. m. _ electrochem . commun . _ * 2007 * , _ 9 _ , 1857 1861 lynden - bell , r. m. _ j. phys . chem . b _ * 2007 * , _ 111 _ , 10800 10806 khan , a. ; lu , x. ; aldous , l. ; zhao , c. _ j. phys . chem . c _ * 2013 * , _ 117 _ , 18334 18342 rollins , j. b. ; conboy , j. c. _ j. electrochem . soc . _ * 2009 * , _ 156 _ , b943 b954 rogers , e. i. ; silvester , d. s. ; aldous , l. ; hardacre , c. ; compton , r. g. _ j. phys . chem . b _ * 2008 * , _ 112 _ , 6551 6557 marcus , r. a. _ j. chem . phys . _ * 1956 * , _ 24 _ , 966978 marcus , r. a. _ j. chem . phys . _ * 1956 * , _ 24 _ , 979 989 marcus , r. a. _ j. chem . phys . _ * 1957 * , _ 26 _ , 867 871 marcus , r. a. _ j. chem . phys . _ * 1957 * , _ 26 _ , 872 877 marcus , r. a. _ annu . rev . phys . chem . _ * 1964 * , _ 15 _ , 155 196 hush , n. s. _ j. chem . phys . _ * 1958 * , _ 28 _ , 962 972 chidsey , c. e. d. _ science _ * 1991 * , _ 251 _ , 919 922 feldberg , s. w. _ anal . chem . _ * 2010 * , _ 82 _ , 5176 5183 henstridge , m. c. ; laborda , e. ; rees , n. v. ; compton , r. g. _ electrochimica acta _ * 2012 * , _ 84 _ , 12 20 gavaghan , d. j. _ j. electroanal . chem . _ * 1998 * , _ 456 _ , 1 12 gavaghan , d. j. _ j. electroanal . chem . _ * 1998 * , _ 456 _ , 13 23 gavaghan , d. j. _ j. electroanal . chem . _ * 1998 * , _ 456 _ , 25 35 crank , j. ; nicolson , e. _ proc . _ * 1947 * , _ 43 _ , 50 67 britz , d. _ digital simulation in electrochemistry _ ; springer - verlag : heidelberg , 2005 press , w. h. , teukolsky , s. a. , vetterling , w. t. , flannery , b. p. , eds . _ numerical recipes : the art of scientific computing _ ; cambridge university press , 2007 ward , k. r. ; lawrence , s. ; hartshorne , r. s. ; compton , r. g. _ j. electroanal . chem . _ * 2012 * , _ 14 _ , 7264 7275 molina , a. ; gonzalez , j. ; barnes , e. o. ; compton , r. g. _ j. phys . c. _ * 2014 * , _ 118 _ , 346 356 henstridge , m. c. ; ward , k. r. ; compton , r. g. _ j. electroanal . _ * 2014 * , _ 712 _ , 14 18 ktelhn , e. ; barnes , e. o. ; krause , k. j. ; wolfrum , b. ; compton , r. g. _ chem . lett . _ * 2014 * , _ 595 _ , 31 34 shoup , d. ; szabo , a. _ j. electroanal . chem . _ * 1982 * , _ 140 _ , 237 245 oldham , k. b. ; zoski , g. _ j. electroanal . chem . _ * 1988 * , _ 256 _ , 11 19 cardwell , t. j. ; mocak , j. ; santos , j. h. ; bond , a. m. _ analyst _ * 1996 * , _ 121 _ , 357 362 rogers , e. i. ; streeter , i. ; aldous , l. ; hardacre , c. ; compton , r. g. _ j. phys . chem . c _ * 2008 * , _ 112 _ , 10976 10981 okoturo , o. o. ; vandernoot , t. j. _ j. electroanal . chem . _ * 2004 * , _ 568 _ , 167 181 yuan , x .- z . ; alzate , v. ; xie , z. ; ivey , d. g. ; qu , w. _ j. electrochem . soc . _ * 2014 * , _ 161 _ , a451 a457 monaco , s. ; arangio , a. m. ; soavi , f. ; mastragostino , m. ; paillard , e. ; passerini , s. _ electrochim . acta _ * 2012 * , _ 83 _ , 94 104 gritzner , j. , g. ; kuta _ pure appl . chem . _ * 1984 * , _ 56 _ , 461 466 omahony , a. m. ; silvester , d. s. ; aldous , l. ; hardacre , c. ; compton , r. g. _ j. chem . data _ * 2008 * , _ 53 _ , 28842891 mcdowell , m. s. ; espenson , j. h. ; bakac , a. _ inorg . chem . _ * 1984 * , _ 23 _ , 2232 2236 anthony , j. l. ; maginn , e. j. ; brennecke , j. f. _ j. phys . chem . b _ * 2002 * , _ 106 _ , 7315 7320 henstridge , m. c. ; rees , n. v. ; compton , r. g. _ j. electroanal . chem . _ * 2012 * , _ 687 _ , 79 83 laborda , e. ; henstridge , m. c. ; batchelor - mcauley , c. ; compton , r. g. _ chem . rev . _ * 2013 * , _ 42 _ , 4894 4905 rogers , e. i. ; huang , x. j. ; dickinson , e. j. f. ; hardacre , c. ; compton , r. g. _ j. phys . c _ * 2009 * , _ 113 _ , 17811 17823

let be an oriented surface of genus with punctures where . we show that if then for every compact subset of the moduli space for there is a closed teichmller geodesic in which does not intersect .

math0511349

let @xmath0 be a _ nonexceptional _ oriented surface of finite type , i.e. @xmath0 is a closed surface of genus @xmath1 from which @xmath2 points , so - called _ punctures _ , have been deleted , and @xmath3 . the _ teichmller space _ @xmath7 of @xmath0 is a smooth manifold diffeomorphic to @xmath8 . it can be represented as the quotient of the space of all complete hyperbolic metrics on @xmath0 of finite volume under the action of the group of all diffeomorphisms of @xmath0 which are isotopic to the identity . the group of all isotopy classes of orientation preserving diffeomorphisms of @xmath0 is called the _ mapping class group _ @xmath9 of @xmath0 . it acts smoothly and properly discontinuously on @xmath7 preserving a complete finsler metric , the so - called _ teichmller metric_. the quotient orbifold @xmath10 , equipped with the projection of the teichmller metric , is a complete noncompact geodesic metric space . even though the teichmller metric is not non - positively curved in any reasonable sense , it shares many properties with a riemannian metric of non - positive curvature . for example , any two points in @xmath7 can be connected by a unique teichmller geodesic , and _ closed _ geodesics in the moduli space @xmath6 are in one - to - one correspondence with the conjugacy classes of the so - called _ pseudo - anosov _ elements of the mapping class group . however , unlike in the case of negatively curved riemannian manifolds of finite volume , closed geodesics may escape into the end of @xmath6 . namely , we show . * theorem :* _ if @xmath4 then for every compact subset @xmath5 of @xmath6 there is a closed teichmller geodesic which does not intersect @xmath5 . _ the organization of this note is as follows . in section 2 we summarize the properties of the train track complex @xmath11 which are needed for our purpose . in section 3 we use train tracks and splitting sequences to show theorem a. in this section we summarize some results and constructions from @xcite which will be used throughout the paper ( compare also @xcite ) . let @xmath0 be an oriented surface of genus @xmath1 with @xmath2 punctures and where @xmath3 . a _ train track _ on @xmath0 is an embedded 1-complex @xmath12 whose edges ( called _ branches _ ) are smooth arcs with well - defined tangent vectors at the endpoints . at any vertex ( called a _ switch _ ) the incident edges are mutually tangent . through each switch there is a path of class @xmath13 which is embedded in @xmath14 and contains the switch in its interior . in particular , the branches which are incident on a fixed switch are divided into `` incoming '' and `` outgoing '' branches according to their inward pointing tangent at the switch . each closed curve component of @xmath14 has a unique bivalent switch , and all other switches are at least trivalent . the complementary regions of the train track have negative euler characteristic , which means that they are different from discs with @xmath15 or @xmath16 cusps at the boundary and different from annuli and once - punctured discs with no cusps at the boundary . we always identify train tracks which are isotopic . a _ trainpath _ on a train track @xmath14 is a @xmath13-immersion @xmath17\to \tau\subset s$ ] which maps each interval @xmath18 $ ] @xmath19 onto a branch of @xmath14 . the integer @xmath20 is then called the _ length _ of @xmath21 . we sometimes identify a trainpath on @xmath0 with its image in @xmath14 . each complementary region of @xmath14 is bounded by a finite number of trainpaths which either are simple closed curves or terminate at the cusps of the region . a _ subtrack _ of a train track @xmath14 is a subset @xmath22 of @xmath14 which itself is a train track . thus every switch of @xmath22 is also a switch of @xmath14 , and every branch of @xmath22 is an embedded trainpath of @xmath14 . we write @xmath23 if @xmath22 is a subtrack of @xmath14 . a train track is called _ generic _ if all switches are at most trivalent . the train track @xmath14 is called _ transversely recurrent _ if every branch @xmath24 of @xmath14 is intersected by an embedded simple closed curve @xmath25 which intersects @xmath14 transversely and is such that @xmath26 does not contain an embedded _ bigon _ , i.e. a disc with two corners at the boundary . a _ transverse measure _ on a train track @xmath14 is a nonnegative weight function @xmath27 on the branches of @xmath14 satisfying the _ switch condition _ : for every switch @xmath28 of @xmath14 , the sum of the weights over all incoming branches at @xmath28 is required to coincide with the sum of the weights over all outgoing branches at @xmath28 . the train track is called _ recurrent _ if it admits a transverse measure which is positive on every branch . we call such a transverse measure @xmath27 _ positive _ , and we write @xmath29 . if @xmath27 is any transverse measure on a train track @xmath14 then the subset of @xmath14 consisting of all branches with positive @xmath27-weight is a recurrent subtrack of @xmath14 . a train track @xmath14 is called _ birecurrent _ if @xmath14 is recurrent and transversely recurrent . a _ geodesic lamination _ for a complete hyperbolic structure on @xmath0 of finite volume is a _ compact _ subset of @xmath0 which is foliated into simple geodesics . a geodesic lamination @xmath30 is called _ minimal _ if each of its half - leaves is dense in @xmath30 . thus a simple closed geodesic is a minimal geodesic lamination . a minimal geodesic lamination with more than one leaf has uncountably many leaves and is called _ minimal arational_. every geodesic lamination @xmath30 consists of a disjoint union of finitely many minimal components and a finite number of isolated leaves . each of the isolated leaves of @xmath30 either is an isolated closed geodesic and hence a minimal component , or it _ spirals _ about one or two minimal components @xcite . a geodesic lamination is _ finite _ if it contains only finitely many leaves , and this is the case if and only if each minimal component is a closed geodesic . a geodesic lamination is _ maximal _ if its complementary regions are all ideal triangles or once punctured monogons . a geodesic lamination @xmath30 is called _ complete _ if @xmath30 is maximal and can be approximated in the _ hausdorff topology _ by simple closed geodesics . the space @xmath31 of all complete geodesic laminations equipped with the hausdorff topology is a compact metrizable space . every geodesic lamination @xmath30 which is a disjoint union of finitely many minimal components is a _ sublamination _ of a complete geodesic lamination , i.e. there is a complete geodesic lamination which contains @xmath30 as a closed subset @xcite . a train track or a geodesic lamination @xmath22 is _ carried _ by a transversely recurrent train track @xmath14 if there is a map @xmath32 of class @xmath13 which is isotopic to the identity and maps @xmath22 into @xmath14 in such a way that the restriction of the differential of @xmath33 to the tangent space of @xmath22 vanishes nowhere . note that this makes sense since a train track has a tangent line everywhere . we call the restriction of @xmath33 to @xmath22 a _ carrying map _ for @xmath22 . write @xmath34 if the train track or the geodesic lamination @xmath22 is carried by the train track @xmath14 . if @xmath22 is a train track which is carried by @xmath14 , then every geodesic lamination @xmath30 which is carried by @xmath22 is also carried by @xmath14 . a train track @xmath14 is called _ complete _ if it is generic and transversely recurrent and if it carries a complete geodesic lamination . every complete train track is recurrent . the set of all complete geodesic laminations which are carried by a complete train track @xmath14 is open and closed in @xmath31 . in particular , the space @xmath31 is totally disconnected @xcite . a half - branch @xmath35 in a generic train track @xmath14 incident on a switch @xmath36 of @xmath14 is called _ large _ if every trainpath containing @xmath36 in its interior passes through @xmath35 . a half - branch which is not large is called _ small_. a branch @xmath24 in a generic train track @xmath14 is called _ large _ if each of its two half - branches is large ; in this case @xmath24 is necessarily incident on two distinct switches , and it is large at both of them . a branch is called _ small _ if each of its two half - branches is small . a branch is called _ mixed _ if one of its half - branches is large and the other half - branch is small ( for all this , see @xcite p.118 ) . there are two simple ways to modify a complete train track @xmath14 to another complete train track . first , we can _ shift _ @xmath14 along a mixed branch to a train track @xmath37 as shown in figure a below . if @xmath14 is complete then the same is true for @xmath37 . moreover , a train track or a lamination is carried by @xmath14 if and only if it is carried by @xmath37 ( see @xcite p.119 ) . in particular , the shift @xmath37 of @xmath14 is carried by @xmath14 . note that there is a natural bijection of the set of branches of @xmath14 onto the set of branches of @xmath37 . second , if @xmath38 is a large branch of @xmath14 then we can perform a right or left _ split _ of @xmath14 at @xmath38 as shown in figure b. note that a right split at @xmath38 is uniquely determined by the orientation of @xmath0 and does not depend on the orientation of @xmath38 . using the labels in the figure , in the case of a right split we call the branches @xmath39 and @xmath40 _ winners _ of the split , and the branches @xmath41 are _ losers _ of the split . if we perform a left split , then the branches @xmath41 are winners of the split , and the branches @xmath42 are losers of the split . the split @xmath37 of a train track @xmath14 is carried by @xmath14 , and there is a natural choice of a carrying map which maps the switches of @xmath37 to the switches of @xmath14 . the image of a branch of @xmath37 is then a trainpath on @xmath14 whose length either equals one or two . we call this carrying map the _ canonical carrying map_. it induces a natural bijection of the set of branches of @xmath14 onto the set of branches of @xmath37 which maps the branch @xmath38 to the _ diagonal _ @xmath43 of the split . the split of a maximal transversely recurrent generic train track is maximal , transversely recurrent and generic . if @xmath14 is complete and if @xmath44 is carried by @xmath14 , then there is a unique choice of a right or left split of @xmath14 at @xmath38 with the property that the split track @xmath37 carries @xmath30 . we call such a split a _ @xmath30-split_. the train track @xmath37 is complete . in particular , a complete train track @xmath14 can always be split at any large branch @xmath38 to a complete train track @xmath37 ; however there may be a choice of a right or left split at @xmath38 such that the resulting train track is not recurrent any more ( compare p.120 in @xcite ) . the reverse of a split is called a _ collapse_. denote by @xmath11 the directed graph whose vertices are the isotopy classes of complete train tracks on @xmath0 and whose edges are determined as follows . the train track @xmath45 is connected to the train track @xmath37 by a directed edge if and only @xmath37 can be obtained from @xmath14 by a single split . the graph @xmath11 is connected @xcite . as a consequence , if we identify each edge in @xmath11 with the unit interval @xmath46 $ ] then this provides @xmath11 with the structure of a connected locally finite metric graph . thus @xmath11 is a locally compact complete geodesic metric space . in the sequel we always assume that @xmath11 is equipped with this metric without further comment . the mapping class group @xmath9 of @xmath0 acts properly and cocompactly on @xmath11 as a group of isometries . in particular , @xmath11 is @xmath9-equivariantly quasi - isometric to @xmath9 equipped with any word metric @xcite . in this section we use train tracks and splitting sequences to investigate closed _ teichmller geodesics _ in _ moduli space _ @xmath6 of our nonexceptional surface @xmath0 of finite type and show theorem a from the introduction . such closed geodesics are the projections of those teichmller geodesics in the teichmller space @xmath7 which are invariant under a pseudo - anosov element of @xmath9 . recall that since the euler characterstic of @xmath0 is negative by assumption , @xmath7 can be identified with the space of all marked complete hyperbolic metrics on @xmath0 of finite volume ( with the usual identification of the elements contained in an orbit for the action of the group of diffeomorphisms of @xmath0 isotopic to the identity ) . a _ measured geodesic lamination _ is a geodesic lamination @xmath30 with a translation invariant transverse measure @xmath47 such that the @xmath47-weight of every compact arc in @xmath0 with endpoints in @xmath48 which intersects @xmath30 nontrivially and transversely is positive . we say that @xmath30 is the _ support _ of the measured geodesic lamination . the space @xmath49 of measured geodesic laminations equipped with the weak@xmath50-topology admits a natural continuous action of the multiplicative group @xmath51 . the quotient under this action is the space @xmath52 of _ projective measured geodesic laminations _ which is homeomorphic to the sphere @xmath53 . every simple closed geodesic @xmath40 on @xmath0 defines a measured geodesic lamination . the geometric intersection number between simple closed curves on @xmath0 extends to a continuous bilinear form @xmath54 on @xmath49 , the _ intersection form_. we say that a pair @xmath55 of measured geodesic laminations _ jointly fills up _ @xmath0 if for every measured geodesic lamination @xmath56 we have @xmath57 . this is equivalent to saying that every complete simple ( possibly infinite ) geodesic on @xmath0 intersects either the support of @xmath30 or the support of @xmath47 transversely . the unit cotangent bundle of the teichmller space with respect to the teichmller metric can naturally be identified with the space @xmath58 of unit area marked holomorphic _ quadratic differentials _ on our surface @xmath0 . such a quadratic differential @xmath59 is determined by a pair @xmath60 of measured geodesic laminations on @xmath0 which jointly fill up @xmath0 and such that @xmath61 by our area normalization . the _ horizontal _ measured lamination @xmath62 for @xmath59 corresponds to the equivalence class of the horizontal measured foliation of @xmath59 ( compare @xcite and @xcite for a discussion of the relation between measured geodesic laminations and equivalence classes of measured foliations on @xmath0 ) . the quadratic differential @xmath59 determines a singular euclidean metric on @xmath0 of unit area . for every simple closed curve @xmath40 on @xmath0 , the _ @xmath59-length _ of @xmath40 is defined to be the infimum of the lengths with respect to this metric of any curve which is freely homotopic to @xmath40 . the bundle @xmath58 admits a smooth action of the group @xmath63 . the restriction of this action to the one - parameter subgroup of diagonal matrices of @xmath63 is the _ teichmller geodesic flow _ @xmath64 on @xmath58 . if a quadratic differential @xmath59 is given by the pair @xmath60 of transverse measured geodesic laminations with @xmath61 then @xmath65 is given by the pair @xmath66 . the action of @xmath63 commutes with the natural action of the mapping class group @xmath9 on @xmath58 and hence the flow @xmath64 projects to a flow on the quotient @xmath67 , again denoted by @xmath64 . a flow line of this flow projects to a teichmller geodesic in moduli space @xmath68 . for every complete train track @xmath14 , the convex cone @xmath69 of all transverse measures on @xmath14 can naturally be identified with the set of all measured geodesic laminations whose support is carried by @xmath14 . a _ tangential measure _ on a complete train track @xmath14 assigns to a branch @xmath24 of @xmath14 a weight @xmath70 such that for every complementary triangle @xmath71 of @xmath14 with sides @xmath72 we have @xmath73 ( here indices are taken modulo 3 ) . let @xmath74 be the convex cone of all tangential measures on @xmath14 . by the results from section 3.4 of @xcite , there is a one - to - one correspondence between tangential measures on @xmath14 and measured geodesic laminations which _ hit @xmath14 efficiently _ , i.e. measured geodesic laminations whose support @xmath30 intersects @xmath14 transversely up to isotopy and is such that @xmath75 does not contain any embedded bigon . in the sequel we often identify a measure @xmath76 ( or @xmath77 ) with the measured geodesic lamination it defines . with this identification , the bilinear pairing @xmath78 defined by @xmath79 is just the restriction of the intersection form on @xmath80 ( see section 3.4 of @xcite ) . denote by @xmath81 the set of all pairs @xmath82 where @xmath83 is a transverse measure on @xmath14 of total weight one and where @xmath77 is a tangential measure with @xmath84 and such that @xmath85 jointly fill up @xmath0 . every pair @xmath86 defines a quadratic differential @xmath87 of area one . if @xmath88 is any splitting sequence , if @xmath89 and if @xmath30 is carried by @xmath90 for some @xmath91 then for every @xmath92 there is a number @xmath93 such that @xmath94 . denote by @xmath95,[\nu]\in { { \mathcal{p}}{\mathcal{m}}{\mathcal{l}}}$ ] the projective classes of @xmath85 and for @xmath96 define @xmath97,[\nu])(\tau(i))=a(i+1)/a(i)$ ] . we call @xmath21 a _ roof function _ for the pair @xmath98 . we have . let @xmath99 be obtained from @xmath14 by a single split at a large branch @xmath38 and let @xmath27 be a transverse measure on @xmath100 of total weight one . via the natural carrying map @xmath101 , the measure @xmath27 defines a transverse measure @xmath102 on @xmath14 . by the definition of a roof function we have to show that the total weight of @xmath102 is bounded from above by universal constant . however , this is immediate from the fact that if @xmath42 are the losing branches of the split connecting @xmath14 to @xmath100 and if @xmath43 is the diagonal of the split then @xmath103 and @xmath104 for every branch @xmath105 of @xmath14 and its corresponding branch of @xmath100 . this shows the lemma . for @xmath106 let @xmath107 be the @xmath108-thick part of teichmller space consisting of all marked hyperbolic metrics whose _ systole _ ( i.e. the length of the shortest closed geodesic ) is at least @xmath108 . for sufficiently small @xmath108 the set @xmath107 is an @xmath9-invariant contractible subset of @xmath7 on which the mapping class group acts cocompactly . let @xmath109 be the canonical projection and define @xmath110 . the sets @xmath111 are invariant under the action of @xmath9 . their projections @xmath112 to @xmath113 are compact and satisfy @xmath114 for @xmath115 and @xmath116 . call a finite splitting sequence @xmath117 _ tight _ if the natural carrying map @xmath118 maps every branch @xmath24 of @xmath119 _ onto _ @xmath120 . note that every splitting sequence which contains a tight subsequence is tight itself . we have [ lemma52 ] let @xmath121 by any finite splitting sequence such that for some @xmath122 , both sequences @xmath123 and @xmath124 are tight . then there is a number @xmath125 depending on our sequence such that for all @xmath126 with the additional property that @xmath30 is carried by @xmath119 the following holds . let @xmath117 be a finite splitting sequence such that for some @xmath128 the sequences @xmath129 and @xmath124 are tight . extending the distance function on the _ curve graph _ of @xmath0 , define a distance @xmath130 on the space @xmath131 of minimal geodesic laminations which do _ not _ fill up @xmath0 ( i.e. which admit a complementary component which is neither a disc nor a once punctured disc ) by requiring that the distance between @xmath132 equals one if and only if @xmath133 can be realized disjointly . by the choice of @xmath131 , the distance between any two elements @xmath134 is finite . let @xmath135 be a minimal geodesic lamination which is carried by @xmath119 and let @xmath136 be such that @xmath137 ; we claim that @xmath138 is carried by @xmath139 . namely , since our splitting sequence @xmath124 is tight by assumption , the lamination @xmath30 fills @xmath139 , i.e. any transverse measure for @xmath30 defines a _ positive _ transverse measure on @xmath139 . assume first that @xmath138 is a simple closed curve on @xmath0 with @xmath140 . since @xmath30 is minimal there is a sequence @xmath141 of simple closed curves which converge to @xmath30 in the hausdorff topology @xcite . now @xmath30 is disjoint from @xmath138 and therefore we can choose the sequence @xmath141 in such a way that each of the curves @xmath142 is disjoint from @xmath138 . for sufficiently large @xmath54 , the curve @xmath142 is carried by @xmath119 ( see lemma 2.4 of @xcite ) and fills @xmath139 , i.e. a carrying map @xmath143 is surjective . by lemma 4.4 of @xcite , since @xmath144 are disjoint the curve @xmath138 is carried by @xmath139 . now the space of geodesic laminations carried by @xmath139 is closed with respect to the hausdorff topology and therefore @xmath139 carries every geodesic lamination @xmath136 with @xmath140 . let again @xmath30 be a minimal geodesic lamination which is carried by @xmath119 and let @xmath47 be a minimal geodesic lamination which hits @xmath120 efficiently . we claim that @xmath30 and @xmath47 jointly fill up @xmath0 . namely , otherwise there is a simple closed curve @xmath40 which is disjoint from both @xmath85 . but this means that the distance between @xmath30 and @xmath40 is at most one and hence @xmath40 is carried by @xmath139 by our above consideration . the same argument , applied to @xmath40 and @xmath47 and the splitting sequence @xmath129 , shows that @xmath47 is carried by @xmath120 which contradicts our assumption that @xmath47 hits @xmath120 efficiently . let @xmath27 be any transverse measure for @xmath119 and let @xmath145 is a branch of @xmath146 . let @xmath102 be the transverse measure on @xmath120 induced from @xmath27 by the carrying map @xmath118 . since the splitting sequence @xmath147 is tight , the @xmath102-weight of a branch @xmath24 of @xmath120 is not smaller than @xmath148 , and it is not bigger than @xmath149 where @xmath150 only depends on our splitting sequence @xmath147 but not on @xmath27 . as a consequence , there is a universal constant @xmath91 not depending on @xmath27 with the property that if the measure @xmath27 on @xmath119 is normalized in such a way that the total weight of @xmath102 on @xmath120 equals one , then the minimum of the weights that the measure @xmath102 disposes on the branches of @xmath120 is not smaller than @xmath28 . this shows the second statement in our lemma . now for any tangential measure @xmath47 on @xmath120 we have @xmath151 ( where as before we identify the measures @xmath152 with the measured geodesic laminations they define ) and therefore if @xmath153 then the _ maximal _ weight disposed on a branch of @xmath120 by the tangential measure @xmath47 is not bigger than @xmath154 . since @xmath155 and @xmath156 are closed subsets of @xmath49 and the intersection form is continuous , this implies that the set @xmath157 of pairs @xmath158 such that @xmath27 defines a transverse measure of total weight one on @xmath120 and such that @xmath153 is a compact subset of @xmath159 . by our above consideration , every pair @xmath160 jointly fills up @xmath0 and hence @xmath157 is a compact subset of @xmath161 . since the assignment which associates to a quadratic differential its horizontal and vertical measured foliation is injective , continuous and open with respect to the smooth topology on @xmath58 and the weak topology on @xmath162 the set @xmath163 is compact . therefore there is some @xmath106 such that @xmath164 which shows the lemma . for the proof of the theorem from the introduction , we use biinfinite splitting sequences @xmath165 to construct closed orbits of the teichmller flow . for this recall that there are only finitely many orbits of complete train tracks under the action of @xmath9 . call a biinfinite splitting sequence @xmath88 _ periodic _ if there is a number @xmath166 and some @xmath167 such that for every @xmath168 we have @xmath169 . we then call the mapping class @xmath170 a _ period map _ of the sequence with _ period _ @xmath171 . a periodic splitting sequence can easily be constructed by choosing any infinite splitting sequence @xmath172 and some @xmath173 such that @xmath174 for some @xmath167 ; then @xmath175 @xmath176 defines a periodic splitting sequence with period map @xmath170 . note that a period map of any periodic splitting sequence is an element of @xmath9 of infinite order . we say that a measured geodesic lamination @xmath30 _ fills up _ @xmath0 if for every simple closed curve @xmath40 on @xmath0 we have @xmath177 . the support of such a measured geodesic lamination is necessarily connected . if @xmath178 is such that its projectivization @xmath95\in { { \mathcal{p}}{\mathcal{m}}{\mathcal{l}}}$ ] is a fixed point of a pseudo - anosov mapping class then @xmath30 fills up @xmath0 . recall that for every complete train track @xmath14 we view @xmath69 as the space of all measured geodesic laminations which are carried by @xmath14 . we have let @xmath181 be a biinfinite periodic splitting sequence with period map @xmath167 . then there is a number @xmath182 such that @xmath183 and @xmath184 . assume first that @xmath170 is pseudo - anosov . since @xmath170 acts on the thurston boundary @xmath52 of teichmller space with north - south - dynamics and since the projectivizations @xmath185 of the cones @xmath186 are closed disjoint subsets of @xmath52 with non - empty interior , the attracting fixed point @xmath62 for the action of @xmath170 on @xmath52 is contained in @xmath187 . but @xmath62 is the projectivization of a measured geodesic lamination @xmath30 which fills up @xmath0 and hence @xmath188 is a lamination as required in the lemma . now assume that there is some @xmath189 which fills up @xmath0 . the set @xmath190 is a closed non - empty @xmath170-invariant subset of @xmath49 . by theorem 8.5.1 of @xcite , it can be represented as a join @xmath191 where @xmath192 is the space of measured geodesic laminations whose support is contained in a fixed geodesic lamination @xmath193 on @xmath0 and where @xmath194 is a space of measured geodesic laminations whose support is carried by a train track on @xmath0 contained in the complement of @xmath193 . a measured geodesic lamination whose support contains more than one connected component does not fill up @xmath0 . but @xmath195 fills up @xmath0 by assumption and therefore the set @xmath194 is empty and @xmath190 is the space of measured geodesic laminations supported in the single geodesic lamination @xmath193 . moreover , @xmath193 is minimal and connected , with complementary components which are topological discs or once punctured topological discs . every mapping class which preserves the set @xmath190 also preserves the geodesic lamination @xmath193 and hence it is pseudo - anosov by thurston s classification of elements of @xmath9 . but @xmath190 is invariant under the period map @xmath170 and hence @xmath170 is pseudo - anosov as claimed . let @xmath196 be the number of branches of a complete train track on @xmath0 . let @xmath197 and number the branches of @xmath14 with numbers @xmath198 . if @xmath37 can be obtained from @xmath14 by a single split at a large branch @xmath38 , then the numbering of the branches of @xmath14 naturally induces a numbering of the branches of @xmath37 ( compare the discussion in section 5 of @xcite ) . in other words , every splitting sequence @xmath88 together with a numbering of the branches of @xmath120 determines uniquely a _ numbered splitting sequence _ , i.e. a splitting sequence together with a consistent numbering of the branches of the train tracks in the sequence which is determined by @xmath88 up to a permutation of the numbering of the branches of @xmath120 . let @xmath199 be the standard basis of @xmath200 . a numbering of the branches of @xmath120 then defines an embedding of @xmath201 onto a closed convex cone in @xmath200 determined by the switch conditions by associating to a measure @xmath202 the vector @xmath203 where we identify a branch of @xmath120 with its number . if @xmath37 is obtained from @xmath14 by a split at the large branch @xmath38 with number @xmath204 and if @xmath205 are the losing branches of the split connecting @xmath14 to @xmath37 then the transformation @xmath206 induced by the canonical carrying map @xmath207 is just the restriction to @xmath208 of the unique linear map @xmath209 which satisfies @xmath210 and @xmath211 for @xmath212 . a change of the numbering of the branches of @xmath120 results in replacing @xmath190 by its conjugate under a _ permutation map _ , i.e. a linear isometry of @xmath213 corresponding to a permutation of the standard basis vectors . similarly , for every finite splitting sequence @xmath147 the canonical map @xmath214 corresponds to a @xmath215-matrix @xmath216 with non - negative entries which is uniquely determined by the sequence up to conjugation with a permutation matrix . by the perron frobenius theorem , a @xmath215-matrix @xmath190 with positive entries admits up to a multiple a unique eigenvector with positive entries . the corresponding eigenvalue @xmath217 is positive , and its absolute value is bigger than the absolute value of any other eigenvalue of @xmath190 . moreover , the generalized eigenspace for @xmath217 is one - dimensional . we call an eigenvector with positive entries for the eigenvalue @xmath217 a _ perron frobenius eigenvector_. we have . [ lemma54 ] let @xmath88 be a periodic splitting sequence with period map @xmath170 . assume that @xmath218 is such that @xmath219 and that @xmath220 is tight ; then @xmath170 is pseudo - anosov and there is some @xmath221 such that its attracting fixed point is the projectivization of a perron frobenius eigenvector of the matrix @xmath222 . let @xmath220 be a tight splitting sequence and assume that there is some @xmath167 such that @xmath219 . we obtain a biinfinite splitting sequence @xmath88 by defining @xmath223 for @xmath224 and @xmath225 . by our above discussion , with respect to some numbering of the branches of @xmath120 the self - map @xmath226 which is the composition of the map @xmath227 induced by @xmath170 and the transformation @xmath228 induced by the canonical carrying map @xmath229 can be represented by the restriction of a linear map @xmath230 given with respect to the standard basis by a matrix @xmath231 . since @xmath220 is tight by assumption , the entries of the matrix @xmath231 are positive and therefore by the perron frobenius theorem , our map uniformly contracts the projectivization of the closed cone of vectors with nonnegative entries into its interior . as a consequence , the intersection @xmath232 consists of a single ray spanned by the positive measure on @xmath120 which corresponds to a perron frobenius eigenvector of @xmath231 ( compare with the beautiful argument in @xcite ) . this ray defines a projective measured geodesic lamination @xmath233 which is _ uniquely ergodic _ and fills up @xmath0 . similarly , the intersection @xmath234 consists of a single ray which defines a projective measured geodesic lamination @xmath235 which fills up @xmath0 . by lemma 5.3 , the mapping class @xmath170 is pseudo - anosov and its attracting fixed point equals @xmath62 , its repelling fixed point equals @xmath236 . recall that a numbering of the branches of @xmath120 induces a numbering of the branches of @xmath237 . there is a second numbering of the branches of @xmath237 induced from the numbering of the branches of @xmath120 via the map @xmath238 . these two numberings differ by a permutation @xmath22 of @xmath239 . let @xmath221 be such that @xmath240 ; then @xmath241 as numbered train tracks . it is then immediate from our above discussion that the attracting fixed point of @xmath170 corresponds to the projectivization of a perron frobenius eigenvector for the matrix @xmath222 which shows the lemma . in a riemannian manifold @xmath242 of bounded negative curvature and finite volume with fundamental group @xmath243 , every closed geodesic intersects a fixed compact subset of @xmath242 . moreover , we can construct an infinite sequence @xmath244 of pairwise distinct closed geodesics which intersect every compact subset @xmath5 of @xmath242 in arcs of uniformly bounded length as follows . choose a hyperbolic element @xmath170 in @xmath243 and a parabolic element @xmath245 ; then up to replacing @xmath245 by a conjugate , the elements @xmath246 are hyperbolic for all @xmath218 and they define an infinite family of pairwise distinct closed geodesics with the required properties . in the next lemma , we point out that a similar construction for a pseudo - anosov element of @xmath9 and a dehn twist about a suitably chosen simple closed curve ( viewed as a parabolic element in @xmath9 ) yields sequences of closed geodesics in @xmath6 with similar properties . 1 . there is a fixed compact subset @xmath247 of @xmath6 which is intersected by @xmath248 for every @xmath54 . 2 . the geodesics @xmath248 intersect every compact subset @xmath249 in arcs of uniformly bounded length . recall from section 3 the definition of a train track @xmath22 in special standard form for a framing @xmath33 of our surface @xmath0 with pants decomposition @xmath250 . we first claim that there is a train track @xmath14 which is shift equivalent to a such a train track in special standard form and there is a tight splitting sequence @xmath220 issuing from @xmath251 such that @xmath252 for some @xmath167 . for this let @xmath100 by any train track which is shift equivalent to a train track in special standard form for a framing @xmath33 of @xmath0 with pants decomposition @xmath250 . let @xmath253 be the unique special geodesic lamination for @xmath250 which is carried by @xmath100 and let @xmath254 be any tight splitting sequence issuing from @xmath255 . since every orbit for the action of the mapping class group on the space @xmath31 of complete geodesic laminations on @xmath0 is dense and since the set of all complete geodesic laminations on @xmath0 which are carried by @xmath256 is open in @xmath31 , there is an element @xmath257 such that the special geodesic lamination @xmath258 for the pants decomposition @xmath259 is carried by @xmath256 . by the considerations in section 4 , the train track @xmath256 is splittable to a train track @xmath260 which is shift equivalent to the train track @xmath261 for some @xmath262 . note that every splitting sequence connecting @xmath100 to @xmath260 is tight and therefore if @xmath263 then such a splitting sequence has the desired properties . otherwise we apply this construction to @xmath260 and find a tight splitting sequence connecting @xmath260 to a train track @xmath264 which is shift equivalent to @xmath265 for some @xmath266 . since there are only finitely many train track in the shift equivalence class of @xmath100 , after finitely many steps we find some @xmath267 such that @xmath268 and hence @xmath269 is splittable to @xmath270 with a tight splitting sequence . now let @xmath220 be a tight splitting sequence as above with @xmath271 for some @xmath167 and such that @xmath120 is shift equivalent to a train track in special standard form for a framing @xmath33 with pants decomposition @xmath250 . by lemma 5.4 , @xmath170 is pseudo - anosov . for @xmath272 define @xmath273 . by the definition of a train track in special standard form , every component @xmath217 of @xmath250 defines an embedded simple closed trainpath on @xmath120 , and there is a number @xmath274 ( depending on @xmath217 ) and a dehn twist @xmath245 about @xmath217 with the property that @xmath275 is splittable to @xmath120 with a splitting sequence of length @xmath28 . each split of this sequence is a split at a large branch contained in @xmath217 . in other words , for every @xmath276 the train track @xmath277 is splittable to @xmath278 with a tight splitting sequence @xmath279 of length @xmath280 . by lemma 5.4 , the mapping class @xmath281 is pseudo - anosov . its axis is the teichmller geodesic which is defined by a quadratic differential whose horizontal foliation is contained in the ray @xmath282 and whose vertical foliation is contained in the ray @xmath283 . for @xmath284 let @xmath285 be such that the quadratic differential @xmath286 is a cotangent vector of the axis of @xmath287 . for @xmath288 $ ] let @xmath289 be such that @xmath290 , i.e. that the total weight of the transverse measure on @xmath291 which corresponds to the measured geodesic lamination @xmath292 equals one . then the function @xmath293 is increasing , and @xmath294 is the length of the closed teichmller geodesic @xmath295 in moduli space which is the projection of the axis of @xmath287 . by lemma [ lemma51 ] , the numbers @xmath296 are bounded independent of @xmath297 , and by lemma [ lemma52 ] there is a number @xmath298 such that @xmath299 for every @xmath276 . the projection @xmath300 of @xmath301 to @xmath113 is compact and therefore the geodesics @xmath295 have property 1 ) stated in the lemma . moreover , since @xmath220 is tight by assumption , by lemma 5.2 there is a number @xmath302 such that for _ every _ @xmath276 the @xmath303-weight of _ every _ branch @xmath24 of @xmath120 is at least @xmath304 ; then the maximal weight that the tangential measure @xmath305 disposes on a branch of @xmath120 is at most @xmath306 . to show that the geodesics @xmath295 also satisfy property 2 ) stated in the lemma , we show that @xmath307 @xmath308 and that moreover for every @xmath106 there is a number @xmath309 which is independent of @xmath297 and such that for all @xmath310 $ ] the @xmath311-length of the curve @xmath217 is not bigger than @xmath108 . to show that such a constant exists , denote by @xmath312 the branches of @xmath120 which are incident on a switch in the embedded trainpath @xmath217 but which are not contained in @xmath217 . note that @xmath313 , i.e. there are at least two such branches . by the considerations in the proof of lemma 2.5 of @xcite we have @xmath314 for every @xmath202 . in particular , the intersection number @xmath315 is bounded from above independent of @xmath297 . choose a numbering of the branches of @xmath120 so that @xmath316 is the branch with number @xmath54 and such that the branches with numbers @xmath317 are precisely the branches contained in the embedded trainpath @xmath217 . then the linear self - map of @xmath200 which defines the natural map @xmath318 with respect to this numbering maps for each @xmath319 the @xmath54-th standard basis vector @xmath320 of @xmath200 to @xmath321 , and for @xmath322 it maps @xmath320 to itself . for @xmath302 as above , the sum of the @xmath303-weights of the branches @xmath316 @xmath323 is bounded from below by @xmath324 and hence for every @xmath325 we have @xmath326 independent of @xmath297 . in particular , @xmath327 tends to zero as @xmath328 . however , @xmath329 for all @xmath150 and therefore for every @xmath106 there is a number @xmath330 not depending on @xmath297 such that for @xmath331 $ ] the intersection with @xmath217 of the horizontal measured geodesic lamination of @xmath332 does not exceed @xmath333 . we observed above that the maximal weight that the tangential measure @xmath305 disposes on a branch of @xmath120 is bounded from above by @xmath306 . since the natural transformation @xmath334 can be represented by a fixed linear map ( with the interpretation discussed above ) and the functions @xmath335 are non - decreasing , there is then a number @xmath336 with the property that for _ every _ @xmath276 the maximal weight that the measure @xmath337 disposes on a branch of @xmath338 is bounded from above by @xmath339 . however , the tangential measure @xmath340 on @xmath341 is the image of the tangential measure @xmath337 on @xmath338 under the transformation @xmath342 which identifies @xmath338 with @xmath341 and therefore the maximal weight that @xmath340 disposes on a branch of @xmath341 is bounded from above by @xmath339 . since @xmath217 is an embedded trainpath in @xmath341 this just means that the intersection with @xmath217 of the vertical geodesic lamination of @xmath343 is bounded from above by a universal constant not depending on @xmath297 . in other words , there is a constant @xmath344 not depending on @xmath297 such that for every @xmath345 $ ] , the intersection with @xmath217 of the vertical measured geodesic lamination of @xmath311 does not exceed @xmath333 . the @xmath311-length of @xmath217 is bounded from above by twice the sum of the intersection of @xmath217 with the horizontal and the vertical geodesic lamination of @xmath311 ( compare e.g. @xcite ) and hence for every sufficiently large @xmath297 and for every @xmath346 $ ] the @xmath311-length of @xmath217 does not exceed @xmath108 . now for every @xmath347 there is a number @xmath348 such that the set @xmath349 is contained in the set of all quadratic differentials @xmath59 with the property that the @xmath59-length of _ every _ essential simple closed curve on @xmath0 is at least @xmath350 . this then yields that the geodesics @xmath295 intersect a fixed compact subset @xmath5 of @xmath6 in arcs of uniformly bounded length , independent of @xmath297 . the lemma follows . assume that @xmath4 ( which excludes a sphere with at most 6 punctures , a torus with at most 3 punctures and a closed surface of genus 2 ) and let @xmath250 be a pants decomposition of @xmath0 . we require that if the genus of @xmath0 is at least 2 then @xmath250 contains two non - separating simple closed curves @xmath352 such that the surface @xmath353 obtained by cutting @xmath0 open along @xmath352 is connected and that moreover the bordered surface @xmath354 consists of @xmath355 pairs of pants and 2 forth punctured spheres containing @xmath352 as essential curves in their interior . if the genus of @xmath0 is at most one then @xmath0 contains at least 4 punctures and we choose @xmath250 in such a way that it contains two separating simple closed curves @xmath352 with the property that the surface obtained by cutting @xmath0 open along @xmath352 consists of three connected components , where two of these components are pairs of pants . in other words , @xmath356 contains a unique connected component @xmath353 of euler characteristic at most @xmath357 , and if we replace the boundary circles of this component by cusps then the resulting surface is nonexceptional . let @xmath14 be a train track in standard form for a framing with pants decomposition @xmath250 and with only twist connectors . then the curves @xmath352 are carried by @xmath14 and define embedded simple closed trainpaths on @xmath14 . we require that a branch of @xmath14 which is contained in the closure of @xmath353 , which is incident on a switch contained in @xmath358 and which is _ not _ contained in @xmath359 is a small branch . by our choice of @xmath250 and the discussion on p.147 - 48 of @xcite , such a complete train track exists . after removing all branches of @xmath14 which are incident on a switch contained in @xmath359 or which are contained in @xmath360 we obtain a train track @xmath361 on the subsurface @xmath353 . if we identify @xmath353 with the surface of finite type obtained by replacing the boundary circles by cusps then @xmath361 is a complete train track on @xmath353 which is in standard form for a framing with pants decomposition @xmath362 . similarly , for @xmath363 let @xmath364 be the connected component of @xmath365 which contains @xmath353 as a subsurface . the train track @xmath366 on @xmath367 which we obtain from @xmath14 by removing all branches which are incident on a switch in @xmath248 or which are not contained in @xmath367 is complete ( in the interpretation as above ) . note that we have @xmath368 , i.e. every branch of @xmath14 is either a branch of @xmath369 or a branch of @xmath370 . moreover , a branch of @xmath14 which is incident on a switch in @xmath366 and which is not contained in @xmath366 is a small branch contained in a once punctured monogon component @xmath231 of @xmath371 , and it is the unique branch of @xmath372 in this component . for @xmath373 choose a tight splitting sequence @xmath374 issuing from @xmath375 with the property that there is a pseudo - anosov element @xmath376 of @xmath377 such that @xmath378 is shift equivalent to @xmath379 . the existence of such a splitting sequence follows from the arguments in the proof of lemma [ lemma55 ] . we view @xmath376 as a reducible element of @xmath9 which can be represented by a diffeomorphism of @xmath0 fixing @xmath248 ( or @xmath380 and @xmath381 for @xmath382 ) pointwise . define @xmath383 ; then @xmath372 is _ carried _ by @xmath14 and contains @xmath384 as a subtrack . moreover , there is a carrying map @xmath385 which maps every branch @xmath24 of @xmath386 _ onto _ @xmath366 and which maps @xmath387 bijectively onto @xmath388 . in the sequel we always assume that our carrying maps have these properties . for @xmath218 define @xmath389 ( where @xmath390 means composition , i.e. @xmath391 represents the mapping class obtained by applying @xmath24 first followed by an application of @xmath39 ) . we claim that for every @xmath218 the mapping class @xmath392 is pseudo - anosov . for this note first that @xmath393 is carried by @xmath14 for all @xmath204 . thus it follows as in lemma [ lemma53 ] that @xmath392 is pseudo - anosov if and only if @xmath394 consists of a single ray which fills up @xmath0 . by the considerations in the proof of lemma 5.4 , this is the case if a carrying map @xmath395 maps every branch of @xmath396 _ onto _ @xmath14 ( compare also @xcite ) . now let @xmath24 be any branch of @xmath396 ; then @xmath397 for @xmath398 or @xmath399 . assume first that @xmath400 . since the splitting sequence @xmath401 is tight by assumption , the image of @xmath24 under a carrying map @xmath402 equals the subtrack @xmath403 of @xmath404 . in particular , it contains the subgraph @xmath405 . on the other hand , for the same reason every branch of @xmath406 is mapped by a carrying map @xmath407 _ onto _ @xmath408 , and its maps @xmath405 bijectively onto @xmath409 . it follows that a carrying map @xmath410 maps @xmath24 _ onto _ @xmath393 . now if @xmath411 then the same argument yields that the image of @xmath24 under a carrying map @xmath412 contains the subtrack @xmath408 of @xmath393 . in particular , it contains a branch @xmath413 which is mapped by a carrying map @xmath414 _ onto _ @xmath14 . as before , we conclude that a carrying map @xmath395 maps every branch @xmath24 of @xmath396 onto @xmath14 and consequently each of the maps @xmath392 is pseudo - anosov . to establish our proposition it is enough to show that for every @xmath106 there is a number @xmath415 with the following property . for every @xmath416 , the periodic orbit for the teichmller geodesic flow on @xmath113 which corresponds to the conjugacy class of @xmath392 is entirely contained in the subset of @xmath113 of all quadratic differentials @xmath59 which admit an essential simple closed curve of @xmath59-length at most @xmath108 ( compare the proof of lemma [ lemma55 ] ) . by lemma [ lemma52 ] , since for @xmath363 the splitting sequence @xmath374 is tight there is a number @xmath150 with the following property . let @xmath27 be a measured geodesic lamination on @xmath367 which is carried by @xmath378 and which defines the transverse measure @xmath417 ; then @xmath418 for any two branches @xmath419 of @xmath366 . now the preimage of @xmath366 under any carrying map @xmath420 equals the subtrack @xmath384 of @xmath421 and the restriction of a suitably chosen carrying map to @xmath422 is injective . this implies that for _ every _ measured geodesic lamination @xmath27 which is carried by @xmath421 and defines a transverse measure on @xmath421 which is not supported in @xmath422 , the measure @xmath102 on @xmath14 induced from @xmath27 by a carrying map @xmath420 satisfies @xmath418 for all @xmath423 . 1 . the total weight of @xmath47 is one and the sum of the @xmath47-weights over all branches of @xmath14 which are _ not _ contained in @xmath366 is at most @xmath108 . 2 . for any two branches @xmath419 of @xmath425 we have @xmath426 . for a transverse measure @xmath27 on @xmath14 denote as before by @xmath427 the total weight of @xmath27 . we claim that for every @xmath106 the transformation @xmath428 which is the composition of the map @xmath429 with the map @xmath430 induced by a carrying map @xmath431 and with the normalization map @xmath432 maps @xmath433 into to @xmath424 provided that @xmath204 is sufficiently large ( indices are taken modulo 2 ) . we show our claim for the map @xmath434 , the claim for @xmath435 follows in exactly the same way . note first that by the choice of @xmath40 , the image under @xmath434 of _ every _ transverse measure on @xmath14 which does not vanish on @xmath369 satisfies property 2 ) above for @xmath398 . thus since a measure from the set @xmath436 does not vanish on @xmath437 , property 2 ) holds for all measures in @xmath438 and all @xmath218 . to establish property 1 ) for sufficiently large @xmath204 , let @xmath439 , viewed as a measured geodesic lamination , let @xmath218 and let @xmath440 . for @xmath441 let @xmath442 be the transverse measure on @xmath443 of total weight one which defines a multiple of the measured geodesic lamination @xmath27 . then there is some @xmath444 such that @xmath445 as measured geodesic laminations . by the definition of the set @xmath436 , there is a number @xmath446 only depending on @xmath108 and @xmath40 such that the total @xmath27-weight of the subtrack @xmath447 of @xmath448 is not smaller than @xmath449 . now the splitting sequence @xmath450 is tight and therefore a carrying map @xmath451 strictly increases the total weight of a transverse measure by at least a factor @xmath452 . hence for every @xmath453 the total @xmath27-weight of @xmath454 is not smaller than @xmath455 . moreover , the sum of the @xmath27-weights of the branches of @xmath456 which are not contained in @xmath457 is independent of @xmath453 . thus the sum of the @xmath458-weights of the branches of @xmath456 which are _ not _ contained in @xmath457 is smaller than @xmath459 . as a consequence , if @xmath460 then the @xmath461-weight of @xmath462 does not exceed @xmath108 and the same holds true for the @xmath463-weight of @xmath464 which shows our above claim . together we deduce the existence of a number @xmath415 such that for @xmath416 the set @xmath465 is invariant under the map @xmath21 which assigns to a measured geodesic lamination @xmath466 the normalized image of @xmath467 under a carrying map @xmath468 . now if @xmath469 and if @xmath470 is such that @xmath471 is a cotangent vector of the axis of @xmath392 then @xmath472 spans the ray @xmath394 and therefore necessarily @xmath473 . let @xmath416 and let @xmath474 be such that the total weight of the transverse measure on @xmath475 defined by the measured geodesic lamination @xmath476 equals one . recall from lemma 2.5 of @xcite that the intersection of @xmath476 with the embedded trainpath @xmath380 on @xmath475 is bounded from above by the sum of the @xmath476-weights of the branches of @xmath477 . by our above consideration , we may assume that for @xmath416 this weight is bounded from above by @xmath108 . as a consequence , for every @xmath478 the intersection of @xmath380 with the horizontal measured geodesic lamination of the quadratic differential @xmath479 is at most @xmath108 . similarly , let @xmath480 be such that the total weight of @xmath481 on @xmath448 equals one , that the total weight of @xmath482 on @xmath483 equals one , that the total weight of @xmath484 on @xmath485 equals one and that the total weight of @xmath486 on @xmath393 equals one . note that @xmath487 is the length of the periodic orbit of the teichmller flow defined by the conjugacy class of @xmath392 . using once more our above consideration we conclude that for @xmath416 and every @xmath488 the intersection of the horizontal measured geodesic lamination of the quadratic differential @xmath479 with the curve @xmath489 is bounded from above by @xmath108 . our proposition now follows if for sufficiently large @xmath204 we can control the intersections of the curves @xmath490 with the vertical measured geodesic laminations for the quadratic differentials @xmath479 . for this let again @xmath491 be such that @xmath470 and @xmath492 . since a carrying map @xmath493 maps every branch @xmath24 of @xmath494 _ onto _ @xmath495 and since @xmath496 by our above consideration , the @xmath481-weight of every branch of @xmath448 which is contained in @xmath497 is bounded from below by a universal constant @xmath302 not depending on @xmath204 . then the @xmath498-weight of every such branch is bounded from above by @xmath306 and hence there is a number @xmath336 not depending on @xmath204 which bounds from above the @xmath499-weight of every branch of @xmath500 . now the curve @xmath501 is an embedded trainpath in the train track @xmath500 and hence our upper bound for the values of @xmath499 on the branches of @xmath494 implies that the intersection between @xmath501 and @xmath499 is uniformly bounded . as we increase @xmath204 , the ratios @xmath502 tend to infinity ( compare the above consideration ) and hence after possibly increasing @xmath503 we may assume that for every @xmath416 we have @xmath504 . by invariance of the intersection form under the action of @xmath9 we conclude that for every @xmath416 and every @xmath505 the intersection between the vertical measured geodesic lamination of @xmath479 and @xmath380 is bounded from above by @xmath108 . this then shows that for every @xmath506 $ ] the sum of the intersection numbers between @xmath380 and the vertical and the horizontal measured geodesic lamination of @xmath507 does not exceed @xmath508 . in other words , for every such @xmath509 the @xmath479-length of @xmath380 is bounded from above by @xmath510 . the same argument shows that after possibly increasing @xmath503 once more we may assume that for @xmath416 and every @xmath511 $ ] the @xmath479-length of @xmath512 is bounded from above by @xmath510 . by periodicity , we conclude that for @xmath416 the periodic orbit of @xmath64 which corresponds to the conjugacy class of @xmath392 is entirely contained in the set of quadratic differentials @xmath59 which admits an essential simple closed curve of @xmath59-length at most @xmath510 . this completes the proof of our proposition . * remark : * 1 ) a pseudo - anosov element @xmath170 acts as an isometry on the _ curve graph _ @xmath513 of @xmath0 . by a result of bowditch @xcite , for every @xmath514 the limit @xmath515 exists and is independent of @xmath40 . this limit is called the _ stable length _ for this action . the stable length of each of the ( infinitely many ) pseudo - anosov elements @xmath392 constructed in the proof of proposition 5.6 is at most 2 . moreover , let @xmath516 be the teichmller geodesic in @xmath7 which is invariant under @xmath392 . then there is a number @xmath106 such that for sufficiently large @xmath204 , the set of essential simple closed curves @xmath40 on our surface @xmath0 for which the minimum of the hyperbolic lengths of @xmath40 along @xmath516 is smaller than @xmath108 is precisely an orbit under the action of @xmath392 of a pair of disjoint essential simple closed curves on @xmath0 . \2 ) we believe that proposition [ prop56 ] is valid for _ every _ nonexceptional surface of finite type , with pseudo - anosov mapping classes which can be constructed as the once in the proof our proposition . however we did not attempt to carry out the details . r. canary , d. epstein , p. green , _ notes on notes of thurston _ , in `` analytical and geometric aspects of hyperbolic space '' , edited by d. epstein , london math . lecture notes 111 , cambridge university press , cambridge 1987 . u. hamenstdt , _ train tracks and the gromov boundary of the complex of curves _ , in `` spaces of kleinian groups '' ( y. minsky , m. sakuma , c. series , eds . ) , london math . notes 329 ( 2006 ) , 187207 , cambridge university press , cambridge 2006 .

we have found a remarkably large response of the transition temperature of cafeas single crystals grown out of excess feas to annealing / quenching temperature . whereas crystals that are annealed at 400 c exhibit a first order phase transition from a high temperature tetragonal to a low temperature orthorhombic and antiferromagnetic state near 170 k , crystals that have been quenched from 960 c exhibit a transition from a high temperature tetragonal phase to a low temperature , non - magnetic , collapsed tetragonal phase below 100 k. by use of temperature dependent electrical resistivity , magnetic susceptibility , x - ray diffraction , mssbauer spectroscopy and nuclear magnetic resonance measurements we have been able to demonstrate that the transition temperature can be reduced in a monotonic fashion by varying the annealing / quenching temperature from 400 to 850 c with the low temperature state remaining antiferromagnetic for transition temperatures larger than 100 k and becoming collapsed tetragonal / non - magnetic for transition temperatures below 90 k. this suppression of the orthorhombic / antiferromagnetic phase transition and its ultimate replacement with the collapsed tetragonal / non - magnetic phase is similar to what has been observed for cafeas under hydrostatic pressure . transmission electron microscopy studies indicate that there is a temperature dependent , width of formation of cafeas with a decreasing amount of excess fe and as being soluble in the single crystal at lower annealing temperatures . for samples quenched from 960 c there is a fine ( of order 10 nm ) , semi - uniform distribution of precipitate that can be associated with an average strain field whereas for samples annealed at 400 c the excess fe and as form mesoscopic grains that induce little strain throughout the cafeas lattice .

1101.4595

cafe@xmath0as@xmath0 manifests an extreme example of the coupled magnetic / structural phase transition that epitomizes the physics of the undoped parents of the feas - based superconductors.@xcite the strongly first order transition at ambient pressure from a high temperature , tetragonal , paramagnetic phase to a low temperature , orthorhombic , antiferromagnetic phase takes place near 170 k in single crystals grown out of sn flux and manifests a hysterisis of several degrees as seen in thermodynamic , transport , and microscopic measurements.@xcite + cafe@xmath0as@xmath0 is also the most pressure sensitive of the @xmath2fe@xmath0as@xmath0 and 1111 compounds with its structural / magnetic phase transition being initially suppressed by over 100 k per gpa . @xcite as pressure increases a non - magnetic , collapsed tetragonal phase that is stabilized by @xmath3 gpa intersects and terminates the lower pressure orthorhombic / antiferromagnetic phase line near 100 k and 0.4 gpa and rises to 300 k by @xmath4 gpa . @xcite in addition to this extreme pressure sensitivity , cafe@xmath0as@xmath0 is also very sensitive to non - hydrostaticity . @xcite if the pressure medium solidifies before the structural phase transitions , then the anisotropic changes in the unit cell lead to non - hydrostatic ( by definition ) stress , which in turn leads to dramatically broadened transitions and a structurally mixed phase sample in the 0.4 gpa pressure region . this mixed phase includes a small amount of strain stabilized , high temperature tetragonal phase which superconducts at low temperature . @xcite the use of helium as a pressure medium allows for a minimization of these non - hydrostatic effects and has allowed for the determination of the @xmath5 phase diagram . @xcite + sn grown single crystals of cafe@xmath0as@xmath0 are highly deformable and join the @xmath6sb@xmath0 and @xmath6bi@xmath0 compounds @xcite as rare examples of a malleable intermetallic compounds . single crystal plates can be bent , and , to some extent , even rolled by simply grasping with tweezers and applying minor torques across the sample length by pressing one end of the crystal downward on the surface of the lab bench , or microscope stage . this malleability can lead to extreme broadening of features in ground samples , as were seen in early attempts at powder x - ray diffraction . @xcite + cafe@xmath0as@xmath0 samples were initially grown out of sn and characterized in single crystal form . @xcite sn grown crystals are well formed , faceted plates that generally have planar dimensions of several mm and thicknesses between 0.1 and 0.5 mm . @xcite for measurements that require larger sample volumes pseudo - polycrystalline @xcite or oriented single crystalline assemblies ( see figure 1 in ref . ) can be used . larger single crystals of cafe@xmath0as@xmath0 have been grown out of ternary melts rich in feas . @xcite in order for these larger crystals to manifest a structural / magnetic phase transition similar to that seen in the smaller sn grown crystal they were annealed at 500@xmath1 c for 24 hours ( a temperature similar to the decanting temperature of the sn grown samples ) . without this annealing the larger , feas - grown , samples had dramatically suppressed transition temperatures . + given recent observations of small shifts in the structural and magnetic transition temperatures of bafe@xmath0as@xmath0 samples , and of the superconducting transition in doped bafe@xmath0as@xmath0 , as well as sharpenings of their signatures in thermodynamic and transport data , @xcite we undertook a systematic study of the effects of post growth thermal treatment of feas grown single crystals of cafe@xmath0as@xmath0 . we have discovered that once again cafe@xmath0as@xmath0 is the extreme case in the @xmath2fe@xmath0as@xmath0 series , manifesting a surprisingly large suppression of the structural / magnetic transition temperature in `` as grown '' samples ( nearly 50% ) that , even more remarkably , can be systematically changed from @xmath7 k to below 100 k with the lowest transition temperature samples having a transition into the non - magnetic , collapsed tetragonal state , but at ambient pressure . + in order to characterize and understand the effects of temperature treatment , as well as the nature of the low temperature state , we have performed a wide variety of thermodynamic , transport , microscopic , and spectroscopic measurements . temperature dependent electrical resistivity and magnetic susceptibility measurements were used to determine a transition temperature ( @xmath8 ) annealing temperature ( @xmath9 ) phase diagram as well as identify similarities between the collapsed tetragonal phase and the low temperature state of feas - grown cafe@xmath0as@xmath0 crystals quenched from temperatures between 850 and 960@xmath1 c. for annealing temperatures @xmath10@xmath1 c the @xmath11 phase diagram is found to be remarkably similar to the @xmath8-pressure ( @xmath12 ) phase diagram , bringing up the question of what could the relationship between @xmath9 and @xmath12 be ? temperature dependent single crystal x - ray diffraction measurements were then employed to unambiguously show that the crystallographic phase transition in as grown samples quenched from 960@xmath1 c is one to a collapsed tetragonal state that is in qualitative as well as quantitative agreement with what is found for sn - grown samples under applied pressures of @xmath13 gpa . temperature dependent mssbauer spectroscopy measurements showed that the low temperature magnetic state of annealed feas - grown cafe@xmath0as@xmath0 single crystals remains antiferromagnetic until the transition temperature is suppressed to below 100 k when the low temperature ground state becomes non - magnetic , a result confirmed by nuclear magnetic resonance ( nmr ) measurements . finally , transmission electron microscopy ( tem ) measurements revealed that there is a small , temperature dependent width of formation for cafe@xmath0as@xmath0 , allowing for a solid solubility of excess fe and as in the single crystals that decreases with temperature . as the quenching temperature is reduced from 960@xmath1 c to 400@xmath1 c the initially fine precipitate coarsens , decreasing the degree of strain detected in the sample . single crystals of cafe@xmath0as@xmath0 were grown out of excess feas by rapidly cooling a melt of cafe@xmath14as@xmath14 from 1180@xmath1 c to 1020@xmath1 c over 3 hours , slowly cooling from 1020@xmath1 c to 960@xmath1 c over 35 hours and then decanting off the excess liquid , essentially quenching the samples from 960@xmath1 c to room temperature . these samples will be referred to as , `` as grown samples '' . post growth , thermal treatments of samples involve the following variables : annealing temperature , annealing time , and annealing environment . annealing environment refers to either ( i ) annealing a whole , unopened , decanted growth ampoule , or ( ii ) annealing individual crystals that have been picked from a growth and resealed in evacuated silica tubes . for studies of the effects of annealing temperature , we seal several crystals into an evacuated silica tube and anneal for 24 hours in a furnace stabilized at the specified temperature . the sample is placed into the hot furnace and , after annealing it is quenched to room temperature . longer time anneals ( seven days ) were used to prepare whole , unopened batches of samples . in order to study the effects of annealing on feas grown samples with transition temperatures ( and features ) like the sn - grown crystals , samples that had been annealed for a week at 400@xmath1 c were subsequently sealed into an evacuated silica tube and annealed for 24 hours in a furnace stabilized at the specified temperature . although a detailed study of the annealing time dependence of sample changes will need to be done in the future , we found that , for example , at 450@xmath1 c a one hour anneal is not enough to effect complete change , but anneals longer than 4 hours do not lead to any further significant changes in sample behavior ; at 800@xmath1 c annealing appears to be completed in under 0.5 hour . + temperature dependent magnetization measurements were made in a quantum design ( qd ) mpms unit . temperature dependent electrical resistivity was measured in a four probe configuration , with pt wires attached to the samples by du pont 4929n ag - paint ( cured at room temperature ) , in a qd ppms unit . although normalized resistivity values are plotted , the resistivity values of samples did not vary outside of the uncertainty associated with a combination of geometric error ( associated with measuring dimensions of the sample ) and difficulties associate with sample exfoliation . the average , room temperature resistivity of as grown , 700@xmath1 c annealed and 400@xmath1 c annealed samples was @xmath15 m@xmath16 cm ( a 20% variation ) . + in order to identify the nature of the structural transition in the as - grown cafe@xmath0as@xmath0 crystal ( quenched from 960@xmath1 c ) and to determine the temperature dependence of the lattice parameters , high - energy x - ray diffraction measurements ( @xmath17 kev ) using an area detector were performed on the 6-id - d station in the mucat sector at the advanced photon source . at this high energy , x - rays probe the bulk of a crystal rather than just the near - surface region and , by rocking the crystal about both the horizontal and vertical axes perpendicular to the incident x - ray beam , an extended range of a chosen reciprocal plane can be recorded . @xcite for the measurements , the horizontal angle , @xmath18 was scanned over a range of @xmath19 for each value of the vertical angle , @xmath20 , between @xmath19 with a step size of @xmath21 . the two - dimensional scattering patterns were measured by a mar345 image - plate positioned 1503 mm behind the sample . the crystal was mounted on the cold finger of a closed - cycle refrigerator surrounded by a beryllium heat shield and vacuum containment . additionally , the crystal was mounted such that there was access to @xmath22 , @xmath23 , and @xmath24 reciprocal lattice planes . the @xmath25 and @xmath26 peaks were fit for lattice parameter determination and , for these measurements , the total exposure time for each frame was 383 s. + the mssbauer absorbers were prepared by attaching several single crystal plates to a 12 mm diameter disc of 100 @xmath18 m thick kapton foil using ge-7031 varnish . the spaces between the crystals were filled with a radio - opaque paint prepared by mixing 1 - 5 @xmath18 m tungsten powder ( obtained from alpha - aesar ) with diluted ge-7031 varnish . the absence of gaps in the completed mosaic was confirmed first visually and then by looking for transmission of the 6.4 kev fe - k@xmath27 x - ray from the @xmath28co mssbauer source . in the configuration used , the crystalline @xmath29 -axis was parallel to the mssbauer @xmath30-beam . + the mssbauer spectra were collected on a conventional spectrometer using a 50 mci @xmath28co / rh source mounted on an electromechanical drive operated in constant acceleration mode . the spectrometer was calibrated against @xmath31-fe metal at room temperature . temperatures down to 5 k were obtained using a vibration - isolated closed - cycle refrigerator with the sample in a partial pressure of helium to ensure thermal uniformity . spectra were fitted using a conventional , non - linear , least - squares minimization routine to a sum of equal - width lorentzian lines . the line positions for the magnetic sextets observed in the ordered state were calculated assuming first - order perturbation in order to combine the effects of the magnetic hyperfine field and the electric field gradient . as the samples were oriented mosaics rather than powders , the line intensities were constrained to be in the ratio @xmath32 ( following the conventional practice of labeling the lines from negative to positive velocity ) @xcite with the intensities of the two @xmath33 lines being variable ( `` @xmath6 '' ) to allow for the expected magnetic texture . @xmath34 would correspond to the moments being parallel to the mssbauer @xmath30-beam , whereas @xmath35 indicates that the moments are perpendicular to the beam . + nuclear magnetic resonance ( nmr ) measurements were carried out on @xmath36as ( @xmath37 ; @xmath38 mhz / t ) by using a homemade phase - coherent spin - echo pulse spectrometer , to investigate the magnetic and electronic properties of differently treated cafe@xmath0as@xmath0 crystals from a microscopic point of view . @xmath36as - nmr spectra at a resonance frequency of 51 mhz were obtained by sweeping the magnetic field . + tem samples were prepared by mechanically polishing the single crystal to @xmath39 @xmath18 m thick along the c - axis and then ion milling to perforation using 3 kev @xmath40 incident angle and following up with 30 min at 500 ev at @xmath41 to remove milling damage . all milling was performed using a liquid n@xmath0 cooled stage ( sample @xmath42 k ) . samples were analyzed using a philips cm30 tem operated at 300 kev . energy dispersive spectroscopy ( eds ) and selective area diffraction patterns ( sadp ) were also performed on the samples in the tem . figure [ f1 ] presents the resistivity and magnetic susceptibility for cafe@xmath0as@xmath0 single crystals grown out of sn and for cafe@xmath0as@xmath0 single crystals grown out of excess feas . two data sets are shown for feas grown crystals : one data set shows measurements on an as grown crystal that was decanted at , and quenched from , 960@xmath1 c ; the other data set shows measurements on a sample from a batch that was subsequently annealed at 400@xmath1 c for a week . the sn - grown single crystal and the feas grown sample that has been annealed at 400@xmath1 c are quite similar , both manifesting similar , modest increases in resistivity and decreases in susceptibility associated with the phase transition near 170 k. @xcite on the other hand , the feas sample that was quenched from 960@xmath1 c shows a significantly larger , very sharp drop in magnetization occurring well below 100 k. the electrical resistivity also drops discontinuously at this temperature , associated with the sample suddenly undergoing a violent structural phase transition that often ( usually ) leads to shattering along the length and width of the bar , as well as loss of contacts . + in addition to the quantitative differences shown in fig . [ f1 ] , there is a qualitative difference between the as grown , cafe@xmath0as@xmath0 single crystals from feas solution and the single crystals grown from sn . whereas the sn - grown single crystals are malleable and can easily be bent and deformed , the crystals quenched from a 960@xmath1 c feas solution are brittle and tend to shatter if bending is attempted . the feas grown crystals that have been annealed at 400@xmath1 c , however , recover some of the malleability of the sn grown ones and can deform a little without shattering . + given the dramatic difference in transition temperature , as well as the different signatures of the transition in resistivity and magnetization , several questions arise . among them we consider : ( i ) what is the nature of the phase transition in the as grown sample and ( ii ) can the transition in annealed samples be varied from near 170 k to below 100 k in a systematic manner ? we will address the latter question first and return to the former after the creation of a transition temperature ( @xmath8 ) - annealing temperature ( @xmath9 ) phase diagram . + in order to assess the extent to which the 170 k phase transition that occurs in sn - grown , as well as annealed feas - grown , samples of cafe@xmath0as@xmath0 can be systematically shifted down to below 100 k we measured the temperature dependent susceptibility and resistivity of as grown samples that were annealed for 24 hours at temperatures ranging from 250 to 850@xmath1 c. figure [ f2 ] presents magnetic susceptibility and resistivity data for representative annealing temperatures . the decrease in susceptibility ( or increase in resistivity ) can be shifted down in temperature by choosing an appropriate annealing temperature between 400 and 800@xmath1 c. for annealing between these temperatures the transitions , particularly as seen in the resistivity data , remain quite sharp and shift in a systematic manner . whereas the size of the jump in the magnetization remains fairly constant in the samples annealed in this temperature region , there is a monotonic increase in the magnitude of the increase in the resistivity ( see fig . [ f7 ] below ) . + such a clear temperature dependence of the effects of annealing , over such a wide temperature range , begs the question of what the annealing time dependence of these effects is . in other cases of clear annealing effects , both time and temperature cuts through phase space are needed to establish unambiguous annealing protocols.@xcite in fig . [ f3 ] we show the evolution of the magnetic susceptibility for different annealing times . at 450@xmath1 c , 0.5 hr is insufficient time to effect any significant change ; 1.0 hr leads to split , broadened features with drops in susceptibility below both 170 and 100 k ; 3.0 hr leads to a single , sharp feature near 170 k , comparable to what is seen for 24 hour anneals . this progression shows that for 450@xmath1 c , 24 hours is longer than the salient time scale for annealing . as would be expected , for higher temperatures the salient time scale is even shorter . in fig . [ f3](b ) samples from a batch that had been annealed for a week at 500@xmath1 c , with a transition temperature above 150 k , were annealed at 800@xmath1 c for representative times . as can be seen , even a 0.5 hour anneal causes the sample to behave in a manner similar to the as grown ( quenched from 960@xmath1 c ) samples . + figure [ f2 ] also demonstrates that 24 hour anneals at temperatures of 300@xmath1 c or lower temperatures do not change the temperature dependence of the as grown samples . the data from the sample annealed at 350@xmath1 c for 24 hours shows somewhat broadened drops in susceptibility near both 170 and 100 k , similar to what was seen for a 1.0 hr anneal at 450@xmath1 c ( fig . [ f3]a ) , indicating that at 350@xmath1 c 24 hours is comparable to , but less than the salient time scale . although longer annealing times for @xmath43@xmath1 c may lead to a sharp , single transition near 170 k ( as is seen for the 400 and 500@xmath1 c , 24 hr anneals ) the time needed to achieve this state is anticipated to become exponentially long . the one other data point we can add to this is the fact that 20@xmath1 c ( room temperature ) anneals approaching @xmath44 hours have not led to significant changes in behavior of as grown samples . + a 24 hour anneal at 850@xmath1 c does not significantly change the transition temperature from that measured for the as grown samples quenched from 960@xmath1 c ( perhaps not too surprisingly since 850@xmath1 c is approaching the 960@xmath1 c quench temperature ) ; the resistivity data for this sample , though , can be collected below the transition temperature , showing that the low temperature state has a lower resistivity , leading to a downward jump in resistivity when cooling through the transition temperature . + in order to see if similar changes in transition temperature could be induced by annealing samples that started with transitions near 170 k ( i.e. started with transitions similar to those found in sn grown cafe@xmath0as@xmath0 ) we annealed an entire batch of crystals at 400@xmath1 c for a week . the resistivity and susceptibility data for these samples are also shown in fig . [ f2 ] and are essentially the same as that found for the 24 hr anneal of individual crystals . single crystals from this , `` 400@xmath1 c anneal for one week '' batch were then separately sealed in silica ampoules and annealed for 24 hours at temperature ranging from 500@xmath1 c to 800@xmath1 c. the temperature dependent resistivity and susceptibility for these samples are shown in fig . as was the case for the as grown samples , sharp features in both resistivity and susceptibility systematically shift to lower temperature when the sample is annealed at higher temperature . the sample annealed at 800@xmath1 c shows the larger drop in susceptibility and broke on cooling through its transition , making it appear to be similar to the as grown , quenched from 960@xmath1 c , samples . + figure [ f5 ] presents the transition temperature - annealing temperature , @xmath45 , plot . figure [ f6 ] illustrates how values for @xmath8 , as well as the error bars , were inferred from the resistivity and susceptibility data . as can be seen in figs . [ f2 ] and [ f4 ] , for @xmath46@xmath1 c , there is a systematic progression of fairly sharp transitions downward for increasing @xmath9 . figure [ f5 ] illustrates that , ( i ) there is some scatter in @xmath8 for a given @xmath9 , but ( ii ) that there is also a fairly well defined suppression of @xmath8 with increasing @xmath9 , e.g. a 400@xmath1 c anneal gives a very different transition temperature from a 700@xmath1 c anneal , which is itself different from a 850@xmath1 c anneal or the as grown sample . in addition , annealing at a given @xmath9 leads to a @xmath8 value , regardless of whether the sample starts from a 170 k or @xmath47 k transition state , i.e. this final anneal determines @xmath8 regardless of sample history . + the @xmath11 phase diagram presented in fig . [ f5 ] shows that cafe@xmath0as@xmath0 grown out of feas can have the temperature of its phase transition modified in an essentially continuous manner from near 170 k to below 100 k. for transitions with @xmath8 between 170 k and 100 k the magnetic signature of the transition is essentially unchanged and the resistive signature evolves gradually with the jump in resistivity , @xmath48 , becoming larger as @xmath8 decreases . for the lowest @xmath8 values , below 100 k , there is a significantly larger drop in susceptibility and , when it can be measured , the jump in resistivity is downward on cooling rather than upward . these observations are quantified in fig . as discussed in the experimental methods section , above , the room temperature resistivity of as grown samples as well as those annealed at 400 and 700@xmath1 c for a week all have room temperature resistivity values of @xmath15 m@xmath16-cm . this invariance , within experimental resolution , allows for conversion of these jumps to absolute resistivity as needed . + the evolution of the transition temperature ( fig . [ f5 ] ) as well as the evolution of the resistive and magnetic signature of the phase transition ( figs . [ f1 ] - [ f4 ] ) make it plausible that for 170 k @xmath49 k the transition is similar to that seen in sn - grown cafe@xmath0as@xmath0 : a transition from a high temperature tetragonal , paramagnetic state to a lower temperature orthorhombic , antiferromagnetic state . on the other hand , dramatic changes in the resistive and magnetic signature associated with the as grown sample as well as samples annealed near 850@xmath1 c are consistent with our current understanding of the collapsed tetragonal phase , a phase that up until this time was associated with cafe@xmath0as@xmath0 under pressures of 0.35 gpa or higher . the change in the resistive signature of the transition from a sharp increase to a sharp decrease , as well as the sudden increase in the size of the drop in susceptibility upon cooling are very similar to the changes seen in ref . under hydrostatic pressure applied with helium . for that matter , the basic phase diagram proposed in refs is remarkably similar to the @xmath46@xmath1 c part of the @xmath11 phase diagram presented in fig . [ f5 ] with @xmath9 playing the role of pressure , or more precisely stated somehow parameterizing the amount of stress in the sample . this similarity can be seen in fig . [ f5]b which directly plots @xmath8 as a function of @xmath12 and @xmath9 . for annealing temperatures that allow for the achievement of equilibrium by 24 hours ( i.e. 400@xmath1 c or greater ) there is a remarkable agreement between the effects of @xmath9 and @xmath12 on @xmath8 , at least as long as the transition is from high temperature tetragonal to low temperature orthorhombic / antiferromagnetic . once the low temperature state is the non - magnetic , collapsed tetragonal phase the annealing temperature does not seem to affect @xmath8 in the same manner as @xmath12 . + before progressing too much further , though , we need to examine ( and , as will be shown , verify ) several of the hypotheses outlined above . first we should parameterize and understand the nature of the low temperature ground state in the as grown feas crystals , once that is done we can return to the question of what may be causing the systematic changes we see in the @xmath11 phase diagram . + the collapsed tetragonal , ct , phase was identified @xcite in cafe@xmath0as@xmath0 by scattering measurements made on samples under hydrostatic pressure using he as a pressure medium as part of a comprehensive effort to better understand the details of the cafe@xmath0as@xmath0 @xmath5 phase diagram . @xcite as cafe@xmath0as@xmath0 transforms from the high temperature tetragonal phase into the low temperature , collapsed tetragonal phase the @xmath29-lattice parameter changes from @xmath50 to @xmath51 , a remarkably large , @xmath39% decrease while the @xmath52-lattice parameter increases by @xmath53% , leading to an @xmath54% decrease in the unit cell volume . @xcite in order to see if the as grown sample manifests such a striking change in lattice parameters , high - energy single - crystal x - ray diffraction data were collected as a function of temperature . figure [ f8 ] displays the results of fits to these data to extract the lattice parameters [ figs . [ f8](a ) and ( b ) ] as well as the unit cell volume [ fig . [ f8](c ) ] . the data from the original sn - flux grown samples , at an applied pressure of 0.63 gpa , @xcite are also included in fig . [ f8 ] for direct comparison . these data clearly indicate that , structurally , the as grown crystals of cafe@xmath0as@xmath0 transform into a collapsed tetragonal phase below 100 k at ambient pressure . in particular , the inset to fig . [ f8](b ) shows the diffraction image of the @xmath55 bragg reflection at 6 k , the base temperature of our measurement . within our resolution , no splitting of the peak is evident as would be expected for an orthorhombic unit cell . furthermore , we find that the temperature dependence of the lattice parameters and unit cell volume are consistent with what was observed for the pressure induced , collapsed tetragonal phase for @xmath56 gpa where the tetragonal - to- ct phase transition has moved above 150 k. + although the sharp drop in susceptibility certainly suggests that the non - magnetic phase is associated with the collapsed tetragonal state , it is prudent to examine the magnetic properties of this state more closely with microscopic measurement techniques . mssbauer spectroscopy measurements were carried out on three feas grown samples at selected temperatures . composite samples , with aligned @xmath29-axes , were made from single crystals from whole batches treated in the following manner : as grown ( quenched from 960@xmath1 c ) , annealed for a week at 400@xmath1 c , annealed for a week at 700@xmath1 c. as shown in fig . [ f5 ] , these three annealing temperatures produce samples representative of the whole range of behavior observed . whereas similar sample masses were used for the two extreme samples , a smaller sample mass was used for the sample annealed at 700@xmath1 c , leading to slightly poorer signal and statistics . + the spectra of the three samples taken at 295 k ( fig . [ f9 ] ) are essentially indistinguishable , showing a barely resolved quadrupole split doublet with the high velocity line slightly more than twice the intensity of the lower velocity line in each case . the location of the fe atoms in the thcr@xmath0si@xmath0-type structure requires that the principal axis of the local electric field gradient tensor lies along the @xmath29-axis , so our oriented mosaic should yield an intensity ratio of 3:1,@xcite rather than the 2.3(3):1 , 2.0(2):1 and 2.4(2):1 observed here for the 400@xmath1 c annealed , 700@xmath1 c annealed and as grown samples respectively . it is unlikely that the reduced ratio is due to misalignment of the crystal plates as the intensities in the magnetic patterns are consistent with almost perfect alignment . it is possible that a minor impurity is the source of the reduced ratio , however no identifiable impurity contribution was found in the residual patterns of the fitted spectra , except in the case of the 700@xmath1 c annealed sample , where about 4% of the total iron was found to be associated with a non - magnetic phase at 5 k. the quadrupole splittings for the as grown and 700@xmath1 c annealed samples were slightly higher ( + 0.236(5 ) mm / s ) than that in the 400@xmath1 c annealed sample ( + 0.202(12 ) mm / s ) probably reflecting more distorted local fe environments . the sign of the electric field gradient can not normally be determined from a @xmath28fe mssbauer spectrum of a powder sample but the combination of oriented thcr@xmath0si@xmath0-type single crystal samples with the observed intensity ratio makes the sign determination unambiguous . + cooling to low temperatures makes the differences between the three samples strikingly obvious . the two annealed samples undergo sudden transitions near 170 k ( annealed at 400@xmath1 c ) and 130 k ( annealed at 700@xmath1 c ) and by 10 k the sample annealed at 400@xmath1 c has developed a clear magnetic splitting of 10.03(3 ) t whereas the sample annealed at 700@xmath1 c exhibits a slightly smaller hyperfine field of 9.51(3 ) t. fitting the intensities of the two @xmath57 lines ( lines 2 and 5 ) in both magnetic patterns yields @xmath58 , implying an almost perfect ( better than @xmath41 ) alignment of the crystal @xmath29-axis with the @xmath30-beam , @xcite as expected from the construction of the sample , and confirming that the ordering direction of the iron moments in both ordered samples lies in the @xmath59-plane . by contrast , the spectrum of the as grown sample is almost unchanged . the quadrupole splitting increases slightly , to 0.272(4 ) mm / s , and there is also a small increase in linewidth ( from 0.143(4 ) mm / s to 0.170(3 ) mm / s ) , possibly reflecting some increased disorder or strain . the largest change appears in the line intensity ratio , which drops to 1.70(7):1 on cooling to 5 k , suggesting either a reduction in quality of the @xmath29-axis alignment or a tilting of the principal axis of the local electric field gradient away from the c - axis . remarkably , the original intensity ratio is recovered on warming back to 295 k , so the change is fully reversible . visual inspection of the three mosaics following several thermal cycles between room temperature and the base temperature of the cryostat did not reveal any apparent damage . + the temperature dependence of the hyperfine field for the two annealed samples , shown in fig . [ f10 ] , reveals that not only is the iron moment probably slightly smaller in the sample annealed at 700@xmath1 c , but the temperature dependence of the hyperfine field ( @xmath60 ) is visibly stronger , consistent with a lower ordering temperature . although it is not possible to determine the ordering temperature of the hypothetical , second order phase transition of an antiferromagnetically ordered , orthorhombic phase transforming into a paramagnetic , orthorhombic phase directly , since in reality the sample undergoes a strong first order transition to the tetragonal phase on warming , we can make an estimate by fitting the observed temperature dependence of @xmath60 to a brillouin function as is shown in figure [ f10 ] . this procedure yields estimated ordering temperatures of 300(10 ) k for the sample annealed at 400@xmath1 c , and 230(10 ) k for the sample annealed at 700@xmath1 c , with the errors dominated by an uncertainty in the effective total angular momentum quantum number , @xmath61 , used in the fits . we note that whereas the absolute values of the nel temperatures did depend somewhat on @xmath61 , the difference between the values of @xmath62 for the two samples did not , and was consistently 70(3 ) k. the magnetic ordering in the sample with the lower structural transition temperature is definitely weaker , involving slightly smaller iron moments . + nmr measurements were also carried out on the as grown and 400@xmath1 c annealed samples . [ f11](a ) shows @xmath36as - nmr spectra at @xmath63 k for two magnetic field directions of @xmath64-axis and @xmath65-plane for the 400@xmath1 c annealed crystal . the observed quadrupole - split nmr spectra are well reproduced by a simple nuclear spin hamiltonian @xcite @xmath66 $ ] , where @xmath67 is the effective field at the as site ( summation of external field @xmath68 and the hyperfine field @xmath69 ) , @xmath70 is planck s constant and @xmath71 is nuclear quadrupole frequency which is proportional to the electric field gradient ( efg ) at the as site ( an assymetric parameter of efg is assumed to be zero for simplicity ) . the blue lines in the figure show simulated spectra calculated from the simple hamiltonian . below 160 k , each nmr line for @xmath64-axis splits into two lines due to internal field @xmath69 ( parallel or antiparallel to @xmath68 ) which is produced by the fe spin ordered moment . a typical example of the split nmr lines for @xmath64-axis is shown at the bottom of fig . [ f11](a ) . the spectrum is reproduced well by @xmath72 t and @xmath73 mhz at @xmath74 k. these values are in good agreement with previously reported values for @xmath36as - nmr of single crystals ( @xmath75 k ) grown out of sn flux , @xcite once again indicating that the sample annealed at 400@xmath1 c is essentially the same as previously reported ones grown out of sn . the temperature dependence of the @xmath36as spin lattice relaxation rates ( @xmath76 ) measured at center line for @xmath64-axis is also in good agreement with previous work . @xcite + similar quadrupole - split nmr spectra are observed in the as - grown cafe@xmath0as@xmath0 sample as shown in fig . [ f11](b ) , but the observed @xmath77 mhz at @xmath78 k is larger than that in the annealed crystal . the @xmath36as - nmr satellite linewidth , which reflects the distribution of efg , is significantly larger than in the annealed sample , indicative of higher degree of inhomogeneity of the local as environment due to strains , defects , or lattice distortion in the as - grown sample . below the transition temperature , @xmath79 k , no splitting of the nmr lines is observed ( indicating that there is no antiferromagnetic order ) but @xmath71 is found to change dramatically : from @xmath80 mhz to @xmath81 mhz , as is shown at the bottom in fig . [ f11](b ) . such a drastic change of @xmath71 ( more than 230% ) can not be explained by thermal expansion of lattice ( at most few % ) but is attributed to a structural phase transition . the value @xmath82 mhz is also confirmed by the observation of nuclear quadrupole resonance ( nqr ) spectrum at zero magnetic field at @xmath83 k ( fig . [ f11](c ) ) . the peak position in the nqr spectrum for the as - grown sample is higher than the 25 and 30.4 mhz for the tetragonal and collapsed tetragonal phases respectively in cafe@xmath0as@xmath0 under high pressure . @xcite the combination of no splitting of the nmr lines with the large shift in @xmath71 are further confirmation that , for the as grown sample , there is only a structural phase transition without any magnetic phase transition . + the combination of x - ray diffraction , mssbauer and nmr data unambiguously identify the low temperature state of the as grown ( quenched from @xmath84 c ) sample as being non - magnetic and also having a collapsed tetragonal unit cell that is remarkably similar to what has been found for sn - grown cafe@xmath0as@xmath0 under hydrostatic pressure . for that matter , the evolution of the temperature dependent resistivity as well as magnetic susceptibility are both qualitatively similar to the evolutions found when pressure is applied as hydrostatically as possible , i.e. with he as a pressure medium . @xcite at this point , having not only created a @xmath11 phase diagram that looks a lot like the @xmath5 phase diagram ( for @xmath46@xmath1 c ) , fig . [ f5 ] , but also having clearly identified the phases associated with this phase diagram , it is appropriate to investigate the possible physical origin , or mechanism , for this apparent similarity between pressure applied to a sn - grown crystal of cafe@xmath0as@xmath0 and annealing of feas grown crystals . + a starting point for this search for a mechanism can be found in a subset of the observations made above . the as grown crystals from feas solution are far more brittle than either the sn - grown crystals or the feas grown crystals after a 400@xmath1 c anneal . this qualitative observation hints at some higher concentration of defects in the as grown crystals that lead to embrittlement . in addition , both the mssbauer and nmr measurements find broader line shapes associated with the spectra from the as grown samples , indicating that there may be a greater degree of disorder in them than in the crystal annealed at 400@xmath1 c. + in order to examine the distribution of defects at a nano - scale level , tem measurements were carried out on both as grown , quenched from 960@xmath1 c , samples as well as samples that had been annealed at 400@xmath1 c for a week . the as - grown sample ( fig . [ f12](a ) ) shows a pervasive tweed - like pattern with @xmath85 nm separation of features . the selected area diffraction pattern , ( inset , fig . [ f12](a ) ) shows only the @xmath86 $ ] zone axis pattern consistent with the cafe@xmath0as@xmath0 compound and no additional reflections or streaking indicative of a superlattice or a highly defective ( intercalated ) lattice . these very long but thin features are orthogonal and are approximately parallel to the @xmath87 planes as best can be determined in this orientation of the sample . the thinnest regions of the sample did not exhibit these features , consistent with the ease at which these samples were damaged by ion milling . ( milling above 3 kev and not cooling with liquid nitrogen resulted in significant milling artifacts . ) the thickness of the foils where these features are present and the lattice strain they cause , prevent atomic resolution imaging at this point . however , tilting experiments and imaging with the principle reflections did reveal the two - dimensional nature of these thin lamellae . these features were consistent throughout all the thin area of the sample although in some regions one variant may dominate over the other and in some regions interpenetrating lamellae were observed as shown in fig . [ f12](a ) . occasional dislocations were observed , but did not dominate the microstructure . + the sample annealed at 400@xmath1 c for one week appears completely different . here we observed a very smooth contrast across the thin region when tilting and uniformly distributed small lenticular precipitates about 25 to 100 nm in width and with a length to width aspect about 5:1 ( fig . [ f12](b ) ) . these precipitates are also fairly uniformly separated , @xmath88 to 1,000 nm , and have their long axis parallel to the @xmath87 , as was observed in the tweed pattern of the 960@xmath1 c quenched sample . dislocations in the matrix are commonly observed to emanate from the interface between the precipitates and the matrix typically near the ends of the precipitate where stresses would be higher if there are differences in coefficients of thermal expansion . the sadp is nearly identical to the as grown sample but here the precipitates are large enough for diffraction analysis . the convergent beam electron diffraction ( cbed ) pattern , rhs inset fig . [ f12](b ) , produces disks rather than spots due to how the pattern is formed , but it is clear that the pattern is identical to that of the matrix . the fact that the sadp did not show any splitting of spots when including the precipitates and the matrix in the same sample area would suggest that in this orientation the two lattices are nearly coherent . eds also indicated that the precipitate phase does not contain ca . efforts to identify the precise chemistry and structure of this second phase are on - going but it should be noted that the basal plane dimensions of the cafe@xmath0as@xmath0 and the tetragonal asfe@xmath0 ( space group @xmath89 , # 129 ) are within 5 % and have very similar @xmath86 $ ] diffraction patterns . a gross estimate of the impurity phase area in less than 5 % of the total sample , giving a gross idea of how much extra fe and as is trapped in the sample when it is initially quenched from 960@xmath1 c. + the similarity in the orientation and lack of distinguishing features in the diffraction between the as grown samples and the samples that were annealed suggests that there is a similarity in chemistry / structure between the tweed strain fields and coarser precipitates in these two samples and the difference is simply one of length scale . an epitaxial relationship would lower the energy barrier for nucleation and allow a second phase to form more readily if thermodynamically stable . annealing at moderate temperatures but within a two - phase field , would promote growth of the second phase to reduce the excess energy due to interfaces ( i.e. , oswald ripening ) . the observations here are consistent with an increase in the width of formation of cafe@xmath0as@xmath0 with respect to excess as and fe at elevated temperatures which decreases monotonically with temperatures below 960@xmath1 c. + such a temperature - dependent solid solubility of excess fe and as leads to the following , plausible , scenario . when the feas flux grown crystal is first quenched , there is little time for the excess as and fe to come out of solution . in these single crystals , the grain dimensions ( often mm to cm ) are simply too large for diffusion to allow for the expulsion of these species to the grain boundaries . energetically , it appears easiest to exsolve excess as and fe epitaxially along the @xmath87 planes . differences in their unit cell size as well as their coefficients of thermal expansion ( cte ) can lead to significant stresses at the interfaces between the cafe@xmath0as@xmath0 majority phase and the finely dispersed fe / as based second phase . if , as fig . [ f12]a would suggest , domains of cafe@xmath0as@xmath0 , about 40 nm on a side , are surrounded by nearly coherent second phase resulting in a significant volume fraction of interfaces or regions strained by interfaces , then the magnitude of the stress would be dependent on the volume fraction of the cafe@xmath0as@xmath0 in these strained regions . + if there is a temperature dependence of the solubility of the excess as and fe , then quenching from lower temperatures would result in a smaller fraction of finely dispersed second phase , the remaining excess as and fe being sequestered in larger precipitates whose insignificant surface to volume ratio would have little impact on the matrix ( as is the case in fig . [ f12]b ) . this smaller amount of finely dispersed precipitate would lead to a smaller average strain ( or pressure ) on the sample , leading to an effective correlation between @xmath12 and @xmath9 . since the initial quench of the large crystals from the flux essentially locks in the excess as and fe , the subsequent processing history determines the size and distribution of the second phase and thereby determines the amount of strain in the sample . + further exploration of this hypothesis requires confirmation of the second phase crystal structure and its chemistry and determination of the cte and bulk moduli of these two phases . in addition , returning to the initial motivations for this study : given that annealing of as grown samples seems to lead to small changes in the transition temperatures of bafe@xmath0as@xmath0 based compounds @xcite and given that cafe@xmath0as@xmath0 is much more pressure / strain sensitive than bafe@xmath0as@xmath0 , it is worth exploring the implications of our current findings . if we speculate that a similar width of formation exists in bafe@xmath0as@xmath0 ( or for that matter srfe@xmath0as@xmath0 ) and that low temperature annealing can lead to similar effects as those we present here , then , based on the existing @xmath90 phase diagrams , @xcite an effective pressure of @xmath13 gpa on bafe@xmath0as@xmath0 or srfe@xmath0as@xmath0 would only lead to shifts in @xmath8 of a few k , consistent with what has been observed . @xcite based on this analysis , tem measurements on as grown and annealed crystals of bafe@xmath0as@xmath0 , as well as srfe@xmath0as@xmath0 , to check for similar , annealing temperature dependent microstructure are in order . we have found a remarkably large response of the transition temperature of cafe@xmath0as@xmath0 single crystals grown out of excess feas to annealing / quenching temperature . whereas crystals that are annealed at 400@xmath1 c exhibit a first order phase transition from a high temperature tetragonal to a low temperature orthorhombic and antiferromagnetic state near 170 k ( similar to what has been found in the original sn grown single crystals @xcite ) , crystals that have been quenched from 960@xmath1 c exhibit a transition from a high temperature tetragonal phase to a low temperature , non - magnetic , collapsed tetragonal phase below 100 k. we have been able to demonstrate that the transition temperature can be reduced in a monotonic fashion by varying the annealing / quenching temperature from 400 to 850@xmath1 c with the low temperature state remaining antiferromagnetic for transition temperatures larger than 100 k and becoming collapsed tetragonal / non - magnetic for transition temperatures below 90 k. this suppression of the orthorhombic / antiferromagnetic phase transition and its ultimate replacement with the collapsed tetragonal / non - magnetic phase is similar to what has been observed for sn - grown single crystals of cafe@xmath0as@xmath0 under hydrostatic pressure . @xcite this similarity is summarized in fig . [ f5]b . tem studies of the as grown ( quenched from 960@xmath1 c ) and annealed crystals indicate that there is a temperature dependent , width of formation of cafe@xmath0as@xmath0 with a decreasing amount of excess fe and as being soluble in the single crystal at lower annealing temperatures . on one extreme , samples quenched from 960@xmath1 c have , finely divided , strain structure with characteristic length scales and spacings of less than 50 nm . on the other extreme , samples annealed at 400@xmath1 c have clearly identifiable , ca free , crystalline inclusions with dimensions of @xmath91 nm@xmath92 that are separated by 500 to 1,000 nm . these images make it clear that when the sample is quenched from 960@xmath1 c it is possible to think of some average , near uniform strain throughout the sample associated with the overlapping strain fields of this fine precipitate . it is this strain that appears to be giving rise to the dramatic suppression of @xmath8 , in , apparently , a similar manner that hydrostatic pressures of @xmath13 gpa can . finally , it is worth noting that , as was the case with hydrostatic pressure applied by helium pressure medium , @xcite there is no indication of superconductivity , even in trace amounts . this would be consistent with the idea that superconductivity comes from the poorly defined strains associated with the onset of the collapsed phase in a confined ( solidified pressure medium ) volume . given that these measurements are at ambient pressure , this is not the case . by the same token , though , given that we can essentially mimic the key effects of hydrostatic pressure via variation of @xmath9 , the collapsed tetragonal phase , as well as the orthorhombic / antiferromagnetic phase with intermediate @xmath8 values , is open to measurements at ambient pressure . this should make detailed thermodynamic , microscopic and spectroscopic measurements across the phase space outlined in fig . [ f5]b much more accessible . we thank d. robinson for his excellent technical support of the x - ray diffraction study . this work was supported by the u.s . department of energy , office of basic energy science , division of materials sciences and engineering . ames laboratory is operated for the u.s . department of energy by iowa state university under contract no . de - ac02 - 07ch11358 . work at mcgill university was supported by grants from the natural sciences and engineering research council of canada and fonds qubcois de la recherche sur la nature et les technologies . a. kreyssig , m. a. green , y. lee , g. d. samolyuk , p. zajdel , j. w. lynn , s. l. budko , m. s. torikachvili , n. ni , s. nandi , j. b. leo , s. j. poulton , d. n. argyriou , b. n. harmon , r. j. mcqueeney , p. c. canfield , and a. i. goldman , phys . b * 78 * , 184517 ( 2008 ) . a. i. goldman , a. kreyssig , k. prokes , d. k. pratt , d. n. argyriou , j. w. lynn , s. nandi , s. a. j. kimber , y. chen , y. b. lee , g. samolyuk , j. b. leo , s. j. poulton , s. l. budko , n. ni , p. c. canfield , b. n. harmon , and r. j. mcqueeney , phys . rev . b * 79 * , 024513 ( 2009 ) . d. k. pratt , y. zhao , s. a. j. kimber , a. hiess , d. n. argyriou , c. broholm , a. kreyssig , s. nandi , s. l. budko , n. ni , p. c. canfield , r. j. mcqueeney , and a. i. goldman , phys . rev . b * 79 * , 060510 ( 2009 ) . c. r. rotundu , b. freelon , t. r. forrest , s. d. wilson , p. n. valdivia , g. pinuellas , a. kim , j .- w . kim , z. islam , e. bourret - courchesne , n. e. phillips , and r. j. birgeneau , phys . b * 82 * , 144525 ( 2010 ) . ( color online ) temperature dependent ( a ) magnetic susceptibility and ( b ) normalized electrical resistivity of cafe@xmath0as@xmath0 for three differently prepared single crystals : squares - sn grown ; circles - as grown ( quenched from 960@xmath1 c ) from feas ; triangles - feas grown , annealed for a week at 400@xmath1 c. note : when the as grown sample from feas melt is cooled below the transition temperature near 90 k it shattered , making further lower temperature resistivity measurements impossible . , title="fig:",width=377 ] ( color online ) temperature dependent ( a ) magnetic susceptibility and ( b ) normalized electrical resistivity of cafe@xmath0as@xmath0 for three differently prepared single crystals : squares - sn grown ; circles - as grown ( quenched from 960@xmath1 c ) from feas ; triangles - feas grown , annealed for a week at 400@xmath1 c. note : when the as grown sample from feas melt is cooled below the transition temperature near 90 k it shattered , making further lower temperature resistivity measurements impossible . , title="fig:",width=377 ] ( color online ) temperature dependent magnetic susceptibility and normalized electrical resistivity of as grown cafe@xmath0as@xmath0 single crystals annealed for 24 hours at temperature , @xmath9 . susceptibility data in ( a ) have been offset from each other by an integer multiple of @xmath93 emu / mole for clarity . data for a one week anneal of a whole batch at 400@xmath1 c is shown for comparison . the resistivity data ( b ) for the as grown sample could not be measured below the transition temperature due to sample breakage , but for the sample annealed at 850@xmath1 c resistivity could be measured through the transition . , title="fig:",width=377 ] ( color online ) temperature dependent magnetic susceptibility and normalized electrical resistivity of as grown cafe@xmath0as@xmath0 single crystals annealed for 24 hours at temperature , @xmath9 . susceptibility data in ( a ) have been offset from each other by an integer multiple of @xmath93 emu / mole for clarity . data for a one week anneal of a whole batch at 400@xmath1 c is shown for comparison . the resistivity data ( b ) for the as grown sample could not be measured below the transition temperature due to sample breakage , but for the sample annealed at 850@xmath1 c resistivity could be measured through the transition . , title="fig:",width=377 ] ( color online ) temperature dependent magnetic susceptibility of ( a ) as grown cafe@xmath0as@xmath0 single crystals annealed at 450@xmath1 c for representative times and ( b ) as grown caas@xmath0fe@xmath0 single crystals that have been annealed for a week at 500@xmath1 c and then annealed at 800@xmath1 c for representative times . data in both panels have been offset from each other by an integer multiple of @xmath94 emu / mole for clarity . , title="fig:",width=377 ] ( color online ) temperature dependent magnetic susceptibility of ( a ) as grown cafe@xmath0as@xmath0 single crystals annealed at 450@xmath1 c for representative times and ( b ) as grown caas@xmath0fe@xmath0 single crystals that have been annealed for a week at 500@xmath1 c and then annealed at 800@xmath1 c for representative times . data in both panels have been offset from each other by an integer multiple of @xmath94 emu / mole for clarity . , title="fig:",width=377 ] ( color online ) temperature dependent ( a ) magnetic susceptibility and ( b ) normalized electrical resistivity of as grown cafe@xmath0as@xmath0 single crystals that were first annealed for a week at 400@xmath1 c and then annealed for 24 hours at temperature , @xmath9 . susceptibility data in ( a ) have been offset by an integer multiple of @xmath94 emu / mole from each other for clarity . , title="fig:",width=377 ] ( color online ) temperature dependent ( a ) magnetic susceptibility and ( b ) normalized electrical resistivity of as grown cafe@xmath0as@xmath0 single crystals that were first annealed for a week at 400@xmath1 c and then annealed for 24 hours at temperature , @xmath9 . susceptibility data in ( a ) have been offset by an integer multiple of @xmath94 emu / mole from each other for clarity . , title="fig:",width=377 ] ( color online ) ( a ) transition temperature ( @xmath8 ) annealing temperature ( @xmath9 ) phase diagram . open symbols are inferred from resistivity data and filled symbols are inferred from susceptibility data . stars are the as grown samples ( quenched from 960@xmath1 c data and are also shown as 20@xmath1 c anneals ) ; squares are as grown samples that have been annealed for 24 hours at @xmath9 and quenched to room temperature ; circles are as grown samples that were first annealed for a week at 400@xmath1 c and then annealed for 24 hours at @xmath9 and quenched to room temperature ; diamonds are as grown samples that have been annealed for a week as whole , unopened batches at @xmath9 . ( b ) @xmath8 as a function of pressure from ref . and @xmath8 as a function of @xmath9 for @xmath95@xmath1 c@xmath96@xmath1 c. , title="fig:",width=377 ] ( color online ) ( a ) transition temperature ( @xmath8 ) annealing temperature ( @xmath9 ) phase diagram . open symbols are inferred from resistivity data and filled symbols are inferred from susceptibility data . stars are the as grown samples ( quenched from 960@xmath1 c data and are also shown as 20@xmath1 c anneals ) ; squares are as grown samples that have been annealed for 24 hours at @xmath9 and quenched to room temperature ; circles are as grown samples that were first annealed for a week at 400@xmath1 c and then annealed for 24 hours at @xmath9 and quenched to room temperature ; diamonds are as grown samples that have been annealed for a week as whole , unopened batches at @xmath9 . ( b ) @xmath8 as a function of pressure from ref . and @xmath8 as a function of @xmath9 for @xmath95@xmath1 c@xmath96@xmath1 c. , title="fig:",width=377 ] ( color online ) temperature dependent ( a ) magnetic susceptibility and ( b ) normalized electrical resistivity and their respective temperature derivatives for an as grown crystal annealed at 500@xmath1 c for 24 hours . @xmath8 is inferred from the temperature of the extremum in the derivative . full width at half maximum of derivative is used to define error bars shown in fig . values of the jump in magnetic susceptibility ( maximum and minimum values ) and resistivity are inferred as shown . , title="fig:",width=377 ] ( color online ) temperature dependent ( a ) magnetic susceptibility and ( b ) normalized electrical resistivity and their respective temperature derivatives for an as grown crystal annealed at 500@xmath1 c for 24 hours . @xmath8 is inferred from the temperature of the extremum in the derivative . full width at half maximum of derivative is used to define error bars shown in fig . values of the jump in magnetic susceptibility ( maximum and minimum values ) and resistivity are inferred as shown . , title="fig:",width=377 ] ( color online ) size of jump in susceptibility and normalized resistivity as a function of transition temperature for feas grown cafe@xmath0as@xmath0 crystals annealed at temperatures shown in figures [ f1 ] - [ f4 ] . room temperature resistivity of samples with @xmath8 values of @xmath47 , 130 and 170 k all fall within the @xmath15 m@xmath16 cm range . , width=453 ] ( color online ) values for ( a ) the @xmath29-lattice paramter , ( b ) @xmath52-lattice parameter and ( c ) unit cell volume as a function of temperature for an unannealed feas flux - grown cafe@xmath0as@xmath0 sample determined from high - energy x - ray diffraction measurements . the open squares denote the results of measurements performed on a polycrystalline sample under applied hydrostatic pressure of 0.63 gpa from reference . the inset to the middle panel is the image of the @xmath55 diffraction peak taken from the two - dimensional x - ray detector as described in the text . note the absence of any splitting that would signal a transition to an orthorhombic phase [ the two open circles illustrate the expected distance between split bragg peaks due to the `` usual '' orthorhombic distortion ] . , width=264 ] @xmath28fe mssbauer spectra of ab - plane single crystal mosaic of samples ( from left ) annealed at 400@xmath1 c , annealed at 700@xmath1 c and as grown ( quenched from 960@xmath1 c ) . in each case the upper spectrum was taken at 295 k whereas the lower spectrum was taken at 10 k or 5 k. only the as grown sample shows no magnetic ordering at base temperature . the solid lines are fits as described in the text . , width=453 ] ( color online ) temperature dependence of the magnetic hyperfine fields ( @xmath60 ) for the two annealed samples shown in figure [ f9 ] . the solid lines are fits to brillouin functions ( as described in the text ) used to estimate the nel temperatures of the low temperature antiferromagnetic orthorhombic form of each sample . in both cases the samples undergo a first order structural transition on warming effectively truncating the @xmath97 data . , width=529 ] ( color online ) @xmath36as - nmr spectra measured at @xmath98 mhz for ( a ) @xmath99 c annealed cafe@xmath0as@xmath0 crystal and ( b ) for the as - grown cafe@xmath0as@xmath0 crystal . black and blue lines are observed and simulated spectra , respectively . expected lines above 9 t are not measured due to the limited maximum magnetic field for our sc magnet . ( c ) @xmath36as nqr spectrum at @xmath100 k and @xmath101 t. , width=453 ] ( a ) tem micrograph of the as - grown and quenched sample . the fine tweed - like pattern is due to thin platelets of a second phase which are coherent with the matrix and parallel the @xmath87 planes . the lower right - hand side inset is the sadp from this image area showing expected lattice reflections for a @xmath86 $ ] zone axis for the 122 compound . note the absence of streaking or extra reflections as discussed in the text . ( b ) tem micrograph of the sample annealed at 400@xmath1 c for 1 week at the same magnification and zone axis orientation as the as - grown sample above . note that the rectangular precipitates are oriented along the same @xmath87 planes as the finer features above . the left - hand inset shows the sadp of the matrix and precipitates while the right - hand inset is a cbed of only the precipitate phase showing nearly the same orientation and d - spacing as the matrix 122 phase . , title="fig:",width=340 ] ( a ) tem micrograph of the as - grown and quenched sample . the fine tweed - like pattern is due to thin platelets of a second phase which are coherent with the matrix and parallel the @xmath87 planes . the lower right - hand side inset is the sadp from this image area showing expected lattice reflections for a @xmath86 $ ] zone axis for the 122 compound . note the absence of streaking or extra reflections as discussed in the text . ( b ) tem micrograph of the sample annealed at 400@xmath1 c for 1 week at the same magnification and zone axis orientation as the as - grown sample above . note that the rectangular precipitates are oriented along the same @xmath87 planes as the finer features above . the left - hand inset shows the sadp of the matrix and precipitates while the right - hand inset is a cbed of only the precipitate phase showing nearly the same orientation and d - spacing as the matrix 122 phase . , title="fig:",width=340 ]

in order to study precession and interstellar magnetic field variations , we measured the polarized position angle of 81 pulsars at several - month intervals for four years . we show that the uncertainties in a single - epoch measurement of position angle is usually dominated by random pulse - to - pulse jitter of the polarized subpulses . even with these uncertainties , we find that the position angle variations in 19 pulsars are significantly better fitted ( at the 3 level ) by a sinusoid than by a constant . such variations could be caused by precession , which would then indicate periods of d and amplitudes of degrees . we narrow this collection to four pulsars that show the most convincing evidence of sinusoidal variation in position angle.also , in a handful of pulsars , single discrepant position angle measurements are observed which may result from the line of sight passing across a discrete ionized , magnetized structure . we calculate the standard deviation of position angle measurements from the mean for each pulsar , and relate these to limits on precession and interstellar magnetic field variations .

1008.0454

many pulsars are among the most highly polarized astrophysical sources of radiation . one polarized parameter , the position angle of linear polarization @xmath3 ( also called the ppa polarized position angle ) is a particularly useful quantity : it carries information about the intrinsic geometry of the emitting region at the pulsar [ e.g. , via the rotating vector model @xcite ] ; it also probes the magnetized plasma along the line of sight through faraday rotation measurements [ e.g. , @xcite ] . the purpose of this paper is to study the behavior of the polarized position angle on timescales of @xmath4 d , in order to investigate time - variable behaviors of pulsars magnetospheric geometry and the interstellar medium . for example , precession of the neutron star or other long - timescale oscillatory behavior affecting the emission beam will change the ppa , as will variations in the magnetized interstellar plasma along the line of sight . ninety - eight pulsars were observed at 21 cm from arecibo observatory every few months for several years from 1989 to 1993 . the resulting grand average polarized pulse profiles were analyzed in @xcite , where observing details can also be found . in the present work , we found that we had adequate data on eighty - one of the pulsars to search for temporal variations of polarized position angle . for all of them , the polarized position angles were originally measured relative to an unknown origin . consequently , in the present work , in order to search for temporal _ changes _ in position angle , we needed to reference our data to a position - angle standard . to do so , we selected two calibrator pulsars , b0540 + 23 ( j0543 + 2329 ) and b1929 + 10 ( j1932 + 1059 ) . the calibrators were chosen on the basis of the relative constancy of their ppa s during full - sky arecibo observing tracks . we then compared the measured polarized position angle of a target pulsar on a given observing session to the ppa of one of the two calibrator pulsars . while the _ absolute _ position angle is still unknown , this referencing procedure enables us to be sensitive to position angle _ changes _ over time , assuming only that the position angle of the calibrators remains fixed over time . the procedures are described in the following sections ; mathematical details are given in appendix a. in the rare cases where a calibrator pulsar was not observed during a daily session , we created an artificial `` meshed '' calibrator pulsar by determining its expected position angle on that session from observations of all other pulsars during that session . pulsars tend to be very highly linearly polarized objects . the observed linear polarization originates at or near the star , often following the magnetic field lines at the point of origin @xcite , and is then modified by various propagation effects . we define the generic polarized position angle curve " for a pulsar at epoch @xmath5 , a function of pulsar longitude @xmath6 at fixed @xmath5 , as the set of polarized position angles @xmath7 for @xmath8 . ( see the bottom panels of fig . [ fig : obspas ] for examples of ppa curves . ) the measured ppa curve at epoch @xmath5 can be expressed as the sum of five terms an ensemble average ( denoted below by angular brackets ) for the indicated pulsar longitude @xmath6 and epoch @xmath5 which reflects the pulsar s intrinsic orientation , an interstellar faraday rotation term , an instrumental origin , and a term arising from errors due to radiometer noise and one from intrinsic pulse - to - pulse fluctuations in amplitude and phase of individual pulses ( called `` jitter ; '' see @xcite and further discussion below ) : @xmath9 the first two terms on the right - hand side of the above equation hold the physics of interest in this experiment namely the temporal behavior of the pulsar s orientation or magnetosphere , and of the interstellar magnetic field through which its signals propagate , respectively . the third term defines the instrumentally measured ppa corresponding to an absolute ppa on the sky of zero . the latter two terms , though having zero mean , represent phenomena that add variance to the measurements , thereby limiting our precision in specifying the first two . we now discuss each of these five terms in order . * ( a ) precession : * precession would lead to a periodic wobble of the pulsar orientation . most pulsar magnetospheric models indicate that the polarized emission is tied fundamentally to the geometry of the magnetic field in the emitting regions . hence it is expected that a periodic wobble in pulsar orientation translates directly into a periodic variation of similar amplitude in polarized position angle . therefore , a search for temporal variations in ppa is a sensitive probe for the presence of precession . the simplest model is one where the wobble would result from the free precession of an asymmetric isolated pulsar , although there also exist models where the precession is forced by an external disk or companion @xcite , or by the star s own radiation @xcite . we assume that the neutron star is approximately spherical , with a slight axially symmetric oblateness about a symmetry axis . hence the three moments of inertia have the relationship @xmath10 , where @xmath11 is the moment parallel to the symmetry axis . ( for extension of the precession model to a triaxial body , see @xcite ) . @xcite and @xcite show that the motion of a point on the star can be decomposed into a quick rotation about the total angular momentum vector plus a slow precession about the body symmetry axis . for a sketch of the process and a link to an animation , [ fig : animation ] ; for readers wishing to modify the geometrical parameters of the animation , the mathematica source code is also available online at the mathsource library @xcite . the angle @xmath12 between the angular momentum vector and symmetry axis leads to a `` wobble '' at both the rotation and precession frequencies , with the latter leading directly to a change in ppa of similar amplitude , is magnified by a factor of @xmath13^{-1}$ ] ; .i.e , @xmath14^{-1}$ ] , where @xmath15 is the ( generally unknown ) latitude of the magnetic axis with respect to the symmetry equator . ] on precession timescales . very slow precession , with a period of many years , would produce just a linear trend in ppa with epoch . * ( b)magnetospheric adjustments : * as we will describe below , there are theoretical difficulties ( though not insoluble ones ) with the picture of precession - induced pulse shape variations , centered primarily on the long observed timescales and relatively large inferred amplitudes . @xcite have proposed an alternate model whereby the pulse shape and timing variations are caused by slowly rotating magnetospheric currents rather than by precession of the whole star . while this is an interesting alternative interpretation of the observations , more work needs to be done to flesh out the model and to verify that it can explain the observed timescales and amplitudes . in this section , we focus on studying variations in the magnetized ism plasma via a search for temporal variations in ppa . the second term of eq . [ eqn : basicpsi ] quantifies faraday rotation of the ppa in the interstellar medium , @xmath16 , at some epoch , @xmath5 . the rotation measure , @xmath17 , is defined so as to specify the angle of ppa faraday rotation at wavelength @xmath18 : @xmath19 with angle in radians , @xmath18 in m , and @xmath17 in rad m@xmath20 . a measured change in ppa , @xmath21 , then implies a change in faraday rotation measure , @xmath22 : @xmath23 in order to relate @xmath24 to the ism , we express @xmath25 as a function of fundamental properties along the pulsar earth path : @xmath26 where @xmath27 is the magnetic field parallel to the line of sight in @xmath28 g , @xmath29 is the electron density in electrons@xmath30 , and @xmath31 is a differential path element whose integrated distance is @xmath32 @xcite . each of the variables in eq . [ eqn : faradayb ] may be a function of time . while this time - variability could result directly from _ temporal _ changes in the ism , it would more likely be induced by the pulsar s motion rapidly sweeping the line of sight across a presumed _ spatially _ inhomogeneous ism . since measurements show that temporal variations of dispersion measure @xmath33 , are generally quite small @xcite , we further expect that any observed variations in @xmath25 would originate principally from the pulsar carrying the line of sight across an inhomogeneity in @xmath27 . therefore , we model the rm change as resulting from the line of sight passing across a discrete region having a variation in the parallel field , @xmath34 , occupying a fraction @xmath35 of the pulsar - earth path , with _ no _ associated @xmath29 variation . in this case , @xmath24 , the @xmath25 change due to the discrete region is @xmath36 finally , we can relate an observed change in ppa , @xmath37 , to the properties of the discrete region by combining eqs . [ eqn : deltafaradayrot ] and [ eqn : deltarmfromgrad ] : @xmath38 or , expressed as a fractional variation : @xmath39 where @xmath40 is the line of sight average . the third term in eq . [ eqn : basicpsi ] represents a conversion of absolute ppa on the sky , to ppa measured in the instrumental frame . the instrumental frame involves geometrical terms associated with telescope orientation as well as instrumental phase delays during a given session . as noted above in [ sec : obsana ] , we did not directly determine the instrumental ppa but rather referenced our measured ppa to the ppa of a calibrator pulsar , which has its own variance ( see [ sec : offset ] ) . the noise term in eq . [ eqn : basicpsi ] results from the finite sampling statistics of a noisy signal and is familiar to all radioastronomers . its contribution to the variance , @xmath41 , was determined empirically by measuring the variance off - pulse . the jitter term in eq . [ eqn : basicpsi ] is unique to pulsars and requires further explanation , especially because it often dominates the error budget . we characterize as `` jitter '' all ppa changes on pulse - to - pulse timescales that are counterparts to the arrival - time jitter of individual subpulses . these random ppa variations occur due to various propagation effects in the pulsar magnetosphere analogous to faraday rotation . these fluctuations might be correlated , for example , for the duration of a single subpulse in a given pulse period , but would be uncorrelated _ between _ subpulses and _ across _ pulse periods . random variations in _ altitude _ of emission would produce a ppa curve that is shifted earlier or later for higher or lower altitudes , respectively . this kind of variation would also be highly correlated across a single subpulse . the variance of @xmath3 due to jitter , @xmath42 , is calculated as follows . [ sec : jittervar ] by definition , uncertainties @xmath43 and @xmath44 in polarized parameters @xmath45 ( total power at longitude bin @xmath6 ) and @xmath46 ( linearly polarized power at longitude bin @xmath6 ) due to random pulse - to - pulse jitter are characterized by @xmath47 and @xmath48 where @xmath49 is the modulation index " of polarized parameter @xmath50 , a measure of its pulse - to - pulse modulation . in our experiment , each polarized profile was integrated online for two minutes , thereby attenuating these pulse - to - pulse modulations . therefore , the jitter - induced fluctuations in the measurement of these quantities after @xmath51 pulses , will be @xmath52 and @xmath53 at high @xmath54 and under the reasonable assumption that @xmath55 , @xmath56 propagates to an uncertainty in position angle @xmath57 : @xmath58 @xcite . therefore , the jitter - induced uncertainty in the measurement of ppa in an integration of @xmath51 pulses is @xmath59 as we shall see below in [ sec : offset ] , we must ultimately determine @xmath60 , the jitter - induced fluctuation in the ppa integrated not only over @xmath51 pulses but also over the full range of longitudes . with the subpulse as the basic unit of correlated emission , the jitter - induced fluctuations in the measurement of ppa after @xmath51 pulses , each containing @xmath61 subpulses , will be characterized by @xmath62 since we could not directly measure the quantities @xmath63 and @xmath61 with our two - minute integrations , we surveyed the literature for typical values . many observations suggest that @xmath64 @xcite and the few extant polarized modulation index observations ( e.g. , @xcite ) support the reasonable assumption that @xmath65 as well . single - pulse observations indicate that @xmath61 is also approximately 1 . finally , we found that constant-@xmath3 fits to the full set of pulsars had a mean @xmath66 when @xmath67 in eq . [ eqn : sigmapsijitt ] , so we adopted this value ; i.e. , @xmath68 we empirically tested the relative contributions of radiometer noise and pulse jitter to the ppa uncertainty in the following tests on one of the two calibrator pulsars , b0540 + 23 . we observed it for several @xmath69two - hour stretches with a series of standard two - minute integrations identical in form to all others in this experiment , and analyzed the resulting data as follows . first , a linear trend in ppa , presumably due to variable ionospheric faraday rotation , was fitted and subtracted out . for psr b0540 + 23 , the uncertainty in ppa resulting from radiometer noise was @xmath70 , while the _ observed _ standard deviation of two - minute integrations from the linear trend was a far larger @xmath71 . pulse jitter considerations , however , lead via eq . [ eqn : jitterempirical ] to an estimated @xmath72 . while the jitter uncertainty estimated via eq . [ eqn : jitterempirical ] appears to be about twice the observed standard deviation , it is important to bear in mind that the simplifying assumptions used in the jitter calculation can easily lead to factor of @xmath73 errors in either direction for a particular pulsar , as they apparently have in this case . ( see further discussion in [ sec : upperlimits ] ) . nevertheless , the important conclusion of this exercise is that the actual ppa uncertainty is far larger than that due to radiometer noise alone , and that pulse jitter appears to be the predominant source of the observed noise , with an amplitude in rough agreement with our model . note also that the jitter model should give excellent results for a given pulsar if all factors in eq . [ eqn : sigmapsijitt ] are directly measured , rather than using an average over our pulsar sample , as in eq . [ eqn : jitterempirical ] . in order to study temporal variations in a pulsar s orientation or in the intervening medium , we investigate the ppa as a function of time . as noted in [ sec : obsana ] , the instrumental ppa origin , @xmath74 of eq . [ eqn : basicpsi ] , was not explicitly measured . instead , we reference our target pulsar ppa s to `` calibrator '' pulsars b0540 + 23 or b1929 + 10 , which can be considered ppa beacons " of constant but unknown ppa . we do so in a three - step process : first we find the offset of the target pulsar s ppa curve at epoch @xmath5 from a high @xmath54 `` template '' ppa curve of the same pulsar from @xcite ( see fig . [ fig : obspas]a and eq . [ eqn : deltapsipsr ] ) . next we do the same procedure at the same epoch on the calibrator pulsar b0540 or b1929 + 10 ( see fig . [ fig : obspas]b and eq . [ eqn : deltapsical ] ) . finally , to yield the desired offset of the target pulsar from the calibrator pulsar at epoch @xmath5 , @xmath75 , we difference the above two results ( see eq . [ eqn : deltapsiwhole ] ) . the variance in a target pulsar - template ppa difference measurement , @xmath76 , is the sum of the noise and jitter contributions from the target pulsar at epoch @xmath5 ( the template contributes negligibly ) : @xmath77 a similar equation holds for the variance in a calibrator pulsar - template ppa difference measurement , @xmath78 . however , we are ultimately interested in the variance of the position angle of a pulsar with respect to the calibrator pulsar , ( _ cf . _ . [ eqn : deltapsiwhole ] ) . this variance , @xmath79 , will then be the sum of the above - discussed variances in the pulsar - template and calibrator - template ppa difference measurements : @xmath80 we searched for evidence of temporal variations of polarized position angle in each of the 81 target pulsars , with an emphasis on a few likely signatures : ( 1 ) sinusoidal variations in ppa due to periodic phenomena at the pulsar such as precession , as described in [ sec : orientation ] ; ( 2 ) discontinuous changes in ppa caused by interstellar propagation effects , such as the line of sight crossing a region with significantly different magnetic field , as discussed in [ sec : faraday ] ; and ( 3 ) ramps in ppa caused by phenomena such as those listed above , having a characteristic time longer than our experiment s . for each pulsar , the dataset consists of the measured ppa offsets , @xmath75 ( _ cf . _ eq . [ eqn : deltapsiwhole ] ) , and their estimated uncertainties , @xmath81 ( _ cf . _ [ eqn : sigmapsiml ] ) , at a set of epochs , @xmath5 . in the following sections , we give an overview of these results . see the discussion ( [ sec : discussion ] ) for exploration of the astrophysical implications . to test for possible periodic variations in each pulsar s ppa , we fit a sine function to @xmath82 . this sine function requires four free parameters : the zero - point @xmath83 about which the sinusoid oscillates ; the amplitude @xmath84 of the sinusoid about @xmath85 ; the period @xmath86 of the sinusoid ; and a phase offset @xmath87 : @xmath88 . \label{eq : sineeq}\ ] ] in fitting with these free parameters , we had two chief concerns : that our best - fit results for this function s parameters not be dependent on initial guesses for those parameters , and that we only accept the sinusoidal fit s results when that fit represents a statistically significant improvement over the fit of a constant , @xmath89 . to ensure that the fit results were not dependent on initial guesses , we ran trial fits on a large number of initial values for both sinusoidal period @xmath86 and amplitude @xmath84 . the initial - parameter search space for the amplitude was set to lie in the interval [ @xmath90 . meanwhile , the maximum sinusoidal period was set to @xmath91 , where @xmath92 is the total timespan of the observations , in order to look for long - timescale variations . the minimum sinusoidal period tested was to @xmath93 where @xmath94 is the number of observation sessions for each pulsar . this range of sinusoidal amplitude and period was then used to define a linearly - spaced grid of @xmath95 initial guesses . a separate non - linear least - squares fit was run for each set of initial parameters , where all four parameters were allowed to `` float '' freely ; the only exception to this was that the period was not allowed to fall below the minimum period already defined . of all of the resultant converged fits , the sinusoid with the minimum @xmath96 was selected as the best fit for that pulsar . as a final check , we re - ran the analysis with a @xmath97 grid of initial guesses ; no significant changes to our fit results were found . to check that the best sinusoidal model significantly improves the fit when compared to the constant - ppa model , we applied the f - test ( see 11.4 , * ? ? ? * ) . for the best - fit sinusoid for each pulsar , we calculated the @xmath98 statistic , where @xmath99 . here , @xmath100 is the difference in @xmath96 between the best - fit constant ppa and the best - fit sinusoid , and @xmath101 is the reduced @xmath96 value for the best - fit sinusoid . to assess its significance , one calculates the probability that the sinusoidal model s improved fit would be achieved with random data . to accept it as formally significant , we ask that the sinusoidal fit be an improvement over the constant - ppa fit at the @xmath102 level ( approximately @xmath103 ) ; i.e. , we require that the calculated improvement in @xmath96 would be less than @xmath104 probable , given a random sample of data points . we find that nineteen of our pulsars meet or exceed this sinusoidal vs. constant - ppa f - test significance level , indicating statistically significant sinusoidal ppa variations with time . it is important , however , to recognize that a particular pulsar s f - test significance calculation is only advisory , given the uncertainty in the value of the pulse jitter variance ( see [ sec : jittervar ] ) , which enters crucially as a scale factor in the reduced @xmath96 calculation . nevertheless , there is a clear pattern of statistically significant improvement in many fits across our pulsar sample when a sinusoidal model is introduced . the details of the nineteen pulsars statistically significant fits , along with two additional ones ( see below ) , are listed in table [ table : sinefits ] . to highlight our assessment of the quality of the fits , we group the 81 sinusoidal fits into four different classes ( see table [ table : sinefits ] for designations of all pulsars in the three highest classes ) . the top two classes have formally significant ( @xmath105 ) fits , while the bottom two do _ not . `` class i '' _ pulsars are those whose sinusoidal fits we find particularly convincing . the data and fits for these four pulsars are shown in fig . [ fig : classifits ] . _ `` class ii '' _ consists of those pulsars where we find the fit less convincing , although still formally significant , for two principal reasons . ( i ) if there is a large spread in the ppa values that seems to not be accounted for by the sinusoid , possibly indicating non - sinusoidal variation in the ppa ( see an example in fig . [ fig : classiifit ] ) ; or ( ii ) if the _ constant_-ppa fit itself appears sufficient @xmath106 , therefore calling into question the need for a sinusoid . _ `` class iii '' _ pulsars are those that we judge might have sinusoidal variation , although the fit was not a formally significant improvement over the constant - ppa fit . both such pulsars fits are displayed in fig . [ fig : classiiifit ] . _ `` class iv '' _ encompasses the remaining 60 pulsars , which have neither formal nor apparent sinusoidal tendencies . see figure [ fig : classivfit ] for an example . in our sinusoidal fits ( _ cf . _ [ sec : sine ] ) , none of the trial fits with long - period variations ( longer than the total timespan of observations ) is a significant improvement over the constant ppa fit at the @xmath103 level . therefore , there is no evidence in our data of variations in ppa , including ramps , on these ultralong timescales . [ sec : upperlimits ] for a discussion of upper limits on ppa variations from a ( constant ) mean on shorter timescales . table [ table : big ] presents a full summary of our ppa versus time measurements for the 81 pulsars . the table is organized into three principal categories . ( i ) _ pulsar details : _ the first seven entries for a given pulsar list catalog information such as pulsar name and @xmath107 and @xmath25 ; ( ii ) _ data catalog : _ the next four entries detail our observations , listing total timespan , total number of sessions , and the number rejected due to the ppa being many @xmath108 from the mean ( see [ sec : discont ] for further details on the rejects ) ; and ( iii ) _ retained session results _ : these results represent important upper limits derived from the `` sessions retained '' after the editing process , and are discussed further below . the standard deviation of the position angle of a given pulsar , called @xmath57 in the `` retained session results '' section of table [ table : big ] , is one of the most important quantities measured in this experiment . it represents the typical deviation of any single ppa measurement from the mean position angle . is uniquely determined from the set of measurements over all @xmath46 retained epochs , @xmath109 . this contrasts with the situation for the individual session error bars , @xmath108 of eq . [ eqn : sigmapsiml ] , which depend on the assumptions discussed in [ sec : jittervar ] regarding pulse jitter that are untestable with the current data . ] ( for the pulsars exhibiting possible sinusoidal variations , the standard deviation from the sinusoid , @xmath110 , is also presented in table [ table : sinefits ] ) . together , these results represent , to our knowledge , the only published systematic measurements of time variations ( or their upper limits ) of pulsar polarized position angles . as noted in [ sec : upperlimits ] , there are a handful of measured @xmath75 lying many @xmath108 away from the mean value for a given pulsar . ( see the `` sessions manually rejected '' column of table [ table : big ] for a listing of the pulsars having `` manually '' rejected sessions , and table [ table : outliers ] for details on these sessions properties . ) there are only six such rejected sessions among all 81 pulsars , and never more than one per pulsar . we have examined the polarized profile in each such case and see no evidence for instrumental problems . these puzzling events , if real , indicate the passage of ionized , magnetized structures across the line of sight on timescales shorter than the several - month intervals between observing sessions . as noted above , polarized position angle variations could be caused either by phenomena at the pulsar ( e.g. , precession or magnetospheric adjustments in the emission beam geometry ) ; or by interestellar faraday rotation variability . the observed measurements and upper limits on position angle changes reported in tables [ table : sinefits ] - [ table : outliers ] can be translated ( 1 ) directly into changes or upper limits thereof in projected spin axis or magnetospheric orientation ; or to ( 2 ) changes or upper limits thereof in interstellar magnetic fields via faraday rotation . we now discuss each in turn . much theoretical work suggests that the amplitude of the precession would be smaller than a degree , which would place it below the threshold of measurability for most pulsars in this experiment ( _ cf . _ table [ table : big ] ) , and furthermore that it would quickly damp , at least in the presence of rigid pinning or even the slow creep of superfluid vortices [ e.g. , @xcite ] . nevertheless , there are several observations of periodic changes in pulse shape and/or in arrival times of isolated pulsars in the literature , along with ours , that seem to indicate the presence of precession . in a few of the other cases , the derived amplitude is on the order of one to several degrees and the period is in the few - yr range , which are similar to our results . the best documented example is psr b1828 - 11 , whose arrival times and pulse shape vary periodically and in a correlated fashion , indicating precession with a period on the order of 1 yr and @xmath111 amplitude @xcite , or @xmath112 @xcite or possibly more @xcite . psr b1642 - 03 also exhibits both phenomena @xcite , which have been modeled as stemming from precession with an amplitude of @xmath113 over several yr @xcite.@xmath114 @xcite discovered pulse shape and timing variations in psr b2217 + 47 , which they modeled as induced by precession with a timescale @xmath115 yr . psr b1557 - 50 shows a several - year timing and dispersion measure periodicity ( the latter interpreted as possible frequency - dependent temporal pulse shape variations ) which is modeled as precession with an amplitude of @xmath116 @xcite . @xcite observed quasiperiodic timing and profile variations in psr b0959 - 54 which could be caused by precession with a period @xmath117 2500 d and amplitude @xmath118 . quasiperiodicities on several hundred day timescales in crab nebula pulsar arrival times may also indicate precession @xcite . @xcite observed pulse shape , ppa , and timing variations in the vela pulsar which they suggested could result from free precesssion with a 330-d period ; and @xcite further elaborated the precession model to explain variations in x - ray morphology of the vela pulsar s pulsar wind nebula . @xcite have proposed that the galactic center radio transient gcrt j1745 - 3009 could be a neutron star precessing by @xmath119 , although an experiment like the current one would be insensitive to its 77-min period . finally , the x - ray emitting isolated neutron star rx j0720.4 - 3125 exhibits a wide variety of correlated periodic phenomena indicative of precession on a 7 - 8 yr timescale @xcite . our analysis shows that 19 of our 81 pulsars exhibit ppa variations that are significantly better fit ( at approximately the 3@xmath0 level ) by a sinusoidal ppa function than by a constant ppa . of these , we judge four pulsars to exhibit the most convincing evidence for sinusoidal variations in ppa . ( these are our `` class i '' pulsars : b0523 + 11 , b0611 + 22 , b0656 + 14 , and b2053 + 21 ; see fig . [ fig : classifits ] ) . our results indicate that variations , and even sinusoidal variations , are occasionally present in the pulsar population . as a group , these 19 pulsars display sinusoidal periods mostly grouped around 185 to 450 days , with a peak at 200 days and outliers at approximately 790 , 1050 , and 1250 days . the amplitude of the variation ranges from 1@xmath120 to 8@xmath120 for most of the significant fits , with a peak at 2@xmath120 and an outlier at 12@xmath120 . note that these sinusoidal results do not include the six large ppa excursions , rejected from the fits in table [ table : big ] and detailed in table [ table : outliers ] ; as these large excursions do not carry the signature of precession even if they are real . in [ sec : orientation]b , we briefly discussed rotating magnetospheric currents as a possible source of periodic ppa changes . in the absence of well - developed models , we only note here that such phenomena might provide an alternate explanation for the observed periodic ppa variations . table [ table : outliers ] lists the properties of those few ppa measurements lying far from the mean . as noted above , these could indicate the presence of magnetized variations in the ism , although they might conceivably result instead from an experimental problem . under the first assumption ( magnetized ism variations causing the deviations ) , we calculate and list in table [ table : outliers ] the quantities @xmath121 ( eq . [ eqn : deltafaradayrot ] ) and @xmath122 ( eq . [ eqn : deltarmfrac ] ) . however , the great majority of our pulsars shows no evidence whatsoever of ppa variations above the noise . under the second assumption ( that the few large ppa deviations observed are spurious ) , our measured upper limits on temporal variations of position angle among `` retained sessions '' ( the quantity @xmath57 of table [ table : big ] ) can be interpreted as upper limits on magnetized variations , again via eqs . [ eqn : deltafaradayrot ] and [ eqn : deltarmfrac ] where the quantities @xmath21 and @xmath24 are replaced by their respective statistical analogs , @xmath57 and @xmath123 . table [ table : big ] lists these derived upper limits on @xmath123 and @xmath124 ) . there are occasional reports in the literature of temporal variations in interstellar magnetic fields , some of which may have origins similar to ours . first , there are the cases of the vela and crab nebula pulsars @xcite . the variations of @xmath25 ( and also of @xmath107 ) are so large and frequent that they are almost certainly associated with passage of inhomogeneous nebular snr material across the line of sight . this interpretation is easy to reconcile with these pulsars relative youth and the observed presence of an snr . there are other reports in the literature of @xmath25 variations toward ordinary pulsars on multiyear timescales , sometimes accompanied by @xmath107 variations as well @xcite , which appear to be caused by the passage of a cloud across the line of sight . @xcite measured many pulsar @xmath25s and compared their results with earlier work wherever possible . they noted numerous cases of @xmath25s changing significantly with respect to values measured @xmath125 yr earlier , while the @xmath107 tended to vary far less on similar timescales . yet our sample of pulsars exhibits only rare evidence for @xmath25 variations over the four years of observation , and _ no _ evidence for sustained changes or long - term trends . we can suggest several possible resolutions to the disagreement between the rather common @xmath25 changes in the literature and our finding of essentially steady @xmath25s in our sample : ( 1 ) @xmath25 variations tend to be small on the four - year timescale of our experiment , but grow on decade - long scales . this would have interesting implications for the fluctuation spectrum of interstellar plasma . however , recall that we have some sensitivity to changes on timescales longer than our dataspan , but no evidence was found for variations on scales beyond @xmath126 d. ( 2 ) a subtle nonorthogonal emission mode competition process can lead to spurious apparent @xmath25 variations @xcite , which might grow with time . ( 3 ) one of the previously published multiepoch @xmath25 measurements was incorrect . this latter possibility can not be ruled out , since @xmath25 measurements are difficult , and since the pairs of observations were frequently performed by different groups on different telescopes . further measurements are required in order to determine the correct explanation . we have studied the temporal behavior of polarized position angle in 81 pulsars over a @xmath127 yr period . we demonstrate that the intrinsic pulse - to - pulse jitter of polarized subpulses frequently dominates the uncertainty in position angle determination . we searched particularly for sinusoidal variations , discontinuous changes , and linear trends in the data . nineteen pulsars have variations in ppa that were significantly better fit ( at the 3@xmath0 level ) by a sinuoidal function than by a constant ; such variations could be caused by precession or other cyclical phenomena at the pulsar . these pulsars ( and especially the four pulsars in our `` class i '' ) are good candidates for further study . a handful of discontinuous changes were observed in the sample , which could be due to the passage of the line of sight across an interstellar region with significantly different magnetized properties . no linear variations ( i.e. , `` ramps '' ) of ppa across the timespan of observations were detected . for all 81 pulsars , we also calculate upper limits on random variations of ppa from the mean , which provide numerical constraints on precession and magetized interstellar variations in our sample . * appendix * at an observing session at epoch @xmath5 , we measure the position angle curves of target and calibrator pulsars as a function of longitude @xmath6 , @xmath7 and @xmath128 , respectively . the high @xmath54 position angle curves of @xcite , which are coherent sums over all sessions , serve as templates against which to measure shifts in position angle for a given pulsar on a particular observing session . the position angle curve of these target and calibrator templates as a function of longitude @xmath6 , is @xmath129 and @xmath130 , respectively . ( see fig . [ fig : obspas ] for examples of session and template position angle curves of a target and a calibrator pulsar ) . we first determine @xmath131 , the target pulsar s weighted , account for @xmath54 varying with longitude bin . ] , longitude - averaged difference between ppa at an observing session at epoch @xmath5 and ppa of its summed template ( see fig . [ fig : obspas]a for a target pulsar s session and template ppa curves ) : @xmath132 } { \sum_{r=1}^{r } w(\phi ) } , \label{eqn : deltapsipsr}\ ] ] where the summations are over @xmath133 longitude bins . similarly , we also determine @xmath134 , the calibrator pulsar s weighted@xmath114 , longitude - averaged difference between ppa at an observing session at epoch @xmath5 and ppa of its summed template ( see fig . [ fig : obspas]b for a calibrator pulsar s session and template ppa curves ) : @xmath135 } { \sum_{r=1}^{r } w(\phi ) } . \label{eqn : deltapsical}\ ] ] note then that the desired @xmath75 of a target pulsar is ultimately determined via a double - differencing procedure on the various ppas first both the target and calibrator pulsars ppas at epoch @xmath5 are differenced from their templates , as in eqs . [ eqn : deltapsipsr ] and [ eqn : deltapsical ] ; and finally the resulting target and calibrator epoch - template differences are differenced from one another , _ cf . _ eq . [ eqn : deltapsiwhole ] . as pointed out in [ sec : obsana ] ( see also eq . [ eqn : basicpsi ] ) , all of the above measured position angles contain an unknown instrumental origin @xmath74 . note that care was taken to measure both the target and the calibrator pulsar with an identical instrumental configuration and hence identical instrumental origin throughout a session at some epoch @xmath5 . therefore @xmath74 drops out of eq . [ eqn : deltapsiwhole ] due to the differencing of two terms , each containing this quantity . as a result , temporal _ variations _ in ppa are accessible even if the absolute ppas themselves are not . and @xmath140 with the summations being over all observing sessions . the horizontal solid line placed at @xmath141 in figs . [ fig : classifits ] and [ fig : classiifit ] then delineates the mean measured position angle in the new coordinate system defined by eq . [ eqn : deltahatpsi ] . the error bar on each individual measurement is @xmath142 , as discussed above in [ sec : offset ] and defined in eq . [ eqn : sigmapsiml ] . everett , j. e. , & weisberg , j. m. 2007 , the free precession of an oblate star , mathematica notebook on the mathsource library of the world wide web , http://library.wolfram.com/infocenter/mathsource/552/ lyne , a. g. , & graham - smith , f. 2006 , pulsar astronomy , 3rd ed . , by a.g . lyne and f. graham - smith . cambridge astrophysics series . cambridge , uk : cambridge university press , 2006 isbn 0521839548 , p. 264 . llcrccrcrrrrrr j0525 + 1115 & b0523 + 11 & & 1.92 & & 0.03 & 435 & 6532 & 1250 & 12.46 & 0.47 & 312.51 & 0.9968 & i + j0614 + 2229 & b0611 + 22 & & 1.01 & & 0.63 & 187 & 6546 & 313 & 2.14 & 1.17 & 10.73 & 0.9982 & i + j0629 + 2415 & b0626 + 24 & & 0.63 & & 0.14 & 228 & 5692 & 356 & 2.33 & 0.55 & 35.25 & 0.9997 & ii + j0659 + 1414 & b0656 + 14 & & 1.38 & & 0.47 & 285 & 7120 & 1047 & 2.47 & 0.85 & 20.24 & 0.9985 & i + j1136 + 1551 & b1133 + 16 & & 1.37 & & 0.75 & 234 & 7603 & 364 & 2.81 & 1.47 & 12.97 & 0.9987 & ii + j16070032 & b160400 & & 3.95 & & 0.48 & 324 & 5667 & 380 & 12.42 & 3.14 & 46.31 & 0.9948 & iii + j1740 + 1311 & b1737 + 13 & & 1.05 & & 0.69 & 106 & 6615 & 185 & 2.97 & 1.80 & 15.94 & 1.0000 & ii + j1805 + 0306 & b1802 + 03 & & 1.98 & & 1.70 & 114 & 7118 & 209 & 2.46 & 2.73 & 6.95 & 0.9980 & ii + j1823 + 0550 & b1821 + 05 & & 1.25 & & 0.90 & 143 & 7129 & 206 & 4.62 & 2.92 & 10.40 & 0.9995 & ii + j1844 + 1454 & b1842 + 14 & & 0.60 & & 0.36 & 158 & 7118 & 221 & 1.67 & 1.22 & 14.60 & 0.9999 & ii + j1857 + 0057 & b1854 + 00 & & 6.08 & & 4.86 & 150 & 7129 & 270 & 7.67 & 7.21 & 7.54 & 0.9974 & ii + j1915 + 1009 & b1913 + 10 & & 0.41 & & 0.25 & 190 & 7123 & 443 & 1.69 & 1.06 & 12.62 & 0.9993 & ii + j1916 + 0951 & b1914 + 09 & & 0.23 & & 0.06 & 265 & 6634 & 454 & 1.92 & 0.59 & 27.47 & 0.9993 & ii + j1920 + 2650 & b1918 + 26 & & 2.25 & & 0.26 & 345 & 5179 & 869 & 15.38 & 2.68 & 41.51 & 0.9764 & iii + j1932 + 2220 & b1930 + 22 & & 0.82 & & 0.44 & 168 & 7126 & 787 & 1.92 & 1.23 & 16.91 & 0.9999 & ii + j1954 + 2923 & b1952 + 29 & & 0.83 & & 0.39 & 158 & 7126 & 213 & 4.00 & 1.84 & 22.24 & 1.0000 & ii + j2022 + 2854 & b2020 + 28 & & 0.89 & & 0.47 & 160 & 7604 & 176 & 4.21 & 1.57 & 18.84 & 1.0000 & ii + j2037 + 1942 & b2034 + 19 & & 0.75 & & 0.49 & 186 & 6521 & 208 & 6.92 & 3.69 & 9.72 & 0.9974 & ii + j2055 + 2209 & b2053 + 21 & & 2.69 & & 1.42 & 201 & 6521 & 444 & 5.75 & 3.29 & 13.78 & 0.9990 & i + j2124 + 1407 & b2122 + 13 & & 2.13 & & 1.24 & 186 & 6034 & 289 & 8.95 & 4.66 & 11.50 & 0.9980 & ii + j2212 + 2933 & b2210 + 29 & & 0.70 & & 0.23 & 201 & 6045 & 310 & 3.65 & 1.29 & 26.25 & 0.9998 & ii + [ table : sinefits ] j0304 + 1932 & b0301 + 19 & 15.74 & 1 & -8.30 & 0.30 & 6 & 3.6 & 13 & 0 & 13 & 1.49 & 0.59 & 0.071 & 0.6 & 0 + j0525 + 1115 & b0523 + 11 & 79.34 & 1 & 35.00 & 3.00 & 7 & 3.6 & 6 & 0 & 6 & 3.73 & 1.48 & 0.042 & 1.9 & 0 + j0528 + 2200 & b0525 + 21 & 50.94 & 1 & -39.60 & 0.20 & 8 & 3.6 & 16 & 0 & 16 & 3.74 & 1.48 & 0.037 & 0.6 & 0 + j0614 + 2229 & b0611 + 22 & 96.91 & 1 & 69.00 & 2.00 & 6 & 3.6 & 14 & 0 & 14 & 1.75 & 0.69 & 0.010 & 1.0 & 0 + j0629 + 2415 & b0626 + 24 & 84.19 & 1 & 69.50 & 0.20 & 7 & 3.6 & 10 & 0 & 10 & 1.51 & 0.60 & 0.009 & 0.6 & 0 + j0659 + 1414 & b0656 + 14 & 13.98 & 1 & 23.50 & 0.40 & 7 & 3.9 & 10 & 0 & 10 & 1.88 & 0.74 & 0.032 & 1.4 & 0 + j0754 + 3231 & b0751 + 32 & 39.95 & 1 & -7.00 & 5.00 & 9 & 3.9 & 11 & 0 & 11 & 4.06 & 1.61 & 0.230 & 1.8 & 1 + j0826 + 2637 & b0823 + 26 & 19.45 & 1 & 5.90 & 0.30 & 6 & 3.9 & 8 & 0 & 8 & 1.56 & 0.62 & 0.105 & 0.6 & 0 + j0837 + 0610 & b0834 + 06 & 12.89 & 1 & 23.60 & 0.70 & 9 & 3.6 & 8 & 0 & 8 & 2.52 & 1.00 & 0.042 & 1.0 & 0 + j0922 + 0638 & b0919 + 06 & 27.27 & 1 & 32.00 & 2.00 & 9 & 3.9 & 10 & 0 & 10 & 2.08 & 0.82 & 0.026 & 1.6 & 0 + j0943 + 1631 & b0940 + 16 & 20.32 & 1 & 53.00 & 12.00 & 9 & 3.9 & 5 & 0 & 5 & 2.69 & 1.06 & 0.020 & 0.8 & 0 + j0953 + 0755 & b0950 + 08 & 2.96 & 1 & 1.35 & 0.15 & 10 & 3.9 & 10 & 0 & 10 & 1.71 & 0.68 & 0.501 & 1.2 & 0 + j1136 + 1551 & b1133 + 16 & 4.86 & 1 & 3.90 & 0.20 & 8 & 4.2 & 13 & 0 & 13 & 2.39 & 0.95 & 0.243 & 1.4 & 1 + j1239 + 2453 & b1237 + 25 & 9.24 & 1 & -0.33 & 0.06 & 10 & 3.8 & 9 & 0 & 9 & 1.87 & 0.74 & 2.243 & 0.8 & 0 + j1532 + 2745 & b1530 + 27 & 14.70 & 1 & 1.00 & 0.30 & 7 & 3.3 & 7 & 0 & 7 & 11.76 & 4.65 & 4.654 & 9.0 & 2 + j1543 + 0929 & b1541 + 09 & 35.24 & 1 & 21.00 & 2.00 & 9 & 3.1 & 8 & 1 & 7 & 3.41 & 1.35 & 0.064 & 1.3 & 0 + j1607 - 0032 & b1604 - 00 & 10.68 & 1 & 6.50 & 1.00 & 6 & 3.1 & 7 & 0 & 7 & 9.71 & 3.84 & 0.591 & 3.9 & 0 + j1614 + 0737 & b1612 + 07 & 21.39 & 1 & 40.00 & 4.00 & 9 & 3.6 & 11 & 0 & 11 & 6.50 & 2.57 & 0.064 & 1.8 & 1 + j1635 + 2418 & b1633 + 24 & 24.32 & 1 & 31.00 & 4.00 & 7 & 3.6 & 10 & 0 & 10 & 8.74 & 3.46 & 0.112 & 3.4 & 1 + j1740 + 1311 & b1737 + 13 & 48.67 & 1 & 64.40 & 1.60 & 7 & 3.6 & 26 & 1 & 25 & 2.44 & 0.97 & 0.015 & 1.1 & 0 + j1805 + 0306 & b1802 + 03 & 80.86 & 1 & 100.00 & 100.00 & & 3.9 & 25 & 0 & 25 & 3.21 & 1.27 & 0.013 & 2.0 & 1 + j1823 + 0550 & b1821 + 05 & 66.78 & 1 & 145.00 & 10.00 & 9 & 3.9 & 20 & 0 & 20 & 3.85 & 1.52 & 0.011 & 1.2 & 0 + j1825 + 0004 & b1822 + 00 & 56.62 & 1 & 21.00 & 13.00 & 7 & 3.1 & 5 & 0 & 5 & 7.40 & 2.93 & 0.139 & 1.3 & 0 + j1841 + 0912 & b1839 + 09 & 49.11 & 1 & 53.00 & 5.00 & 9 & 3.9 & 16 & 0 & 16 & 3.89 & 1.54 & 0.029 & 2.4 & 1 + j1844 + 1454 & b1842 + 14 & 41.51 & 1 & 121.00 & 8.00 & 9 & 3.9 & 18 & 0 & 18 & 1.79 & 0.71 & 0.006 & 0.6 & 0 + j1848 - 0123 & b1845 - 01 & 159.53 & 1 & 580.00 & 30.00 & 10 & 3.9 & 10 & 0 & 10 & 1.28 & 0.51 & 0.001 & 0.2 & 0 + j1851 + 0418 & b1848 + 04 & 115.54 & 1 & 100.00 & 100.00 & & 3.6 & 4 & 0 & 4 & 16.11 & 6.38 & 0.064 & 3.7 & 0 + j1851 + 1259 & b1848 + 12 & 70.61 & 1 & 158.00 & 16.00 & 11 & 3.9 & 12 & 0 & 12 & 5.27 & 2.09 & 0.013 & 1.4 & 0 + j1850 + 1335 & b1848 + 13 & 60.15 & 1 & 146.00 & 8.00 & 11 & 3.9 & 17 & 0 & 17 & 1.64 & 0.65 & 0.004 & 0.5 & 0 + j1856 + 0113 & b1853 + 01 & 96.74 & 1 & -140.00 & 30.00 & 12 & 3.6 & 14 & 0 & 14 & 12.48 & 4.94 & 0.035 & 3.4 & 2 + j1857 + 0057 & b1854 + 00 & 82.39 & 1 & 104.00 & 19.00 & 7 & 3.9 & 19 & 0 & 19 & 9.08 & 3.59 & 0.035 & 6.1 & 6 + j1901 + 0156 & b1859 + 01 & 105.39 & 1 & -122.00 & 9.00 & 12 & 3.9 & 8 & 0 & 8 & 8.14 & 3.22 & 0.026 & 1.9 & 1 + j1901 + 0331 & b1859 + 03 & 402.08 & 1 & -237.40 & 1.50 & 9 & 3.6 & 9 & 1 & 8 & 1.05 & 0.42 & 0.002 & 0.1 & 0 + j1903 + 0135 & b1900 + 01 & 245.17 & 1 & 72.30 & 1.00 & 9 & 3.6 & 10 & 0 & 10 & 1.94 & 0.77 & 0.011 & 0.4 & 0 + j1902 + 0556 & b1900 + 05 & 177.49 & 1 & -113.00 & 11.00 & 9 & 3.6 & 10 & 0 & 10 & 3.06 & 1.21 & 0.011 & 0.5 & 0 + j1902 + 0615 & b1900 + 06 & 502.90 & 1 & 100.00 & 100.00 & & 3.6 & 7 & 0 & 7 & 9.77 & 3.87 & 0.039 & 1.5 & 0 + j1905 - 0056 & b1902 - 01 & 229.13 & 1 & 100.00 & 100.00 & & 3.9 & 7 & 0 & 7 & 10.11 & 4.00 & 0.040 & 2.2 & 0 + j1906 + 0641 & b1904 + 06 & 472.80 & 1 & 100.00 & 100.00 & & 3.3 & 9 & 0 & 9 & 1.85 & 0.73 & 0.007 & 0.5 & 0 + j1909 + 0254 & b1907 + 02 & 171.73 & 1 & 100.00 & 100.00 & & 1.3 & 5 & 0 & 5 & 5.79 & 2.29 & 0.023 & 0.5 & 0 + j1910 + 0358 & b1907 + 03 & 82.93 & 1 & -127.00 & 7.00 & 9 & 3.9 & 17 & 0 & 17 & 6.53 & 2.58 & 0.020 & 1.1 & 0 + j1909 + 1102 & b1907 + 10 & 149.98 & 1 & 540.00 & 20.00 & 9 & 3.9 & 11 & 0 & 11 & 2.13 & 0.84 & 0.002 & 0.7 & 0 + j1912 + 2104 & b1910 + 20 & 88.34 & 1 & 148.00 & 10.00 & 9 & 2.5 & 7 & 0 & 7 & 5.39 & 2.13 & 0.014 & 0.5 & 0 + j1914 + 1122 & b1911 + 11 & 100.00 & 2 & 100.00 & 100.00 & & 3.9 & 9 & 0 & 9 & 7.58 & 3.00 & 0.030 & 2.1 & 0 + j1913 + 1400 & b1911 + 13 & 145.05 & 1 & 435.00 & 30.00 & 9 & 3.9 & 13 & 0 & 13 & 3.25 & 1.29 & 0.003 & 0.8 & 0 + j1915 + 1009 & b1913 + 10 & 241.69 & 1 & 431.00 & 22.00 & 9 & 3.9 & 15 & 0 & 15 & 1.61 & 0.64 & 0.001 & 0.4 & 0 + j1916 + 0951 & b1914 + 09 & 60.95 & 1 & 100.00 & 6.00 & 9 & 3.6 & 10 & 0 & 10 & 1.47 & 0.58 & 0.006 & 0.2 & 0 + j1916 + 1312 & b1914 + 13 & 237.01 & 1 & 280.00 & 15.00 & 9 & 3.9 & 12 & 0 & 12 & 2.97 & 1.18 & 0.004 & 1.1 & 0 + j1917 + 1353 & b1915 + 13 & 94.54 & 1 & 233.00 & 8.00 & 9 & 3.9 & 13 & 0 & 13 & 2.05 & 0.81 & 0.003 & 0.8 & 0 + j1918 + 1444 & b1916 + 14 & 27.20 & 1 & 100.00 & 100.00 & & 3.6 & 12 & 1 & 11 & 4.13 & 1.63 & 0.016 & 1.6 & 0 + j1919 + 0021 & b1917 + 00 & 90.31 & 1 & 120.00 & 7.00 & 9 & 2.8 & 5 & 0 & 5 & 12.19 & 4.82 & 0.040 & 2.2 & 0 + j1920 + 2650 & b1918 + 26 & 27.62 & 1 & 100.00 & 100.00 & & 2.8 & 6 & 0 & 6 & 7.91 & 3.13 & 0.031 & 2.2 & 0 + j1921 + 1419 & b1919 + 14 & 91.64 & 1 & 100.00 & 100.00 & & 3.9 & 9 & 0 & 9 & 4.93 & 1.95 & 0.020 & 1.3 & 0 + j1921 + 2153 & b1919 + 21 & 12.46 & 1 & -16.50 & 0.50 & 9 & 2.8 & 7 & 0 & 7 & 3.53 & 1.40 & 0.085 & 0.4 & 0 + j1922 + 2110 & b1920 + 21 & 217.09 & 1 & 282.00 & 14.00 & 9 & 2.5 & 7 & 0 & 7 & 10.13 & 4.01 & 0.014 & 2.2 & 0 + j1926 + 0431 & b1923 + 04 & 102.24 & 1 & 0.00 & 11.00 & 9 & 2.8 & 8 & 0 & 8 & 2.89 & 1.14 & & 0.3 & 0 + j1926 + 1648 & b1924 + 16 & 176.88 & 1 & 320.00 & 14.00 & 9 & 3.6 & 16 & 1 & 15 & 1.33 & 0.53 & 0.002 & 0.2 & 0 + j1932 + 2020 & b1929 + 20 & 211.15 & 1 & 10.00 & 6.00 & 9 & 2.3 & 4 & 0 & 4 & 10.28 & 4.07 & 0.407 & 1.4 & 0 + j1933 + 1304 & b1930 + 13 & 177.90 & 3 & 100.00 & 100.00 & & 3.6 & 5 & 0 & 5 & 10.09 & 3.99 & 0.040 & 2.4 & 0 + j1932 + 2220 & b1930 + 22 & 219.20 & 1 & 173.00 & 11.00 & 9 & 3.9 & 17 & 0 & 17 & 1.93 & 0.76 & 0.004 & 0.8 & 0 + j1935 + 1616 & b1933 + 16 & 158.52 & 1 & -1.90 & 0.40 & 8 & 3.6 & 8 & 0 & 8 & 0.69 & 0.27 & 0.144 & 0.1 & 0 + j1937 + 2544 & b1935 + 25 & 53.22 & 1 & 100.00 & 100.00 & & 3.9 & 11 & 0 & 11 & 1.25 & 0.49 & 0.005 & 0.2 & 0 + j1946 + 1805 & b1944 + 17 & 16.22 & 1 & -28.00 & 0.40 & 9 & 3.6 & 5 & 0 & 5 & 3.52 & 1.39 & 0.050 & 1.9 & 0 + j1946 + 2244 & b1944 + 22 & 140.00 & 4 & 2.00 & 20.00 & 7 & 2.3 & 4 & 0 & 4 & 7.29 & 2.89 & 1.443 & 0.4 & 0 + j1948 + 3540 & b1946 + 35 & 129.07 & 1 & 116.00 & 6.00 & 9 & 3.3 & 9 & 0 & 9 & 1.51 & 0.60 & 0.005 & 0.2 & 0 + j1952 + 3252 & b1951 + 32 & 45.01 & 1 & -182.00 & 8.00 & 7 & 3.6 & 11 & 0 & 11 & 6.55 & 2.59 & 0.014 & 2.8 & 1 + j1954 + 2923 & b1952 + 29 & 7.93 & 1 & -18.00 & 3.00 & 9 & 3.9 & 18 & 0 & 18 & 3.04 & 1.20 & 0.067 & 0.8 & 0 + j2004 + 3137 & b2002 + 31 & 234.82 & 1 & 30.00 & 6.00 & 9 & 3.6 & 15 & 0 & 15 & 2.89 & 1.14 & 0.038 & 0.3 & 0 + j2018 + 2839 & b2016 + 28 & 14.17 & 1 & -34.60 & 1.40 & 8 & 3.9 & 20 & 0 & 20 & 3.02 & 1.20 & 0.035 & 0.9 & 0 + j2022 + 2854 & b2020 + 28 & 24.64 & 1 & -74.70 & 0.30 & 6 & 4.2 & 19 & 0 & 19 & 2.42 & 0.96 & 0.013 & 0.9 & 0 + j2030 + 2228 & b2028 + 22 & 71.83 & 1 & -192.00 & 21.00 & 7 & 3.6 & 10 & 0 & 10 & 4.58 & 1.81 & 0.009 & 1.2 & 0 + j2037 + 1942 & b2034 + 19 & 36.00 & 5 & -97.00 & 10.00 & 7 & 3.6 & 14 & 0 & 14 & 5.38 & 2.13 & 0.022 & 0.7 & 0 + j2037 + 3621 & b2035 + 36 & 93.56 & 1 & 100.00 & 100.00 & & 3.6 & 7 & 0 & 7 & 4.69 & 1.86 & 0.019 & 1.4 & 0 + j2046 + 1540 & b2044 + 15 & 39.84 & 1 & -100.00 & 5.00 & 7 & 3.6 & 12 & 0 & 12 & 4.48 & 1.77 & 0.018 & 1.0 & 0 + j2055 + 2209 & b2053 + 21 & 36.36 & 1 & -80.50 & 3.00 & 7 & 3.6 & 14 & 1 & 13 & 5.45 & 2.16 & 0.027 & 2.7 & 1 + j2055 + 3630 & b2053 + 36 & 97.31 & 1 & -68.00 & 4.00 & 9 & 3.6 & 13 & 0 & 13 & 2.55 & 1.01 & 0.015 & 1.4 & 1 + j2113 + 2754 & b2110 + 27 & 25.11 & 1 & -37.00 & 7.00 & 7 & 3.6 & 15 & 0 & 15 & 4.88 & 1.93 & 0.052 & 1.2 & 0 + j2116 + 1414 & b2113 + 14 & 56.15 & 1 & -25.00 & 8.00 & 9 & 3.6 & 12 & 0 & 12 & 7.57 & 3.00 & 0.120 & 2.2 & 1 + j2124 + 1407 & b2122 + 13 & 30.12 & 1 & -57.00 & 8.00 & 7 & 3.6 & 13 & 0 & 13 & 7.31 & 2.89 & 0.051 & 2.1 & 0 + j2212 + 2933 & b2210 + 29 & 74.50 & 1 & -168.00 & 5.00 & 7 & 3.3 & 12 & 0 & 12 & 2.80 & 1.11 & 0.007 & 0.7 & 0 + j2305 + 3100 & b2303 + 30 & 49.54 & 1 & -75.50 & 4.00 & 7 & 3.3 & 10 & 0 & 10 & 3.57 & 1.41 & 0.019 & 1.3 & 0 + j2317 + 2149 & b2315 + 21 & 20.91 & 1 & -37.00 & 3.00 & 9 & 3.3 & 8 & 0 & 8 & 2.79 & 1.10 & 0.030 & 0.7 & 0 + [ table : big ] j1543 + 0929 & b1541 + 09 & 48870.4 & -31.4 & 3.0 & -10.5 & & -12.4 & -0.590 + j1740 + 1311 & b1737 + 13 & 48678.1 & 7.8 & 1.9 & 4.1 & & 3.1 & 0.048 + j1901 + 0331 & b1859 + 03 & 49040.5 & 27.4 & 3.8 & 7.2 & & 10.9 & 0.046 + j1918 + 1444 & b1916 + 14 & 48675.1 & -11.0 & 2.5 & -4.4 & & -4.4 & -0.044 + j1926 + 1648 & b1924 + 16 & 49159.4 & -20.8 & 2.8 & -7.4 & & -8.2 & -0.026 + j2055 + 2209 & b2053 + 21 & 48466.8 & 18.5 & 3.2 & 5.8 & & 7.3 & -0.091 + [ table : outliers ]

we measure the clustering of galaxy groups in the 2dfgrs percolation - inferred galaxy group ( 2pigg ) catalogue . the 2pigg sample has 28877 groups with at least two members . the clustering amplitude of the full 2pigg catalogue is weaker than that of 2dfgrs galaxies , in agreement with theoretical predictions . we have subdivided the 2pigg catalogue into samples that span a factor of in median total luminosity . our correlation function measurements span an unprecedented range of clustering strengths , connecting the regimes probed by groups fainter than galaxies and rich clusters . there is a steady increase in clustering strength with group luminosity ; the most luminous groups are ten times more strongly clustered than the full 2pigg catalogue . we demonstrate that the 2pigg results are in very good agreement with the clustering of groups expected in thecdm model . galaxies : haloes , galaxies : clusters : general , large - scale structure of universe

astro-ph0402577

galaxy groups are important tracers of the matter distribution in the universe , offering a powerful alternative to individual galaxies . part of the appeal of galaxy groups lies in the simplicity of their relation to the dark matter . theoretically , each dark matter halo yields a single group of galaxies . this is a clear advantage over the use of galaxies to trace mass , as the occupation of haloes by galaxies depends upon halo mass ( e.g. benson et al . 2000 ; berlind et al . typical groups are less elitist than the richest clusters which only account for a few percent of the mass of the universe ( e.g. jenkins et al . 2001 ) . the measurement of the clustering amplitude of groups as a function of their mass is an important probe of hierarchical clustering ( governato et al . 1999 ; colberg et al . 2000 ; padilla & baugh 2002 ) . previous attempts to measure the clustering of groups have been hampered by the small size of the samples . for example , jing & zang ( 1988 ) measured the redshift space correlation function of a sample of 135 groups extracted from the cfa survey by geller & huchra ( 1983 ) . by the end of the 1990s , this situation had only improved slightly with an analysis of @xmath3 groups in the updated zwicky catalogue by merchn , maia & lambas ( 2000 ) . perhaps unsurprisingly , these early studies were inconclusive . one dispute concerned the relative strength of the auto - correlation function of groups and of galaxies . kashlinsky ( 1987 ) produced an analytic argument suggesting that a sample of groups corresponding to haloes less massive than a characteristic mass @xmath4 ( see section 2.3 for a definition of @xmath4 ) should display a weaker clustering signal than that of their constituent galaxies . similar conclusions can be reached from the calculation of the bias of dark matter haloes first carried out by cole & kaiser ( 1989 ) , and subsequently by mo & white ( 1996 ) and sheth , mo & tormen ( 2001 ) . however , the first results comparing the clustering of groups and galaxies both confirmed ( jing & zang 1988 ; maia & da costa 1990 ) and contradicted ( ramella , geller & huchra 1990 ; girardi , boschin & da costa 2000 ) this theoretical prediction . an accurate measurement of the clustering of groups is now particularly timely as theoretical models have advanced to the point where detailed predictions can be made for the clustering of galactic systems . techniques have been developed recently that allow high resolution n - body simulations of hierarchical clustering to be populated with galaxies using semi - analytic models ( kauffmann et al . 1999 ; benson et al . 2000 ; helly et al . the models make firm predictions for how the luminous baryonic component of the universe is partitioned between dark matter haloes . a significant advance in statistical studies of group properties was made possible by the two - degree field galaxy redshift survey ( 2dfgrs ; colless et al . 2001 ; 2003 ) . merchn & zandivarez ( 2002 ) used the 2dfgrs 100k data release to construct a group catalogue , the 2dfggc , consisting of 2198 groups with at least four members . the correlation function of the 2dfggc in redshift space can be described by a power law with correlation length @xmath5mpc and slope @xmath6 ( zandivarez , merchn & padilla 2003 ) . the clustering amplitude of 2dfggc groups is significantly higher than that measured for 2dfgrs galaxies , for which @xmath7mpc ( hawkins et al . 2003 ) . the completion of the 2dfgrs has made it possible to construct a much larger group catalogue , the 2dfgrs percolation inferred galaxy group catalogue ( hereafter the 2pigg catalogue ; eke et al . the gain is bigger than the simple factor of just over @xmath8 increase in the number of galaxy redshifts ; the improved homogeneity and higher spectroscopic completeness means that group finding is far more efficient in the final 2dfgrs catalogue . to quantify this improvement , the number of groups with a minimum of four members in the 2pigg catalogue is 7200 , over three times the number of comparable groups detected by merchn & zandivarez ( 2002 ) . moreover , eke et al . demonstrate that the group finding algorithm can be pushed further still to extract groups containing two or more members , which greatly increases the number of groups . in this paper , we study the clustering of groups in the 2pigg catalogue . the properties of the catalogue are summarized in section 2 , along with an explanation of how we construct subsamples of groups defined by mass . the key role played by mock catalogues of groups is also outlined in section 2 . the estimation of the correlation function of groups is set out in section 3 . a number of assumptions need to be made in order to measure the clustering of groups and these issues are dealt with in section 4 , in which we use mock catalogues extensively to quantify the likely random and systematic errors on our measurements . the clustering measurements for 2dfgrs groups are presented in section 5 . a summary of our conclusions is given in section 6 . ccccccccc sample & luminosity cut & median lum . & @xmath9 & @xmath10 & median mass & number & @xmath11 & @xmath12 + i d & @xmath13 & @xmath14 & mean[median ] & ( kms@xmath15 ) & @xmath16 & of groups & ( @xmath17mpc ) & ( @xmath17mpc ) + @xmath18 & all & 1.8 & @xmath19 $ ] & @xmath20 & 5.0 & @xmath21 & @xmath22 & @xmath23 + @xmath24 & @xmath25 & 3.0 & @xmath26 $ ] & @xmath27 & 9.4 & @xmath28 & @xmath29 & @xmath30 + @xmath8 & @xmath31 & 5.5 & @xmath32 $ ] & @xmath33 & 22 & @xmath34 & @xmath35 & @xmath36 + @xmath37 & @xmath38 & 12 & @xmath39 $ ] & @xmath40 & 64 & @xmath41 & @xmath42 & @xmath43 + @xmath44 & @xmath45 & 19 & @xmath46 $ ] & @xmath47 & 110 & @xmath48 & @xmath49 & @xmath50 + @xmath51 & @xmath52 & 30 & @xmath53 $ ] & @xmath54 & 200 & @xmath55 & @xmath56 & @xmath57 + @xmath58 & @xmath59 & 42 & @xmath60 $ ] & @xmath61 & 310 & @xmath62 & @xmath63 & @xmath64 + [ tab : subsamples ] in this section we describe the catalogue from which measurements of the clustering of groups are made , the role played by mock datasets in our analysis and the definition of subsamples of groups . we use the 2pigg catalogue constructed by eke et al . ( 2004a ) . the 2pigg sample is the largest and most homogenous group catalogue currently available and therefore provides a unique tool with which to study the clustering of galactic systems . the properties of the 2pigg are outlined in section [ ssec:2pigg ] . an important and novel feature of the eke et al . analysis is the extensive use of realistic , physically motivated mock catalogues to calibrate the performance of the group finding algorithm and to understand how the recovered groups relate to the underlying distribution of dark matter . we utilize these catalogues to assess possible systematic effects in the measurement of clustering from the 2pigg sample and to estimate the errors on our results ; a brief outline of the mocks is given in section [ ssec : mocks ] . we present the rationale behind our definition of subsamples extracted from the 2pigg in section [ ssec : subsamples ] . finally , the choice of relation between an observed group property and the underlying group mass is justified in section [ ssec : mass ] . the 2pigg catalogue was constructed from the completed 2dfgrs as described by eke et al . ( 2004a ) . the group finding algorithm requires a small number of parameters to be set . the motivation for the adopted parameter values and the consequences of these choices for the accuracy and completeness of the catalogue are discussed at length by eke et al . ( 2004a ) . eke et al . find that after applying the identification algorithm to the 2dfgrs ( which contains @xmath65 galaxies in the two large contiguous regions ) , @xmath66 percent of the galaxies are grouped into @xmath67 groups containing at least two members . these groups have a median velocity dispersion of @xmath68 ; if groups of four or more members are considered , this value changes to @xmath69 . in both cases , the median redshift is @xmath70 . the 2pigg catalogue is sufficiently large that it may be divided into subsamples in order to measure trends in clustering strength with group properties ; the construction of subsamples from the 2pigg catalogue is set out in section [ ssec : subsamples ] . two types of mock catalogue are used in this paper : semi - analytic mocks and hubble volume mocks . these mocks have different underlying physics and play different roles in our analysis . * _ semi - analytic mocks_. we use the mock 2pigg catalogues constructed by eke et al . ( 2004a ) . these mocks are produced from a high resolution n - body simulation which is populated with galaxies using the semi - analytic model , galform ( cole et al . 2000 ; benson et al . 2002 , 2003 ) . the simulation uses standard @xmath2cdm parameters , with the normalization of the density fluctuations given by @xmath71 and a primordial spectral index of @xmath72 , in line with the constraints from the first year of data from wmap ( spergel et al . the n - body simulation box is @xmath73mpc on a side and contains @xmath74 dark matter particles each of mass @xmath75 . we use the @xmath76 simulation output , which is close to the median redshift of the 2dfgrs . dark matter haloes are identified using a friends - of - friends algorithm with a linking length of @xmath77 times the mean interparticle separation ( davis et al . + the halo resolution limit is taken to be @xmath78 . the galform code is run for each halo and galaxies are assigned to a subset of dark matter particles in the halo following the technique described by benson et al . ( 2000 ) . around @xmath79 of central galaxies brighter than @xmath80 are expected to be in haloes resolved by the simulation . the effective limit of the catalogue is extended to @xmath81 using a separate galform calculation for a grid of halo masses below the resolution limit of the n - body simulation . these galaxies are assigned to particles that are not identified as part of a resolved dark matter halo ; such particles account for approximately @xmath82 of the dark matter in the @xmath76 simulation output . the luminosity function predicted by the semi - analytic model is close to that estimated for the 2dfgrs ( norberg et al . 2002b ) : however , we force the luminosity function of the model galaxies to agree with the 2dfgrs luminosity function by rescaling the luminosity of the model galaxies . a mock 2dfgrs is then extracted by placing an observer at a random location within the simulation cube and applying the radial and angular selection function of the 2dfgrs ( norberg et al . the simulation box needs to be replicated several times to cover the full 2dfgrs volume and geometry . for our absolute magnitude limit , the mock 2dfgrs is complete above @xmath83 . the mock 2dfgrs is then degraded by applying the redshift completeness mask , and the magnitude and velocity errors of the 2dfgrs ( norberg et al . the eke et al . group finding algorithm is applied to the mock 2dfgrs catalogue in exactly the same way and with the same parameter values as for the real data . we will henceforth refer to the resulting mock 2pigg catalogue as the galform mock . + the galform mock has three roles to play in our analysis : ( 1 ) to determine which of the observed characteristics of 2pigg systems provides the most reliable estimate of the true mass of the underlying dark matter halo in which the group resides . ( 2 ) to assess the systematic errors in our clustering estimates , by comparing measurements from a 2pigg mock with the measurement for an equivalent sample of galaxy groups extracted from the simulation cube ( i.e. before applying the 2dfgrs selection criteria and mask ) . this allows us to quantify how well our clustering estimator compensates for the selection function of the survey and its redshift incompleteness . the results of these comparisons are used to define sample selection criteria that , when applied to the 2pigg catalogue , will yield the most robust and reliable clustering measurements . the construction of the equivalent samples is outlined below . ( 3 ) to cast the predictions of the @xmath2cdm model for the clustering of galactic systems in a form that can be confronted directly with the observations . + finally , we explain the criteria used to construct samples of groups from the n - body simulation cube that can be compared directly with the groups taken from the galform mock ; these samples will be denoted as equivalent samples . we will employ two approaches to construct equivalent samples . in the first approach , the spatial abundance of groups identified in the galform mock is reproduced in the simulation cube by selecting an appropriate fraction of the groups at each mass . the second method is in two stages . first , an effective bias is computed for the groups in the galform mock , using eq . [ eq : beff ] below , and the true mass of each group . second , groups more massive than some lower mass limit in the simulation cube are selected , such that the effective bias of these groups matches that computed for the mock 2pigg sample in the first stage . the correlation functions measured from these equivalent samples are compared with direct measurements from the galform mock in section [ ssec : xissample ] . + .3 cm * _ hubble volume mocks_. eke et al . ( 2004a ) also made use of mock 2dfgrs catalogues extracted from the @xmath2cdm hubble volume simulation ( jenkins et al . 2001 ; evrard et al . 2002 ) . in this case , the `` galaxies '' are dark matter particles that are selected in order to have a clustering strength similar to that measured for galaxies in the flux limited 2dfgrs ( hawkins et al . the selection is based upon the smoothed density of the dark matter distribution ( cole et al . 1998 ; see also norberg et al . the large volume of the hubble volume simulation ( @xmath84 ) allows a high number of independent mock versions of the 2dfgrs to be extracted ; the ensemble that we consider contains 22 2dfgrs mocks . the 2pigg group finding algorithm is run on these hubble volume mocks to produce a set of groups in each mock . the primary aim of these mocks is to provide an estimate of the error on measured clustering statistics ; these errors naturally include the contribution from sample variance due to large - scale structure . we compute the _ rms _ scatter over the ensemble of 22 mocks in the manner described by norberg et al . ( 2001 ) . to recap , the correlation function measured from one of the mocks is considered as the `` mean '' and the scatter of the remaining mocks around this mean is computed . this process is repeated for each mock in turn . the _ rms _ scatter is the resulting mean scatter . this procedure gives an _ rms _ scatter that is larger than the formal variance over the ensemble of mocks , since , to some extent , it takes into account the covariance between the correlation function measurements in different bins . the fractional _ rms _ scatter obtained from the mocks is taken as an estimate of the statistical error and is applied to the measured correlation functions . one could consider dividing the 2pigg catalogue into either cumulative or differential bins in a property related to group mass . naively , one might anticipate that a division into differential mass bins would give a cleaner trend of clustering amplitude varying with increasing sample mass , since the clustering signal from cumulative samples might be dominated by the most massive objects . to investigate this prejudice , we use theoretical models of the clustering of dark matter haloes and ascertain which of these two alternatives is the best way to split up the 2pigg catalogue . the theoretical predictions for halo clustering are based upon the formalism developed by cole & kaiser ( 1989 ) and mo & white ( 1996 ; hereafter mw ) . these authors computed an asymptotic bias for dark matter haloes , using extended press & schechter ( 1974 ) theory and the spherical collapse model . sheth , mo & tormen ( 2001 ; hereafter smt ) improved upon this calculation by incorporating an ellipsoidal collapse model . both approaches have been tested extensively against direct predictions from n - body simulations ( governato et al . 1999 ; colberg et al . 2000 ; smt ; padilla & baugh 2002 ) . = 8.truecm the effective bias for a sample of dark matter haloes is computed by weighting the bias factor associated with each individual halo , @xmath85 , by its abundance in the sample ( e.g. baugh et al . 1999 ; padilla & baugh 2002 ) : @xmath86 where @xmath87 is the space density of haloes in the mass interval @xmath88 to @xmath89 . in the case of cumulative mass bins , the lower mass limit , @xmath90 , defines the sample , as @xmath91 . [ fig : bias ] shows the theoretical predictions for the effective bias as a function of the median mass of the sample . for differential mass samples , an interval of one decade in mass is assumed . to make these predictions , we have adopted the linear theory power spectrum used in the n - body simulation described in section [ ssec : mocks ] . the clustering amplitude of the overall dark matter distribution corresponds to @xmath92 . the mw theory predicts that haloes with mass below a characteristic value @xmath93 ( defined as the mass contained within a sphere for which the linear theory variance is equal to the threshold for collapse ) will display a weaker clustering signal than the overall mass distribution , i.e. these haloes will have @xmath94 . the dynamic range in mass of the 2pigg samples that we consider is @xmath95h@xmath96 to @xmath97h@xmath96 . over this interval , the theoretical predictions for cumulative and differential mass bins have similar shapes , so there is no clear advantage in using differential mass bins . moreover , the cumulative samples benefit from better statistics . for these reasons , we will subdivide the 2pigg sample using cumulative bins in a group property related to mass ( see section [ ssec : mass ] ) . = 8.5truecm our goal in this paper is to estimate the clustering of subsamples of the 2pigg catalogue , defined by group mass . we therefore need to find an observed property of the 2piggs that displays a tight correlation with the underlying or `` true '' group mass . following eke et al . ( 2004a ) , we have considered two possibilities : ( i ) a dynamical mass estimate that depends upon the _ rms _ radius of the group and its estimated velocity dispersion . ( ii ) an indirect estimate in which the corrected total group luminosity is used to infer a mass from the estimated median mass to light ratio of 2piggs . from the galform mock catalogue , it is possible to find the true " mass of any group by associating it with a dark matter halo in the simulation box . the group s true " mass is simply the mass of the halo , as computed by summing the particles identified as members of the halo by a friends - of - friends group finding algorithm . [ fig : ml ] shows the relation between corrected total group luminosity and the true group mass for the galform mock . the total group luminosity is obtained by summing the luminosity of the galaxies identified as members of each group , and then applying a correction to account for missing galaxies that are fainter than the apparent magnitude limit of the 2dfgrs . this correction factor is redshift dependent and reaches a factor of @xmath98 at @xmath99 . in fig . [ fig : ml ] , we only show groups with @xmath100 . the shading represents the abundance of groups for a given total luminosity and true mass ; the darker shading indicates a higher number density of groups . the solid line shows the median true mass to total luminosity relation , and the dashed lines enclose @xmath101 of the groups in each mass bin . this relation is reasonably tight over more than two decades in true mass . the equivalent comparison between the dynamical mass estimate and the true mass exhibits much more scatter ( see eke et al . the dynamical mass estimator is a particularly poor indicator of true mass for objects less massive than @xmath102 . we therefore choose to employ the corrected total group luminosity as the indicator of the true group mass . however , caution is advisable when dealing with very massive groups , since , as is apparent in fig . [ fig : ml ] , the interval containing @xmath101 of the groups with masses @xmath103 broadens considerably . this is due to the fact that such large groups , because of their rarity , are more likely to be found at higher redshifts , and therefore require a larger correction to be applied to the observed group luminosity to infer the total group luminosity . = 8.5truecm further evidence supporting our choice of the total group luminosity as a mass indicator is given in fig . [ fig : massdist ] . here , we show the true " mass distributions of samples from the galform mock selected according to two criteria : ( i ) cumulative bins in total luminosity ( top panel ) , and ( ii ) cumulative bins in dynamical mass ( lower panel ) . the sample definitions are given by the key in each panel . these distributions are for groups with at least @xmath8 members . the different lines in each panel show the corresponding distributions of true group mass for each subsample . the points with error bars show the median and the range containing @xmath104 of the groups for the true mass distributions . it is clear from this plot that the distribution of true mass for samples selected by total group luminosity spans a larger dynamic range in mass , and the distributions have tighter percentile intervals . we have performed the same exercise for groups with a minimum of four members and reach similar conclusions . however , we note that in the case of groups with 4 or more members , there are fewer low mass groups as expected . we now investigate how the clustering signal from group samples varies with the minimum number of group members , @xmath105 , looking at the cases where @xmath106 or @xmath44 . we compute an effective bias for each sample using eq . [ eq : beff ] . in fig . [ fig : bias ] , the effective bias is plotted at a representative mass for each sample ( indicated by the lines which intersect the bias curves and run parallel and perpendicular to the mass axis ) . this mass is obtained using the galform mocks by finding the halo mass above which the effective bias of haloes in the simulation cube matches that of the haloes recovered from the mock group catalogue . we show the effective bias for the brightest and faintest cumulative luminosity bins to illustrate the dynamic range of our clustering measurements . the dashed lines show the results of this calculation for @xmath106 and the solid lines show the case where @xmath107 . the range of effective masses ( x - axis ) and clustering strengths ( y - axis ) probed by the different samples is widest when considering groups with @xmath106 . we now outline the method followed to measure the clustering of the 2pigg samples . in a sample of groups extracted from a flux limited galaxy redshift survey , the space density of groups is a function of radial distance . an accurate estimate of the radial selection function of the sample is required to make a robust measurement of the clustering of groups . the procedure that we follow to obtain the selection function from the redshift distribution of groups is set out in section [ ssec : nofz ] . the weighting scheme used to approximate a minimum variance estimate is given in section [ ssec : xi ] , along with the estimator used to infer the correlation function . an accurate estimate of the selection function of groups in the 2pigg samples is required to permit an analysis of the clustering signal of the groups . the most straightforward way of doing this is to use the luminosity function of groups to obtain an estimate of the selection function which is unaffected by large scale structure . however , eke et al . ( 2004c ) show , using the galform mock , that estimates of the group luminosity function are affected by errors in the determination of total group luminosity and by contamination , leading to systematic errors in the number density of groups of up to a factor of @xmath44 . the alternative is to estimate the selection function directly from the observed redshift distribution . the concern in this case is that the redshift distribution displays features that are due to large scale structure ( see the histogram in fig . [ fig : nz ] ) . care must be taken when fitting a parametric form to the observed redshift distribution to avoid ` over - fitting ' the peaks and troughs , thereby inadvertently removing some of the clustering signal from the estimated correlation function . with this concern in mind , we explore two procedures to describe the shape of the observed redshift distribution : fitting parametric forms or smoothing the observed redshift distribution . we will explore the influence on the measured correlation function of these different ways of fitting the redshift distribution in section [ ssec : optimal ] . we adopt a fit to @xmath108 , the observed 2pigg redshift distribution , of the form @xmath109 where @xmath110 and @xmath111 are parameters set by minimising the quantity @xmath112 alternatively , we have experimented with including a weight proportional to the volume of each redshift bin of width @xmath113 , @xmath114 the effect on the best fit parameter values of weighting by volume is small . from now on , best fit parameter values @xmath110 and @xmath111 obtained using eq . [ eq : chi ] ( and the redshift distribution constructed using them ) will be referred to as unweighted fits , while those found using eq . [ eq : chiv ] will be referred to as volume - weighted fits . the resulting best fits for the full 2pigg catalogues are shown in fig . [ fig : nz ] , as indicated by the key in the upper panel . the parametric form adopted for the fit is particularly attractive since it can be integrated analytically to give the cumulative redshift distribution @xmath115 the smoothed redshift distribution is obtained by smoothing the observed redshift distribution , which is tabulated in bins of width @xmath116 , with a top - hat window in redshift . the smoothed redshift distribution in fig . [ fig : nz ] was produced by smoothing the observed distribution with a top - hat of width @xmath117 . = 8.truecm to obtain a minimum variance estimate of the correlation function , each group is assigned a weight given by ( efstathiou 1988 ) : @xmath118 where we assume @xmath119h@xmath120mpc@xmath121 , @xmath122 is the angular completeness of the 2dfgrs and @xmath123 is the space density of groups at redshift @xmath124 . in section [ ssec : optimal ] we explore the impact of the choice of @xmath125 on the measured correlation function . the space density of groups , @xmath126 , is obtained by integrating over the product of the cosmological volume element and the adopted description of the redshift distribution ; as explained above , this could be a smoothed version of the redshift distribution or a parametric fit . a catalogue of unclustered points is produced with the same radial and angular selection functions as the data . the number of random points is set to be @xmath127 times the number of data points , with a lower limit of @xmath128 . in the case of the unclustered catalogue , the weights are scaled in order to ensure that their sum matches that of the weights of the observed groups : @xmath129 where @xmath130 and @xmath131 indicate unclustered points and data groups respectively . we adopt the landy & szalay ( 1993 ) estimator for calculating the redshift - space correlation function , @xmath132 where @xmath133 with @xmath134 and @xmath135 representing the actual groups ( @xmath131 ) and/or points in the random catalogue ( @xmath130 ) ; @xmath136 and @xmath137 are the indices for each individual group / random point , and @xmath138 corresponds to the weight defined above . we use the same estimator to obtain the correlation function as a function of pair separation perpendicular ( @xmath139 ) and parallel ( @xmath140 ) to the line - of - sight , @xmath141 . from this correlation function , we can define the projected correlation function @xmath142 this quantity is free from redshift - space effects arising from peculiar motions and can be related directly to the real space correlation function ( e.g. norberg et al . ( 450,580 ) ( 0,275 ) ( 245,275 ) ( 0,-10 ) ( 245,-10 ) the measurement of the clustering of groups requires a number of choices to be made that can have an impact upon the estimated correlation function . in this section , we carry out a careful examination of our method for measuring the clustering of the mock groups , with the goal of determining a correlation function that is accurate , i.e. with minimal systematic errors , and with random measurement errors that are as small as possible . the mock catalogues described in section [ ssec : mocks ] play a key role in this exercise . the galform mock is used to assess systematic errors by comparing the correlation functions measured from the semi - analytic mock group catalogue with the correlation function of an equivalent sample in the simulation cube ( the construction of such a catalogue is described in section [ ssec : mocks ] ) . we stress that the goal of quantifying the systematic errors is to devise a clustering measurement algorithm that minimizes these errors rather than to correct for them . the hubble volume mocks are used to estimate the random measurement errors , which automatically include the contribution of sample variance from large scale structure . an important test for systematic errors in our clustering measurement is to compare the correlation function recovered from a mock group catalogue with that of a comparable , equivalent sample drawn from the distribution of groups in the simulation cube . the rest of this section is split into two parts . in the first , we present a systematic study of how various assumptions and approximations affect the recovered correlation function . we focus attention on the size of the random measurement errors and on minimizing systematic errors in the estimated correlation function . in the second part of this section , we build on the conclusions of the first part and apply an optimal method to measure the clustering of mock 2piggs to determine how well we can recover trends in clustering amplitude for different subsamples of groups . the main results of this subsection are presented in fig . [ fig : err ] . each plot in this figure is divided into two panels . the top panel in all cases compares the correlation function recovered from a galform mock group catalogue with the correlation function of an equivalent sample , drawn from the simulation cube , of groups generated by galform . the errors on this ratio are the _ rms _ scatter over the ensemble of 22 hubble volume group mocks . note that this estimate of errors includes cosmic variance , and therefore overestimates the error bars in a comparison between results from the mock and the simulation cube , since both come from the same simulated volume . in the absence of systematic errors in the clustering measurement , this ratio would be consistent with unity , indicated by the horizontal line . the lower panel in each plot shows the logarithm of the relative error on the measured correlation function , as deduced from the _ rms _ scatter over the hubble volume mocks . the horizontal line indicates an error of @xmath143 . to improve the clarity of the presentation , we have used a power law fit to the correlation function for separations larger than @xmath144 in the definition of the relative error , which avoids dividing by small or negative numbers . the correlation function of the equivalent sample of groups falls below a power law on large scales , typically @xmath145 , so the plotted error is a lower limit to the true relative error on these scales . we now go through the set of choices or assumptions that needs to be made in our clustering measurement method , directing the reader to fig . [ fig : err ] where appropriate . \(i ) _ choice of 2dfgrs region . _ the differences between the correlation functions measured in the ngp and sgp regions provide a crude estimate of the errors ; these differences are comparable to the errors inferred from the hubble volume mocks . we find that the correlation functions measured in mock ngp and sgp 2dfgrs regions are consistent within the errors . combining the pair counts of groups in the two regions reduces the poisson noise and sample variance in the clustering measurement . henceforth , our results will be for combined ngp and sgp samples . \(ii ) _ choice of @xmath125 in the radial weight . _ the weight assigned to each group to compensate for the radial selection function requires a value for the integral of the two - point correlation function , @xmath146 ( eq . [ eq : w ] ) . we have explored using constant values in the range @xmath147h@xmath120mpc@xmath148 , and find that the changes in the estimated correlation function are small and within the statistical errors . we adopt @xmath149h@xmath120mpc@xmath121 , which corresponds to the typical asymptotic value of @xmath146 for separations @xmath150h@xmath15mpc in the hubble volume mocks . \(iii ) _ sample definition . _ fig . [ fig : err](a ) compares the correlation function measured for samples defined in different ways : by setting ( i ) a lower limit on total group luminosity , @xmath151h@xmath152 ( solid line ) , ( ii ) a lower limit on dynamical mass , @xmath153 ( dotted line ) , which , for the median mass - to - light ratio of the 2piggs ( eke et al . 2004b ) , corresponds to the luminosity cut in ( i ) , and ( iii ) a differential luminosity sample defined by @xmath154h@xmath155 ( dashed line ) . the correlation function measured for the sample of groups brighter than the luminosity threshold in ( i ) has the smallest systematic and random errors . this adds further weight to the preference arrived at for cumulative bins in luminosity in section [ ssec : mass ] . the three remaining panels , fig . [ fig : err](b),(c ) and ( d ) show the errors on @xmath156 for samples of mock groups with @xmath151h@xmath152 ( shown by solid lines ) , and , in some cases for a brighter sample with @xmath157 ( indicated by dashed lines ) . \(iv ) _ minimum number of group members_. we compare the accuracy of the correlation function for groups with a minimum number of members of @xmath106 ( solid lines ) and @xmath107 ( dashed lines ) in fig . [ fig : err](b ) . the systematic errors on the correlation functions recovered from the mocks are roughly comparable in the two cases . the random errors , however , are smaller for the case of @xmath106 . this is expected since this sample contains more groups . we recall that samples of groups with @xmath106 span a wider range in mass ( see fig [ fig : massdist ] in section [ ssec : mass ] ) . the dashed lines in fig . [ fig : err](b ) show the results for a brighter sample of groups , as explained at the end of ( iii ) above . this sample contains fewer groups and so the random errors are somewhat larger than before , particularly at small separations . the systematic errors in the correlation function recovered from the mock 2pigg also become significant below @xmath158mpc for the brighter sample . = 8.truecm \(v ) _ selection function model . _ the impact on the correlation function of using different approximations to the form of the redshift distribution of groups to model their selection function is shown in fig . [ fig : err](c ) . the solid line shows how well the correlation function can be measured using a parametric fit to the redshift distribution of groups in the mock , as described in section [ ssec : nofz ] ; the dotted line shows the results when a volume weighting is applied to determine the best fit parameters . on small scales , @xmath159mpc , the systematic and random errors are very similar in these two cases . on larger scales however , the unweighted fit results in a more faithful and less noisy measurement of the correlation function . the evidence in this panel shows that the approach of using the smoothed redshift distribution of groups to model their selection function ( dashed line ) is clearly flawed , leading to a significantly discrepant estimate of the correlation function for separations @xmath160mpc . ( 450,450 ) ( 0,225 ) ( 0,0 ) ( 225,225 ) ( 225,0 ) \(vi ) _ sample redshift limit . _ an increase in the redshift limit of the sample leads to a larger volume and the associated increase in the size of the sample . however , these additional groups come at the expense of requiring a larger correction to the observed group luminosity to deduce the total group luminosity . the consequences of varying the maximum redshift limit of the sample are presented in fig . [ fig : err](d ) , where the cases @xmath161 ( light lines ) and @xmath162 ( dark lines ) are considered . there is a negligible difference in the random measurement errors for these redshift limits . the correlation function is more accurately recovered for @xmath161 . the conclusions of this subsection are that an optimal measurement of the clustering of the 2pigg catalogue will be obtained under the following conditions : ( i ) the pair counts in the ngp and sgp regions of the 2dfgrs are combined . ( ii ) the value of @xmath125 is taken to be @xmath163h@xmath120mpc@xmath121 , though the results are fairly insensitive to the exact value . ( iii ) cumulative bins in luminosity are used to define subsamples of groups from the 2pigg catalogue . ( iv ) @xmath106 is used , which leads to the most reliable measurements . ( v ) the most accurate model of the selection function is obtained by a simple parametric fit to the observed group redshift distribution . ( vi ) a conservative redshift of @xmath161 is used . = 16.truecm we now investigate how well we can distinguish between the clustering signals displayed by different samples of groups defined by mass . we also test our clustering measurement algorithm to see how faithfully the clustering signal can reproduce an idealized measurement . this is an important consideration , as this determines the strength of the constraints we can place on theoretical models of the clustering of galactic systems . the errors on the measured correlation function are obtained from the _ rms _ scatter over the ensemble of hubble volume mock 2pigg catalogues . we show these relative errors in fig . [ fig : xierr ] . the horizontal line marks an error of @xmath143 . this plot indicates that we should be able to get a robust measurement of the correlation function out to pair separations of @xmath164mpc . the small scale limit depends more upon the sample under consideration , due to the changing space density of groups . for the full group sample , the correlation function can be measured down to separations below @xmath165mpc ; for the most luminous sample , which has a much lower space density , the smallest pair separation that can be probed is @xmath166mpc . to test the accuracy of our clustering measurements , we have extracted samples of groups from the galform mock catalogue defined by a series of lower limits in total group luminosity as discussed earlier . the clustering measured for these samples is compared with that of samples drawn from the simulation cube using two different criteria to construct equivalent samples as discussed earlier . the samples drawn from the simulation cube are idealized in the sense that they do not have a radial or angular selection function imposed upon them as is the case for the mock 2pigg catalogue . moreover , the groups occupy dark matter haloes identified by a friends - of - friends group finder . the mass of a group in the simulation cube is known very accurately : it is simply the sum of the mass of the dark matter particles connected by the group finder . the correlation functions measured for mock group subsamples defined by a lower limit on total group luminosity are compared with the results for the equivalent samples drawn from the simulation cube in fig . [ fig : ximockcube ] . the top row of this figure shows the comparison when the equivalent samples are set up to reproduce the mass function of groups recovered from the mock group catalogue for each luminosity subsample . the bottom row shows the comparison for equivalent samples constructed to match the effective bias estimated from the mock subsamples . the left hand column shows the results for groups with a minimum membership @xmath106 , and the right hand column for @xmath107 . the clustering measurements from the mock subsamples are in impressively good agreement with the results obtained for the equivalent samples in the simulation cube , particularly for the case of @xmath106 . the equivalent samples constructed to reproduce the effective bias of mock group subsamples are in somewhat better agreement with the measurements from the mocks for the brighter group samples . the agreement in clustering amplitude between samples of groups taken from a mock and those extracted from an idealized simulation is significant ; it means that we fully understand the impact of making a mock group catalogue on the measurement of the clustering of groups and can use the clustering of 2pigg samples to constrain theoretical models of structure formation . ( 450,250 ) ( 0,0 ) ( 225,0 ) in this section we measure the clustering amplitude of the real 2piggs for samples defined by total group luminosity . some basic properties of the samples are given in table 1 . we apply the optimal clustering measurement algorithm set out at the end of section [ ssec : optimal ] . we combine the pair counts in these regions to estimate the correlation function . we have checked that the measurements for the individual regions agree within the errors . an important test of the integrity of a group catalogue is the form of the correlation function measured in bins of pair separation parallel ( @xmath140 ) and perpendicular ( @xmath139 ) to the line of sight , @xmath141 . padilla & lambas ( 2003a , b ) explored different techniques used in the construction of group and cluster catalogues . they found that the shape of @xmath167 is a powerful way to reveal spurious agglomerations of galaxies . a tell - tale sign of problems with the composition of galaxy groups is a significant enhancement of the correlation function , @xmath167 , in the @xmath140 direction , as seen in the early analyses of abell catalogue clusters ( e.g. bahcall & soneira 1983 ; sutherland 1988 ) . such a distortion of the clustering pattern is seen for 2dfgrs galaxies , due to the large peculiar motions inside massive virialised structures ( peacock et al . 2001 ; hawkins et al . however , for the case of groups and clusters , such virialised structures are not predicted in viable models of structure formation ( kaiser 1987 ; padilla et al . 2001 ; padilla & baugh 2002 ) . the correlation function , @xmath167 , for samples of 2piggs is shown in fig . [ fig : dist ] . there is no enhancement of the clustering signal in the @xmath140 direction . the contours of constant clustering amplitude are , however , affected by peculiar motions , showing the flattening due to the infall motions expected in the gravitational instability paradigm ( padilla et al . 2001 ; peacock et al . there is also a clear increase in clustering amplitude with sample luminosity . the projected correlation function , @xmath168 , is the integral of @xmath141 over the @xmath140 direction ( e.g. norberg et al . 2001 ) . this statistic is unaffected by peculiar motions . in fig . [ fig : si4 ] , we show the projected correlation functions for selected samples of 2piggs . the black line shows the best fit to the projected correlation function of galaxies in the 2dfgrs for comparison ( hawkins et al . we find that the lowest luminosity sample plotted is less strongly clustered than the galaxy distribution , a point to which we return later on in this section . the brightest sample of groups displays a clustering amplitude that is an order of magnitude stronger than that measured for 2dfgrs galaxies . the projected correlation function , due to the way it is calculated , can be unreliable if the correlation function is noisy on large scales . a more robust quantity in such cases is the redshift space correlation function , @xmath156 . this quantity is the average of @xmath141 within annuli centred on @xmath169 . the redshift space correlation function measured for 2pigg samples with membership @xmath170 is plotted in the left - hand panel of fig . [ fig : xi4 ] . for reference , we also plot the redshift space correlation function of 2dfgrs galaxies measured by hawkins et al . . the correlation function of the group samples has a similar shape to that obtained for galaxies , on scales where a comparison is reliable . there is , however , a dramatic change in clustering amplitude with group luminosity , mirroring that seen for the projected correlation function in fig . [ fig : si4 ] . these points are emphasized in the right hand panel of fig . [ fig : xi4 ] , in which we plot the group correlation function divided by the galaxy correlation function . there is a relative bias between the spatial distribution of groups and galaxies . for the full 2pigg sample , and also for the next faintest luminosity sample , this relative bias is actually an _ anti - bias _ ; these groups are more weakly clustered than the galaxies . the brightest groups , on the other hand , are almost ten times more strongly clustered than the galaxy distribution . a useful summary of the clustering measurements can be made by plotting the correlation length , @xmath12 , as a function of the mean group separation , @xmath11 , which is related to the space density of groups , @xmath171 , by @xmath172 . the space density of groups is estimated using the cumulative @xmath173band luminosity function of 2piggs out to @xmath174 ( eke et al . the space density of each subsample is simply the luminosity function evaluated at the appropriate lower limit in luminosity . we estimate the correlation length ( defined by @xmath175 ) by fitting a power law to the measured redshift space correlation function for pair separations in the range @xmath176h@xmath15mpc@xmath177 . our results are fairly insensitive to perturbations to this range . the clustering strength of 2pigg samples versus mean group separation is plotted in fig . [ fig : s0dccomp ] . there is an increase in clustering strength with the mean separation of groups in each sample ; the clustering strength increases slightly less rapidly than the change in group separation . there is very good agreement between the results for @xmath106 and @xmath107 for group separations for which a comparison is possible . [ fig : s0dccomp ] also shows the @xmath12-@xmath11 relation in the galform mock 2pigg catalogue with @xmath178 , which is in very close agreement with the results from the real 2pigg samples . the solid and dotted lines show the measurements for equivalent samples of groups drawn from the simulation cube . the solid line gives the @xmath12-@xmath11 relation in redshift space and the dotted line shows the real space results . the clustering amplitude is typically @xmath179-@xmath180 weaker in real space than in redshift space . = 8.truecm = 8.truecm = 8.truecm the 2pigg results are compared with a selection of measurements taken from the literature in fig . [ fig : s0dcdata ] . this comparison between datasets should be treated with caution due to a number of differences in the way the various samples have been defined and analysed : namely , in decreasing order of seriousness : ( i ) the technique used to identify groups and clusters , ( ii ) the derivation of a value for the mean inter - group separation and ( iii ) the calculation of the correlation length . zandivarez et al . ( 2003 ) measured the clustering of groups in a catalogue constructed from the 100k release of 2dfgrs data by merchn & zandivarez ( 2002 ) . the two most abundant samples of groups analysed by zandivarez et al . have a significantly higher clustering amplitude than we recover for 2pigg groups of comparable abundance . zandivarez et al . estimated @xmath11 from a dynamical estimate of the group mass which they translated into a spatial abundance using the measured mass function of groups ( martnez et al . moreover , zandivarez et al . only consider groups with @xmath107 ; such groups are probably only associated with low mass systems through errors in the dynamical mass estimates . the abundance of groups in the updated zwicky catalogue was estimated in a similar way by merchn et al . bahcall et al . ( 2003 ) have applied a photometric technique ( annis et al . 2004 ) to a subset of the sloan survey data to obtain a sample of clusters that overlaps with the most luminous 2pigg groups and with apm survey clusters ( croft et al . 1997 ; dalton et al . the sloan measurements are in reasonable agreement with the results from apm clusters and are marginally lower than the 2pigg values at comparable abundances . however , both the sdss and apm clusters are identified in projection . the trend of clustering strength versus abundance found for 2piggs agrees quite well with the measurements from x - ray selected cluster samples ( abadi , muriel & lambas 1998 ; collins et al . 2000 ; lee & park 2002 ) . the dotted line in fig . [ fig : s0dcdata ] shows the @xmath181-@xmath182 relation advocated by bahcall & west ( 1992 ) , whereas the dot - dashed line is the relation proposed by bahcall et al . ( 2003 ) ; the dashed line is the fit of zandivarez et al . ( 2003 ) to their data . the analogous fit to the 2pigg results is plotted in fig . [ fig : s0dcgal ] for clarity . if the faintest groups are ignored , the 2pigg results are described by the relation @xmath183 . this is somewhat steeper than the fit of bahcall et al . ( 2003 ) . we compare the clustering of 2pigg groups with 2dfgrs galaxies in fig . [ fig : s0dcgal ] . the clustering amplitude of 2dfgrs galaxies is taken from norberg et al . ( 2002a ) , who analysed volume limited samples defined by galaxy luminosity . the curve plotted over the galaxy clustering measurements shows the predictions of the simple relation between relative clustering strength and luminosity deduced by norberg et al . the full sample of 2piggs is more weakly clustered than @xmath1 and brighter galaxies . on the other hand , the trend of clustering strength with pair separation is stronger for groups than it is for galaxies . we have measured the clustering of groups in the 2pigg sample constructed from the 2dfgrs by eke et al . ( 2004a ) . the group catalogue , made up of galactic systems with a minimum of two members , contains @xmath18429,000 groups , out of which @xmath18416,000 are at @xmath100 . the sample is sufficiently large and homogenous that a robust measurement of clustering is possible for subsamples of groups defined by their corrected total luminosity . our analysis has relied extensively on the use of various mock group catalogues constructed using high resolution n - body simulations populated with galaxies using the galform semi - analytic model ( cole et al . 2000 ; benson et al . 2000 , 2003 ) . in summary , the main conclusions we have reached are the following : 1 . the main goals of this paper were to make a robust measurement of the clustering of galaxy groups with the aim of exploring the dependence of the clustering amplitude on a property related to group mass ( for example the total group luminosity ) and to compare the results with theoretical predictions . we have tested our algorithm for estimating the correlation function by comparing results obtained from a mock 2pigg catalogue with those derived from equivalent samples drawn from the n - body simulation cube . the success of this comparison is significant for two reasons . firstly , in order to make a robust measurement of the clustering of groups from a flux limited galaxy survey , it is necessary to make a number of approximations and to choose certain parameter values ( see section [ ssec : optimal ] ) . we have tuned the algorithm for measuring the clustering of 2pigg groups by requiring a close match between the correlation functions estimated from the mocks and the original results from the simulation cube . secondly , the ability to extract equivalent samples of groups from a simulation volume with the same clustering as samples taken from mock 2pigg catalogues makes it possible to interpret the 2pigg clustering results in the context of structure formation models . we find that clustering amplitude increases substantially with total group luminosity . the 2pigg catalogue allows clustering measurements to be made for samples spanning a factor of @xmath185 in median total luminosity . the correlation function increases in amplitude by an order of magnitude over this luminosity interval . another way of expressing this is that the redshift space correlation length changes by a factor @xmath186 from the faintest to the brightest sample . the most luminous groups we consider have a mean pair separation that is @xmath187 times greater than that of the full group sample . there is little change in the shape of the correlation function with group luminosity . the shape of the redshift space correlation function of groups is very similar to that measured for 2dfgrs galaxies on the scales for which a comparison is possible . the full 2pigg sample has a _ weaker _ clustering amplitude than is measured for 2dfgrs galaxies ; the correlation length of the 2pigg sample with @xmath106 is @xmath188mpc , whilst that of 2dfgrs galaxies is @xmath189mpc ( hawkins et al . 2003 ) . however , the clustering amplitude of brighter samples of groups is much greater than that of 2dfgrs galaxies . the correlation functions measured for 2pigg samples are in very good agreement with the predictions of a semi - analytic model of galaxy formation , the galform model ( cole et al . 2000 ; benson et al . 2003 ) , in a cold dark matter universe with a cosmological constant ( @xmath2cdm ) . previous work has compared the predictions of the @xmath2cdm model with measurements of the correlation length versus abundance relation for rich clusters ( governato et al . 1999 ; colberg et al . 2000 ) . to extend the theoretical calculations down to group - sized systems , it is essential to extend the predictions of the spatial distribution of dark matter haloes in the @xmath2cdm cosmology with a model for galaxy formation . the galaxy formation model predicts how the dark matter haloes should be populated with galaxies . for this reason , the clustering of groups provides a more stringent test of theories of galaxies formation than the clustering of the richest clusters . the trend of correlation length ( measured in redshift space ) against mean inter - group separation can be quantified as : @xmath190 . the 2pigg catalogue has made possible a robust measurement of the clustering of galactic systems , ranging from poor groups to rich clusters , from one sample for the first time . our results are in general agreement with those obtained previously for rich groups and clusters . galaxies and groups of galaxies trace the underlying distribution of dark matter haloes in different ways . galaxies trace the halo distribution in a complex way . the halo occupation distribution ( hod ) , which depends strongly on halo mass , has been used extensively to constrain theoretical models ( e.g. benson et al . 2000 ; peacock & smith 2000 ; seljak 2000 ; berlind et al . 2003 ) . in principle , the group sample should trace the halo distribution in a simple fashion , with every halo above some mass spawning one galaxy group . however , in practice , the identification of groups in real surveys is complicated , making this ideal difficult to attain . because of this it is essential to devise a scheme to quantify the fidelity of a group catalogue and to uncover any biases in comparisons with theoretical models by applying the group finder to mock catalogues ( eke et al . 2004a ) . the difference between the clustering amplitude measured for the full 2pigg and the full 2dfgrs galaxy samples can be readily understood in terms of the hod . ideally , the groups have a one - to - one correspondence with the underlying dark matter haloes within some mass interval . the galaxies obviously sample haloes in the same mass range , but with a weight that increases with halo mass , since the most massive haloes host more than one galaxy . thus , the more massive haloes , which are the more strongly clustered , make a larger contribution to the correlation function of 2dfgrs galaxies than they do in the case of the full 2pigg sample . the clustering of galaxy groups as a function of group luminosity therefore provides an important new test of theories of galaxy formation . this work was supported by a pparc rolling grant at the university of durham . ndp acknowledges receipt of a british council - fundacin antorchas fellowship . cmb was supported by a royal society university research fellowship . abadi m.g . , lambas d.g . , muriel h. , 1998 , apj , 507 , 526 . annis j. , et al . , 2004 , in preparation bahcall n.a . , soneira r.m . , 1983 , apj , 270 , 20 . bahcall n.a . , west m.j . , 1992 , apj , 392 , 419 . bahcall n.a . , dong f. , hao l. , bode p. , annis j. , gunn j. e. , schneider d. p. , 2003 , apj , 599 , 814 . baugh c. m. , benson a.j . , cole s. , frenk c.s . , lacey c.g . , 1999 , mnras , 305 , l21 . benson a.j . , cole s. , frenk c.s . , baugh c.m . , lacey c.g . , 2000 , mnras , 311 , 793 . benson a.j . , lacey c.g . , baugh c.m . , cole s. , frenk c.s . , 2002 , mnras , 333 , 156 . benson a.j . , bower r.g . , frenk c.s . , lacey c.g . , baugh c.m . , cole s. , 2003 , apj , 599 , 38 . berlind a.a . , weinberg d.h . , benson a.j . , baugh c.m . , cole s. , dav r. , frenk c.s . , jenkins a. , katz n. , lacey c.g . , 2003 , apj , 593 , 1 . colberg j. m. , et al . , 2000 , mnras , 319 , 209 . cole s. , kaiser n. , 1989 , mnras , 237 , 1127 . cole s. , hatton s. , weinberg d.h . , frenk c.s . , 1998 , mnras , 300 , 945 . cole s. , lacey c.g . , baugh c.m . , . , 2000 , mnras , 319 , 168 . colless m. et al . ( the 2dfgrs team ) , 2001 , mnras , 328 , 1039 colless m. et al . ( the 2dfgrs team ) , 2003 , astro - ph/0306581 collins c. a. , et al . , 2000 , mnras , 319 , 939 . croft r.a.c . , dalton g.b . , efstathiou g. , sutherland w.j . , maddox s.j . , 1997 , mnras , 291 , 305 . dalton g.b . , maddox s.j . , sutherland w.j . , efstathiou g. , 1997 , mnras , 289 , 263 . davis m. , efstathiou g. , frenk c.s . , white s.d.m . , 1985 , apj , 292 , 371 . efstathiou g. , 1988 , in comets to cosmology , ed a. lawrence , proc 3rd iras conference , springer - verlag , berlin . ( the 2dfgrs team ) , 2004a , mnras , 348 , 866 . ( astro - ph/0402567 ) . et al . ( the 2dfgrs team ) , 2004b , mnras , submitted , astro - ph/0402566 . et al . ( the 2dfgrs team ) , 2004c , in preparation . evrard , a.e . , macfarland , t.j . , couchman , h.m.p . , colberg , j.m . , yoshida , n. , white , s.d.m . , jenkins , a. , frenk , c.s . , pearce , f.r . , peacock , j.a . , thomas , p.a . , 2002 , apj , 573 , 7 . geller m.j . , huchra j.p . , 1983 , apjs , 52 , 61 . girardi m. , boschin w. , da costa l.n . , 2000 , a&a , 353 , 57 . governato f. , babul a. , quinn t. , tozzi p. , baugh c.m . , katz n. , lake g. , 1999 , mnras , 307 , 949 . hawkins e. , et al . ( the 2dfgrs team ) , 2003 , mnras , 346 , 78 . helly , j.c . , cole , s. , frenk , c.s . , baugh , c.m . , benson , a. , lacey , c. , 2003 , mnras , 338 , 903 . jenkins a. , frenk c.s . , white s.d.m . , colberg j.m . , cole s. , evrard a.e . , couchman h.m.p . , yoshida n. , 2001 , mnras , 321 , 372 . jing y. , zhang , j. , 1988 , a&a , 190 , l21 . kaiser n. , 1987 , mnras , 227 , 1 . kashlinsky a. , 1987 , apj , 317 , 19 . kauffmann g. , colberg j.m . , diaferio a. , white s.d.m . , 1999 , mnras , 303 , 188 . landy s.d . , szalay a.s . , 1993 , apj , 412 , 64 . lee s. , park c. , 2002 , j. korean astron . , 35 , 111 . maia m.a.g . , da costa l.n . , 1990 , apj , 352 , 457 . martnez h.j . , zandivarez a. , merchn m.e . , domnguez m.j.l . , 2002 , mnras , 337 , 1441 . merchn m.e . , maia m.a.g . , lambas d.g . , 2000 , apj , 545 , 26 . merchn m. , zandivarez a. , 2002 , mnras , 335 , 216 . white s.d.m . , 1996 , mnras , 282 , 347 . norberg p. , et al . , ( the 2dfgrs team ) , 2001 , mnras , 328 , 64 . norberg p. , et al . , ( the 2dfgrs team ) , 2002a , mnras , 332 , 827 . norberg p. , et al . , ( the 2dfgrs team ) , 2002b , mnras , 336 , 907 . padilla n.d . , merchn m.e . , valotto c.a . , lambas d.g . , maia m.a.g . , 2001 , apj , 554 , 873 . padilla n.d . , baugh c.m . , 2002 , mnras , 329 , 431 . padilla n.d . , lambas d.g . , 2003a , mnras , 342 , 519 . padilla n.d . , lambas d.g . , 2003b , mnras , 342 , 532 . peacock j.a . , smith r.e . , 2000 , mnras , 318 , 1144 . peacock j.a . , et al . ( the 2dfgrs team ) , 2001 , nature , 410 , 169 . press w.h . , schechter p. , 1974 , apj , 187 , 425 . ramella m. , geller m.j . , huchra j.p . , 1990 , apj , 353 , 51 . seljak u. , 2000 , mnras , 318 , 203 . sheth r.k . , mo h.j . , tormen , g. , 2001 , mnras , 323 , 1 . spergel d.n . , et al . ( the wmap team ) , 2003 , apjs , 148 , 175 . sutherland w. , 1988 , mnras , 234 , 159 . zandivarez a. , merchn m.e . , padilla n.d . , 2003 , mnras , 344 , 247 .